DMTD board

[5065AGuru]

The boards for the DMTD mixer board arrived!

I'll be populating and testing and will post updates.

Cheers!

Corby

[mark03]

What are all those holes for??  Seem too big to be vias.

 

[tkamiya]

Will you share the design of the circuit?  I've been wanting to do this myself.

[5065AGuru]

Yes I will be sharing the schematic/parts list/and expressPCB file.
Once assembled I'll be testing it and determining the noise floor.
I'll share some plots taken with it and some pointers on how to get the best results from the particular frequency counter you use with it.
I'm also in the process of designing a counter to go along with it so you don't have to tie up your fancy counter.

Just populated the front end and mixers and have good signals out of the mixers. Will be doing the rest of the board soon. I need to breadboard the op amp portion first so I can determine what bandwidth I want to use.

Cheers,

Corby

[notfaded1]

Looks awesome Corby!  You know I'm all over this!  I'm so psyched after the bad experience with the TURN Dynamic Systems DMTD I bought!

Best Regards,

Bill

[notfaded1]

The best part about this is we could setup multiple ones if the counter works for monitoring multiple oscillators!  Good stuff!

Bill

[5065AGuru]

OK! Fully populated.
Just need to mount it in a frame with power supplies for testing.

[notfaded1]

Dang Corby... that's some really cool stuff brother! 

Bill

[rubidium]

Just curious: Were you driven to thru-hole because of your choice of passives?  It will be interesting to see the test results.
Jim

[thinkfat]

Huh, is that really a MC1651L at the input? Any chance to use something more contemporary? I'm not too keen on buying that device from Rochester Electronics...

[tkamiya]

Looks like major components are "pulls". 

[5065AGuru]

This was only to scratch an itch of mine to see how the Efratom TS105 stages would work as a standalone Dual Mixer.
I was not trying to build "modern"!
The 1650 or 1651 were purchased on ebay for reasonable prices.
I also wanted something that could be easily assembled as Sometimes that scares people off.
Once I have done comparative testing with different types of op amps and comparators we will see if it has reasonable performance for entry level DMTD work.
First Time interval count and dirty layout for testing PIX.

[5065AGuru]

tkamiya,

Only pulls are the 2 Mixers and the 4 inductors.

My inductor kit was at work and rather than wait a day I used pulls.

Not that I think pulls are bad!

Cheers,

Corby

[tkamiya]

So....  how is it?  I've been wanting to do this myself but been unable to get my feet wet.

[thinkfat]

This was only to scratch an itch of mine to see how the Efratom TS105 stages would work as a standalone Dual Mixer.
I was not trying to build "modern"!
The 1650 or 1651 were purchased on ebay for reasonable prices.
I also wanted something that could be easily assembled as Sometimes that scares people off.
Once I have done comparative testing with different types of op amps and comparators we will see if it has reasonable performance for entry level DMTD work.
First Time interval count and dirty layout for testing PIX.

No problem with "not modern", I was just hoping to have something that is more easily reproducible without resorting to finding "new old stock" on ebay. I don't usually buy parts from unknown sources, I stick to DigiMouser or maybe LCSC for Chinese chip vendors that are not stocked by my usual sources.

Still, when you publish the schematic at a point, I can likely come up with a "modernized" version, easier though if you provide some insight in what parameter was key to selecting a certain part over another.

For the MC1651, I'm guessing any reasonably fast dual channel comparator with differential output will do?
Do you need the latch function of the MC1651 (I'm guessing not)?

Without seeing the schematic, I figure the idea is to drive the mixer from the differential outputs of the comparator, first channel with the 10MHz from the DUT, the second channel from the transfer oscillator? The mixer output is then going into an amplifier and limiter and from there into a logic gate, probably some HCT or AHCT to have a TTL output for the counter.

[5065AGuru]

I have 4 other DMTD units and all of them have many more components and some have a lot of surface mount. One is an NBS 106D that was used in the testing of oscillators for the original GPS satellites.
I admired Efratom's design as it has minimal parts count and can tolerate a wide input amplitude range.
I wanted to offer an entry level version that would not intimidate a home brew hobbyist. Here is the first plot of an HP5065A against an FTS1200. I'll run and post a noise floor/ baseline plot next.

Cheers,

Corby

[5065AGuru]

OK, Here is a plot of the noise floor of the simple dual mixer.
This is good enough for pretty much any TimeNut!
Bill of materials will follow, Then the schematic, and then a corrected express PCB file.
(I need to enlarge a couple holes.)
The prototype is spoken for but the other two bare boards including some of the harder to find parts included will be offered for sale. (Pretty much at cost!)
This will be after I have a chance to do some more testing to make sure I'm happy with the design.

Cheers,

Corby

[tkamiya]

Pretty impressive!

Can I ask the board to have little more edge so parts aren't so close to the very edge of the board and mounting hole?  Please?  Whatever you end up with, I'll buy three.

[5065AGuru]

Unit is working fine but looking into a glitch in the outputs. Will update status when solved!

Cheers,

Corby

[Frex]

Hello,

I don't have found many clear info about this with a Google search, so I ask you :
Does somebody can explain what is the purpose and applications of this DTMT board ?
How it work ?
Regards.

Frex

[thinkfat]

Measure phase between two clocks with sub picosecond precision

[5065AGuru]

Frex,

Thinkfat is correct.

The utility of the device is to be able to compare frequency standards/oscillators and be able to reach Allan Deviation levels that most counters themselves cannot.

I just did a test a friend requested.

I wrapped 4 feet of RG58C/U around a coffee cup and measured the delay at ambient and then poured boiling water into the cup.

The cable delay changed by 40 PicoSeconds.

Resolution is 2 FemtoSeconds but noise level is in the 20 to 50 FemtoSecond range.

Ignore the scale on the left as it is not correct!

Still working on testing, BOM, and schematic.

Cheers,

Corby

[thinkfat]

At 1300s, what did you pour into the cup, liquid nitrogen? That would explain the condensation on the outside of the cup and the cable

[5065AGuru]

Let everything stabilize till 650 Sec and then poured in boiling water.
Then at 1300 Sec dumped water and then filled cup with ice cubes.
PIX of cup with condensation was just after I dumped the ice cubes.

Cheers,

Corby

[KE5FX]

Very nice.  What's an MC1651L?  Google says it's an A/D...?  I see, it's a dual comparator.  Haven't seen those referred to as "A/D converters" before, but I guess it's technically correct (the best kind of correct.)

The noise floor looks better than some of the more complex DMTDs I've seen.  It's clearly a good part for the role.

[5065AGuru]

From the data sheet:

"The MC1650 and the MC1651 are very high speed comparators utilizing differential amplifier inputs to sense analog signals above or below a reference level. An output latch provides a unique sample-hold feature. The MC1650 provides high impedance Darlington inputs, while the MC1651 is a lower impedance option, with higher input slew rate and higher speed capability."

The nice thing about using them as an input stage is the wide input range you get. 100mv to over 6v.
The other dual mixers I have must be carefully fed a specific level for best performance and to prevent damage to the mixers.
Also they have balanced outputs so you can drive the mixers that way.
Not sure what that does for performance???

So far testing is looking good with a couple minor changes made to the circuit.
I'd say another week of testing and  I'll be happy with it.

Cheers,

Corby

[notfaded1]

OK, Here is a plot of the noise floor of the simple dual mixer.
This is good enough for pretty much any TimeNut!
Bill of materials will follow, Then the schematic, and then a corrected express PCB file.
(I need to enlarge a couple holes.)
The prototype is spoken for but the other two bare boards including some of the harder to find parts included will be offered for sale. (Pretty much at cost!)
This will be after I have a chance to do some more testing to make sure I'm happy with the design.

Cheers,

Corby

The noise floor figures look totally legit!  It's hard to argue with that. 

[notfaded1]

Hello,

I don't have found many clear info about this with a Google search, so I ask you :
Does somebody can explain what is the purpose and applications of this DTMT board ?
How it work ?
Regards.

Frex

Close Frex but search for DMTD instead... easy to mix these letters up... I've done it myself before that's how I know.  I think maybe it's old DTMF push button phone tones ringing in the back of head after all these years.  Brings back memories... we used take the old hand held DTMF phone dialers from Radio Shack and replace the timing crystal in the back.  This made one * star a nickel, 2 * stars a dime, and 5 * stars a quarter sound out of the speaker on the back.  Perfect portable Red box for pay phones... back when we had pay phones!

Bill

[tkamiya]

I still can't remember that 4 letter combination!  I have to look it up every time.  I'd say there are too many acronyms in this world.

[rubidium]

Close Frex but search for DMTD instead... easy to mix these letters up... I've done it myself before that's how I know.  I think maybe it's old DTMF push button phone tones ringing in the back of head after all these years.  Brings back memories... we used take the old hand held DTMF phone dialers from Radio Shack and replace the timing crystal in the back.  This made one * star a nickel, 2 * stars a dime, and 5 * stars a quarter sound out of the speaker on the back.  Perfect portable Red box for pay phones... back when we had pay phones!

Bill

Ha! You're dating yourself, buddy! Back in the day, we used those recordable Hallmark greeting cards when they first came out.   

[awallin]

The noise floor figures look totally legit!  It's hard to argue with that. 

almost too good to be true!?
Riley reports ca 5e-13 @ 1s http://www.stable32.com/A%20Small%20DMTD%20System.pdf
3120A/TimePod or the newer 53100A is around  3...4e-14 @ 1s http://www.miles.io/PhaseStation_53100A_user_manual.pdf
Ettus SDR around 7e-14 @1s  https://arxiv.org/abs/1605.03505

[KE5FX]

In principle, if the front-end noise is low enough (and white enough), it's possible to see useful results all the way down at this level.  Just a matter of setting the appropriate measurement bandwidth... presumably 0.5 Hz or so for your current setup, right, Corby?

Taking full advantage of a floor that low can be tricky, as even your cables will conspire against you.  At 3E-15 @ t=1s, you probably need 140+ dB of port isolation to avoid beatnotes between closely-spaced 10 MHz sources.  Definitely achievable, but it usually doesn't happen by accident!

[thinkfat]

The real-world sanity check would be the "RG58 hot vs. cold". Not sure about the effect, but I guess the cable simply got longer due to thermal expansion? If the noise floor is off by a magnitude, this should reveal it. The cable changed length corresponding to a 40ps phase shift.

PS: looks like phase shift is not linear with temperature ("Teflon Knee")

[awallin]

The real-world sanity check would be the "RG58 hot vs. cold". Not sure about the effect, but I guess the cable simply got longer due to thermal expansion? If the noise floor is off by a magnitude, this should reveal it. The cable changed length corresponding to a 40ps phase shift.

in the coffee-cup post "noise level is in the 20 to 50 FemtoSecond range." - but I don't understand how this goes together with the ADEV-floor which seems to be plotted at ca 3e-15 @ 1s (if I read the figure correctly...).
EDIT: the post on page 1 has now been edited to show a noise-floor of ca 2e-13 @ 1s, a mere 100-fold de-rating from the initial figure...

For white-noise 20 fs amplitude (or peak-to-peak?) @ 1s I would assume there's a sqrt(3) factor to give an ADEV-floor around 3.5e-14 @ 1s, which is very much in line with the references posted above, but 10-fold higher than the ADEV-floor figure from the post above.

[thinkfat]

I'd still take it. I doubt I could create an environment silent enough to realize a 3e-15 noise floor.

[FriedLogic]

in the coffee-cup post "noise level is in the 20 to 50 FemtoSecond range." - but I don't understand how this goes together with the ADEV-floor which seems to be plotted at ca 3e-15 @ 1s (if I read the figure correctly...).

Just a guess: one is a basic test through what does not look like super high quality cable and connectors, and the other likely done with very much more suitable cabling. I'm sure we'll hear anyway.

[notfaded1]

There are some cables that don't exhibit the same issues PTFE dielectric has.  See attached used in radar and other applications where phase is of extreme importance.

Bill

[5065AGuru]

OK, My bad somewhere!

You are correct. The Noise floor plot is too good!

2.67X10-15th at 1 Sec is not realistic.

I'm not sure if I messed up the scaling or what.

I don't have the file with me but I'll check tomorrow and see what happened.

The SMDfirstrun AD plot however has correct scaling.

Cheers,

Corby

[Mrt12]

Hi Corby,
so do you really drive the mixers with square waves from those ECL drivers?
I knew that it is possible to use a square wave for the LO input, but does it work equally well if both the RF and the LO are square waves? is the performance comparable or even better than if sine waves were used? and: how does the signal at the IF port look like? I am really curious for the schematics.

[5065AGuru]

MRT12,

Yes the mixers are driven by square waves.
This was how it was done in the EFRATOM TS105A, so I know it works.
Not sure how it would compare to sine inputs but this whole exercise was to make a low parts count unit that had a wide input amplitude range for an entry level Dual Mixer unit.
Still testing and hope to have some good data by the first of next week.
Schematics and Express PCB file will follow.
Next post I hope to have a composite plot that compares my 3 current Dual mixer units, one I got rid of, another complex unit, and the simple dual mixer.
That will let us see how it stacks up.

Cheers,

Corby

[thinkfat]

For the mixers its all the same I guess. They will just see a LO and RF that are rich in tones and happily multiply them all. As long as the high-order products are far enough away from the fundamental, the low-pass after the mixer will get rid of them.

[ch_scr]

I do not know the schematics but if only a "fast" comperator with differential out is needed maybe the LM360 could be used?

[thinkfat]

AD8612 would be suitable, too, I guess. But I'm more concerned about the mixers. The MiniCircuits stuff is really nice, but I need to find a distributor that stocks them. Digikey doesn't have them at all and Mouser refuses to sell them in Germany 

[notfaded1]

Now I had to look it up... the front panel display is so frickin' cool!  Dang!  How can you not love a classic display technology like that on a DMTD!  I have a collection of Nixie tube frequency counters including a Dekatron one.  I'll post some pics later.  Really neat devices!    I love that classic neon glow.

Bill

[tkamiya]

Corby,

Since you said "hard to get parts" are involved, I would like to see just a partial parts list (less jellybean parts) to get head start on sourcing them.  Would you mind doing that?

[FriedLogic]

Mini-Circuits appear to distribute from the UK now too, so that should be easier for those in the EU - for a couple of months anyway....

[thinkfat]

We may have to organize a group buy. Some parts have MOQ in the twenties.

[Mrt12]

But I am still wondering whether it is really a smart idea to drive the mixers with square waves. Assume you drive a mixer with sine waves, 10MHz and 10MHz+1Hz for the LO and RF.
The mixing products will be: 1Hz and 20MHz+1Hz.

Now assume that the mixer is driven with the same frequencies as above, but with square waves. The square waves each have the following spectral components:
10MHz, 30MHz, 50MHz for one signal, 10MHz+1Hz, 30MHz+3Hz and 50MHz+5Hz for the other signal.
When these two signals are mixed, the result at the IF output will contain the following frequencies (I list the difference frequencies only, as the sum frequencies are all well above 10MHz and can be easily filtered out):

1Hz (=10MHz+1Hz - 10MHz)
3Hz (=30MHz+3Hz - 30MHz)
5Hz (=50MHz+5Hz - 50MHz)

I don't think the IF output will be a square wave, but it will definitely contain a set of harmonics, so I wonder how this will work - because in this very case, we are actually only interested in the 1Hz frequency.

[tkamiya]

You are right about heterodyning square waves.  All of the harmonics will be mixed together and create whole bunch of spectrum impurities.  But levels are much lower.  I read somewhere DBMs work much more efficiently with square wave than sine wave due to switching action of the diode ring. 

Maybe someone can simulate it or run it through in lab.... 

Here's some reference:
https://www.microwaves101.com/encyclopedias/double-balanced-mixers

[notfaded1]

If it worked well for Efratom I'm not sure why it wouldn't work here?    Can't wait to see how it stacks up to other DMTD's.

I know how long it takes to run some of these tests.  I pretty much have at least one frequency experiment running at all times in my lab.  Lately I've been comparing the 5065 against 5061B, 5071A, and various Rb's I have.  I really like the old classic Efratom Rb's.  They really knew how to build a device in those days with nice large rubidium lamps in them.  I have an FRK-L and two M-100's.  Easy access to the Rb lamp too!  It's just a big machined screw on the back of the unit which, believe it or not, I believe was made so the lamp could be replaced while it's running although I've never tried this.

Bill

[tkamiya]

If I had a schemetic, I'd try mixing those frequencies in square wave and sine myself in my lab....  Maybe I'll just take a wild guess and try it.

I think, he's mixing 10MHz and 9.999999MHz, and get 1Hz, then feeding two of these to start and stop of interval counter each.  So RF and Local to each side of DBM via transformer, and center-tap goes to the counter?  I'll have to lower the frequency so I can use function generators rather than RF synthesizers.  I don't think that matters as long as difference is 1Hz.

That would be fun while I impatiently wait Coby's results....

I have few each of popular GPSDO, Rb, OCXO, and an old Cs in my lab.  What I lack is a good reliable reference and measurement method with fine enough resolutions.  I've tried comparing every permutations until I got to a point I started questioning my results.  Sure enough, results don't make much sense.  I'm trying to regroup myself (and lab) and hoping a new DMTD will fix some of the issues I had.

I looked up MC1450L and 1451L.  I really don't want to go eBay route as performance of these chips will be the key to success.  I found DigiKey had them via third party seller, but minimum quantity is like 30 pcs for close to $300. 

Anyone know what mixer Coby is using?  (Coby?  Hello!)

[thinkfat]

Not so easy to simulate at a 1Hz (or 10Hz) beat frequency. You'll want an FFT with sub-Hz bin width so that you can see all the mixing products, at the same time you have a 10MHz and a 10MHz+10Hz input, so you need to simulate at a quite high resolution. You'll need to generate a lot of samples to have enough waveforms at the beat frequency.

Anyway I don't think it matters. You get additional tones which "deform" the resulting waveform, but that's OK, since you'll be deforming it a lot more anyway with the adjacent limiter stages. What's important is that there the mixer will not introduce additional uncertainty (passive mixer has really good noise figure).

[tkamiya]

I have an HP3561A for FFT.  It can do sub milli-Hz resolution. 

I know it won't be an issue in Coby's implementation.  I'm just curious as questions were raised.

[5065AGuru]

Just a note,

The output of the mixer is filtered to remove any high frequencies and the BW of the op amp is around 35Hz So the 1Hz or 10Hz beat note should be suitably clean!

BOM finalized should post tomorrow.

Cheers,

Corby

[rubidium]

Regarding the discussion of square waves vs sine waves into a mixer, is it not true that we have become conditioned, i.e. by block diagrams of entities that have an "X" in them, to think that mixers act as multipliers of 2 waveforms in the strict mathematical sense when this is not the case with real circuits? To the contrary, I thought that most practical mixer circuits actually perform periodic inversion of the RF input based on zero crossings of the LO input, in which case the IF output is the same regardless of the LO being square (with 50% duty) or sine. Of course, having less jitter in the zero crossings of a square wave in noise over that of a sine in noise is the difference. I don't have any simple mixers lying around to play with, but I suspect that the IF output of one with a simple sine RF input and sine LO is more than just 2 lines, at (f_RF+f_LO) and (f_RF-f_LO), in the frequency domain due to this action. All of the odd harmonics of the LO frequency are mixed in there, regardless of LO waveform, so9 long as the zero crossings of the waveform are equally spaced. With suitable filtering, the fundamental (f_RF+f_LO) or (f_RF-f_LO) is what we care about.
Jim

[thinkfat]

I've attached a LTspice file that simulates a DBM with a beat frequency of 1000Hz. Maybe this will help for illustration. You can play with it, just set different amplitudes for V2. Start with 50mV, increase and observe the waveform between C1 and R2.

[notfaded1]

You're right Jim, zero crossing of square and sine wave at the same frequency is the same.

Bill

[5065AGuru]

Here is the BOM.

Schematic soon!

Cheers,

Corby

            Bill of materials for Simple Dual Mixer

All ICs DIP
2 MC1650L or 1651L
2 LT1008
2 LM311
2 74HCT14

Sockets
2 16 pin
2 14 pin
4 8 pin

2 SBL-1 mixers
1 circuit board
3 SMA to right angle PC connectors (long thread)

All capacitors radial .1" spacing
24 .1uf bypass
4 .1uf COG or film
2 .01uf COG
2 100pf

All resistors 1/4W physical sized 1% metal film
6 1K
12 51.1 Ohm
8 221 Ohm
2 100 Ohm
2 68.1 Ohm
3 16.7 Ohm


Chokes 1/4W physical size
4 3.9uh

All surface mount sized 1206 except noted
2 4700 pF COG
2 .0068 uF film
2 .1 uF film (1210) PPS dielectric (Digi-Key 338-3627-1-ND) for low TC
4 10uF Tantalum

2 1K
2 100K
2 47.5K

[tkamiya]

What is COG?

Is MC1651L necessary better for this application than MC1650L?

[rubidium]

What is COG?

Is MC1651L necessary better for this application than MC1650L?

C0G (AKA NP0) (note the zero, not an uppercase “O”) is an extremely stable dielectric, and is used to make capacitors that are not significantly affected by temperature, applied voltage, or aging. You will find that cheaper multilayer ceramic capacitors using barium titanate based dielectrics (like X7R or X5R) can turn out to have as low as 20% of their rated capacitance once sufficient DC biases are imposed over them.  When I learned that I got hooked on C0G's for critical applications. They cost more, generally require a larger volume for a given capacitance, and you're not going to find values greater than about 0.1uF. (If you do, be prepared to pay dearly). For something like 4700pF, the extra cost is not anything to be concerned about.
Jim

[tkamiya]

Oh....   I C....  C zero G....

I knew that...  cough, cough, cough....

[5065AGuru]

Both 1651 and 1650 tested the same so whatever is easier to source.

Cheers,

Corby

[Mrt12]

By the way, has anybody ever tried to use a Gilbert cell, like an NE612 or similar instead of a diode ring mixer?

[5065AGuru]

See change to BOM, corrected a capacitor value and added digi-key PN.

Cheers,

Corby

[tkamiya]

This is great.  I'm turning my eyes into sourcing them.

MC1650L and MC1651L, I found both at Digikey but they are from third party seller.  Nothing on Mouser.  Lots on eBay.  Any word on what to look out for?  I would imagine eBay stuff can be questionable.  I'd rather buy from someone reputable.

SBL-1, current models are SBL-1-1+ (0.1-400MHz, 7db), SBL-1+ (1-500MHz 7db), SBL-1X+ (10-1000MHz, 7db)
I would guess, first two are suitable?

Just requesting clarifications,
1)  last few items with K as units, they are registers, right?
2)  capacitors, 0.1 inch spacing:  what types are these?  ceramics?
3)  I'm guessing socketing ICs are not recommended, right?

[thinkfat]

MC1651L you'll have to buy from Ebay, Aliexpress or from Rochester Electronics, which will ask for your firstborn in exchange.

[thinkfat]

By the way, has anybody ever tried to use a Gilbert cell, like an NE612 or similar instead of a diode ring mixer?

I did some Googling for this, the only references that came up were time-nuts email threads (seems you were even involved in one ).

I guess they must be lacking some key qualities, noise and isolation come to mind.

[tkamiya]

I saw Rochester Electronics when I queried Digikey.  Do you know this company?  What is its reputation?

[thinkfat]

Oh the reputation is very good. They specialize in buying stock of end-of-life components and then selling them at a premium. You turn to Rochester Electronics when you're desperate to fulfill contractual obligations for military or aviation gear that cannot be upgraded or re-certified.

[tkamiya]

I've shot them an email asking if they would be willing to sell smaller quantities.  I don't need twenty of them.

[5065AGuru]

I bought 4 1650 and 4 1651 on eBay from US sellers and am satisfied with them.

Corby

[tkamiya]

Would you be comfortable telling me the seller's name?

[5065AGuru]

markltulsaok 1650L HAS MORE THAN 10 AVAILABLE FREE SHIPPING AND $6.75 EACH

The 1651L seller is out of stock.

There is another seller hodge16 that has 7 1651L available at $4.99 each and $3.99 shipping.

Cheers,

Corby

[5065AGuru]

Here is the schematic!

Also look back on page 1 for the "real" noise floor plot.

ExpressPCB file will post soon.

Once that last step is done I am going to start a new post detailing how to operate a DMTD system, showing how to get the best results and covering some of the little covered quirks and theory!

Cheers,

Corby

[tkamiya]

Corby,

What are you using for your offset oscillator?  It'd be difficult to shift OCXO for 5Hz, for example.  1Hz is possible with HP10811.  I was thinking of using my HP signal generator HP8644B, maybe.

[5065AGuru]

I'll get into more detail on the offset oscillator in the tutorial post but you want to have an offset oscillator that has very good short term stability.
I use an FTS 1200 with a 1HZ offset at 5 Mhz.
The 10811 at 10Mhz can be used also with a 1HZ offset or a 10Hz offset.
I almost never use a 10Hz offset but you can do it.

Corby

[chuckb]

Corby
Thanks for the design. I will be building one. I pulled ten Efratom FRK Rbs from surplussed VLF Navigation equipment many years ago. 

On the schematic the output pairs of the MC1650 chips are terminated with a 51 ohm and a 220 res network. Three of the four pairs use 0.1u filtering cap. The 2nd from the top uses 0.01u caps. Is this correct?

The lowpass filter after the DBM has a RLC filter. Are those 0.1u caps C0G dielectric? I have had distortion issues with X7R caps used in filters in the past. Because of their nonlinear characteristics.

My favorite C0G cap right now is TDK. Other brands can have issues.
https://www.eevblog.com/forum/metrology/film-and-ceramic-capacitor-leakage-current/msg3190618/#msg3190618

[5065AGuru]

chuckB,

Corrected the schematic!

I used the bypass .1 caps for those 4 also but I did match them. Probably would be good to change them to film or other type.

Had to add the expressPCB file as a ZIP file. Once unzipped you can load it into the  express PCB program to inspect it. From there you can order boards from them, 3 boards for around $78.00 with shipping.

Cheers,

[tkamiya]

Two 0.01 C0G....  are they the one at positive input of LT1008?

4007pf COG, they are at output of DBM, correct?

Power to MC1650L, positive side is going through an inductor and bypassed by two caps, where as negative side isn't.  Is this on purpose?

[5065AGuru]

Yes,

Yes the only two .01uf on the board are at the input pins of the LT1008.
Yes the 4700 are across the output pins of the DBM.
The positive into the 165X is bypassed by a .1uf ceramic and a 10uf Tantalum.
The minus into the 165X only has the .1uf and no inductor.

Cheers,

Corby

[chuckb]

Corrected the schematic!
I used the bypass .1 caps for those 4 also but I did match them. Probably would be good to change them to film or other type.

Just tried to add the expressPCB file but it won't take the .pcb file extension anyone know how to post the file?

Looks like the system will accept an attached ZIP file

[chuckb]

chuckB,

Corrected the schematic!

I used the bypass .1 caps for those 4 also but I did match them. Probably would be good to change them to film or other type.

Had to add the expressPCB file as a ZIP file. Once unzipped you can load it into the  express PCB program to inspect it. From there you can order boards from them, 3 boards for around $60.00

Cheers,

I added part Values onto the PCB to help me with assembly. I took a guess for the 68 ohm terminations on the output HEX Buffers. Let me know if I messed anything up.

[thinkfat]

Going by the currently attached schematic, did you intend to just short all the HCT14 outputs together? There should be a small series resistor after each of the inverters before you link them to OUTA and OUTB, no?

[Mrt12]

Going by the currently attached schematic, did you intend to just short all the HCT14 outputs together? There should be a small series resistor after each of the inverters before you link them to OUTA and OUTB, no?

I agree to that. Besides that, the output impedance of  a HC14 is not 50 Ohm but maybe around 30 Ohm or so. If you short 5 outputs together, this results in perhaps 6 Ohms output impedance.
It would be better to use approx. 220 Ohms series resistors at each output. This would result in a total output impedance of 250 Ohms for each output, and then 5 of them im parallel brings you to 50 Ohms.

[thinkfat]

What I'm more worried about is that the top mosfet of one output and the bottom mosfet of another output might be both conducting during a transition. That will give quite some spikes on the supply rail.

[Mrt12]

What I'm more worried about is that the top mosfet of one output and the bottom mosfet of another output might be both conducting during a transition. That will give quite some spikes on the supply rail.

I would not worry too much about this. In my GPSDO I used a 74ACT541MTC with paralleled outputs and that works fine.

[Mrt12]

Corby,
now I looked at the schematics. Thank you.
I saw that you actually use the differential outputs of the ECL comparators to drive the two transformes inside the mixers in a differential manner. This is indeed very interesting! are those internal connections of the SBL-1 documented somewhere or did you find it yourself? besides that, the Minicircuits datasheet reads that the ground pins need to be connected externally, which you don't (because you use the primary windings of the transformers of the LO and RF inputs in the differential configuration). Do you know whether the mixer's isolation or other characteristics degrade somehow in this configuration?

I would like to use ADE-1+ or ADE-2+ in my DMTD, and for these, there is, similar to the SBL-1, no official document from miniricuits where I can see the actual connections of the internal transformers. However, here

https://www.on1bes.be/sdr_up_conv_v1.0_ade1_125_en.html

I found a schematic where the internal connections are visible, so it appears as if this mixer could be used in the same way as you do: input the RF at pins 3 and 4, and LO at 1 and 6. What do you think?

Further I thought about using a PECL or LVPECL comparator to drive the mixer inputs. Or would it even work with an LVDS line driver? Just directly connect the differential outputs of the driver to the mixer transformers.
For instance, would an LMH7220 be usable?

[thinkfat]

Corby,
now I looked at the schematics. Thank you.
I saw that you actually use the differential outputs of the ECL comparators to drive the two transformes inside the mixers in a differential manner. This is indeed very interesting! are those internal connections of the SBL-1 documented somewhere or did you find it yourself? besides that, the Minicircuits datasheet reads that the ground pins need to be connected externally, which you don't (because you use the primary windings of the transformers of the LO and RF inputs in the differential configuration). Do you know whether the mixer's isolation or other characteristics degrade somehow in this configuration?

I would like to use ADE-1+ or ADE-2+ in my DMTD, and for these, there is, similar to the SBL-1, no official document from miniricuits where I can see the actual connections of the internal transformers. However, here

https://www.on1bes.be/sdr_up_conv_v1.0_ade1_125_en.html

I found a schematic where the internal connections are visible, so it appears as if this mixer could be used in the same way as you do: input the RF at pins 3 and 4, and LO at 1 and 6. What do you think?

Further I thought about using a PECL or LVPECL comparator to drive the mixer inputs. Or would it even work with an LVDS line driver? Just directly connect the differential outputs of the driver to the mixer transformers.

The main difference between the ADE-1 and the SBL-1 seems that ADE-1 just bonds out the center tap of the secondary winding while the SBL-1 has the individual secondary transformer coils available.

[ch_scr]

View inside "HPF-505, SRA-1, SBL-1, IE-500" http://www.qrp4u.de/docs/en/Components/index.htm

[Mrt12]

The main difference between the ADE-1 and the SBL-1 seems that ADE-1 just bonds out the center tap of the secondary winding while the SBL-1 has the individual secondary transformer coils available.

Indeed. But that would be not much of a problem because these centre taps are connected together anyways, are they.
What could be the reason that Minicircuits does not provide these official pinouts but instead labels all of these pins  as ground?

[chuckb]

Updated the unofficial schematic parts layout. I had the polarity of a 10u tant backwards in the lower left corner.

[chuckb]

Now the board passes DRC for Express PCB

[thinkfat]

Corby, any indication how the design would perform with just +/- 5V power rails? It will impact the LT1008 and the LM311 of course, but is it relevant?

[5065AGuru]

Thinkfat,

Original design ran with +5.0 -5.2 and +- 15VDC.
I ran the prototype that way except for +- 12VDC as I had a packaged module of that voltage.
I also ran it a -5.0 to see what happened and performance looked OK.
I haven't considered trying it at +- 5 only.

The current draws are 10mA on the +- 12, 30mA on the +5 and 200mA on the -5.
So you could use a 5V regulator off the +12 to get the +5.
The same for the -5 if your +12 supply would take the load. You would need a heat sink!

Cheers,

Corby

[5065AGuru]

On my other dual mixer setups I use a dual monostable like the 74C221 to drive "heart beat" LEDS. One for the start channel and one for the stop.
The monitor outputs on the board are provided to do that. I usually just wire it up on a perf board and select the one shots time to give a good visible pulse. It's handy as it will easily show if you are missing an input.
I use the long threaded SMAs to mount the card to a front panel and run the output pulses to BNCs on the front panel.

Corby

[Mrt12]

Dear colleagues

I investigated a bit more concerning those MC1650 and MC1651 devices because I wondered whether there is a more modern alternative to them.
In this document here

https://cddis.nasa.gov/lw11/docs/slrpap1.pdf

on pages 2 and 3, there is a recommended circuit which shows how to AC couple a sine into an ECL gate. The gate used in this example is an MC10EL16, which, fortunately, is still available and even sold by Mouser.
The authors of this document claim that the minimum required input amplitude is around 800mVpp, which is around 1 to 2 dBm into 50 ohms.
However I don't know whether it will be a problem if a higher amplitude would be used.
However to me it looks like this device could be used alternatively to the MC1651. What do you guys think?

Yesterday evening I did a simple test with an ADE-1+ mixer, and connected two square waves to its input. Indeed it works just as fine as with sines, and also the LTSpice simulation confirms this. So I think this is a really smart idea Corby had there, because it basically makes the DMTD completely independent of the input amplitude!

(Next question would then be whether this approach is also usable for phase noise measurement.)

[tkamiya]

I was getting ready to order some boards and noticed 68 ohm register at the output exists on the board but not on circuit diagram.  Which one represents the latest design?  (I'm so excited!)

By the way....  3 for $51 pricing does not appear to include solder masking or silk screening.  With solder masking the order is +20 dollars  Am I doing this wrong??

[chuckb]

There are two options for the PCB assemblies. I did the $70 version with solder mask because I know how bad my soldering is.

[5065AGuru]

Taka,

Nice catch,

added the 0utput resistors on the schematic.

With the solder mask (recommended!) and top silk screen parts total came to $79.00 including the shipping.

Cheers,

Corby

[tkamiya]

I have a question for the group about offset oscillators.

What type of precision/accuracy/stability do I need?  Granted better is.... BETTER but having have to have slight offset to normalized frequencies provide extra challenges.  Some products such as "Small DMTD" by Mr. Riley uses one chip DDS source with an external reference. 

I can easily tune HP10811 off 1Hz by coarse tuning, but any more than 1Hz, I have good possibility of pulling it too far for a stable operation.  I plan to do 0.1 second reading, which requires 10x output per second, which in turn requires 10Hz offset.  I don't know of any good way to achieve this.  Given input splits and goes to both mixers, theory says unless extreme, changes doesn't matter as much.  It goes on to explain, such is an advantage of dual mixer scheme compared to single mixer. 

SO...  is something like well behaving HP5644A/B sufficient?  Anyone have comment?

[chuckb]

Make sure to space the mixer slightly off the PCB when soldering it. If you don't have a slight gap it's easy to add to much solder and it will flow and short to the case. I know this from experience. Somebody makes a dedicated spacer for this but you could use a folded business card to hold 0.030" gap while you solder.
I ordered the pcbs but some of the parts are coming in from china so it will be a few weeks yet.

[rubidium]

I have a question for the group about offset oscillators.

What type of precision/accuracy/stability do I need?  Granted better is.... BETTER but having have to have slight offset to normalized frequencies provide extra challenges.  Some products such as "Small DMTD" by Mr. Riley uses one chip DDS source with an external reference. 

I can easily tune HP10811 off 1Hz by coarse tuning, but any more than 1Hz, I have good possibility of pulling it too far for a stable operation.  I plan to do 0.1 second reading, which requires 10x output per second, which in turn requires 10Hz offset.  I don't know of any good way to achieve this.  Given input splits and goes to both mixers, theory says unless extreme, changes doesn't matter as much.  It goes on to explain, such is an advantage of dual mixer scheme compared to single mixer. 

SO...  is something like well behaving HP5644A/B sufficient?  Anyone have comment?

This is tied to a more general question that I have. What happens to the stability of any good crystal oscillator, nominally rated for frequency F if you tune it to F +/- dF, assuming dF is within the oscillator's EFC range?

[thinkfat]

Might be a good reason/opportunity to buy one of Leo Bodnars mini precision frequency standards.

[tkamiya]

Once I have DMTD working with a good reference, I will be able to measure that.   

[5065AGuru]

Taka,

The HP10811 is specified to tune  + - 10Hz so I don't see a problem there!
I'll see if I can measure a 10811 stability at +10 and -10Hz this weekend.
Remember you can use either the + or the - offset, whichever you can reach.
My offset oscillator is very stable and I use a 1Hz offset. (4.999999 MHz) It runs 24/7.
When I start the DMTD tutorial post (soon) one of the things covered will be the offset oscillator.

Cheers,

Corby

[tkamiya]

Appreciate it, Corby! 

[Pipelie]

I'm new in time-nuts, while I'm looking what's DMTD means,
I found  an article that looks promising
design detail, schematic, and test results are all in the article.


A Small Dual Mixer Time Difference (DMTD) Clock Measuring System
http://www.wriley.com/A%20Small%20DMTD%20System.pdf

[Pipelie]

Spec:

[FriedLogic]

  I checked with ExpressPCB, and the cost for 3 boards including delivery, customs, etc. for the UK would be something like $140.
  At some point I would like to convert it to KiCad anyway, but for now, are there any gerber files available? I'd like to build at least one of the original ones as a reference.

[tkamiya]

When finalizing your order, did you pick "mini" PCB spec?  There are two choices in "mini" type.  One has solder mask and silk screening, and the other doesn't.  Also, "mini" are limited to 3 boards.  Otherwise, cost will significantly rise.

[thinkfat]

Having Gerber files would still be better. 3 boards done by Aisler will cost significantly less.

[chuckb]

Having Gerber files would still be better. 3 boards done by Aisler will cost significantly less.

Here is what Express PCB sent to me. Hope it works for you!

The gerber files are formatted as RS-274-X files.  The drill file includes the hole size definitions.  The diameters listed refer to the finished hole sizes after plating, not the drill size.
Each file can be identify by it's extension:
    .TSK = top silkscreen layer
    .TSM = top solder mask layer
    .TOP = top copper layer
    .GTP = top paste layer
    .IPT = top inner copper layer
    .BOT = bottom copper layer
    .IPB = bottom inner copper layer
    .BSM= bottom solder mask layer
    .GBP = bottom paste layer 
    .DRI = drill file
    .OLN = board outline
    .CSV = pick and place

[thinkfat]

Thanks, the Gerbers look good! KiCADs "GerbView" program can open them just fine.

[FriedLogic]

When finalizing your order, did you pick "mini" PCB spec?  There are two choices in "mini" type.  One has solder mask and silk screening, and the other doesn't.  Also, "mini" are limited to 3 boards.  Otherwise, cost will significantly rise.

The problem is more the extra costs for international - much higher postage, then clearance fees and VAT.
I feel like a bit of a cheapskate for asking, but it all really adds up.
Anyway, thanks to chuckb for posting them.

[thinkfat]

Aisler quotes around 30€ for 3 boards. Didn't check about the shipping rate, but they're in EU, can't be that much.

[5065AGuru]

Attached is a PIX showing the bottom mounted surface mount stuff.

starting just to the right of the yellow 100pfd cap the surface mount stuff starts.

47.5K and .1uf PPS cap in parallel then a 1K then 100K and 6800pfd in parallel.

There are also the two not shown, the 4700pf across the output pins of the mixer

The yellow cap is the 100pf.

I'll post a stuffing diagram for the top soon.

Cheers,

Corby

[5065AGuru]

Taka,

Ran a test of the 10811 at 10HZ offset and it's stability compared to the no offset is just a tiny bit worse but not something that would effect using it as the L.O. in the dual mixer.

Cheers,

Corby

[tkamiya]

Corby,

I appreciate that very much.  Thanks!

I am planning to box together 11801 units to generate 1Hz, 5Hz, and 10Hz.  What's the output level required for this DMTD board?  I'm guessing it'd be a good idea to follow this with a low phase noise buffer amplifier.

[notfaded1]

So what does that mean that 1Hz or 10Hz is better offset?  I have a couple 105B's (both contain 10811's) and some loose 10811's.  I've been running a 96 hour test of my 5061B as DUT using the 5065 as the reference after it burned in for a couple weeks.  Everything is getting better and better.  I'm really excited to think the 5065 could be my primary reference vs. running the Cs all the time.  Also the 5065 isn't that far off my Cs in frequency and is looking more and more consistent!  This morning when I left for work I noticed ADEV had dropped into the 10^14.  This is using the 53132 to measure.  Next I'm going to try the TURN Dynamic Systems DMTD.  I ordered some solid SMA to BNC connectors to convert the TURN DMTD to BNC since I need some longer cables.  When I tested some other devices using the TURN (that has SMA connectors) I used semi rigid SMA cables that were factory made by Tek (exactly the same) to an SRO-100 Rb and reference that both had SMA outputs or was forced to use my RF converter kits.  Those are the ones where you connect two different RF types together with the center generic connector.  I don't think those kits actually are as good as using the factory made ones?  Maybe it doesn't matter... idk yet?

Bill

[tkamiya]

It's not better or worse.  Just different purpose.

I have SRS PRS-10, the rubidium standard.  One of the characteristic of this unit is a HUMP around 0.5 to 1 second.  If I use 1Hz offset, converted signal is 1Hz.  That would mean measurement is made once every second.  Clearly, that's not fine grained enough to detect something happening a sub-second interval.  Raising it to 10Hz offset produces 10Hz.  I can measure at 0.1 second interval.  That's what I need.

Also, one of my interest is Crystal oscillators, including ovenized ones.  Almost all crystal oscillators fall apart, Adev wise, at beyond 300 seconds.  At 100 seconds most will start to show problems.  I want to be able to measure better than once a second.

As far as I know, de-facto standard for DMTD is 1Hz.  This is a special case.

[notfaded1]

I like the idea of 10 measurements per second too.  That's part of the problem with many counters... you need longer sample periods.  That's fine for longer term Rb and Cs measurements.
 But it's why we really need this DMTD for shorter sub second measurement periods in additional to obviously being able to get phase data too.  Like you said for quartz... it's the shorter Tau we're interested in really.

Bill

[tkamiya]

I just tried how far I can pull UCT108663.  Nominal frequency is 10MHz.  On low side, it went to 9,999,995.80.  On high side, it went to 10,000,001.38.  Benefit of course, is size and simplicity of power supply circuit.  I am going to look for modern oscillators and see what's available.

[chuckb]

Just a thought -
When we use a DC voltage to tune an oscillator off frequency the voltage reference can add noise to the RF output signal via the varactor.  When you need to reach a specific frequency we just end up with whatever little extra noise is added and we live with it.

For the DMTD reference frequency it does not matter if the oscillator frequency is high or low. So for lowest noise we should pick a minimum tuning voltage. Ground is the lowest noise voltage in the system.

I wonder how much difference it makes.

[5065AGuru]

Taka,

A lot of people use Dual mixers at 10HZ offset. I usually don't but that's just me!

If you need to look at the stability under 1 second then 10Hz is the way to go.

The data files end up being 10X bigger when you go that route.

The simple dual mixer should work from .2V P-P to >6VP-P on the offset input and .1V to >6V P-P on the RF inputs. So for most oscillators just hook them up!

A low noise buffer is not really needed on the offset input, just hook right in.

Cheers,

Corby

[5065AGuru]

Hi,

Assembled a second unit and the outputs were out of phase (for a common input).

The SBL-1 units on this one did not have the colored insulator to indicate polarity.

Since the board is laid out so that the SBL-1s are installed in opposite directions this matters,

I thought the logo on the top would work so just installed them in opposite directions by that!

Well, that turned out wrong as the mixer output polarity between channels is inverted!

So I'll have to remove a mixer and re-install it in the other orientation.

If your mixers don't have the colored polarity insulator you might want to test them!

Cheers,

Corby

[tkamiya]

I just ran numbers....  Would someone check if this is correct?

If input signal is 10MHz and it drifts 1Hz, that represents 1x10^-7 error.  Now, If this signal is down converted by hetrodyning to 1Hz, period is 1 second.  It would mean error of 1, but it also means 1 second error at 1Hz represents 1x10^-7 error for the original 10MHz.

That would mean 1 milli-second error at 1Hz (10^-3 ) represents 1x10^(-7-3) = 1x10^-10 at 10MHz
That would mean 1 micro-second error at 1Hz (10^-6) represents 1x10(-7-6) = 1x10^-13 at 10MHz
That would mean 1 nano-second error at 1Hz (10^-9) represents 1x10(-7-9) = 1x10^-16 at 10MHz

So, realistically, TICC counter only needs to resolve 100 nano-second to realize 1x10^-14 at 10MHz
(noise level of this DMTD circuit)  Common counter, HP53132 can resolve 500ps. 

Is this correct?

ps.
Corby,  you mentioned you have other DMTD.  What are they?

[ch_scr]

I started putting this circuit into Kicad (only schematic so far), and made the following selfish adjustments (because it fit's my part bin):
LM360 input comperators (comperator to SBL-1 circuit slightly modified), CA3000 as amplifier, fully differential IF filter right into the LM311 (made a LTSpice sim to make the filter act similar to the single ended one). Also added the series resistors on the output gates.
If someone would like to have a look (or use this as a starting point for there own shenanigans in Kicad) feel free.

[tkamiya]

Hi,

Assembled a second unit and the outputs were out of phase (for a common input).

The SBL-1 units on this one did not have the colored insulator to indicate polarity.

Since the board is laid out so that the SBL-1s are installed in opposite directions this matters,

Corby,

How did you test it to find out the polarity is reversed?  What are the symptoms?  I think signals are zero crossing and naturally out of phase.  I'm pretty confused as to what this means.  Thanks!

[5065AGuru]

Taka,

The unit functions a high resolution phase detector so if you feed two in phase RF signals in your outputs should also be in phase.
Of course as the DUT or REF drifts the RF signals will no longer be in phase but the outputs should follow.
Take one RF signal into a tee and feed both inputs from the tee, then on a scope both outputs should be high at the same time and transition low at the same time. With the inverted mixer one was high while the other low!

As far as resolution with an 8 digit counter, 5MHz inputs, and 1 1Hz offset the LSB = 2X10-15th

10Mhz and 1HZ gives and LSB of 1X10-15th

10.000001 00000000

The 1 is the 1HZ offset which is 1X10-7th the 8 digits after are the 8 digits of the counter with each digit one decade lower so -7 + -8 = -15 which is the LSB.

More on this when I start the tutorial post.

Cheers,

Corby

[5065AGuru]

Here is the stuffing diagram.

Blue caps are .1uf bypass caps
yellow caps are .01uf COG
red caps are .1uf COG or film (changed BOM)

[cncjerry]

Corby, Jerry here from time-nuts. 

Nice work, I haven't had a chance to read thru it all yet.  I just built Bill Riley's dual mixer setup using one of Bert's boards and it works great.  Wondering if there was a reason why you didn't use transformers on the input side?

As far as offset generators, I wanted to integrate a DDS into my DMTD system so I didn't have to screw around setting up and/or dedicating a more expensive generator just for testing.  So I bought and built one of the AD9910 DDS "kits" you can find on ebay with the external frequency generator input option.

I wanted to put 10Mhz directly into the AD9910 through a transformer but they way the implemented the chip in the kit software, you have to have a input frequency of 60Mhz or more so I am looking for a 10x digital multiplier to take my 10Mhz reference into 100Mhz.  Alternatively, I can change the software to enable the ref_clk multiplier and then the input frequency can by 3.2Mhz to 60Mhz but that looked like a ton of work.  I found that you can still inject 10Mhz into the board, but it uses the PLL in that mode and the frequency is a little off.  I've been using 10hz offset and instead of for instance, 9,999,990.000 000 the DDS output is measuring 9,999,990.002 53.  From what I understand the offset generator doesn't have to be that critical or phase locked to the reference but darn it, I wanted it to be!

if anyone is interested, the item number like the one I purchased is 64194186839.  DIY KIT for DDS AD9910 Arduino Shield RF Signal Generator.  Nice kit but a lot of smd soldering.  I used chip-quik low temp with a rework heat wand to flow the solder. Took a couple hours to build it even with my tremor.

I scanned the thread really quickly, but I didn't see a reference to Bill Riley's Dual Mixer but I'm sure I overlooked it.  I need another DMTD board so that I can run three-corner analysis.

Compared to my TICC alone and frequency counters including the FA2 and a couple other circuits I have for running stability tests, my DMTD enables me to collect tau's down to .01 or even lower with an offset of 100 and the limiting factor is the TICC.

Nice work,

Jerry

[5065AGuru]

Jerry,

I wanted to evaluate the TS105A circuitry mostly unmodified to see how it would do so did not try to add any bells and whistles!

I did not mention Riley's but probably should have as it has a good writeup.
Those DDS from eBay are nice but I'm pretty married to my 1Hz offset from an FTS 1200.
Does all I want and I don't choke on the big data files!

Cheers,

Corby

[cncjerry]

Corby,  I've heard of the FTS1200 but don't have one.  Can the EFC pull it 1hz (or more)?  I have a bunch of 10811s, you might remember you  sent me connectors, I think, and I've not been able to get the EFC to pull it that much.  I guess I could always do it the manual way and turn the screw... ha!

I wanted to be able to just push a button and do 1hz, 10hz or 100hz offsets.  I didn't think about it, but I wonder if using something that can hold like .5hz would be worth it as then you would get a really big multiplier.

I remember reading in Riley's paper, and then he sent me some emails, that the offset generator doesn't have to be all that accurate or phase locked to the reference.  Also, if you take two of your boards you can do three-hat (as you know) and factor out any of the three signals and that would take the offset out of the picture, I think. I only have tested my DMTD a little as after I built it the smoke up here drove me out of my office.

One parting comment, there was a mixer test done by I think it was NIST for NASA, or maybe the other way around, and a guy used the B-E junction of common transistors for the diode ring.  They then used 5:1 and inversely connected transformers and they claimed with tables and pretty graphs that the phase noise of their mixer whipped all the MCL versions.  I don't remember if the SBL-1 or TUF-3, etc were in that test, also don't know if it matters that much but since you did such a great job bringing the DMTD back to the masses here I thought I would mention it.

Lastly, since I wrote about that issue with the DDS, I found that they have used an external loop filter on the board.  So I'm going to do some reading to see if I can tighten up the PLL a little given I'll only be using the DDS around 10Mhz +/- 100hz.

Jerry

[thinkfat]

Having Gerber files would still be better. 3 boards done by Aisler will cost significantly less.

Here is what Express PCB sent to me. Hope it works for you!

The gerber files are formatted as RS-274-X files.  The drill file includes the hole size definitions.  The diameters listed refer to the finished hole sizes after plating, not the drill size.
Each file can be identify by it's extension:
    .TSK = top silkscreen layer
    .TSM = top solder mask layer
    .TOP = top copper layer
    .GTP = top paste layer
    .IPT = top inner copper layer
    .BOT = bottom copper layer
    .IPB = bottom inner copper layer
    .BSM= bottom solder mask layer
    .GBP = bottom paste layer 
    .DRI = drill file
    .OLN = board outline
    .CSV = pick and place

Took a closer look at the Gerber files, it looks like some component values and text that should have been on a silk screen made it into the bottom copper layer?

Corby, could you show images of the bare PCBs, top and bottom so that I may compare them? For example, there's a "PPS" text that fused with one of the traces on the bottom layer. It doesn't seem to cause any shorts, but doesn't instill a lot of confidence either.

[ch_scr]

[...]
One parting comment, there was a mixer test done by I think it was NIST for NASA, or maybe the other way around, and a guy used the B-E junction of common transistors for the diode ring.  They then used 5:1 and inversely connected transformers and they claimed with tables and pretty graphs that the phase noise of their mixer whipped all the MCL versions.  I don't remember if the SBL-1 or TUF-3, etc were in that test, also don't know if it matters that much but since you did such a great job bringing the DMTD back to the masses here I thought I would mention it.
[...]
Jerry

This seems at least related https://tf.nist.gov/general/pdf/2554.pdf

[tkamiya]

Thinkfat,

That's probably because an older version of (still available) Express PCB does not do screen printing on bottom layer.  New version of the software does. 

Jerry,

Some versions of 11801 has full adjustability via EFC.  Some does 5Hz.  Some does 1Hz.  One for HP5071A (which I have) does 10Hz but it is pain to use EFC as change rate is too fast.

[notfaded1]

Thinkfat,

That's probably because an older version of (still available) Express PCB does not do screen printing on bottom layer.  New version of the software does. 

Jerry,

Some versions of 11801 has full adjustability via EFC.  Some does 5Hz.  Some does 1Hz.  One for HP5071A (which I have) does 10Hz but it is pain to use EFC as change rate is too fast.

So does it make more sense to just use a 10 turn pot to adjust the frequency offset?  Or possibly just find a good OCXO that can be adjusted with it's own built in adj?

Bill

[tkamiya]

I find 10 turns to be still a bit too fast on some, especially when adjustable range is more than 1Hz.  In those cases, I do corse-fine arrangement with multi-turns.  I've tried several arrangements for the latter but none seem to work too well.  Someone suggested putting a second VR on ground side of first.  I'll have to try that next.

[notfaded1]

I've got some really nice Bourns 10 turn pots.  So you're suggesting using two instead of one then?

Bill

[tkamiya]

Depends on what you got for OCXO and its range of EFC.  If it moves more than few Hz, I'd go with two.  Since 10 turns are wire wound, values jump.  It will be very finicky to do with just one and have an effective control.  You could, of course, breadboard it and see how you like it.

[notfaded1]

10 Hz may be more difficult but it sounds like 1 Hz using the built in coarse control on 10811 would be pretty ideal without needing anything else right?  Going for 10 Hz may be a little more tricky but would neat to try.  I've got a literal boatload of different OCXO's many with built in the voltage references.  I have a box of new 10811-60111 and recently just bought one of the original HP counters that used the OCXO boards I have.  Interesting thing... I have large collection of older counters all with Nixie tube displays (because I love the classic neon display technology).  These might actually make a good TIC for a DMTD and would look really cool at the same time.  Attached are pics of the counter... I haven't taken it out of the box yet.  The 10811-60111's look like in attached picture.

Attached are two of my favorites of these because of the tubes they use.  WE LOVE NEON!
https://lh3.googleusercontent.com/-2n0QFSqoqyY/Wm2zTWQi09I/AAAAAAAABjc/-oOk1B6I7SAzaxgsg9NmmfFH47WKP9nQQCLcBGAs/s1600/People%2BLike%2BNeon%2B600%2Bx%2B338.gif
This is gif of a B-7971 tube clock I have called the MOD-SIX.

Bill

[tkamiya]

Yes.  With 1Hz span, one 10-turn is good.  Still touchy but do-able.  Ones that were originally designed to be used in PLL circuit, has 10Hz and 5Hz spans.  Be mindful that initial tuning will be a bear as opening the cabinet to adjust alone will cause a good shift in frequency.  Devising a way to do it through a hole from outside will be ideal.

Colby tested 10MHz shift and he says it is sufficiently stable, so that's what I'm going to use.  My current plan is to have a set of offsets with a selection switch.  I bought mercury wetted reed relay for this.  By always having heater running, it should only take few minutes to settle down to new setting.  Once everything is setup, I'm also going to try using PLL signal generator to see how much difference it actually makes.  My understanding is this is entirely usable.

I grew up using nixie.  As such, I'm not that fond of it myself, but I am aware they are quite popular.  My preference is LED.

Another thing I'm not sure about right now, along with DUT, I need something going into channel A and this needs to be a stable 10MHz.  I don't have anything that good at the moment.  That is going to be a problem....

I like what you did with a steel case.  Is that electrical junction box?  How did you machine it?

[Mrt12]

Another question @Corby.
I see that you have connected the ECL outputs of the MC1650 to the mixer with DC coupling. What happens if you disconnect the input signal at the clock input? I guess Q is low and Qbar is high, so there must be a DC current flowing through the mixer. Doesn't this cause any problems?

[5065AGuru]

The current in the mixer dynamic or static should have the same peak value so will not cause a problem.
Just got three more boards with the updated layout.
I don't see a "PPS" text that fused with one of the traces.
(my original was modified by chuckb so that's where it came from!)
Since I assume that they made the boards with the Gerber file they sent me that should not be an issue.
Top and bottom Pix attached.
If you stay at one offset then just use the mechanical adjust to get you the offset and the EFC for a fine tweak. You don't need ultimate precision for the offset frequency, you just need it stable. I usually end up running at 1.0005Hz or so.
Don't really know it the EFC on an FTS 1200 can pull it 1Hz, the one I use for the offset won't tune to 5Mhz but it will tune to 4.999999! So that's what I use. There is no mechanical adjust on the FTS 1200, just an external EFC input pin.

Cheers,

Corby

[tkamiya]

Corby,

Can we please have a "stuffing chart" for the bottom layer?

Also, on top chart, I see you have naked rounds right below the metal cans of DBM.  That'll be an issue, right?

Question:
There is DUT and there is reference.  (Not the offset input)
For a given DUT, what is the requirement of reference?  10 times better as usual?  So it'll be an issue if say, I want to compare 11801 and all 3 oscillators are 11801, right?

ps.
I just searched on "DMTD" on eBay....  the first item that showed up was a sex toy.

[chuckb]

Having Gerber files would still be better. 3 boards done by Aisler will cost significantly less.

Here is what Express PCB sent to me. Hope it works for you!
...
    .CSV = pick and place

Took a closer look at the Gerber files, it looks like some component values and text that should have been on a silk screen made it into the bottom copper layer?

Corby, could you show images of the bare PCBs, top and bottom so that I may compare them? For example, there's a "PPS" text that fused with one of the traces on the bottom layer. It doesn't seem to cause any shorts, but doesn't instill a lot of confidence either.

When I did the "D" version of the PCB last week I added the component values to the silkscreen on the top side and I added the component value text into the copper ground plane on the back side. This was deliberate. All the bottom side text looks backwards in the Express PCB software because it's an X-Ray view of the pcb. Yes, the first P in PPS overlays a 5V power trace but it will not affect the electrical performance.

I also moved a 51 ohm resistor from between the mixers closer to the center SMA connector because I was concerned about clearance to the mixer case.
ChuckB
 

[thinkfat]

Does anyone know if these mixers (ADE-1, SBL-1 in particular) are symmetric with regards to the LO and RF inputs? Are these interchangeable? That would allow for a very nice, symmetric component placement for the two channels.

[Mrt12]

Does anyone know if these mixers (ADE-1, SBL-1 in particular) are symmetric with regards to the LO and RF inputs? Are these interchangeable? That would allow for a very nice, symmetric component placement for the two channels.

Hi,
I just measured the inductance of the LO and RF inputs of a couple mixers using my HP 4262A.

For ADE-1+: both, LO and RF input, have an inductance of 31.5 uH.
SBL-1X: 10uH for LO, 15uH for IF
SRA-1: both inputs 17.5uH
SRA-3: both inputs 230uH

so, for *most* mixers it looks like the LO and RF inputs are interchangeable, at least when the inductance is considered. Perhaps the turns ratios of the corresponding transformers are not the same, I cannot tell as I didnt want to dismantle my mixers.

[ch_scr]

I have made a layout for the board. It's intended for either powering from µPowerDirect A324R (3W) or Traco TEN4-2422 (4W), whichever makes the cut
(you guessed it, parts bin had both. 3W might be on the edge tho).
It's intended to be milled as 2-sided PCB, the yellow inner layer shows placement of wire links. This is done so a groundplane can be on one side.
I have 4 seperate local regulation +-6V for both inputs (1,2) the offset input (3) and both IF amps (4) with LM337 / LM317 SO-8.
I suppose this is more usefull as a starting point then to be made as a direct copy.

[notfaded1]

I just had a thought for the 5MHz Reference I could use the BVA.  I'd like to figure out a 5MHz and 10MHz setup.  Another question.  If you had two of these setup could you hook it up so you could do three corner hat measurements?  Would you use the same reference or same offset reference in a case like this?  Obviously I'd like to start with one that works but thinking about how you would hook up for three corner hat measurements?  I'm going to learn the other software you're using Corby.  I've used Stable32 and timelab.  I'm not as familiar with your other tool.

Regards,

Bill

[tkamiya]

I'm going to make one as a direct copy of Corby's design, then have another where I will experiment.  As to if DBMs are symmetric, they look that way.  I've taken few apart. 

I've made my power supply with linear regulators.  I wasn't comfortable with ripples and noise in involving switching power supply.  At minimum, if I have to do that, I'll do that on another PCB and well filter it.  It didn't make sense to me to get those expensive low noise OPAmps and use dirty power supplies.  It'll be interesting to see what difference, if any, substitution will make though.

[Mrt12]

I just had a thought for the 5MHz Reference I could use the BVA.  I'd like to figure out a 5MHz and 10MHz setup.  Another question.  If you had two of these setup could you hook it up so you could do three corner hat measurements?  Would you use the same reference or same offset reference in a case like this?  Obviously I'd like to start with one that works but thinking about how you would hook up for three corner hat measurements?  I'm going to learn the other software you're using Corby.  I've used Stable32 and timelab.  I'm not as familiar with your other tool.

Regards,

Bill

Hi Bill
I am also working on a DMTD for a while now. It is actually not a DMTD but a TMTD, a "triple mixer". It uses exactly the approach you mentioned.
There are four inputs: three inputs for the DUTs, and one for the offset oscillator. The zero crossings are then time-stamped and via the appropriate data processing you can do three cornered hat. That is exactly my plan.
I was almost finished with my design when I saw this thread and noticed Corby's excellent idea of using an ECL gate to drive the mixers. That's a brilliant idea because it will make the whole thing less sensitive to the input signal level.
Also I hope that the ECL gates provide a certain degree of isolation.
I ordered a couple MC100EL16 because the MC1650 Corby uses is obsolete and no longer available. The MC100EL16 is a driver, but it can be AC coupled from a sine wave. That's what I plan to do. With some of the ordered devices I want to test first how well that works and what degree of reverse isolation they provide. Next step is then to add them to my design.
However my design is on a 130x80mm PCB and uses SMT components only. For the mixers, I plan to take ADE-1+ or ADE-2+ which will work up to 400MHz or 1GHz, respectively. For me, the high frequency end is almost as important as 10 MHz because I want to characterise 100+ MHz oscillators as well.

[tkamiya]

I haven't done multi-channel DMTD personally but here's a note from someone who has:

https://ftp.tapr.org/tech_docs/multi-TICC_App_Note_2020-01.pdf

When you go there, you might get a warning from your browser.  TAPR is a known ham radio organization.  It's safe.  TICC is a time interval counter.  I have one. 

[notfaded1]

That would be the ultimate right multiple port DMTD with linked TIC's counting together aka poor mans PhaseStation 53100A.

[thinkfat]

I have made a layout for the board. It's intended for either powering from µPowerDirect A324R (3W) or Traco TEN4-2422 (4W), whichever makes the cut
(you guessed it, parts bin had both. 3W might be on the edge tho).
It's intended to be milled as 2-sided PCB, the yellow inner layer shows placement of wire links. This is done so a groundplane can be on one side.
I have 4 seperate local regulation +-6V for both inputs (1,2) the offset input (3) and both IF amps (4) with LM337 / LM317 SO-8.
I suppose this is more usefull as a starting point then to be made as a direct copy.

Is there an inherent advantage in using a differential IF configuration? I figured my parts bin doesn't contain any Tesla stuff but instead of the MA3000, any FDA could be used, like, LTC6362 or THS4131?

The THS4131 could be used with a single 5V supply rail, at the cost of increased THD (haha!).

[5065AGuru]

Taka,

See the PIX showing the SMD on the bottom. That post also has the parts values!

Yes the mixers need to up off the board a tiny bit!

Of course you want one of your best performers for the REF input.

You can test a few 10811 against each other and isolate the best ones.

As good as or better is fine, it would be very hard to find a REF that is 10X better than most of what we see!

I use my HP5065A for almost all measurements, and only switch to my FTS1200 when I'm most concerned about the lower Tau. (about8 Sec and below.)

Cheers,

Corby

[5065AGuru]

Bill,

I'd recommend using TimeLab.
I'm just using PLOTTER in my examples as I use it all the time!
Plotter does not like the commas the 53131A puts out but TimeLab works fine with them.
Also TimeLab has more versatile measurements and plots.

Once we get the custom counter debugged you will not have to tie up a fancy instrument and can just build it into the same box your dual mixer is in!

Cheers,

Corby

[tkamiya]

You can test a few 10811 against each other and isolate the best ones.

And, THIS puts me squarely back to my original problem.  I'm in a position where none is proven to be better than the rest.  There are lots of candidates but not one has been selected.  Given that, how can I accomplish this goal?  DMTD as well as all other measuring method requires one to compare DUT against other to choose the best - which requires one unit that is far better than the rest.  I can compare 11801 against another 11801 but I can't imagine how useful this will be....  as in the end, I will not know if i just measured DUT or the reference.  Or, am I missing something?

[FriedLogic]

Nice work, I haven't had a chance to read thru it all yet.  I just built Bill Riley's dual mixer setup using one of Bert's boards and it works great.  Wondering if there was a reason why you didn't use transformers on the input side?

   I see isolating transformers being used in quite a few projects, but I'm not convinced that it is usually the way to go for best performance.
   I've noticed that isolating transformers can make noise issues worse, and have usually had better results with good solid grounding.
   I expect that there are situations where a DMTD might benefit from them, but in most cases I don't see an advantage. Although not as neat or cheap, they can always be added externally if necessary.

   Some versions will no doubt be done with isolating transformers, so it will be interesting to see the results.

[5065AGuru]

Taka,

You can use a DMTD system to figure out your best oscillator!
Traditionally when you build an ultra-stable oscillator and want to evaluate it you build two.
Then you test them against each other.
You do NOT need one to be MUCH more stable than the other!
I don't remember the exact math but if I remember correctly two identical stability oscillators will give an Allan deviation that needs to be degraded? by 1.4. So if your two oscillator read 5X10-13th then they really are 1.4X5 or 7X10-13th. I may have the maths wrong but if you research DMTD enough there are references to it.
Maybe someone can weigh in on this.

Also three corner measurements can help you measure the real performance.

For my standards I can't really tell if an oscillator is better that around 2.5X10-13th at 1 Second although I can accurately measure down to 7X10-14th at 100Sec.

Cheers,

Corby

[tkamiya]

Right.  But how you tell which is which?

Say I have OSC1 and OSC2.  OSC1 into DUT port, OSC2 into REF port.  DMTD will output the time between a pulse from OSC1 and next pulse from OSC2.  One output per cycle.  How do you tell which one is better?  I can understand it will give aggregate of both - say both are at some degraded number from actual reading.  But my need is to compare.  Keep comparing until one remains as a winner.  How can THAT be done?  I have more than a dozen 11801 ready for this process.

This has been a problem since beginning.  I was getting ready to buy a brand new OCXO so I can trust the factory data.  But 11801 are incredibly good.  To one up on it, I'll have to spend quite a bit.  Can I come over to your house post-Covid with stacks of sandwiches?  Will you let me measure against your Rubidium?

[5065AGuru]

Taka,

Just test them pairwise against each other.

As you fine pairs that test better than others keep going until a couple clear winners are found.

If you have HP 5065A, 5061A, 5061B, or 105B that have the 10811-60109 then use those instruments outputs (in open loop) at first.

Most 10811-60109 give a performance of between 4 and 7X10-13th in the low Tau region.

They are the best version out there. The ones in the 5071A can also be good.
I do however have a 10811-60111 that also gives that level of performance but that is rare!

The 5065A (locked) usually has an AD of between 1 and 2X10-13 at 100Sec so could be used to start picking out the best out the pile of oscillators!

Everyone has there "golden" or favorite oscillator but it's a never ending search!

Cheers,

Corby

[tkamiya]

Corby,

How about something like this.  Say I will build and set up two complete DMTD system.  They will share difference ref.  Take a reference oscillator and divide that signal into 2.  Each of the leg goes to ref port of DMTD1 and DMTD2.  DUT will have OSC1 and OSC2.  Take Adev measurement on both simultaneously.  Then, I will be able to pick which one is better relative to each other, although actual reading will be irrelevant.  I can keep going like this 2 OSC at a time.  Then gather better of two as a set.  Now, repeat the process until one is a winner.  I should, actually, have enough parts to make 3 complete set.

Now, take 2 of the best.  Place the best one as common REF and test original REF and the second best.  After this step, I can tell if REF is any good....

[chuckb]

With the Three Cornered Hat technique, you measure the three pairs of oscillators, A-B, B-C, C-A, develop the AD curves for the pairs, put that data into the TCH algorithm and it will provide the unique curves for A, B and C. The Stable32 software will do all of this for you. I don't know about other software packages.

I used the technique a month ago and it worked well.

I got my Corby DMTD PCBs today. I'm still waiting on a few parts from China. I have two nice Datum / FTS 1130 10MHz osc that I want to test. They were pulled from some Telecom freq references. I have not found any data on these but they are in the same package as the high end Datum 1000 osc.

The Datum crystal osc will probably be my best reference up to 100-1000 sec. Then the GPS locked Rb standard will be best for more than 1000 sec.

[edpalmer42]

I have two nice Datum / FTS 1130 10MHz osc that I want to test. They were pulled from some Telecom freq references. I have not found any data on these but they are in the same package as the high end Datum 1000 osc.

I found a small note on the FTS 1130.  Looks very nice!

A Stratum Three module uses an AT-cut FTS 2510
OCXO as the oscillator. The Stratum 2.1 module uses
a SC-cut FTS 1000B oscillator. Its performance greatly
exceeds the Stratum 3 module, but falls short of the
Stratum Two requirements for thermal stability. The
Stratum Two module uses an FTS 1130 oscillator. The
1130 oscillator utilizes a double oven technique to
provide the required thermal stability. The inner oven
is hermetically sealed to provide immunity to humidity
changes [5]. Table 3 lists the specifications for the
different oscillator modules.

Table 3. Oscillator Module Specifications
                                            Str 3      Sir 2.1         Sir 2
Tuning Range                        10e-6     3e-7            3e-7
Thermal stability(per ' C)         7e-9       1.7e-11       3e-12
Aging (per day)                      1e-8       1e-10          5e-11
Time constant range(secs)     20-100    100-3,000   100-10,000

https://archive.org/download/DTIC_ADA272017/DTIC_ADA272017.pdf

[ch_scr]

Is there an inherent advantage in using a differential IF configuration? I figured my parts bin doesn't contain any Tesla stuff but instead of the MA3000, any FDA could be used, like, LTC6362 or THS4131?

The THS4131 could be used with a single 5V supply rail, at the cost of increased THD (haha!).

MA3000 is clone of RCA CA3000. But there are still companys around that have stock of MA3000. In principle any FDA could be used.
The FDA should amplify a slow changing voltage (1-10Hz) -> low input offset voltage, input offset voltage thermal drift and voltage noise
It should attenuate residual high frequency signal coming through filter -> high CMRR at Frequency in 10Mhz range.

Differential signal path has the benefit of not beeing referenced to ground (currents through the groundplane can't inject noise).
When using a single ended topology you have to make sure all components (opamp inputs, filter components) see the "same" ground -
for MHz you want a groundplane, for Hz you want star ground (generally speaking). Here you have both so it might be complicated.

I have opted to just reference all the circuit to the groundplane, in a single ended configuration this would not be optimal.
In my circuit the only (DC) connection of the IF signal to ground is at one point, and not directly as something to compare against as well (opamp/comperator input).
Also, if noise couples into the differential path from the outside (hopefully evenly) it is rejected by the opamp (to the best (CMRR) of it's ability (inside it's bandwith))

[thinkfat]

I've been experimenting a bit, simulating the design in LTSpice. I've added the current schematic as an attachment.
I like the idea of using the MC10EL16 as input buffer, though it seems a bit tricky to terminate ECL gates properly (schematic also attached).

All the FDAs I found have a Vocm voltage output, which you can use to set the center threshold for a comparator or DAC.
The output of the mixer is floating, isn't it? It's not ground referenced. Does it make sense to loosely bias it to Vocm (R14/R15 in my schematic) to prevent the differential pair from drifting outside of the valid input range? Or are the amplifier inputs always biased against the Vocm anyway?

I as far as components go, I just used what seemed appropriate in the LTspice library. As far as I can see, this design could be run from a single 5V supply rail (or multiple 5V rails).

[Mrt12]

I've been experimenting a bit, simulating the design in LTSpice. I've added the current schematic as an attachment.
I like the idea of using the MC10EL16 as input buffer, though it seems a bit tricky to terminate ECL gates properly (schematic also attached).

All the FDAs I found have a Vocm voltage output, which you can use to set the center threshold for a comparator or DAC.
The output of the mixer is floating, isn't it? It's not ground referenced. Does it make sense to loosely bias it to Vocm (R14/R15 in my schematic) to prevent the differential pair from drifting outside of the valid input range? Or are the amplifier inputs always biased against the Vocm anyway?

I as far as components go, I just used what seemed appropriate in the LTspice library. As far as I can see, this design could be run from a single 5V supply rail (or multiple 5V rails).

Yes the MC10EL16 has the advantage that it is still in production and is available in SMT packages. We could even think about using the MC100EP16VA: this is basically the same, but with a higher gain. This would allow for an even larger dynamic range at the input!

As for the termination of the ECL gates, I read the corresponding application notes from ON Semi. I first went for the Y Termination in my design, the same as you do. However I am unsure; maybe the Thevenin termination is better because the voltage ad the Q and Qbar pins is better defined. Also it is only one additional resistor compared to the Y termination, so the disadvantage is not a big deal I think.

I am still a bit concerned about the DC coupling between the ECL gate and the mixer, though.

[thinkfat]

I've been experimenting a bit, simulating the design in LTSpice. I've added the current schematic as an attachment.
I like the idea of using the MC10EL16 as input buffer, though it seems a bit tricky to terminate ECL gates properly (schematic also attached).

All the FDAs I found have a Vocm voltage output, which you can use to set the center threshold for a comparator or DAC.
The output of the mixer is floating, isn't it? It's not ground referenced. Does it make sense to loosely bias it to Vocm (R14/R15 in my schematic) to prevent the differential pair from drifting outside of the valid input range? Or are the amplifier inputs always biased against the Vocm anyway?

I as far as components go, I just used what seemed appropriate in the LTspice library. As far as I can see, this design could be run from a single 5V supply rail (or multiple 5V rails).

Yes the MC10EL16 has the advantage that it is still in production and is available in SMT packages. We could even think about using the MC100EP16VA: this is basically the same, but with a higher gain. This would allow for an even larger dynamic range at the input!

As for the termination of the ECL gates, I read the corresponding application notes from ON Semi. I first went for the Y Termination in my design, the same as you do. However I am unsure; maybe the Thevenin termination is better because the voltage ad the Q and Qbar pins is better defined. Also it is only one additional resistor compared to the Y termination, so the disadvantage is not a big deal I think.

I am still a bit concerned about the DC coupling between the ECL gate and the mixer, though.

I started routing the input part, I think I'm going to switch to Thevenin. The layout seems cleaner and simpler. With "Y", I didn't find a good way to route the VTT rail without the differential signal crossing over it. The Thevenin doesn't create a separate rail, it's just a voltage divider between VCC and VEE.

I like the idea of using a differential IF signal. It seems a good idea to decouple the mixer output from the ground noise until it is sufficiently amplified.

No opinion yet about the DC coupling of the ECL gate and the mixer. There is nothing than the output drive strength of ECL gate to limit the current...

If you disconnect the input signal though, should not the 50 Ohm resistor across the inputs bring the differential input voltage to 0?

[ch_scr]

[...]

I started routing the input part, I think I'm going to switch to Thevenin. The layout seems cleaner and simpler. With "Y", I didn't find a good way to route the VTT rail without the differential signal crossing over it. The Thevenin doesn't create a separate rail, it's just a voltage divider between VCC and VEE.

I like the idea of using a differential IF signal. It seems a good idea to decouple the mixer output from the ground noise until it is sufficiently amplified.

No opinion yet about the DC coupling of the ECL gate and the mixer. There is nothing than the output drive strength of ECL gate to limit the current...

If you disconnect the input signal though, should not the 50 Ohm resistor across the inputs bring the differential input voltage to 0?

Yes, the input of the buffer / comperator will go to zero. But the outputs are always inverted from each other, always one high the other low.
For the DBM always the RF power and IF maximum current is specified, but not RF coil DC max current.
With 1u / 100n smd 0805 X7R in parallel very wide band ac coupling can be achieved - a jumper can always be placed later if it turns out to be a problem.

[Mrt12]

I've been experimenting a bit, simulating the design in LTSpice. I've added the current schematic as an attachment.
I like the idea of using the MC10EL16 as input buffer, though it seems a bit tricky to terminate ECL gates properly (schematic also attached).

All the FDAs I found have a Vocm voltage output, which you can use to set the center threshold for a comparator or DAC.
The output of the mixer is floating, isn't it? It's not ground referenced. Does it make sense to loosely bias it to Vocm (R14/R15 in my schematic) to prevent the differential pair from drifting outside of the valid input range? Or are the amplifier inputs always biased against the Vocm anyway?

I as far as components go, I just used what seemed appropriate in the LTspice library. As far as I can see, this design could be run from a single 5V supply rail (or multiple 5V rails).

Yes the MC10EL16 has the advantage that it is still in production and is available in SMT packages. We could even think about using the MC100EP16VA: this is basically the same, but with a higher gain. This would allow for an even larger dynamic range at the input!

As for the termination of the ECL gates, I read the corresponding application notes from ON Semi. I first went for the Y Termination in my design, the same as you do. However I am unsure; maybe the Thevenin termination is better because the voltage ad the Q and Qbar pins is better defined. Also it is only one additional resistor compared to the Y termination, so the disadvantage is not a big deal I think.

I am still a bit concerned about the DC coupling between the ECL gate and the mixer, though.

I started routing the input part, I think I'm going to switch to Thevenin. The layout seems cleaner and simpler. With "Y", I didn't find a good way to route the VTT rail without the differential signal crossing over it. The Thevenin doesn't create a separate rail, it's just a voltage divider between VCC and VEE.

I like the idea of using a differential IF signal. It seems a good idea to decouple the mixer output from the ground noise until it is sufficiently amplified.

No opinion yet about the DC coupling of the ECL gate and the mixer. There is nothing than the output drive strength of ECL gate to limit the current...

If you disconnect the input signal though, should not the 50 Ohm resistor across the inputs bring the differential input voltage to 0?

as ch_scr said, the Q and Qbar outputs are always inverted with respect to each other, so there will be definitely a DC current through the mixer if the input signal is disconnected.

I think a series capacitor is a wise idea. I am thinking about using even two capacitors, one for Q and the other for Qbar, to keep the connection as symmetrical as possible. Maybe this is unnecessary?

I ordered a couple of ECL gates on eBay (MC100EL16). I found a bag with 30 pieces for a good price. As soon as they arrive I can make some tests, hopefully ;-)

[Mrt12]

This is my current design of the zero cross detector and the ECL input stage.
What do you guys think, do the ECL gates provide enough isolation between the mixer and the input connector?

As I said I plan to make a 3 channel device which will be suitable for three cornered hat methodology. So the offset oscillator is common to all three channels.
If only an ordinary DMTD is required, just don't use the third channel.

[tkamiya]

I just learned Mini-circuits will sell insulated spacers for SBL-1.  I am requesting price quote and will update this post once I get it.

P/N B14-045-01

[edpalmer42]

I have two nice Datum / FTS 1130 10MHz osc that I want to test. They were pulled from some Telecom freq references. I have not found any data on these but they are in the same package as the high end Datum 1000 osc.

I did some more digging.  No data sheet unfortunately, but better than nothing.

https://web.archive.org/web/19990502022103/http://www.datum.com/prod_fts_1130.html

[chuckb]

I have two nice Datum / FTS 1130 10MHz osc that I want to test. They were pulled from some Telecom freq references. I have not found any data on these but they are in the same package as the high end Datum 1000 osc.

I did some more digging.  No data sheet unfortunately, but better than nothing.

https://web.archive.org/web/19990502022103/http://www.datum.com/prod_fts_1130.html

Thanks for investigating this. I better get the 1130s warmed up for testing!

[thinkfat]

as ch_scr said, the Q and Qbar outputs are always inverted with respect to each other, so there will be definitely a DC current through the mixer if the input signal is disconnected.

I think a series capacitor is a wise idea. I am thinking about using even two capacitors, one for Q and the other for Qbar, to keep the connection as symmetrical as possible. Maybe this is unnecessary?

Wikipedia says ECL gate current is limited to avoid driving the output transistors into saturation and avoiding slow switch-off. I'd therefore say it's not necessary to do any current limiting or DC blocking.

[tkamiya]

I just learned Mini-circuits will sell insulated spacers for SBL-1.  I am requesting price quote and will update this post once I get it.

P/N B14-045-01

I got a quote.  40 cents each.

[notfaded1]

I have two nice Datum / FTS 1130 10MHz osc that I want to test. They were pulled from some Telecom freq references. I have not found any data on these but they are in the same package as the high end Datum 1000 osc.

I did some more digging.  No data sheet unfortunately, but better than nothing.

https://web.archive.org/web/19990502022103/http://www.datum.com/prod_fts_1130.html

It sure looks a lot like the 1000B I have Ed from the Datum 4040A.  What hardware did these FTS 1130 come from?

Bill

[Mrt12]

On a previous page I said that I am making a similar PCB with three channels.
For reasons of symmetry, I now expanded it to four channels :-)
Attached is my layout of the RF part. As one can see it is actually two identical DMTDs which can be used either as two independent DMTDs or as a multi channel mixer. So it is actually a QMTD ;-)
In terms of crosstalk, I don't know how bad this layout will be, but I plan to use a PCB with a solid ground plane and fill unused space with ground such that the isolation should be optimal.

[chuckb]

I have two nice Datum / FTS 1130 10MHz osc that I want to test. They were pulled from some Telecom freq references. I have not found any data on these but they are in the same package as the high end Datum 1000 osc.

I did some more digging.  No data sheet unfortunately, but better than nothing.

https://web.archive.org/web/19990502022103/http://www.datum.com/prod_fts_1130.html

It sure looks a lot like the 1000B I have Ed from the Datum 4040A.  What hardware did these FTS 1130 come from?

Bill

The oscillators were in a master osc module for some rack mounted Stratum 2 Telecom equipment. I just salvaged the osc and SMA coaxes then tossed the rest of the module many years ago. The IEEE proceedings stated they were developed for a Datum FTS 3380 Telecom Timing Signal Generator. Whatever that is.

Each osc has two independent 10Mhz outputs, which is nice. Ten years ago I opened it to figure out which pins got power and ground. I also noticed that the pcb had provisions for two more buffered 10Mhz outputs. I may have to open it again and add some more buffers and coax connectors to the PCB. I will take pictures if I do that.

It looks like they modified/selected the normal FTS1000 osc for 3x better temp co and aging. So I have high hopes for their performance. We will see.

I will probably use my little HP 5316B Counter (100ns single shot resolution) for this testing. I added a small Rb and PS inside to replace the normal osc many years ago. It makes for a small powerful package. It hasn't gotten much use in the last few years. And it's much quieter than the HP 5370B (20ps single shot resolution) counter.
 

[thinkfat]

The whole idea behind the DMTD system is to not need a 20ps counter in the first place

[tkamiya]

Um....  I'm thinking we should leave this thread to topics that are related to DMTD and Corby's design.  It'll be kind of hard to follow in few months otherwise.

[notfaded1]

No problem the dual DMTD board looks pretty neat if it would work.  I don't think I've seen a 4 channel one before.  Although Corby said he had an 8 channel counter so anything is possible.  Seems like you could use two of the original design though couldn't you?  I assume Corby and others have done this for exactly the same reason.  The way I understand it the two references need to be very close to each other in stability so the measured deviation can be corrected for by dividing by the square root of 2.  Many of the videos I've watched the person used two identical/similar reference oscillators to test the DUT.  Bill Wriley did say it could be extended to M clocks.  I don't fully understand negative variances but it's supposed to be a sign the method is failing.  To try and better understand how this works I've been reading this old paper that has a chapter called Decomposition Of Multi-Clock Comparisons.  For some reason this paper seems to explain things better than some others.  To be honest though I'll be happy if I can get one DMTD working well.

Bill

[Mrt12]

If it's interesting for someone: I made some progress with the layout of my quad mixer. In the picture attached is one half. They are almost perfectly symmetrical, I even tried to match the trace lengths between the channels.
I never did a PCB layout with ECL gates before, and I have been warned it will be "extremely difficult", but I don't know to what degree this is true. Therefore I took care to minimise every track's length which has to do with the ECL gates.

@Corby: you posted on page 1 some numbers about the noise floor of your DMTD. But I remember that you said it was "too good". Why? I see that your design uses one OpAmp after the mixer. I decided to do so as well, but I am not sure how much it will affect the noise floor. Can you say something about this?

[Mrt12]

No problem the dual DMTD board looks pretty neat if it would work.  I don't think I've seen a 4 channel one before.  Although Corby said he had an 8 channel counter so anything is possible.  Seems like you could use two of the original design though couldn't you?  I assume Corby and others have done this for exactly the same reason.  The way I understand it the two references need to be very close to each other in stability so the measured deviation can be corrected for by dividing by the square root of 2.  Many of the videos I've watched the person used two identical/similar reference oscillators to test the DUT.  Bill Wriley did say it could be extended to M clocks.  I don't fully understand negative variances but it's supposed to be a sign the method is failing.

Bill

Yes you divide by sqrt(2) if the two DUTs are very similar to each other.
A bit simplified: you take a load of measurements and then you calculate the variance. The variance you get is the sum of the variance of both DUTs (remember, in the ideal case, the offset oscillator drops out).

I don't know how I can enter math here, so I write it like so: the variance of your measured data is

sigma_meas² = sigma₁² + sigma₂²

where sigma² is the variance, and sigma₁ and sigma₂, respectively are the stabilities of your individual DUTs. Now if both DUTs are "the same", e.g. two LPRO-101 against each other, you have

sigma₁ = sigma₂ = sigma₁² = sigma₂² = sigma_dut²

and therefore you can write

sigma_meas² = sigma₁² + sigma₂² = 2 * sigma_dut²

and now you don't want to know the variance, but the standard deviation AKA allan deviation, so take the square root

sigma_meas = sqrt(2 * sigma_dut²) = sqrt(2) * sigma_dut

For the three cornered hat, the methodology is very simple as well. This time you observe three oscillators, so you kind of have three measurements, each of which has its own variance (sigma_xy is the variance when DUTx is measured against DUTy):

sigma₁₂² = sigma₁² + sigma₂²
sigma₁₃² = sigma₁² + sigma₃²
sigma₂₃² = sigma₂² + sigma₃²

and now you can do the math and solve this equation system because you have three equations and three unknowns (the variance of each DUT). For example, take the first equation, subtract the second one and add the third one. You get

sigma₁₂² - sigma₁₃² + sigma₂₃² = sigma₂² - sigma₃² + sigma₂² + sigma₃² = 2 * sigma₂²

Note, on the right hand side, how all DUT variances have dropped out. All what remains is the variance of a single DUT, and on the left hand side, this is what you measured. So you can figure out the unknown variance of one DUT, provided you take at least three different measurements.

The problem with the negative variance arises from numerical problems. Because your measurement contains noise etc., your variances may not behave exactly as they should and the whole thing breaks apart. For instance, with the Riley DMTD, you can only measure two DUTs at a time against each other. So, if you want to do three cornered hat, you measure 1-2 on day 1, 1-3 on day 2 and 2-3 on day 3. However, the DUTs might change slightly during this time, you don't observe them at the same time and this is where the numerical problem may arise.
I would expect that if you had a multi-channel DMTD, where you can measure all three DUTs at the same time, the numerical problems will be less.

[ch_scr]

If it's interesting for someone: I made some progress with the layout of my quad mixer. In the picture attached is one half. They are almost perfectly symmetrical, I even tried to match the trace lengths between the channels.
I never did a PCB layout with ECL gates before, and I have been warned it will be "extremely difficult", but I don't know to what degree this is true. Therefore I took care to minimise every track's length which has to do with the ECL gates.

@Corby: you posted on page 1 some numbers about the noise floor of your DMTD. But I remember that you said it was "too good". Why? I see that your design uses one OpAmp after the mixer. I decided to do so as well, but I am not sure how much it will affect the noise floor. Can you say something about this?

With so many channels you might want to include some passive resistive splitters like the system that was linked before in the thread. (If you have space left on the pcb)

[thinkfat]

If it's interesting for someone: I made some progress with the layout of my quad mixer. In the picture attached is one half. They are almost perfectly symmetrical, I even tried to match the trace lengths between the channels.
I never did a PCB layout with ECL gates before, and I have been warned it will be "extremely difficult", but I don't know to what degree this is true. Therefore I took care to minimise every track's length which has to do with the ECL gates.

The absolute delay between the channels is not all that important, as long as it is constant.

[ch_scr]

If it's interesting for someone: I made some progress with the layout of my quad mixer. In the picture attached is one half. They are almost perfectly symmetrical, I even tried to match the trace lengths between the channels.
I never did a PCB layout with ECL gates before, and I have been warned it will be "extremely difficult", but I don't know to what degree this is true. Therefore I took care to minimise every track's length which has to do with the ECL gates.

The absolute delay between the channels is not all that important, as long as it is constant.

I wondered if instead of having a phase shifter on one channel input a variable trigger level would work? Would have to be edge sensitive tho, so something a bit more clever that a voltage level pot on the negative input of input comperator would be in order. This to-be-concieved solution certainly has more potential to mess things up than some coax-in-a-box...

(If one were to put a divide-by-2 flipflop after the comperator, the signal would always be 50/50 on the pulsewidth, no matter what the input signal was or what the "phase shifter" did to it. Whole shebang would always be running at half the input frequency (and offset))

On another note: Thermally coupled input gates? To better synchronize their thermally induced delay differences?
Maybe clamp a strip of 2mm aluminium over all smd cases?

[5065AGuru]

Mrt12,

Hi,

All the more sophisticated dual mixers I have Do not have that "too good" performance.
After cleaning up some stuff the new noise floor is just about where I expected it.
The fancy ones use more than one stage of Opamps.
One thing to do to get a close stable phase match is to use low TC caps on the op amp stage and match them.
Also match the parallel resistors in that stage and use low TC on them.
On my NBS unit I replaced the original 100K resistor with .01% 1PPM and the capacitor for the 1HZ setting with selected PPS caps.

Cheers,

Corby

@Corby: you posted on page 1 some numbers about the noise floor of your DMTD. But I remember that you said it was "too good". Why? I see that your design uses one OpAmp after the mixer. I decided to do so as well, but I am not sure how much it will affect the noise floor. Can you say something about this?

[thinkfat]

Mrt12,

Hi,

All the more sophisticated dual mixers I have Do not have that "too good" performance.
After cleaning up some stuff the new noise floor is just about where I expected it.
The fancy ones use more than one stage of Opamps.
One thing to do to get a close stable phase match is to use low TC caps on the op amp stage and match them.
Also match the parallel resistors in that stage and use low TC on them.
On my NBS unit I replaced the original 100K resistor with .01% 1PPM and the capacitor for the 1HZ setting with selected PPS caps.

Cheers,

Corby

Corby, how do you select PPS capacitors for matching TC?

@ch_scr: that might be a drawback of a differential signal path. You have to match even more components, not only between the channels but also for symmetry of the positive and negative half of the amplifier. I guess the IF filter is probably not that critical, but definitely the components that set the amplifier gain.

PS: my take on the DMTD with differential IF

[5065AGuru]

I only matched the capacitance of the caps! I relied on the very low TC intrinsic to the PPS caps.

Cheers,

Corby

[tkamiya]

Corby,

I'm amazed by amount of excitement your project has generated.  Thanks!

[Mrt12]

Corby,
you said

After cleaning up some stuff the new noise floor is just about where I expected it.

so what's the corrected noise floor then? do i need just more gain if I want to go towards the 1e-13 range at 1sec?

[tkamiya]

....and what dictates the noise floor? 

I wonder such things as using semi-rigid line for input and output and securing such rigidly would improve noise level?  I plan to use N fitting for both in and out, then use sma-semiRigid-sma assembly for interconnect.  I also plan to secure it to chassis for longer runs.  I just don't trust BNC, and SMA is rather fragile for anything other than internal connections.  I'm afraid it will be just another possible cause for unrepeatable results.  After all, we are seating over less than nano-seconds variance.

[Gerhard_dk4xp]

AD8612 would be suitable, too, I guess. But I'm more concerned about the mixers. The MiniCircuits stuff is really nice, but I need to find a distributor that stocks them. Digikey doesn't have them at all and Mouser refuses to sell them in Germany 

The German distributor for Minicircuits is www.municom.de. They were here already
40 years ago and won't like it to be kicked away by Mouser.
You can get an interesting subset at funkamateur.de who is a NOS reseller, not a distributor.

Cheers, Gerhard

[5065AGuru]

Mrt12,

If you look back on page 1 you will see the plot "realnoisefloor". That is the corrected plot.

You can reduce the BW by increasing the value of the cap around the op amp.

If you reduce it to 10Hz then you can still use 10Hz or 1Hz offset frequencies.

If you reduce it to below 10Hz you can only use the 1Hz offset.

The lower the BW the lower the noise in the system and the lower the noise floor.

Also stable well regulated low noise power supplies are best.

The original NBS106 DMTD system used batteries, the later version (the one I have) has built in AC powered supplies.

I keep my DMTD units powered 24/7 so that they are mostly at thermal equilibrium.

I'm not sure how much you can buy with the simple dual mixer architecture?

For my regular DMTD units running them with 10MHz vice 5MHz inputs also buys a little better performance but unsure if that applies here.
Cheers,

Corby

[thinkfat]

Corby,
you said

After cleaning up some stuff the new noise floor is just about where I expected it.

so what's the corrected noise floor then? do i need just more gain if I want to go towards the 1e-13 range at 1sec?

I'd say the noise floor is mainly dictated by the ZCD. You want a low noise amplifier after the IF filter with moderate gain and small bandwidth and increase the bandwidth in additional stages to have a steep trigger signal eventually.

[thinkfat]

AD8612 would be suitable, too, I guess. But I'm more concerned about the mixers. The MiniCircuits stuff is really nice, but I need to find a distributor that stocks them. Digikey doesn't have them at all and Mouser refuses to sell them in Germany 

The German distributor for Minicircuits is www.municom.de. They were here already
40 years ago and won't like it to be kicked away by Mouser.
You can get an interesting subset at funkamateur.de who is a NOS reseller, not a distributor.

Cheers, Gerhard

I took a leap of faith meanwhile and ordered a few ADE-1 from China. Not sure what they actually sent, they are marked "ADE-1-24" and from the looks they are actually mixers

I've connected them to my AWG and feeding the LO and RF inputs with 10MHz and 10.000,010 MHz indeed gives the expected spectrum on the IF port. So they should be usable at least for an initial test.

[notfaded1]

AD8612 would be suitable, too, I guess. But I'm more concerned about the mixers. The MiniCircuits stuff is really nice, but I need to find a distributor that stocks them. Digikey doesn't have them at all and Mouser refuses to sell them in Germany 

The German distributor for Minicircuits is www.municom.de. They were here already
40 years ago and won't like it to be kicked away by Mouser.
You can get an interesting subset at funkamateur.de who is a NOS reseller, not a distributor.

Cheers, Gerhard

Probably the safest thing would be to just order them from Mini-Circuits.  They can ship to you from US or the UK.  For you probably UK would make the most sense.
https://www.minicircuits.com/WebStore/dashboard.html?model=SBL-1-1%2B

[Gerhard_dk4xp]

I have bought both at funkamateur.de and Municom (just recently 10 GHz stuff
for a commercial project). No problem whatsoever and no ado with customs
(as in 2 months).

Gerhard

[chuckb]

I just learned Mini-circuits will sell insulated spacers for SBL-1.  I am requesting price quote and will update this post once I get it.

P/N B14-045-01

I got a quote.  40 cents each.

Were you able to get those spacers?
Otherwise Mouser has spacers for a single pin - 144-R902-1 that work fine, 2 per mixer,  $0.21 ea.

My board is going together now.

[tkamiya]

Yes, I received them yesterday.  It ended up being a rather expensive purchase.  Shipping was $10!

[cncjerry]

If it's interesting for someone: I made some progress with the layout of my quad mixer. In the picture attached is one half. They are almost perfectly symmetrical, I even tried to match the trace lengths between the channels.
I never did a PCB layout with ECL gates before, and I have been warned it will be "extremely difficult", but I don't know to what degree this is true. Therefore I took care to minimise every track's length which has to do with the ECL gates.

The absolute delay between the channels is not all that important, as long as it is constant.

I just proved this to myself after working for several days to try to figure out why I have a phase shift between the two identical channels of a DMTD.  I averaged the output and subtracted that from the the result channel and plotted it. ADEV was Identical, of course.  I still cant figure out why I have a phase shift before the ZCD in Riley's circuit...  something to do with grounding or lack thereof.

I would shy away from a complex filter or a complex  ZCD as Riley jn subsequent emails had made it clear to me that noise reduction is where you should spend your time.  One thing you might want to try is Rick Campbell's diplexor for mixer termination, used in DC receivers.  But I think you would have to have a 50ohm path from the input through the mixer. The DMTD channels are just DC downconverters.

Also, I suggest you use one reference no matter how many channels with a passive splitter combiner.

Lastly, the old PICTIC works great with this system and can be built cheaply from Riley's paper if you don't have a TIC.  Riley used one for his excellent results.

Jerry

[5065AGuru]

We are working on an INEXPENSIVE and SIMPLE companion counter for the simple dual mixer so I'd not spend any $$ now for a PITIC unless you want it for other uses.
Hope to be able to share in the near future as it's looking really nice!

Cheers,

Corby

[thinkfat]

I have a TAPR TICC, I'm going to use that. I think I can do a three-cornered hat between the two channels and the reference input of the TICC, no?

[notfaded1]

I have a TAPR TICC, I'm going to use that. I think I can do a three-cornered hat between the two channels and the reference input of the TICC, no?

Wouldn't you need two TIC's to do a 3 cornered hat measurement?  I think they have paper on linking multiple TAPR TICC's together for doing this kind of thing?  If you want to do a hack and use the B-A channel to create a fake channel C you could kinda fake a 3 cornered hat measurement but it's not the same as having three separate real time stamped channels:

https://lists.febo.com/pipermail/time-nuts_lists.febo.com/2016-November/084591.html

I found the application note about Mutiple TAPR TICC's and attached it.

Bill

[Mrt12]

Dear colleagues
so I finished my Quad Mixer design or dual DMTD. See the attached PCB layout. It is all SMD. Also I decided to use two cascaded OpAmp stages. For first tests, the 2nd one can be omitted. A LM311 is used as comparator.
I will make a very simple time-stamping counter on a separate PCB.
What do you think of my layout?

N10 in the middle is a +/-15V voltage regulator and besides the SMA connectors the only THT component.

[tkamiya]

I'm wondering....  since the layout is so dense, perhaps some kind of guarding, shielding, or maybe just space would be a good idea? 

Also use of DC/DC converter....  I presume 15V is chosen to drive the opamp.  Wouldn't typical 100mV P-P noise/ripple affect detecting the true zero-cross point?   -  never mind!  I see filtering.

[cncjerry]

We are working on an INEXPENSIVE and SIMPLE companion counter for the simple dual mixer so I'd not spend any $$ now for a PITIC unless you want it for other uses.
Hope to be able to share in the near future as it's looking really nice!

Cheers,

Corby

Based on the TDC7200?  I've ben considering building another TICC dead-bug style as there are only 10 chips on that shield and a couple are level converters.

[notfaded1]

We are working on an INEXPENSIVE and SIMPLE companion counter for the simple dual mixer so I'd not spend any $$ now for a PITIC unless you want it for other uses.
Hope to be able to share in the near future as it's looking really nice!

Cheers,

Corby

Based on the TDC7200?  I've ben considering building another TICC dead-bug style as there are only 10 chips on that shield and a couple are level converters.

I'd wait to see what Corby and Tom come up with.  From what I've been hearing it's going to be pretty legit.  I can't think of anyone here that would know better either.  They both have some of the best references and measurement gear I know.   

Bill

[ch_scr]

Here is my dogmeat "DMTD". Turns out the LM160 (or at least my Tesla MAC160) definitely have a temperature dependant delay.
My time interval reading is drifting around slowly, and I can make it drift (more) if I put my finger on the case of one MAC160.
If I take the finger away, it drifts back. Guess that's whats meant by "temperature compensated" in the ECL buffer datasheets?
Edit: Found an app note that has a relevant graph. Pretty amazing how sensitive this is.

[Mrt12]

I'm wondering....  since the layout is so dense, perhaps some kind of guarding, shielding, or maybe just space would be a good idea? 

Also use of DC/DC converter....  I presume 15V is chosen to drive the opamp.  Wouldn't typical 100mV P-P noise/ripple affect detecting the true zero-cross point?   -  never mind!  I see filtering.

The DC/DC converter operates at around 500kHz, so its frequency is anyways way beyond the bandwidth of the amplifiers. But you are right, I inserted some DC filters.

Concerning the shielding of components - you are right as well, this is the only part where I am a bit worried. On the other hand, the ECL gates have differential outputs, so I don't see that they should cause EMI problems.
By the way I plan to make this as a 4layer board which means I do have a solid ground and 5V plane.

We are working on an INEXPENSIVE and SIMPLE companion counter for the simple dual mixer so I'd not spend any $$ now for a PITIC unless you want it for other uses.
Hope to be able to share in the near future as it's looking really nice!

Cheers,

Corby

Based on the TDC7200?  I've ben considering building another TICC dead-bug style as there are only 10 chips on that shield and a couple are level converters.

The TDC7200 is very good. However I think it is not necessary for a DMTD. 10ns or even 100ns resolution should be good enough; realistically, with the TDC7200, you get 1ns. Perhaps 500ps. I am unsure whether this improvement is so dramatic that it justifies the additional costs and complexity, does it?

[thinkfat]

Dear colleagues
so I finished my Quad Mixer design or dual DMTD. See the attached PCB layout. It is all SMD. Also I decided to use two cascaded OpAmp stages. For first tests, the 2nd one can be omitted. A LM311 is used as comparator.
I will make a very simple time-stamping counter on a separate PCB.
What do you think of my layout?

N10 in the middle is a +/-15V voltage regulator and besides the SMA connectors the only THT component.

If I'm not mistaken, the MC100EL16 is 8V abs max between Vcc and Vee. You'll release magic smoke if you supply it from the +15V rail.

[Mrt12]

Dear colleagues
so I finished my Quad Mixer design or dual DMTD. See the attached PCB layout. It is all SMD. Also I decided to use two cascaded OpAmp stages. For first tests, the 2nd one can be omitted. A LM311 is used as comparator.
I will make a very simple time-stamping counter on a separate PCB.
What do you think of my layout?

N10 in the middle is a +/-15V voltage regulator and besides the SMA connectors the only THT component.

If I'm not mistaken, the MC100EL16 is 8V abs max between Vcc and Vee. You'll release magic smoke if you supply it from the +15V rail.

don't worry, there is also a 5V input ;-)

[cncjerry]

You might find this helpful if using TimeLab.

There is a section titled 'Masks" and I use it to evaluate my boards vs known baselines.  This isn't meant to evaluate one board vs another because the method used to acquire the data might be different for each boards.  Also, these work with 'ADEV' plots with the prefix on the line of 'a'.  If you have other measurement  baseline data like 'p' or 'm' you can add that as well (with the appropriate line prefix character).  I think you will find it interesting to look at Corby's Mask vs some of the other much more expensive options just by turning them on, up to three at once.

I added Corby's mask from the data he published on the first page, I think it was.  If it has been updated, then the pairs below of (tau, value) would need to be corrected.

I also added Riley's baseline.

Add these statements to your "masks.txt" file by going to Masks, edit mask definitions.  The separator commands do just that; organize these within the drop-down list with a separator line.

Code: [Select]

separator

mask Riley DMTD baseline
a (0.1, 5.24e-13) (0.2, 2.73e-13) (0.4, 1.34e-13) (1.0, 5.43e-14) (2.0, 2.74e-14) (4.0, 1.4e-14) (10.0, 5.71e-15) (20, 2.93e-15) (40, 1.44e-15) (100, 5.91e-16) (200, 3.03e-16) (400, 1.57e-16)

mask Corby DMTD baseline
a (1, 2.46e-13) (2, 1.34e-13) (4, 7.97e-14) (8, 4.51e-14) (16, 2.15e-14) (32, 1.07e-14) (64, 5.66e-15) (128, 3.1e-15) (256, 1.62e-15) (512, 7.26e-16) (1024, 3.73e-16) (2048, 3.63e-16)

separator


I put them right after this line:

Code: [Select]


mask Kvarz CH1-76A passive maser (5 MHz typical)
a (1, 4.8E-13) (10, 1.5E-13) (100, 4.5E-14) (1000, 1.5E-14) (3600, 8.5E-15) (86400, 4E-15)
P (10, -110) (100, -125) (1000, -150) (10000, -150)


Jerry

[cncjerry]

The tdc7200 chips and associated circuitry, if operated on 3.3V, cost almost nothing and on 5v you had to add level converters that add complexity in board design and soldering.  A guy I am working with replaced the TIC on the front-end of Lars GFPSDO with one and it was a relatively simple change.  The TAPR TICC, that a lot of people have including me, uses two, one for each channel with 11 digits of resolution, no carefully selected components or special caps, etc.  Only issue is that you must have logic to account the overflows over what, 800ns I think it is, in the specific mode.

From what I found, the biggest problem with the DMTD system is 60/120/180hz ground loop problems and noise.  For instance, I ran a baseline and it was so close to Riley's baseline you can't really see the difference on the plot. I ran this over and over without an issue.  Then I made some changes as I readied to test a GPSDO and now I can't get back there.  I can watch it snap back and forth so clearly either something is failing on the board or a grounding problem has shown-up that eludes me.  So yes, if it took that much longer to implement a tdc7200 TIC vs improving the grounding, coupling, etc., then you would be better off doing that, I guess.

I have to build mine into a box.  The intermodulation of local FM stations can be seen in subharmonics in my case, all over it.  Riley mentions about aliasing in his notes, so yes, you get a lot of accuracy for the money, almost can't be beat, check the masks I created vs much, much more expensive products, but I don't think you can get there just throwing it together on a couple of boards without shielding and careful attention to grounding.  I'm not suggesting anyone would try except for me!

I've worked on DMTD, loose and tight PLL, etc  so long I forgot why I even started.  I think I wanted to update the oscillators in my Lucent GPSDOs and possibly the one in my Cs beam and then couple one to clean-up an Rb unit.

Jerry

[tkamiya]

Corby,

Would you mind giving a newbie a giant favor?  I am done with populating the board except for SMD parts.  Power supply is already done and tested.  1Hz offset signal is ready.  By tomorrow, I will be able to conduct initial functional test.  What kind of wave form should I be seeing on output?  I would imagine nearly 1Hz pulse or square wave, but an actual oscilloscope screenshot will be very helpful.

Like you, I ended up soldering one of DBM backwards.  In process of pulling it out, I damaged the through-hole.  I used the thin wire trick to repair it but I want to make sure it is actually functioning.  The colored pin at bottom is supposed to line up with square copper foil, correct/ 

Just to make sure....on LM311, both in 5 and 6 has connected pads.  Yet, there is nothing on schematic.  What are these for?  Datasheet shows they are for zero adjust.

I'm impressed how you crammed parts in this little board!

[thinkfat]

Corby,

Would you mind giving a newbie a giant favor?  I am done with populating the board except for SMD parts.  Power supply is already done and tested.  1Hz offset signal is ready.  By tomorrow, I will be able to conduct initial functional test.  What kind of wave form should I be seeing on output?  I would imagine nearly 1Hz pulse or square wave, but an actual oscilloscope screenshot will be very helpful.

You should see a 1Hz square wave with 50% duty cycle on the outputs.

[tkamiya]

Ladies and gentlemen! 

My Corby2020 system is doing SOMETHING!  I am getting square waves at both channels.  I will need to properly wire this before going any further but functionality is definitely there.  Yay 

[5065AGuru]

Taka,

Glad you got it built!

Yes the colored pin goes towards the foil square.

On all mine that meant that the "writing" between the tops of the mixers were pointed opposite.

If you use a T to connect both inputs together you should see the 1Hz outputs in phase.
(Both should be high or low at the same time.)

Those pads were in case I used a different comparator. Leave unconnected for the LM311.

Look over the tutorial post for some info on making measurements.

I'd just hook up two oscillators, adjust one of them to make the counts increment slowly and log some data.

With PLOTTER you then open the file just made.

With TimeLab you can do that also but I think there is a mode that takes data in real-time and you can watch the AD develop directly.

Cheers,

Corby

[tkamiya]

Corby,

I also noticed there is an extra output for each of the ports.  No connectors.  Just pads on PCB.  I'm assuming they are there just in case someone wanted an LED or hook up a scope.  Is that right?

[notfaded1]

Ladies and gentlemen! 

My Corby2020 system is doing SOMETHING!  I am getting square waves at both channels.  I will need to properly wire this before going any further but functionality is definitely there.  Yay 

Congrats Taka!!!  That looks like a nice setup.  What case did you put that in?  It looks a little like the one Corby made first.  I'm trying to figure out the long term power supplies for this still.  We need all these different DC voltages and I'm wondering what's really the best way to go about it?  Is there an elegant way to do it with linear PS somehow vs. multiple linear PS's?  Or even some switching I suppose but I've always been kinda adverse to them despite the size and such when dealing with sub ppm measurement applications???

I picked up some 1050A's last night.  They all have the 015 option for disciplining the OCXO.  I'm hoping they'll go hand in hand with the DMTD.  I'm not sure how it work when using the PLL to sync the quartz?  Do they keep the sync after you disconnect the ext. input?  Seems for great S, M, L term stability using a Cesium to sync to the quartz then running off the quartz is the way to go.  Also I'd be interested to see if I can improve on the 5065 output by trying the same thing.  The 5065 has a really good LO but perhaps a well aged 1050A might give it an edge?  Anyhow there's only one way to find out right and I can't think of a better way to check than with this DMTD from Corby.  I don't have the Efratom DMTD documentation but judging from what Corby said it sounds like this design is very similar to it.

Bill

[tkamiya]

Here's my power supply.  After this, there are four 3 terminal regulators.  Transformer puts out 8V per winding.  I just used what I already had....

Case is a shallow 2U aluminum case I bought on eBay.  I can't find a link for it now.  6 panels bolts together.

[Mrt12]

Dear colleagues
today I received my ECL gates: a couple MC100EL16 and MC100EL11. Found them for a good price on eBay.
I immediately wanted to test the gates and see how they perform when fed with a sine wave. So I soldered a couple of them in an ugly dead-bug style to a piece of PCB, attached a Thevenin termination network and a couple SMA connectors.
I measured the output waveforms (Q and Qbar) with the scope with different input frequencies and amplitudes. For the high frequencies, the oscilloscope pictures are not so accurate as my scope's bandwidth is only 1.5GHz. However we can see that the output amplitude stays nicely constant over a wide frequency range.
Further, I also varied the input power. From about -20 to +7dBm, the output waveforms look the same. If the input power is increased above +8dBm, the output waveforms become more and more distorted, and below -20dBm the output amplitude stays more or less constant but the waveform becomes more sine-like.
I also measured the reverse isolation using an HP 8753C network analyzer. Up to 1GHz, the isolation with my crappy soldered test circuit is, according to my measurements, approx. 80dB. I am not sure whether this is acceptable, but I assume it is as most oscillators which we are going to test already have some sort of output amplifier included. For example, my GPSDO has already an OpAmp based buffer amplifier built in, which has a reverse isolation of around 50dB, so the total reverse isolation when the GPSDO is connected to an MC100EL16 is around 130dB, which should be good enough for most measurements.

The output power of the MC100EL16 is around 1dBm (circa 700mV). This is a bit less than specified in the datasheet, but acceptable I'd say for the ADE-1+ mixer, isn't it?

I am thinking about whether I sould use a MC100EL11 for the offset oscillator, as the '11 has two outputs. However I am not sure how well they are isolated from each other and I have only 6 of those ICs, so I don't want to waste too many of them...

But I think from these measurements we can say that the MC100EL16 is an acceptable alternative to the MC1650 and is still available these days. Even at mouser for a decent price.
So I think I am going to finish my PCB and we'll see how it works :-)

[thinkfat]

Can you add a small resistor into the +5V rail and measure the current draw? I've read somewhere that ECL gates are somewhat "quiet" regarding their current draw, unlike high-speed CMOS gates.

[Mrt12]

Didn't need to do that, as my lab power supply does it for me ;-) Including the termination resistors, the current consumption is 65mA. Theory says that ECL current consumption doesn't change with frequency, and this is EXACTLY what I measured here:

DC: 65mA
1MHz: 65mA
10MHz: 65mA
100MHz: 65mA
1GHz: 65mA

So it is indeed quite power hungry, but considering the fact that this is indeed a very simple limiting amplifier with a decent input dynamic range, I would say this is not bad, is it?

What do you think by the way about the output level? 700mV, should be enough to drive the mixers. I assume the MC1650 are not much better. Are they?

[thinkfat]

Didn't need to do that, as my lab power supply does it for me ;-) Including the termination resistors, the current consumption is 65mA. Theory says that ECL current consumption doesn't change with frequency, and this is EXACTLY what I measured here:

DC: 65mA
1MHz: 65mA
10MHz: 65mA
100MHz: 65mA
1GHz: 65mA

So it is indeed quite power hungry, but considering the fact that this is indeed a very simple limiting amplifier with a decent input dynamic range, I would say this is not bad, is it?

What do you think by the way about the output level? 700mV, should be enough to drive the mixers. I assume the MC1650 are not much better. Are they?

I'm more interested in the transients. Did you measure the consumption with the mixer connected?

I think 700mV is enough for a Schottky diode mixer.

[Mrt12]

Ah yes I see. I can try that tomorrow but I doubt my scooe is fast enough to see them.
I had no mixer connected but an AC coupled 50Ohm load.

[tkamiya]

Everything is in aluminum case and seems to be functioning.  One of the 3 terminal regulator is getting awfully hot, so I'll have to look at that later.  I believe it's -5V.  All others are somewhat warm to touch.

I've heard there is a convention folks follow in terms of what goes into Ch A and Ch B.  Anyone know which one is DUT and which one is reference?

If anyone is using TimeLab and TAPR TICC, I'd appreciate sharing of setup parameters.

[5065AGuru]

Taka,

I use the extra outputs to drive a dual one shot that drives an LED for each channel.

As far as which input to use for DUT and REF...

When I run plots on Quartz I like to set the counts at a low value that is slowly going up.

I want the aging to cause the counts to slow and then start back down.

On a good  oscillator you end up with a nice parabola which you can see in plot1 in the DMTD tutorial post.

This way I can stay in the lower end for a longer time.

I pick the DUT/REF connection that causes this to occur.

Cheers,

Corby

[thinkfat]

Ah yes I see. I can try that tomorrow but I doubt my scooe is fast enough to see them.
I had no mixer connected but an AC coupled 50Ohm load.

The scope will do fine I think. You should at least see "something", qualitatively at least.
I'm most interested if you can really drive the mixer, especially the LO capacitively.

[Mrt12]

I'm most interested if you can really drive the mixer, especially the LO capacitively.

Just tested that. Setup was as follows:

HP 33120A function generator, square wave, 600mVpp, 5 MHz into ADE-1+
HP 8663A signal generator, 5MHz, 0dBm, into the ECL gate
output of the ECL gate, capacitive coupling, into ADE-1+
IF output of ADE-1+ to the scope via lowpass filter as shown in my schematic.

Result on the scope: a nice sine-like signal with about 800mVpp amplitude. I can reverse the mixer's RF and LO ports and the result stays the same.
I can vary the signal generator's amplitude between -12dB, and approx. 5dBm withouth any noticeable change on the scope display.
Of course I cannot change the function generator amplitude as much because this one is directly connected to the mixer.

Does that sound acceptable? I can make a scope picture later.

[Mrt12]

Hello again

so as I promised previously, I performed some additional tests with an ADE-1+ mixer. I use the configuration as shown in the schematic attached.
I used the following test equipment:
HP 8341A signal generator with 0dBm output power, connected to LO port of ADE-1
HP 8663A signal generator with 0dBm output power, connected to the ECL gate
the ECL gate is capacitively coupled to the ADE-1
the IF output is measured with the scope.
The signal generators are synchronized together with an external 10MHz reference, such that I can adjust them to exact offset frequencies.

I measured the IF output signal up to 500MHz, the max. specification of the ADE-1. As we can see, at 500MHz, it does have a significant DC offset; however, below 500MHz, the IF signal looks quite acceptable. However I was wrong previously, it is not 800mVpp but merely around maybe 600mVpp. But I guess this should still be acceptable.

We also need to take into account that I currently drive only ONE of the mixer ports with the ECL gate. But I assume the mixer's properties will not change much if both ports are driven like so, as the ECL gates anyway are 50 Ohm sources.



Edit: I needed to attach the 500MHz measurement as JPG. The forum software rejected the PNG picture, for whatever reason :-O

[tkamiya]

A couple of things I'm concerned about....

It is typically known that diode ring DBM needs high IF input, typically in tune of 7dB.  I wonder what effect it will have when it falls below this level?

On my own setup, I didn't use long neck SMA but used a regular one, and connected to front panel mount N/SMA via short piece of coax.  Initially, I was going to use semi-rigid line but I don't have the size I need.  Then I was wondering if this is really necessary.  From there, there will be a pair of longish flexible cable to DUT and REF anyway.  What do you folks think?

Here's front panel of completed setup.  Little silver box is just a boxed up OCXO.  I can adjust +/-4Hz off 10MHz nominal.  I see many examples where DDS was used for Lo.  I'm still not clear what kind of stability is required.  But Bill Riley did it that way on his implementation.  Otherwise, I'll just setup a few HP11801 already set to proper offset.  I'm using TAPR TICC because I already have it handy.

I put LED on front panel driven by extra gate on output side.  If I have to do this over again, I'd probably steal another gate from parallel driver and have Aux output.  During testing, having scope connected to it was "instrumental."

[5065AGuru]

Taka,

Saw your mention of the LED!

Realized since the simple dual mixer monitor outputs are square wave you can probably just drive two LEDs through a resistor directly!
My original NBS106B dual mixer has very narrow pulse outputs so that's why I mentioned the  74XX221 circuit to drive the LEDs.
I'm pretty sure my other ones have square waves out so no need for the one shots, just a buffer.
Silly me!

Your unit looks nice.

Cheers,

Corby

[tkamiya]

Corby,

I don't think I would do that - that is, tap off of main output via register and drive an LED.  I am not sure what kind of load LED would present...  (I'm using 680 ohm series register)  I'm even hesitant to Tee out the output into scope and counter.  Reflection and stray capacitance might have some impact.  I was going to try this today and I got side tracked.  I'd prefer keeping output as straight 50 ohm as I can.

You mentioned NBS106B few times.  I know the design but you have a real unit produced by NIST/NBS? 

[Mrt12]

Hey Taka,

"It is typically known that diode ring DBM needs high IF input, typically in tune of 7dB.  I wonder what effect it will have when it falls below this level?"

well, Minicircuits has an excellent application note on this. They say that it is no problem to operate a mixer with a couple dB lower LO power. As you can see from my tests, the ADE-1+ works well with 0dBm to 1dBm input power. But I agree to your point and maybe an ADE-1L+ would be even better. This is even specified in the datasheet for 0dBm LO. But even for the ADE-1+ I don't see much problems, as we saw from my tests.

[thinkfat]

Driving a mixer too low is mostly impacting the IP3 and linearity, no? That's not a problem for this application. Besides, the ECL gates driving the mixer are really a good idea. That provides a lot of independence.

[Mrt12]

Driving a mixer too low is mostly impacting the IP3 and linearity, no? That's not a problem for this application. Besides, the ECL gates driving the mixer are really a good idea. That provides a lot of independence.

Yes, I totally agree. The ECL gate input buffer is indeed one smart idea from Corby! It allows for a wider input dynamic range and also suppresses amplitude fluctuations (more or less).
The only unknown is whether the ECL gates are noisy (i.e. increase the phase noise) in a manner that influences the measurements. But from the instrument noise floor Corby posted on page 1, it seems quite good.

[notfaded1]

I think this design is very similar to how Efratom did it in their commercial DMTD product so yeah... it's legit. 

[5065AGuru]

Taka,

The monitor outputs are separate from the main outputs!

So you can load them to your hearts content with an LED/resistor or otherwise!

Cheers,

Corby

[Mrt12]

Dear colleagues
during the past days I finished my dual DMTD and the corresponding PCB.
I just optimised the BOM and ordered the PCBs, hopefully they will arrive next week.
The schematic is attached. I am very curious how well it will work and what noise floor I will get!
To get started and save some costs on the OpAmps, I decided to populate only the first OpAmp stage with the LT1007 and skip the 2nd OpAmp stage with the OP37.
Besides that, I will use the HP 5372A Timing Analyzer I have, this would allow to measure 2 oscillators against each other.
Later on, I plan to make a simple time-tag counter with 4 channels. This would then allow to measure 2, 3 or even 4 oscillators against each other!

I also prepared some maths for a "four cornered hat". If four oscillators are heterodyned with one common offset oscillator, I think a similar approach as for the three-cornered hat would be possible and therefore would allow to determine the absolute oscillator stability.
This is because we get 6 different measurements when we time-tag the four outputs:

1 vs 2
1 vs 3
1 vs 4
2 vs 3
2 vs 4
3 vs 4

so we have 6 possible measurements and 4 unknowns (the stabilities of the four oscillators), so we can do the math and solve this overdetermined equation system. The same as for the three-cornered hat! it will be very interesting whether this would work.

Also the inter-channel isolation will be an interesting topic as I tried to make a compact layout. However I decided to use a 4layer PCB with proper power and ground planes, so the EMC should be acceptable - as we already saw that the reverse isolation of the ECL gates is around 60 to 80 dB, even with my crappy manhattan-style test circuit I showed a couple weeks ago.

If there is interest I will keep posting updates on this as well!

[5065AGuru]

Hi,

Here is a comparison test between two BVA oscillators that Tom Van Baak made simultaneously using the Simple Dual Mixer against a TimePod.

The first 2 plots shows ADEV and MDEV as measured by both systems and the second plot shows the frequency "wander" between the BVA oscillators using 10 Second averaging.

Very nice data showing how good the Simple Dual Mixer can perform!

Cheers,

Corby

[Mrt12]

Dear colleagues
today I received my PCBs from Eurocircuits. I immediately decided to populate the input stages with the ECL gates and the mixers, so I could do first simple tests!
Further, I also changed the mixer and decided to use the ADE-2+ instead of the ADE-1+. This will allow only a lower frequency of 5 MHz, but at least 1 GHz for the upper frequency.
Then I tested the beat note output from the mixers after the lowpass filter for various input frequencies. Again I set my signal generators with 10 Hz offset to each other and tested different input frequencies. The output beat note stays quite nice and constant up to 2 GHz input frequency. I didn't test further, as this is definitely more than enough for me! Beat note amplitude around 500mVpp.

Further I tested the isolation between the DUT inputs and also between the DUT input and the offset oscillator input.

The further is very good, around 110 dB (measured with my HP 3577A network analyzer, unfortunately only up to 200 MHz). The isolation between the DUT inputs and the offset oscillator input is a bit worse, around 95 dB. But I guess this is stil acceptable, is it.

The beat note output does not vary with input amplitude as long as it is in the range of -25 dBm to +12dBm. I did not test higher amplitudes. If the input amplitude is lower than -25 dBm, the output becomes more and more sinusoidal (it is normally some sort of triangle wave).

The return loss for the DUT inputs is around -30 dB, whereas the input return loss for the offset oscillator is a bit worse, around -20 dB. I guess this is because I neglected "proper" RF design criteria for the offset oscillator's traces and just used a simple Y connection for them. Still, I consider -20dB good enough for the offset osc.

Unfortunately I don't have suitable scope probes to measure the output signals from the ECL gates - but as the mixer output seems to be quite stable, I assume the ECL gates work as they should.

As I don't have the OpAmps ready for the gain stage, I cannot test further yet, but I think the results so far look promising, do they! Let's see what the noise floor will be. I am very curious.

By the way, the ECL gates are indeed power hungry, this little part with only 4 gates already consumes 270mA! :-D

[tkamiya]

Very nice!

I'm thinking of going WILD and do 8 channel.  A have a questions for you.....
What are you using for off-set oscillators?  Seems you have something frequency agile?
Over the course of thread, some discussed transformers at input.  It appears you opted to go without.  What was your thought process?

[Mrt12]

Hi Taka,
indeed 8 channel would be good as well. We can proof that if there are three or more channels, it is possible using some linear algebra that the absolute allan variances of the individual oscillators can be calculated, so it can indeed make sense to use 8 channels if you have so many oscillators :-)

Concerning the transformers. Indeed I didn't use them. I cannot see any advantage of those. Neither my scope, spectrum analyzer, network analyzer, counter, signal generator or impedance analyzer does have transformers at its input but uses N or BNC connectors connected to ground. So I don't see why exactly here in THIS very application we should have a problem with ground loops?! but, if necessary, it will be easily possible to connect a transformer externally.

However they also have a lot of disadvantages, for example many transformers are not as wideband as my input stage is currently without transformers.

[Mrt12]

Hi all
today I wanted to continue testing of my DMTD, but I am still waiting for the OpAmps I ordered.
However, I found that I do have a bunch of OPA27 in my parts box, and at first glance, it seems that they are an acceptable alternative to the LT1007 I planned originally to use. So instead I populated the OP27. I don't have the LM311 comparators yet, but I will still be able to measure the output signals from the OpAmps!

That's what I did now. I used a HP 8663A signal generator for the offset oscillator, and a HP 8341A signal generator for the DUT input. The other DUT input was terminated with 50 ohms.

I then measured the mixer output signal, as well as the output signal of the first OpAmp gain stage and the 2nd gain stage for different offset frequencies.I set the offset signal generator 10 Hz or 1 Hz low compared to the DUT oscillator.

For the measurements, forget about the Y scale - I used an 1:10 probe but didn't tell the scope, so voltages are 10 times lower.

We can see from the graphs that the slope at the 2nd OpAmp stage is around 2.5 V/ms for 1 Hz beat note, and around 10 V/ms for 10 Hz beat note. When we compare this to Bill Riley's famous "Small DMTD system", it seems that these values are in a similar order of magnitude, and, further, the waveforms look  also quite similar. The slope of my OpAmps is perhaps even a bit steeper.

As soon as I receive the LM311 comparators, I have everything ready so I can generate a proper 5V TTL output which I then can connect to my HP 5335A counter or 5372A timing analyzer.

Again I also tested the output signals of the OpAmps when the input frequency and power is varied. The current design seems to work fine up to 2.5 GHz, I didn't test further. I think so far it looks good ;-) a noise floor test cannot be done yet as I don't have a proper output signal I can connect to my counter.

[tkamiya]

I am working on THE Corby design.  It seems base line graphs are one order of magnitude worse than expected, so I've started poking around.    Input is fed by a common source.  Difference input is 1Hz offset.  Looking at numbers coming out of TIC on a terminal session, I can clearly see fairly repetitive and predictable wobble between channels.  Scoping the output, yup, I see relative shifts of the output, occasionally as much as 100 micro seconds.  Counters are confirmed "not in the equation."   

Looking at power bus, it's incredibly dirty.  Not a regular ripple, but more of white noise at 50mV or so.  I am finding this very odd, since I am using a linear supply I've made just for this setup.  My plan is to try running it with battery and see what happens.  Source of this white noise is unknown at this point.     

Just reporting in....

[tkamiya]

I kind of have my setup stabilized now.  Can someone who is clear headed help me with this?

DUT and reference is 10MHz
Difference oscillator is 1Hz offset
Using TIC, I see 0.nnn,nnx where x is changing about 4 counts.

What does it indicate on 1s Adev? 

[Mrt12]

Taka,
have you found out why your Power Bus was so dirty? was it the power supply? or switching transients from the logic gates?
As for your other question I'd say: you see a change of around 4 microseconds, i.e. 4e-6. Your DMTD mixes 10 MHz down to 1 Hz, this is factor 1e-7, so I'd say your stability is around 4e-13?
Not 100% sure as you would normally take a timeseries of readings and then calculate the AVAR from that.

[tkamiya]

Yes, I did.

A 3 terminal regulator on -5V rail was wired incorrectly.  The power supply was working well when I tested individually.  I must have messed it up when I noticed it was getting quite hot that I removed it from PCB and mounted it on bottom of the case.  Once I corrected the wiring, it settled down to where it should be.  I knew negative regulators are oddball when it comes to pin out, so I was being careful, I thought....

[Mrt12]

I did some first tests with my DMTD board. I connected 10 MHz to one input and 9.99999 MHz to the other input. At the output, I got a nice 5V TTL signal with 10 Hz. I measured the frequency with my HP 5372 timing analyzer. I saw that this frequency was jumping up and down by around 0.0001 Hz. However, I had locked my two signal generators to the same external 10 MHz reference... so I am unsure why I do have this instability. Can it mean that my signal generators are indeed that bad or does it perhaps mean that my circuit is unstable?

If I am correct, this 100 uHz instability translates to 1e-12 instability at 10 MHz at tau=0.1 sec. Is that right?

Further, another interesting question: If I want to compare two oscillators having the same nominal frequency, say 10 MHz. Then I use an offset oscillator which is, say, 10 Hz higher.
However the offset oscillator is maybe not 100% accurately 10 Hz high, so the beat note at the output will perhaps be something like 9.999 Hz or 10.001 Hz. So my tau is not actually 0.1 second, but merely something like 0.10001s or 0.09999s. Does that matter when I want to make an ADEV plot? or can I neglect this tiny error and say this is exactly 0.1s?

[notfaded1]

Yes, I did.

A 3 terminal regulator on -5V rail was wired incorrectly.  The power supply was working well when I tested individually.  I must have messed it up when I noticed it was getting quite hot that I removed it from PCB and mounted it on bottom of the case.  Once I corrected the wiring, it settled down to where it should be.  I knew negative regulators are oddball when it comes to pin out, so I was being careful, I thought....

Taka-

This is exactly where I'm at now trying to figure out the power supply and voltage regulators.  I ordered and received some of long SMA jacks and also got some SMA to Female BNC headers so I could use BNC on the output if I want or stick to SMA.  The  power supplies I got are POWER ONE HTAA-16W-AG linear power supplies.  I got two for a good deal.  It seems the negative voltage regulators can have a real need for a heatsink from what everyone has told me?  Are the negative really any different from the positive if either would be 12V to 5V with LM340T-5.0 vs. LM320T-5.0?  As, you said Taka I also noticed the pinouts are different between the LM340 and LM320.  Also I've noticed people usually put an electrolytic capacitor across the input and output and sometimes also a small ceramic as well on the output around the linear voltage regulator.  What size caps would you suggest I use with 12V to 5V on both sides?

Bill

[tkamiya]

Hi, Bill!

First, the connectors and internal wiring.  I avoided using flexible coax.  On front panel, there are N to sma panel mount adapter.  From there, semi-rigid connector goes to the PCB.  I've noticed there is a lot of noise picked up on output cable.  I suspect ground loop of some kind.  I'm in process of insulating output connector.

On 3 terminal regulators, I used 0.1uF on back side of PCB to reduce inductance, and on front, 10uF aluminum caps.  I've always done it that way.  On 5V negative side,the regulator actually does get hot.  I mounted on bottom of the case.  0.1uF right at the regulator, short piece of wire, and 10uF on PCB.  The reason I use both is that some brand of 3 terminal regulators are prone to oscillation at high frequency where regular aluminum caps are not that effective.

If you look at spec sheet for those regulators, they typically suggest 0.1uf and 1uf.  Somewhere else on the note, those are for tantalum and use of aluminum requires x10 the value.  I just go by my past experience and do what I said above.

My regulators are just typical 78 and 79 series.  I did not use anything special.

For filter caps, I used as large as I could find.  I recall each leg has 6800uF or something like that.  Then, I put two ferrite beads on output - although it does not appear they did anything meaningful.

[tkamiya]

This is something I do and certainly not a universal practice.

I don't trust BNC.  It is because NOT all BNCs are bad, bu there are SO MANY substandard BNC around.  I try to avoid them as much as I can.  I use N mostly.  SMB has 300 connect/disconnect limit.  Fine for internal connections.  I don't know about external ones unless you use connector savers.

[tkamiya]

Oh, by the way....

I noticed there were LOTS of noise generated SOMEWHERE in my lab.  Perhaps it's from LED lighting overhead.  Perhaps all the test equipment in the room, or even dozens of 10MHz sources in the lab.  Something to think about when you are testing yours.  Ultimately, I plan to put everything on custom made 19" aluminum rack, battery powered, and do particularly sensitive measurement outside my main lab and computer room.  USB connection will be via fiber optic.  Perhaps this is over-the-top.  But I've spent enough time looking for noise sources, I learned to not assume anything.

[Mrt12]

Hi guys

I have attached a first proper measurement I made.
I used a 5 MHz FTS1000 as external reference for my HP 8663A signal generator and set the signal generator output frequency to 4.99999 MHz. This produces a 10 Hz beat note at my DMTD output. I measured this frequency with a HP 5335A counter and let TimeLab read the data from the counter via GPIB.
The resulting plot is in the attachment.

a) is it right - I have to scale the whole thing by 10Hz/5MHz = 2e-6? so my stability at 1sec would be around 2e-12?
b) doesn't look terribly good, does it? I guess the HP 8663A is adding a bit of noise, but THAT MUCH?

I am still waiting for some parts so I could not yet make a proper DMTD measurement - still only one channel is fully populated with all ICs :-/

[tkamiya]

You are getting the same problem I did.  That's not right at all.  You should be getting -12 at least.

Instead of recording and graphing at the same time, record it in terminal software first, then feed it to TimeLab.  When I did that, factors were interpreted correctly.  I don't know why that made a difference but it did.

[Mrt12]

Hi Taka
Indeed, TimeLab seems to have a problem with the scaling factor!
For some reason it is correct now. Instead of Frequency measurement, I did period measurement. And for this, the scaling factor works fine!

Interesting: I did a test where I compared a FTS1000 against HP 8663A, and a SLCR-101 (Rb) against HP 8663A. The Rb is more than 10 times worse, but the numbers seem to be in the right order of magnitude: http://www.ke5fx.com/rb.htm

I also tested a HP 33120A function generator and the curve looks sort of good to me - kind of what I expect.
However the numbers for the FTS1000 are much TOO good to believe!

Anyways, I think I am on the right track now! as soon as I have all remaining missing ICs ready I can go for the proper DMTD or dual DMTD and even do cross correlation.

[tkamiya]

Oh, wait!  Did you say you have only one channel completed?  That's why.  You don't have the start>stop time interval, yet. 

I'm doing A-B TI (A->B) (time interval).  Yet, if I set it that way, and do simultaneous measurement, Adev starts from E-7 and goes down from there.  (set up as multiplication factor E-7) I also noticed odd waving motion and abrupt start.  All of that is gone if I record it first and import the data.  I'm running 24 hour test, so I cannot interrupt it.  But I'll have to figure this out.  I doubt TimeLab has a bug like this with so many users involved in testing and using.  I think I'm doing something wrong.  It works with TICC but from HP53132, TimeLab stopped displaying anything.  Yet, from terminal emulator, I see valid data.

I have a big plan for this setup, so I need each piece done correctly and tested. 

By the way, I'm using TAPR TICC but someone has also made a simpler and less expensive version.  I've seen it somewhere.  That may be all we need.  All these digits are mainly wasted anyway.

Have you tested by changing difference oscillator from crystal to HP8663A?  I'm still not clear on how good this diff osc needs to be and how much it affects the result.

[Mrt12]

Taka,
yes, I have only one channel at the moment as I am missing some OpAmps and comparators.
But I think the single-mixer heterodyne technique still should work - I can measure the output frequency or period of the resulting mixing product....
Anyways, I guess tomorrow I will get the missing parts. We'll see how it works ;-)

what is your big plan you have for this device?


Indeed it is not yet very clear how good the difference oscillator needs to be, but from what we can read from Bill Riley the difference oscillator should cancel out IF the zero crossings are CLOSE TOGETHER.
However if a beat note of 10 Hz is used, the zero crossings can be at max. 100ms apart, so I assume the ADEV of the difference oscillator at 100ms is a limiting factor, but I am not sure how much it affects the measurement.

[KE5FX]

You are getting the same problem I did.  That's not right at all.  You should be getting -12 at least.

Instead of recording and graphing at the same time, record it in terminal software first, then feed it to TimeLab.  When I did that, factors were interpreted correctly.  I don't know why that made a difference but it did.

Thanks for the heads-up on this, will take a look. 

[tkamiya]

John,
I was going to retest this and send you a more formal report.  But what I said was what I saw.  Thanks for looking into it.

MRT12
Yes, I was told of this technique, single mixer setup a while back.  It just won't have "cancel out" feature of double setup.  I know if one chooses too narrow of offset and relationship between f1 and f2 changes, then it will return an invalid response.  I am not sure if zero crossing point makes difference.  That happens at mixer stage.

My plan is to expand this to 4 channel minimum, hopefully 8 channel.  Set everything up in one short rack, except for PC and operate it under battery.  Connection between PC and TIC will be an optical fiber.  Earlier in testing, I was amazed how much noise a short piece of coax between counter and DMTD picks up.  I was using good quality TMR240(Times something brand?) coax, too.  Also, isolation between the case and ground side of coax appears to make some difference as well.  So by going all-in-one route, I intend to eliminate as much variable as possible.  I'm hoping I can pick up some more resolution, too.  Still too early in stages to say for sure, but looks like Corby's design has quite a bit more potential than we are seeing right now.

[Mrt12]

Taka,

Still too early in stages to say for sure, but looks like Corby's design has quite a bit more potential than we are seeing right now.

I agree BUT we also shall be a bit cautious: I don't know if you use Corby's original variant with the MC1650 comparator or my variant with the MC100EL16 gate.
For the former, the jitter of the MC1650 is not specified, but for the MC100EL16 a typical jitter of 0.7ps is specified - that's 0.7e-12, I am not sure whether this also scales with 1e-6 - in that case it would be totally fine - or whether it stays like so - in this case the noise floor would be somewhat limited. The MC100EP16 would be the better choice then (0.2ps).
I doubt that the very dated MC1650 are much better than that, so I wonder what limitations arise from this.


Can somebody maybe explain how the jitter of the ECL gate scales when the heterodyning occurs?

[thinkfat]

Taka,

Still too early in stages to say for sure, but looks like Corby's design has quite a bit more potential than we are seeing right now.

I agree BUT we also shall be a bit cautious: I don't know if you use Corby's original variant with the MC1650 comparator or my variant with the MC100EL16 gate.
For the former, the jitter of the MC1650 is not specified, but for the MC100EL16 a typical jitter of 0.7ps is specified - that's 0.7e-12, I am not sure whether this also scales with 1e-6 - in that case it would be totally fine - or whether it stays like so - in this case the noise floor would be somewhat limited. The MC100EP16 would be the better choice then (0.2ps).
I doubt that the very dated MC1650 are much better than that, so I wonder what limitations arise from this.


Can somebody maybe explain how the jitter of the ECL gate scales when the heterodyning occurs?

The jitter will add to the jitter of your source. It's in the same domain (RF), before the mixer stage.

[Mrt12]

The jitter will add to the jitter of your source. It's in the same domain (RF), before the mixer stage.

sure. So that means it is also divided by the heterodyne factor?

[tkamiya]

That's what I was thinking.  Anything before mixer will be RF.  That gets hetrodyned to 1Hz after the mixer so the "dividing effect" takes place.  I think the exception is the offset oscillator signal.  Since they both get applied equally, it isn't critical.  At least that's my understanding.  I really would like a final word on this but so far, I haven't found any material indicating one way or another.

Then...  I wonder if ECL jitter is critical?  I'm thinking if they are from same batch, it may behave similarly enough that end result may be close enough.

[Mrt12]

I think even if the ECL gates are from the same batch, their jitter won't be equal as jitter is a random process.
But I think it is indeed the case that the jitter is in the RF domain and heterodyned down, which means that the 0.7ps Jitter yields a jitter of 0.7e-18 noise floor. So nothing to worry about.

[Gerhard_dk4xp]

> I think the exception is the offset oscillator signal.  Since they
> both get applied equally, it isn't critical.  At least that's my
> understanding.  I really would like a final word on this but so far,
> I haven't found any material indicating one way or another.

The noise of the downconversion oscillator won't cancel completely.

That's for the letters TD in DMTD. The two channels are evaluated at
different times, so they get at least somewhat different noise.

Cheers, Gerhard

[tkamiya]

Right.  But time interval is short enough the amount of drift/jitter should be very very small.  I think that was the point Corby was trying to show us by having DUT and reference pulse as close as possible.

[5065AGuru]

Hi,

Here is more data on how the quality of the offset oscillator affects the performance and also
another illustration of the importance of keeping close to coincidence.
I took a well performing oscillator and injected calibrated amounts of noise and measured
how much the STS was degraded by several fixed amounts.

AD at 1Sec for indicated noise levels
 
8.65X10-13th          Zero
3.10X10-12th          62.5
5.66X10-12th          125
9.72X10-12th          250
1.36X10-11th          375
2.26X10-11th          500

Then I setup to measure the simple dual mixer noise floor with this oscillator as the offset
oscillator.

I measured the noise floor at close to coincidence and also at close to full scale at the 6
levels of noise shown.

The plot shows all 12 300 Sec. measurements with the 6 close to coincidence on the left.
As you can see close to coincidence the noise floor stays fairly constant at all noise levels.
At the midpoint we transition to the 6 close to full scale measurements.
With no noise the first section is not too bad but as the noise is increased you can clearly
see the noise levels rise up.

This should help you select what oscillator to use as your offset oscillator.

I'm currently running some oscillator to oscillator plots with the 6 noise levels and will
post those results soon.

Cheers,

Corby

[Mrt12]

Corby,

I don't understand all of your post. STS? what's that?

Also I cannot yet interpret your chart correctly. Could you again please explain (for the dummies) what the x and y axes are? 

But I guess if the DMTD is adjusted such that the zero crossings are very very close together, the noise of the offset oscillator really cancels out well. There was once a discussion in the time-nuts with Ulrich Bangert about this topic, and he also said that, if the zero crossings are "far" apart, the offset oscillator will change its state in the meantime and therefore it cannot cancel out perfectly.

[5065AGuru]

Mrt12,

STS is "short term stability"

The X axis is in Seconds.

The Y axis is not calibrated as I just wanted to show relative noise levels.

Yes it cancels out well when close, that's what I'm trying to show as well as what the noise level of your offset oscillator does to affect performance.

Here is another plot that shows the AD degradation as the offset oscillator noise increases, all plots close to coincidence.

Cheers,

Corby

[tkamiya]

I'm making a PCB design for HP10811.  Should be handy for offset or reference oscillator.  I'll post here when it's done.

[Mrt12]

Sorry I was wrong yesterday, of course the jitter of the ECL gates is multiplied by the heterodyne factor!

@Corby. Do you have an idea how large the jitter of your MC1650 comparators is?
you said you measured the noise floor of your DMTD and that was quite good, but I wonder how big the ECL jitter contribution is. 0.7ps yiels 0.7e-6 when heterodyned with 10^6, that's quite a lot. On the other hand I can't imagine that your very old MC1650 has much better jitter performance - but as far as I remember you had noise levels in the 1e-13 range, I don't understand yet how that is possible.

The best ECL gates I found do have a jitter of 0.2ps, that would still be 0.2e-6.

Maybe the thing is that this jitter is PER PERIOD and since we are prodicing a 10 Hz beat note the jitter is averaged over 100ms and thus becomes lower?

by the way thanks for your explanations!

[Gerhard_dk4xp]

> I'm making a PCB design for HP10811.  Should be handy
> for offset or reference oscillator.  I'll post here when it's done.

I have made one for 10811A, Morion MV89, MTI-260 and
some less known, up to 100 MHz. It has a coolrunner CPLD
to generate a local 1pps. There is also a post amplifier
to increase power output to ~20dBm, with the option
to use it as a push-pull frequency doubler, with 2 crystal
notches against sub/harmonics. 10 MHz Morions have
an internal doubler and can use that.
There is a 2FF phase comparator in the $3 CPLD and the
low pass filter on the board to lock the oven as a clean-up
oscillator to an incoming reference frequency.
The hardware could also lock to an incoming 1pps, for
GPS etc. This has not yet been completely implemented in
the CPLD. Frequency locking works.
Vtune direction and some common frequencies are strapping
options and can be changed without touching the VHDL
of the CPLD.

Everything is optional. Think of it as a toolbox.
Someone in the US had boards produced locally from the
Gerber set, just for mechanical mounting.

Cheers, Gerhard

[tkamiya]

Gerald,

While yours in a Swiss Army knife, mine is a Ginsu knife.  (American joke, sorry)  I wanted to keep all that separate.  I have an external low noise amplifier board to go with it.  I also didn't use SMD for ease of assembly.

My initial goal was to make a burn-in tool.  But then I decided to make it more general interface board. 

One of the painful part was that commonly available socket for 10811 comes in two different pitch between rows.  I'll see if a fab house can actually do this.

[5065AGuru]

Mrt12,

I don't know the jitter of the 1650/1651 but the noise floor of the simple dual mixer is close to 3.5X10-13th at 1 Sec.

Cheers,

Corby

[tkamiya]

I had finally done a 24 hour noise floor test successfully.  The graph goes straight down to 20,000s and ends with 1E-17 somewhere.  I'd expected to start turning up way before then.  I wonder if this is too good to be true? 

Input is an output from 10MHz OCXO tied right at input of DMTD. 
Reference is 1Hz difference using 108663.  Purchased new but well aged for about a year or so.
Counter is TAPR TICC.
Recorded the data first and imported it into TimeLab, scale factor 10E-7

[awallin]

The graph goes straight down to 20,000s and ends with 1E-17 somewhere.  I'd expected to start turning up way before then.  I wonder if this is too good to be true? 

looks ok, the slope is slightly less than tau^(-1). if you press 't' you get TDEV which should show a slightly rising slope?

[tkamiya]

Tdev shows going as low as 5E-14 at 100s and starts to go up.  I am not familiar with tdev.  Will you give a quick education on this for me?

[Mrt12]

Will you give a quick education on this for me?

indeed this would be interesting - what does a TDEV of 5e-14 at 1sec mean?

[tkamiya]

I've read the definition of it but I am not understanding the difference between Adev and Tdev in practical sense.  In particular, what does it mean by Adev showing continuously declining graph and Tdev shows more less what I expect to see on a typical OCXO. 

[Mrt12]

I've read the definition of it but I am not understanding the difference between Adev and Tdev in practical sense.  In particular, what does it mean by Adev showing continuously declining graph and Tdev shows more less what I expect to see on a typical OCXO. 

I think I do have a simplified explanation of what ADEV is.
Suppose we operate a clock timer with an oscillator. Since the oscillator is not perfectly stable, the time this clock shows us wobbles a bit, i.e. sometimes it is a bit early and sometimes it is a bit late.
In my understanding, if I look at this clock every second, the amount how much too early or too late it is is the ADEV at tau=1sec. Further, if I look at the clock every 10 seconds, the ADEV at tau=10s tells me how much it will be too early or too late. And so on. No?
I know this is not the definition, but a more practical approach. I may be totally wrong.

On the other hand I cannot imagine what TDEV could be then 

[5065AGuru]

Taka,

Nice work!

Your noise floor looks great!

So now you can compare 5065A, test almost any Quartz, test any telecom type Rubidium, test any Cesium.

Of course you could also compare almost the complete range of two active Masers and the complete range of two passive Masers!

Now all you need to do is search through your oscillators to find the best one to use as the reference.

HooRay

Cheers,

Corby

[tkamiya]

Thanks!

I know I have a couple of Masers in my junk drawers somewhere....

[Mrt12]

Thanks!

I know I have a couple of Masers in my junk drawers somewhere....

whaaaaat! everybody seems to have fancy Masers and BVAs and stuff except me 

[tkamiya]

Today's task is going to be including a difference oscillator in the DMTD box.  I am finding I get most use out of 1Hz difference.  So that will be my default.  Then when I need something else, I can feed something else.

Question:
TAPR TICC has a need for 10MHz clock.  Does it make sense to use a good one or feed the same signal that will be used as a reference?  (not the 10MHz - 1Hz, but 10MHz itself)  I'm trying to reduce the cable clutter and fluctuations that will bring.  I want each piece to be self-contained as possible.  The usual lab clock (GPSDO) is useless for this task as short time stability is not that great being GPS driven)

[thinkfat]

Today's task is going to be including a difference oscillator in the DMTD box.  I am finding I get most use out of 1Hz difference.  So that will be my default.  Then when I need something else, I can feed something else.

Question:
TAPR TICC has a need for 10MHz clock.  Does it make sense to use a good one or feed the same signal that will be used as a reference?  (not the 10MHz - 1Hz, but 10MHz itself)  I'm trying to reduce the cable clutter and fluctuations that will bring.  I want each piece to be self-contained as possible.  The usual lab clock (GPSDO) is useless for this task as short time stability is not that great being GPS driven)

The TICC is in the NF domain of the DMTD, so the stability requirements for its clock are not too strict. Long-term stability is probably more important when you're doing time tagging. I'd say using a GPSDO is fine. The counter clock is not going to determine the noise floor of the system. After all, this is what the whole DMTD design is about, right?

This makes the idea of using a Bodnar frequency standard even more attractive. I could get the "big" one, which has two programmable outputs, and use one as the offset oscillator and the other one as the reference clock for the TICC...

[tkamiya]

TICC is used as in interval counter.  For that, TICC counts pulses between start and stop.  If time base moves around too much, even if input signals are constant, it will read differently resulting in "noise".  True, it does not need to be as accurate as if I am counting 10MHz directly but still, too much fluctuation will result in increased inaccuracy.  This is why I was wondering about GPSDO.  It has significant short term "yank" compared to a naked OCXO. 

I've tried using HP8644A with low phase noise option as difference clock.  It didn't go very well.  4 orders of magnitude difference in readings.  I'm still not sure of the requirement of this source.  For now, I'm using UCT108663 OCXO for that.

[Mrt12]

Hi Taka,

TICC is used as in interval counter.  For that, TICC counts pulses between start and stop.  If time base moves around too much, even if input signals are constant, it will read differently resulting in "noise".  True, it does not need to be as accurate as if I am counting 10MHz directly but still, too much fluctuation will result in increased inaccuracy.  This is why I was wondering about GPSDO.  It has significant short term "yank" compared to a naked OCXO. 

I've tried using HP8644A with low phase noise option as difference clock.  It didn't go very well.  4 orders of magnitude difference in readings.  I'm still not sure of the requirement of this source.  For now, I'm using UCT108663 OCXO for that.

basically right, BUT: if you heterodyne your 10MHz down to 1Hz, the gain is 1e7. Therefore, if your TIC has, for instance, a resolution of 100ns, which would be 1e-7, your timing resolution at 10MHz is 1e-14. A very cheap XTAL which is not even ovenized has maybe 50ppm tolerance, this is so tiny that you don't even notice it at 10MHz.

[tkamiya]

I have some TCXO package....  that will be perfect and it doesn't even have to be kept on.

[thinkfat]

The transfer oscillator is a different story, though. As we learned, its noise contribution will not entirely cancel out even though it is equally applied to both channels.

[tkamiya]

"Transfer Oscillator" is the difference signal?

[awallin]

I've read the definition of it but I am not understanding the difference between Adev and Tdev in practical sense.  In particular, what does it mean by Adev showing continuously declining graph and Tdev shows more less what I expect to see on a typical OCXO.

I don't know if there's any deep insight wrt. TDEV, but for a time-interval (or time-difference) measurement system it is maybe easier to look at the TDEV-plot which is mostly flat with a small slope - compared to ADEV where you expect a steep -1 (or so) slope 'forever' (as shown above, although no microwave clocks really go into 1e-17 and below) if the measured time-interval stays bounded.

some playing around with a noise-generator (e.g. https://github.com/aewallin/colorednoise) and an ADEV-library (e.g. https://github.com/aewallin/allantools) is suggested for the adventurous and curious...

[awallin]

"Transfer Oscillator" is the difference signal?

In a DMTD this is the 'offset-oscillator' tuned e.g. to 10MHz-10Hz in order to get ca 10 Hz outputs for the two 10MHz DUTs being compared. It needs to preferably be of similar quality to the DUTs being tested, IIRC.

[thinkfat]

"Transfer Oscillator" is the difference signal?

In a DMTD this is the 'offset-oscillator' tuned e.g. to 10MHz-10Hz in order to get ca 10 Hz outputs for the two 10MHz DUTs being compared. It needs to preferably be of similar quality to the DUTs being tested, IIRC.

Looking back at the DDS the "Small DMTD" by W. Riley used as the TO, the AD9951 that was used has a data sheet phase noise of -132dBc/Hz @1kHz  (best case with refclock multiplier disabled). The Bodnar GPS reference clock is spec'd with -143dBc/Hz @1kHz. Seems they play in a similar league.

[5065AGuru]

Taka,

The reference for the TIC is NOT critical.
I have used a 14 pin DIP OCXO, a nice TXCO, and a cheap crystal and got good results with all of them!
I'd just use what is convenient for you!

Cheers,

Corby

[Mrt12]

Taka,

The reference for the TIC is NOT critical.
I have used a 14 pin DIP OCXO, a nice TXCO, and a cheap crystal and got good results with all of them!
I'd just use what is convenient for you!

Cheers,

Corby

Hey Corby,
that is indeed one interesting result and good news! Have you seen differences in the ADEV or TDEV when your TIC runs from an expensive reference vs. when your TIC uses a cheapo crystal?
I want to make my own TIC as well, but I want to use time tagging. I assume you have used a "normal" TIC. Do you know the quality of the reference clock makes a difference if time tagging or a normal TIC is used?

[notfaded1]

Does anyone know if having a 2.4/5GHz wireless router near any of the gear that we usually use for these kinds of projects involving TIC's, counters, 5 and 10MHz LO's and GPSDO's cause any major effect on frequency nearby?  Or has anyone compared and found that it makes a big difference?

Bill

[tkamiya]

I have not noticed any effect from WiFi or Cellular phones.  But, I did see impact from near-by 10MHz.  Basically, I have my lab wired with lab standard 10MHz from GPSDO.  Also, just about 1/2 of my stuff contains 10MHz OCXO.  There is no concrete proof that indeed, alone, was the problem but I saw something.

Careful wiring internally and externally of the box took care of it.

[Mrt12]

Dear colleagues

finally, I got all my parts from Mouser and I assembled now one of the two DMTDs I have on my PCB. I have exactly the parts that are in the schematic, so I use LT1007 and OPA27. For this high effort I put into the design, I expected way better performance ;-) perhaps I could have even used LM741 as OpAmps.

I tested the noise floor as follows:
HP 8341A, 10 MHz 6dBm output goes on a resistive splitter and goes to the DUT1 and DUT2 inputs (X5 and X7). The HP 8663A signal generator puts 10.00001MHz 0 dBm directly to the offset oscillator input X6.
The HP 5335A counter measures the time interval between the two outputs (X1 and X2, rising edge, auto trigger, it does not make a difference if I use 50 ohm termination or not on the counter).
The settings on the HP 5335A counter are as follows: Gate mode minimum, no delay - this outputs a TI reading 10 times per second.

I adjusted the cable lengths for DUT1 and DUT2 such that the time interval at the output became as short as possible (i.e. such that the zero crossings are close together). Currently I was able to adjust it to around 100 us. However I do not have semi rigid cables but just odrinary Huber+Suhner coax cables, but I arrange them on the table such that they are as stable as possible and don't touch anything during the measurement.

My Timelab settings are the default settings, with a scale factor of 1e-6 because I use 10 Hz beat note. I have attached my noise floor I get with this setup. It looks terrible! if we look at the "phase difference" it looks even worse.

Either my circuit is complete rubbish or I am measuring something wrong - but what?
Please don't tell me that this is because I don't have transformes at the inputs :-D is it?

I tried to repeat the measurement with 1 Hz beat note. There's not much of a difference.

Something is completely wrong here!

[jpb]

I've not built a DMTD system (yet) but I've done lots of experiments with mixing down with a single mixer and generally got similar disappointing results. I think the trickiest part is the zero crossing detection on the very low frequency beat wave - I think it is there that you should concentrate your investigation into your circuit.

Some of the old papers show doing this with multi-stage limiting amps.

My own planned approach is to use autocorrelation (sorry wrong spelling) using an audio interface which gets round the issue of trying to change very low frequency sine waves into nice square waves. A paper was published on this by a Chinese research group.

[tkamiya]

Based on my own result, I can suggest the following, if for nothing else, testing purposes.

Use OCXO detuned for 1Hz for offset oscillator. 
Use coax from DUT to DMTD and split the signal RIGHT AT THE INPUT
Use a counter you can talk SERIALLY and record the session via putty or some other terminal emulator and make sure to REMOVE preambles and garbage.
Play it back on TimeLab by importing it with 10E-7 factor

A lot of this SHOULD NOT have made any difference but it did for my setup.  I still don't have transformers at input.  I am not entirely sure if that is necessary either.

[thinkfat]

The phase difference plot is quite interesting, you seem to have 20-30 picosecond "jumps" that gradually get worse the further the measurement goes. In the beginning they're few and far apart, they gradually get more frequent until they dominate the plot. That doesn't look like flicker noise. It also doesn't look like the noise plots Corby has posted here.

For your measurement setup, I'd not try to exactly phase align the start and stop. How about going for a nice 90° phase offset?

[Mrt12]

The phase difference plot is quite interesting, you seem to have 20-30 picosecond "jumps" that gradually get worse the further the measurement goes. In the beginning they're few and far apart, they gradually get more frequent until they dominate the plot. That doesn't look like flicker noise. It also doesn't look like the noise plots Corby has posted here.

For your measurement setup, I'd not try to exactly phase align the start and stop. How about going for a nice 90° phase offset?

Yup I saw those needle like jumps also. However there could be a ton of reasons for these, I don't know yet where I should start looking!
These spikes could come from:

a) the HP 5335A. I have repaired it once. So far I didn't have any problems with it, but maybe it has an instability which is just so small that I didn't notice  it so far?
b) I have not locked the 10 MHz references of my test instruments together, this means: the two signal generators and the counter do not operate on the same reference. Could this make a difference?
c) I guess the HP 8341A has not a very good stability. Even though the data sheet of this signal generator doesn't specify stability, it specifies phase noise, and this is not very good. Maybe my 8341A is unstable? but again, I thought at least for the noise floor test this should really not matter!
d) maybe it was not a good idea to use switching regulators for the supply rails? I use TRACO TDN5 WI series, they operate at >100kHz and I have added an output filter with ceramic capacitors and inductors. I have tested the output voltages with a HP 34401A, 15.02V and -15.01V and 5.01V. That looks indeed really good.


I don't think there is a problem with the circuit itself, but it is rather a malfunction of some kind I'd say. If I compare my circuit to Corby's, the difference is in the input stage (MC1650 vs. more modern ECL gate MC100EL16). The OpAmp I use (LT1007) has lower noise; besides that I also use the LM311.

I have AC coupled my mixers. Could that make a difference? I used ceramic caps for that. Will try to remove them tomorrow and replace with shorts. However I don't really like the DC current through the mixers! Also my coupling caps to the ECL gates are ceramic. Could this be a problem? it is X7R 100nF. Shall I test C0G?

The fact that the spikes are rare at the beginning and become more frequent the longer I measure indicates that they could be related to some sort of thermal problem. Something in the circuit is cold at the beginning, and as it warms up, it produces the spikes.
I see that the DC/DC converter for the 5V rail is quite warm, even though it is specified for 1A output but it currently needs to deliver only 300mA. Maybe this could be a culprit. On the other hand it has an efficiency of 80%, so it dissipates around 0.5W, should get a bit warm I guess.

I'll try to measure the power supply rails tomorrow with the scope and see whether there is a lot of dirt. And also check without the AC coupled mixers. At the moment I am quite disappointed as I expected much better performance for this effort!

I bet it's the powersupply. Hopefully. This would be a place which would be easy to fix :-)

I'll post my timelab files tomorrow.

[5065AGuru]

Mrt12,

For initial testing use a good OCXO. Once you get a good noise floor you can try other sources.

Here is a PIX of what I use.
- A passive SMA power splitter
- a couple short SMA cables
- an assortment of SMA M to SMA F adaptors (elbows work nice)
- A variable phase shifter from an HP 8405A (You can manage without it!)
 
Feed a good oscillator into the passive power splitter.
Connect the two outputs to the inputs of the DMTD unit.
Observe the count from your counter:
-If it's near full scale switch the start and stop cables
at the counter and or switch the RF input cables.
 
Once you have a count that is in the low end of the count
insert a very short cable or SMA connector into one or the other
RF input channel of the dual mixer.
Inserting the short delay into one of the channels will cause
the count to drop.
Continue adding small delays to that channel until the count
underflows to the full scale region.
Then remove the last delay added.
Then make sure all SMA junctions are tight!!!!
Now go ahead and take data to get your noise floor.
Take data for an hour and avoid drafts and movement around the
system.

If you have not read the DMTD tutorial post I'd recommend doing so.
 
Cheers,
 
Corby

[Mrt12]

Hi together

OK today I solved my problem at least partially!
Initially, I thought I will use ADE-1+ mixers. However I saw that the ADE-2+ is very similar but has a wider frequency range, so I bought that as well.
As I had my DMTD board not yet ready, I made some initial tests with the ADE-1+.  We can connect this mixer indeed as intended, i.e. use the input and output transformers in a differential fashion.
I assumed (but never checked!) that this is also the case for the ADE-2+. Now let's have a look at the mixers from below!



On the left there is the ADE-1+ and on the right the ADE-2+. The very top left and right pins are LO, whereas the very bottom left and right pins are RF. As we can see, the ADE-1+ indeed has only exactly one connection at these pins which goes directly into the LO and RF transformers, so this is fine. However, the ADE-2+ on the right has an additional green wire at the bottom left pin which goes to the center left pin - which is ground! so in my circuit, I have partially grounded my RF inputs. I noticed this problem when I replaced the coupling capacitors to the mixers with 0R resistors - the ECL gates and also the mixers became quite warm after a short while, and the mixers didn't work at all.
I now use ADE-1+ as intended initially, and indeed, this had some influence on the noise floor.



Also the phase difference looks much better now, the mysterious spurs I had previously are gone now!



The noise floor is still not very good, yet. But has improved much! And I think the remaining fluctuation on the phase may be caused by my coax cables and some temperature drift - when I came into my lab this morning, it was quite cool, and switching on the signal generators (with the jet engine fans, you know...) produces a lot of heat which causes the room temperature to increase and therefore the cables to change.
Currently my DMTD is improved much and I almost consider it usable, but getting rid of the remaining phase drift would be interestng!

The phase drift could still be due to a number of reasons; for instance my HP 5335A counter does not have a very good time base. It uses HP 10544A oven which is quite dated and it took months for it to settle and it has perhaps still not completely settled - maybe I should replace this oven someday.

Will do some further tests today or tomorrow, for instance with the FTS1000 which is the most stable and quiet source I have at the moment!

[tkamiya]

Can you not lock HP5335 to a lab standard, or another OCXO or something else that is known stable?

[Mrt12]

Hi Taka
I have only one LPRO-101 that is known stable, but I am unsure whether this is the right reference because I think the LPROs are known for being a bit noisy?
also I don't want to waste precious Rb lamp life, so I try to save it as much as I can. But maybe you are right and I have to use it now!

Maybe the noise floor I have is indeed caused by the jitter of the ECL gates?

[jpb]

I'm wondering if another alternative to the MC1650 might be the LTC6957-1 which is specifically designed as a low noise frequency reference buffer though it is not designed to drive mixers.
https://www.analog.com/media/en/technical-documentation/data-sheets/6957fb.pdf
It is quite pricey though.

[tkamiya]

You don't have any other OCXO?  I have all of my rubidium turned on 24x7.  Ha... 

[thinkfat]

Hi Taka
I have only one LPRO-101 that is known stable, but I am unsure whether this is the right reference because I think the LPROs are known for being a bit noisy?
also I don't want to waste precious Rb lamp life, so I try to save it as much as I can. But maybe you are right and I have to use it now!

Maybe the noise floor I have is indeed caused by the jitter of the ECL gates?

I would not worry about the Rb lamp life too much. Whether the LPRO is a good reference - well, it's not best-in-class but it's a solid performer with a much cleaner output than e.g. a FE5680A (source: http://www.ke5fx.com/rb.htm). You wouldn't be able to pitch it against a good GPSDO for small tau but for larger tau it can easily beat a GPSDO down to maybe 20000 seconds.

[5065AGuru]

Mrt12,

A 10544A (if operating normally) is just fine for the counter timebase!

I'd try a good OCXO as your offset oscillator and see it the noise floor gets better.

Cheers,

Corby

[Mrt12]

Hi together
I just testes these things. Currently I have locked all my equipment to a common HP 10811A. Its frequency may not be accurate, but it should at least be stable. I have no improvement of my noise floor so far!
The ECL gates have a jitter of around 0.7ps. I think with that I should hit the noise floor presumably somewhere at 7e-13. But currently, my noise floor is still around 1.5e-12. Phase is drifting around like crazy! I replaced the input coupling capacitors with C0G. No effect!
I removed the 5V DC/DC converter and use an external power supply (HP E3630A, this should be a really quiet one!). No effect.
I switched off all the lights (I do have LED lights in my apartment). No effect.
Switched off the cellphone. No effect (no big surprise, though).
Replaced one of the coupling capacitors to the mixers. No effect!
I'm going crazy. There is a bug somewhere, but I don't see it right now.
At least I solved the issue with the spikes I had, but that is only a cold comfort.
Currently I hit the 1e-13 noise floor at 100sec. That's too bad I guess!
I'm starting to believe in the transformer things.

[5065AGuru]

Mrt12,

You say "Currently I have locked all my equipment to a common HP 10811A. Its frequency may not be accurate, but it should at least be stable"

Does this mean your offset oscillator is still the 8663A that is driven by your 10811A?

Unless you can measure the AD of your 8663A output you can't really trust it!

In the DMTD tutorial my last post shows the noise floor degradation caused by an offset oscillator that is not at it's best.

You want better than 1.0X10-12th at 1 Sec AD for your offset oscillator.

Last weekend I had locked my HP 3325A to my good OCXO so I could do some tests at a 10Hz offset.
Noise floor was terrible! I then locked my Fluke 6160A to the OCXO and I got an excellent noise floor.

That's why I suggest that you use a good OCXO at 1Hz offset and see if your noise floor improves.

Cheers,

Corby

[KE5FX]

Mrt12,

You say "Currently I have locked all my equipment to a common HP 10811A. Its frequency may not be accurate, but it should at least be stable"

Does this mean your offset oscillator is still the 8663A that is driven by your 10811A?

Unless you can measure the AD of your 8663A output you can't really trust it!

8662A/8663A generators take an unusually long time to warm up because they use narrowband crystal filters in their VHF reference chain.  They eventually achieve decent residual phase stability, but you really need to run them for a couple of hours at a minimum before using them for anything that requires good phase stability.  That would definitely include acting as an offset oscillator for a DMTD.  There will be a lot of drifting and jumping in the first 30-60 minutes.

[Mrt12]

Mrt12,

You say "Currently I have locked all my equipment to a common HP 10811A. Its frequency may not be accurate, but it should at least be stable"

Does this mean your offset oscillator is still the 8663A that is driven by your 10811A?

Unless you can measure the AD of your 8663A output you can't really trust it!

In the DMTD tutorial my last post shows the noise floor degradation caused by an offset oscillator that is not at it's best.

You want better than 1.0X10-12th at 1 Sec AD for your offset oscillator.

Last weekend I had locked my HP 3325A to my good OCXO so I could do some tests at a 10Hz offset.
Noise floor was terrible! I then locked my Fluke 6160A to the OCXO and I got an excellent noise floor.

That's why I suggest that you use a good OCXO at 1Hz offset and see if your noise floor improves.

Cheers,

Corby

Hi Corby

yes I still do use my 8663A as offset oscillator. Why does the offset oscillator influence the noise floor? I thought as long as the zero crossings are closely aligned the offset osc should cancel out.
I'll try again today with some OCXOs!

8662A/8663A generators take an unusually long time to warm up because they use narrowband crystal filters in their VHF reference chain.  They eventually achieve decent residual phase stability, but you really need to run them for a couple of hours at a minimum before using them for anything that requires good phase stability.  That would definitely include acting as an offset oscillator for a DMTD.  There will be a lot of drifting and jumping in the first 30-60 minutes.

aha, this is one interesting result! thanks for that. I didn't consider this. I only know the 8663A has a good phase noise performance.
I definitely didn't let the 8663A warm up for long enough. Just power on and connect and then measure.

[tkamiya]

Corby,

A while back, you've mentioned you are working on a new TIC design.  How is that going?  I'm thinking of starting an 8 channel DMTD (basically 4 of your boards in one box) so I'll have to get 4 of such things.  I have enough HP counters but they are loud!

[notfaded1]

So ideally for my offset for 5Mhz with 1Hz offset with more digits on my Frequency counter would be 4.99999900000.  Used with DUT and REF showing 5.00000000000 or as close as possible.  I've been working on trimming some of the 1050A's I recently acquired.  With the rotary trimmer wheels it's pretty easy to change frequency.  The parts are in 4x10^-12.  My 5065 is a little off from 10MHz so I've been debating what to do about that?  I've been trimming the 1050A's to my 5061B with pretty good luck although after changing the frequency wheels it seems to take a little while to settle on a stable frequency.  I make a change then run timelab vs. the 5061B for a few hours and watch the frequency difference graph in timelab.  It's nice to make a run then start a new one after changing the wheels then start a new run in a different color on the graph to see how it changed.  You can see the frequency change for little while until it mostly stabilizes finally after a few hours.  The bipolar voltages for the board has also been a learning experience for me creating stable +-5 and +-12VDC with a linear power supply.  I'm getting there though.  It helps hearing about what some of you did first.

Bill

[5065AGuru]

Taka,

The counter is looking good.

I've been using the prototype for all my Simple Dual Mixer testing and also with a couple of my other DMTD systems and it works great.

I'm not sure when it will be released, I'll get some feedback on that.

It uses a Parallax Propeller Mini 32150 that you can get direct from Parallax for $24.99 plus shipping.

All you need to add is a couple BNC input jacks, a reset switch, and two resistors.

To interface with USB you will also need to add an Adafruit FTDI Friend @ $14.95.

I think you could also interface via an opto-isolator to an RS232 port.

DigiKey and others also sell them for a couple dollars more.

I'll update when I can.

Cheers,

Corby

[5065AGuru]

Mrt12,

Yes if closely aligned the offset oscillator noise mostly cancels out.
However look at the Nov 28 entry in the DMTD tutorial post.
The oscillators into the dual mixer are an excellent FTS 1200 and a very nice HP 10811.
All plots were run close to coincidence.
The offset oscillator had 6 noise levels added to it from no noise to 2.26X10-11th, and AD plots were made for each noise level as shown.
As you can see for the level of coincidence I was at the noise level makes a big difference.
That's why I'd recommend a noise floor run with a good OCXO as the offset oscillator first, and then you can see if other offset sources will work for you.

Cheers,

Corby

[tkamiya]

Corby,

That's good to hear.  I'll be the backseat child...
Are we there, yet??   

[notfaded1]

I decided to go the other way with the offset oscillator.  This is what I'm looking for correct?  In testing my 1050A's I found one that seemed to possibly be happier higher than 5MHz on the straight from LO 5MHz.  The 1MHz and 10MHz on the 1050A are derived while the 5MHz is directly from the 1000B LO.  Another thing I've noticed is when changing the voltage control on these they seem to drop a little from where they are set after a few hours so it takes a little time (and multiple retries) to get them pretty close to spot on and stable.  One of the 3 1050A I have is really stable at 5MHz.  In timelab after 8 hours the frequency stability measurement still shows in 10^13.  I think this will be my 5MHz REF for short term stability.

Bill

[tkamiya]

Aging OCXOs....    I wonder if 60 days is enough to start comparing??  Some appears to be new while some were well used and possibly sat in warehouse for years. 

I have 8 more boards to make.

[5065AGuru]

Taka,

I usually power them for at least 24 hours, do a first plot, and then comparison plots each day.

After a few days you will be able to select out the more stable performers.

Cheers,

Corby

[notfaded1]

What PCB's you use for the 10811's taka?

Regards,

Bill

[tkamiya]

I made my own.  The circuit is combination of what HP does in their products and what's in the data sheet.  Feed 24V and the rest is regulated down from it.  I'll be happy to provide Gerbers for it, if you'd like.

[notfaded1]

New power supplies I'm going to try.  It should be easy to experiment with different supply voltages.  It's interesting the original design had higher negative 5VDC you said -5.2VDC vs. the +5VDC.  Some progress.

Bill

[tkamiya]

Since this project, I have been paying attention on supply voltages.  I found several HP frequency counters having the same scheme of +5V and -5.2V, so this seems to be a common practice.  I'm thinking it may have something to do with noise with OPamps.  I wonder what happens to dynamic range and NULL value?  Would the null value be a mid point which is 0.1V?  Or would difference in supply voltage appear as different max values/swings?  (having negative swing larger than voltage) but null is still at 0V? 

[tkamiya]

Here's the ZIP for 11801 interface board design.

Please see SEEME_FIX.jpg for jumpers that are required.  These files were created using KiCad.  Gerbers ready for upload is in 10811interfaceV1.4/11801v1_4gerbers/HP11801interfaceV1_4gerbers.zip.  (yes, I misspelled the name)  I used PCBWay as a fabricator.  I've attached a comment to my order stating "Hole overlaps around a connector is intended.  Expect elongated through holes."  Expect the cost to be around $5 for 10 PCB + fees and shipping.

Please note, spacing for capacitors are tight both vertically and horizontally.  I chose smallest 35V type I can get from Digikey.  Or, you could possibly install them on the other side if clearance is an issue.  This board contains circuits in HP's own implementations of 10811, and what's in datasheet.  Please see circuit design for detail.  Goal of this board is to give 24V and create all voltages necessary for 10811 to operate.  EFC is provided on PLUS SIDE ONLY.  Output is not buffered but is DC isolated.  Please note, 30 pin connectors come in two different spacing between rows.  Both 2.54mm and 3.96mm versions work (thus the elongated holes)

Please use this at your own risk.  I am successfully using it (after jumpers) but I do not claim it to be error free.

[tkamiya]

I did my first long run as long as a bunch of short ones.

Please ignore the label.  There is no MV89a involved.  That is a naming mistake.
Reference is 10MHz from HP105B
DUT is HP11801-6011
Local (difference) is Oscilloquarz 8663 which is set to 1Hz over 10MHz.
Data was dumped to ascii file and imported into TimeLab.  Please see the import/acquisition screen.

I did few short runs as you can see.  They pretty much agree, so at least the result is consistent.  An odd thing though, this is not the kind of graph I expected.  First, it is starting somewhere around 6x10E-10.  It will quickly go down to  6x10E-12, then starts to go up.  My understanding was that OCXO are stable for short term, up to 100s or less and it will deteriorate.  The graph shows differently.  As suggested by someone earlier, I also did time deviation.  It pretty much agrees with my assumption... except... (see next)

Another thing is, 6x10E-10 as initial value is not what I expected by order of magnitude.  I expected 10 times better. 

I did multiple pieces (not shown in this post).  They were either almost identical to this, or two orders of magnitudes worse.  I'd imagine those pieces may not even be functioning correctly. 

Sorry folks - I don't remember how to screen dump.  But I hope you'll get the gist.  I'm soliciting any hint, suggestions, etc, etc, etc. 

[KE5FX]

Hint: don't even look at ADEV until you've looked over the frequency and/or phase-difference plots to evaluate the quality of your data.

(Also, ctrl-c will copy the screen to your clipboard, it's also near the top of the File menu.)

[tkamiya]

But what am I looking for? 

This is what it looks like now.  (please don't mind the color change....  blue = red same data)

[KE5FX]

You can tell by looking at the 'p' and/or 'f' traces if the drift or other instability was constant over your test run, or if it came and went for some reason.  For instance, occasional outliers -- anything from excessive popcorn noise to bumping the table -- can explain excessive instability at longer-term taus.  These can be removed with F4 (see the manual) if you're sure you can account for them.

Low-level spikes may suggest EMI that can affect short-term ADEV.  Low-level noise may be OK, if it occurs at the level expected due to counter quantization noise or trigger jitter, but watch out for low-level tones.  These appear as ripple on the ADEV plot and regular cyclic artifacts on the frequency-difference plot, and could suggest external interference or crosstalk from a nearby source.

Note how John Ackermann uses phase- and frequency-difference traces to study the behavior of the HP 5370B.

To look for the lower-level artifacts, hit 'r' to take the linear trend out, so you can see the residual.  The 'r', 'z', and 'y' hotkeys will end up being baked permanently into your muscle memory, or at least they should.

[tkamiya]

John,

Nothing is jumping out at me.  I knew those waviness on Adev curb exists on everyone of my graphs.  With 60Hz and 10MHz everywhere near the testing setup, I can imagine interference is common.  I can set it up in middle of my 3/4 acre yard if I have to.  I don't see any peculiar peaks.  I don't see any abrupt changes.  I've been at this for almost a year now. 

[5065AGuru]

Taka,

Can you share the raw data file for the long (>1day) run.

Also what counter were you using?

I'd like to see what's happening.

Cheers,

Corby

[tkamiya]

Here it is....

I've communicated with John and he gave me wonderful answers.  I'll just see if I can solve the problems he mentioned.

[5065AGuru]

Taka,

It appears that your DUT and REF are WAY TO FAR apart in frequency.

Measure the REF on a good counter and then adjust the DUT on the same counter to be within .001HZ of the REF (1X10-10th)

Then look at the counts on your DMTD counter. You should see the counts either going up or coming down.

Adjust the DUT or REF carefully to make the counts increment as slowly as possible.

Trying to do this with the oscillators adjustment capacitor is not going to work. (With a lot of messing around you might get it to work just to try)
I see your REF is a 105B so the fine adj pot will work with it.

Added coarse and fine EFC pots are the way to go.

Now take data say for 15 minutes and see what you get.

Your phase plot should look smooth with no wraps for such a short plot.

Cheers,

Corby

[tkamiya]

Corby,

I've adjusted DUT to match reference within 0.0xHz.  Here's the results.  Red line is the new trace.  Blue trace is the old.  It's better but in Phase difference and Frequency difference plot, I'm still seeing something funny.  My lab is not a good environment.  10MHz signal everywhere.  I'm going to isolate the system to middle of my living room and see if I can eliminate possible interference.  This will take few days.

I'm also going to have to modify my DMTD to bring out another output so that I can hook up a scope.  Frequency counter won't cut it as its reference is GPDSO.  I'm sure (hope) my OCXO is more stable than that in short term.

[5065AGuru]

Taka, That does look a bit better but I'd need the raw data again so I can see what's going on.

Cheers,

Corby

[tkamiya]

Here it is.

[5065AGuru]

Taka,

Well you are about 6X10-10 offset between the DUT and REF so you are wrapping every 165 Sec or so.

You can't do an Allan Deviation on the whole file because of the wraps!

You can remove all but one cycle and run a plot on that or remove all the wraps.

You need to get them close enough so that they don't wrap so quickly.

Look at the DMTD tutorial post, in the example shown they did not wrap for over 2000 Sec.

Like I said adjust to make the count almost stop changing prior to taking the data.

Here is a phase plot of your file.

What counter are you using?

Cheers,

Corby

P.S. Going offline for tonight.

[KE5FX]

Here's how the data from corby1.txt looks in TimeLab:


This is the original wrapped phase data ('w') view of corby1.txt, which I imported as follows:



With those import settings, TimeLab will unwrap the phase automatically.  Pressing 'p' shows this view of the data:



As you (Corby) note, this is a frequency error of 6E-10, which should be OK.  There is normally no need to avoid phase wrap, as long as it happens slowly enough to obey the usual Nyquist limit.  (If you want to eliminate all relative drift between the reference and DUT, you should use a tight PLL rather than a DMTD!)  It's necessary to look at the residual ('r') to see the fine details:



Or differentiate it, by looking at the effect in the frequency ('f') difference view:



The question I have is, where is the 2ns sawtooth coming from?  The 2ns resolution of the counter should be swamped by a factor of 1E7, taking it out of the picture entirely, but for some reason we still see this 2 ns glitch every couple of minutes when the phase wrap occurs.  If I zoom in on each of those spikes in the frequency-difference view and remove it with F4, I get a more believable result:



Hitting 'z' to position the trace for viewing at a zero baseline:



The phase drift now looks like this, compared to the original data:



Again, a perfectly-reasonable display of two decent 10 MHz oscillators being compared to each other.  ADEV now looks like this:



Once again, very believable.  But what's up with those spikes at the wrap point?  They shouldn't be there, unless the counter has a bug or calibration issue, or I'm overlooking something about how the DMTD works. 

[tkamiya]

My counter is TAPR TICC with all settings at default. 

I'll use my scope to set frequency of DUT and REF exactly and see what happens.  I thought TimeLab can take care of wrapping issue, so I CAN do the Adev....  no?  I do have number of different counters so I can try that, too.  TICC is just easier, so I was staying consistent with it.

Something I forgot to say yesterday.
My difference oscillator is a simple oscilloquartz 8663 set at 10MHz + 1 Hz.  Power supply for this is a switching regulator - the one someone talked about on TimeNuts list.  This is supposed to be a low noise type for pre-amps.  I never bothered to measure it.  So it could be dirty.  It'd be easy to use a linear supply.

By the way....  TICC and CorbyDMTD, and all the oscillators EXCEPT DUT are all in metal case.  Internal wirings are semi-flexible coax (the type with no outside and tin soaked braid is exposed.  Connectors are either N or SMA.  (I don't trust BNC)  Coax is all good to great quality.  TimesMicrowave LMR 200 and some uber expensive ones I got second hand.  I've tried to limit signal leaks.

Thank you very much for looking at my data.  I owe you some beer.

[tkamiya]

About the Adev graph above, is it reasonable that it is staying that low beyond 100 seconds?  I thought it should start rising at around 100s.

John, how are you arriving at 2nS figure?  Will you share how you are reading the saw tooth graph?

[KE5FX]

John, how are you arriving at 2nS figure?  Will you share how you are reading the saw tooth graph?

Just refer to the Y axis labels in the phase-difference plot above.  They are in seconds.

I thought you were using an HP 5334A for some reason, which has a 2-ns single shot resolution, but I believe the TICC is more like 80 ps.  So the ~2ns height of the sawtooth-shaped phase glitch doesn't make much sense either way.  Before the 1E-7 scale factor is applied, this is more like a 20-millisecond error that is showing up every time the counter wraps.  Some sort of dead time effect, perhaps, that occurs whenever the counter thinks its STOP signal is arriving before the START signal.

Note that your data is quite clean-looking once the wrap glitches are removed.  You can stop worrying about the test environment for now, and focus on getting the counter to report valid results in the presence of phase wraps.  Or, as Corby suggests, move the oscillators so close together that they don't wrap at all during your test period.  That obviously puts massive limitations on what the system can do, so it would be better to get the setup working properly with wrapping.

About the Adev graph above, is it reasonable that it is staying that low beyond 100 seconds?  I thought it should start rising at around 100s.

Yes, a good 10811 will easily stay below 1E-11 at t=1000 seconds or more.  E.g.:

[thinkfat]

The TICC is a timestamping counter. There should not be any problem around phase wraps since it just calculates the difference between two consecutive timestamps. It should not matter whether channel A comes before B or vice versa as each of the channels delivers timestamps independently.

@tkamiya can you record a series in timestamping mode, or even better in debug mode that contains at least one wrap? Don't bother importing in TimeLab, just post the raw capture data. This will allow me to have a look at the individual channels and to see if there's any funny business going with the individual timestamps.

[tkamiya]

John,

When Corby suggested, I measured DUT frequency.  It was almost exactly .1Hz off

[tkamiya]

I'm so glad I've included EFC in my 11801 board. 

I am going to try using different counters to see if the wrap-spike is caused by TICC.  I kind of doubt it but I have multiple 53132A, HP5335A, and HP5370B.  Trying one of each should tell me if I'm on the right track.  I've been wanting to separate time-nuts stuff from general lab.  I may just to it sooner than later.  Lab temperature changes too much.

[tkamiya]

So, to get a good count, the following conditions must be met.

DUT and REF has to be very close to each other in frequency.
Start and Stop signal has to occur in shortest period possible.
Wrap should be avoided or minimized.

I have a lot of coax relays.  I could make something like a decade box for adjusting phase.  (I like overkill)

[thinkfat]

I'm really interested in the results you get with the other counters. In theory, the TICC should be ideal for the purpose and I was planning on using it myself. I'm curious what you will find.

[tkamiya]

Oh, me too!  In addition to good functionality, it's low power, quiet (silent) and inexpensive.  I'm really not discounting environmental factors.  It'll take some setup as I took down GPIB from others.  For some reasons, 53132a doesn't like to work with me, either.

I'll post back as soon I get something.

[5065AGuru]

Taka,

The ONLY time I use phase shifters to get to close coincidence is for noise floor tests.

For other tests I just adjust the DUT or REF to bring me down to close coincidence.

Then readjust for as slow as possible increments. Since you would need to do this even with phase shifters why make the setup more complex than needed?

Also on long plots you can run an AD on the whole plot (after unwrapping) this will give results that are good at the higher >10 Sec Taus. Then pick a spot where you are near the lower wrap and do an AD on just that section. This will give good results for the <10Sec Taus.

You can then plot them both on the same chart to get a combined result.

There are dead time issues with the 53131/132 as well I suspect on the 5335 and 5370.

I used a 53131 or SR620 for a few years before I went to a purpose built TIC.

I could work around the dead time easily enough but it's nice not to have to mess with it.

The companion counter (to be released soon) for work with Dual mixer systems has no dead time issue and is cheap/small/easy to work with.

When you take some data with your 53132 let it wrap a few times and send me the file.

I can show you the dead time.

Cheers,

Corby

[tkamiya]

At this time, I have no idea what is going on....

I changed the counter to HP5335A and started experimenting.  Oscillator and reference, and everything else is identical.  Frequency difference between DUT and ref has been brought very close to each other.  Wrapping still appears to be a problem.  In frequency difference screen, I see an abrupt change.  Phase difference does not show the same.  ADev showed somewhere between 1E-11 and 1E-12 until wrapping took place.  Then it went up by a factor of 10 (-11 and -10) once and by factor of 1000 next time.

I have not been able to talk to HP53132A.  (I did this before successfully)
Something is awfully wrong....

One odd thing....  On TimeLab, if I select HP5335A, "Scale Factor" column shows up.  If I select "Talk only" it does not.  I only mention this because my issue with live capture using TL is similar with 5335A but use of scale factor appears to fix this.  I've used input frequency of 10E6 and scale factor of 1E7.  (same thing...)

It's pretty conclusive wrapping is the source of problem.  I have no idea why.

One thing I've noticed.  In Phase Difference screen, unwrapped and original shows identical graph.  Why???  I don't recall it was like this before.

[5065AGuru]

John's (KE5FX) info was spot on.

The wraps can be unwrapped so you can use the longer files.

After unwrapping does your AD look OK? If so are your frequency difference abrupt changes gone?

The 53132 should work OK, RS232 talk only. You do need to turn print on.

There is another parameter (besides baud rate) that needs selecting but I can't remember it offhand.

I'll dig it up and let you know.

Cheers,

Corby

[tkamiya]

I did get the HP53132A to work.  It had to be power cycled.  (Yes, I know about enabling printing)

An odd thing is, unwrapping does not appear to do anything on my TimeLab installation.  The graph is exactly the same.  I remember seeing "saw tooth" like wrapping became straight at one point.  I cannot do it for some reason.

[KE5FX]

I did get the HP53132A to work.  It had to be power cycled.  (Yes, I know about enabling printing)

An odd thing is, unwrapping does not appear to do anything on my TimeLab installation.  The graph is exactly the same.  I remember seeing "saw tooth" like wrapping became straight at one point.  I cannot do it for some reason.

In the actual data, does it still look like the numbers are repeating every so often?  Previously they were cycling every 165 seconds or so. 

If not, you may not have had any phase wraps on that particular run.  In that case the 'p' and 'w' views will be identical.

One odd thing....  On TimeLab, if I select HP5335A, "Scale Factor" column shows up.  If I select "Talk only" it does not.  I only mention this because my issue with live capture using TL is similar with 5335A but use of scale factor appears to fix this.  I've used input frequency of 10E6 and scale factor of 1E7.  (same thing...)

'Scale Factor' in the 5335A dialog is the same as the 'x' part of the "Numeric Field # _____ x ______" fields.  Enter 1E-7 in those, not 1E7.

[tkamiya]

John, Corby, and everyone else

John,
You are right, of course.  There was nothing to unwrap! 

Today, I've managed to fine tune frequency of DUT within 1 mHz of reference, and took nearly 10 minutes of data without wrapping.  With TICC, I have a believable data.  This is still a problem though - as it means I cannot take long run session. 

I'll have to study what EXACTLY TimeLab does to find out why my dataset doesn't unwrap correctly.  I wonder if there is a dead-time when it wraps?  The variation in phase data isn't that much.  Seems when it wraps, phase data gets shortened by few parts in 10E-9.

I wonder if this is something like....  when register in TICC overflows code takes different path and reduces incremental count by just a little?  I mean the "error" is very small.  Execution of few extra steps will explain it.  Hum....

[tkamiya]

Here's a graph with 10Hz difference.  Reading is taken every 0.1 second.  Still having problem with wrapping, so this is just the data before that happens.

DUT 11801
REF  HP105B
LO 11801 at 10Hz offset

[tkamiya]

I am now trying HP5335A with 10Hz difference.

Long story short, it works well except for few problems.  Looking at Frequency Difference chart, where wrapping took place, there are deep "dips".  (see 500 second and 1000 second mark)  I've visually identified wrapping taking place through TIM file.  In addition to this, there was a strike of noise coming in at one point, and with it, Adev dropped by 3 orders of magnitude.  (see 363 second mark)  Before this noise strike, Adev was tracking 10^13 and immediately after, it dropped to 10^10.  Concerning wrapping, unwrapped and wrapped display look identical.  However, Adev appears unaffected.  What's unknown is if its impact is actually buried into poor Adev caused by the "strike".

I'm going to upload TIM and all graphs for reference.  I don't expect anyone to analyze this for me but it might help others identify their problems.  With my setup, taking screenshot is rather cumbersome.  So please excuse use of cellphone snapshots. 

This is rather confusing but an adev graph with two traces, please see red trace.  Blue trace is irrelevant.  After this, I deleted that trace, so blue one is the right one.

<edit>
I also noticed at 330 second mark, and 880 second mark, slope of the curb changes in phase.  They are exactly the same point from t=0 and again same distance from wrapping taking place.  On graph, I see this zoomed in.  (see IMG_4784.jpg)

[tkamiya]

John,

How do I find the exact spot in TIM file where these weird symptoms are appearing in graph?  Having a time stamp doesn't help.

[tkamiya]

Taka's monolog, stardate 20201229....

I am repeating the experiment just to make sure this wasn't a one-off issue.  In a word, it is repeatable.

Unwrapping still results in deep and narrow dip in frequency difference curb.  However, I am realizing the change in slope noted as above is not present in frequency difference curb.  Increase in dy/dt in phase should result in shift in frequency - which is not present.  I am wondering if this is a graphing artifact.  I am also noticing after the wrap, dy appears to be greater than one.  (meaning it takes more than one interval to start positive delta again)  That would mean, unwrapping will leave a hole in same depth as frequency.  Noting phase difference is unit of 2 measures in graph, and frequency difference is also unit of 2 measures.  At this point, I am not sure of transfer function between the two.  However, graphically, it seems to work that way.  If this is true, then the deep drop is an artifact in unwrapping.  Note, in zoomed in graph, increment in dT is 2 seconds.  Visibly, it appears to be taking at least 2/10 seconds to reset to zero before climbing back again.  Zooming further in did not help resolve exact fractional seconds.  What happens in t+1/10?

In other words.... someone who's better than I in math, HELP!

[tkamiya]

It appears to be a known issue according to Mr. Google in mathlab.

https://stackoverflow.com/questions/32442396/phase-unwrap-issue-the-unwrapping-of-the-phases-is-not-correcly

Basically, when data point is shifted in Y direction by 2pi, discontinuity in data will result in missing point remaining in the same location while the rest shifts UP.  That creates a deep null. 

So the issue is of sampling frequency and piloting interval.

[KE5FX]

For the record, this is the code used by TimeLab for phase unwrapping:

Code: [Select]

      //
      // Store wrapped and unwrapped copies of phase sample
      //
      // If data came from a TIC, the maximum supported phase is bounded by the cycle period,
      // and we need to detect wraparounds in both directions.  If it came from a frequency counter
      // or direct-digital test set, the phase is unbounded, so we can inhibit wrap testing
      //

      P[n_acq_points] = phase;                           // store original potentially-wrapped phase ('w' view)
      P_sum += phase;

      if (enable_wrapping && (last_wrap_phase != DBL_MAX) && (full_cycle_period_secs != 0.0))
         {
         DOUBLE d_secs = fabs(phase - last_wrap_phase);  // compare phase difference in seconds with nominal half-period in seconds

         if (d_secs > half_cycle_period_secs)
            {
            if (phase > last_wrap_phase)                 // underflow wrap (e.g., 1 ns to 99 ns)
               phase_wrap -= full_cycle_period_secs;
            else                                         // overflow wrap (e.g., 99 ns to 1 ns)
               phase_wrap += full_cycle_period_secs;
            }
         }

      last_wrap_phase = phase;
      phase += phase_wrap;

      U[n_acq_points] = phase;                           // store unwrapped phase ('p' view)

... i.e., it does nothing except look for successive differences that exceed a half-cycle at the user-specified frequency, and accumulate an additional whole cycle every time that happens. 

Barring any strange bugs that have gone unreported for almost ten years, any artifacts that occur at wrap points are likely to be an issue with the counter or test setup.  Counter support will always be a 'best effort' sort of thing in TimeLab since its main purpose is to support direct-digital hardware, but if anyone can show that TimeLab is introducing any wrapping artifacts on its own, I'm all ears.

John,

How do I find the exact spot in TIM file where these weird symptoms are appearing in graph?  Having a time stamp doesn't help.

That's a bit of an open-ended question; what do you mean?  Just hit 'p' and/or 'f' and look for behavior you don't expect.

[tkamiya]

Thank you!

No, I don't think I've discovered a bug in your code.  I do think there must be an odd combination of corner cases happening.  Either that or my error in defining the acquisition screen.  I was thinking of manually unwrapping or writing a simple code to do the same.

About my question, say I'm looking at a graph and find a spot of interest.  Graph is doing "something funny".  Since the graph is a representation of time variant data, in order for "funny" to show up on a graph, there must be a corresponding data that is responsible for the behavior.  Let's say I save the data into a TIM file.  I want to find a row/data that correspond to that spot on the graph.  How do I do that?  p(phase difference), and f(frequency difference) will take me to the graph.  I can zoom in and see the data I find "funny".  How do I know which row in TIM file that caused/represents it?

Additionally, I see there is a half_cycle_period_secs.  You said it comes from "user specified" period.  Which field in acquisition screen exactly in HP5335 and stream device support?  Where does the following variables come from?  full_cycle_period_secs

How is it they declared? 
The DEFINE DBL_MAX, in the compiler in use for this, what is sizeof(DBL_MAX)?  What is double in this case?  8 bytes?

*something* is causing if(phase > last_wrap_phase) to return false I think.
Thank you very much.

[tkamiya]

Never mind on "how do I find" question.  I just found out I can export it with time stamp.

[KE5FX]

Never mind on "how do I find" question.  I just found out I can export it with time stamp.

Yep, that works.  At some point I'm going to add the ability to create annotations at desired spots in the data set, which will also make it easy to narrow down individual points.

The period_secs variables are based on the reciprocal of whatever you enter in the 'Input Frequency' fields, which are present in both the HP5335 and talk-only/import dialogs. 

Everything in TimeLab uses 64-bit doubles, and DBL_MAX is just used as a flag value that will never show up in real data.  It's either a #define or a const double declaration in float.h or someplace like that. 

[tkamiya]

Excel dump of frequency data shows all rows are unwrapped. 

I took this dump then subtracted n+1 from n in frequency data.  I see 343.4 and 343.5 being 6 to 7 orders of magnitude larger.  This is where wrapping happened.  I don't have what my counter sent.  But these two rows where it represents point of wrapping is unusual. 

Starting from 343.1 and incrementing by 0.1 second per row:

Here's the frequency data:
1.00000000002397E-05
1.00000000002523E-05
1.00000000002439E-05
9.99999900009689E-06
1.00000000001218E-05
1.00000000001228E-05
1.00000000001253E-05

Here's n+1 - n:
6.01901612318906E-18
-1.25699689372538E-17
8.36021518939994E-18
1.00014705108824E-12
-1.00002494281856E-12
-9.97804811865566E-19
-2.50044126029469E-18
-2.5410988417629E-18

Something happened at 343.4 to 343.5 second.  Could very well be my counter reporting something funny.  Wouldn't surprise me....  It's an old counter.  Somehow, I need to ascii capture what HPIB is reading.

Wait a second!  I exported frequency data.  Why am I seeing data in range of 1E-5?  Counter is showing nn.xxxx E-03.

Posting for sake of record keeping...

[tkamiya]

I think there is a major error (on my part) in acquisition screen.

The setting is DMTD 10MHz + 10Hz, 10 readings per second.
Data type is time interval.
Input frequency is 10E-6
Scale factor is 10E-7

This can't be right, right??  What should they be?  I'm always confused with these two.

[KE5FX]

That should be fine (see the image of the acquisition fields I posted above.)

[tkamiya]

What about INPUT FREQUENCY?  I was putting in 10E-6.  You have 10E6.

[tkamiya]

That didn't make any difference in graphs.  Hum.... 

[KE5FX]

Sorry, didn't see the minus sign -- yes, you should give it the frequency at the "DUT" mixer input there.  10E6 or 1E7, take your pick.  Then, with the 1E-6 scale factor, the phase is scaled to account for the heterodyne factor between your 10 Hz beat frequency and the 10 MHz input.

[tkamiya]

UNwrapping worked when I set input frequency to 1E8, and scale factor 1e-7.  I took verbatim of what help says.  Phase swings between 0 and 1E-8.  Recipical of largest phase difference, which is 1E8.

Now trying your suggestion.

[tkamiya]

That worked as well as far as wrapping is concerned.

The Adev graph looks unreasonable though.  Dropping down to 5E-12 at 0.5 second.  Rising there after.  I expected somewhat flat response to 10 second or more.

Light blue trace is what I said before you that worked.
Brown trace is your latest response.

[tkamiya]

Thank you, John.  It's 2am now.  I'll try the same with TICC tomorrow.

Thank you for all of your help.  Great big THANKS!

[KE5FX]

No prob.  5E-12 @ t=0.1s is actually quite reasonable, because your measurement floor will likely get (much) worse at short-term taus below 1s. 

In general, nothing involving 1E8 or 1E-8 should be present, though.  Your heterodyne factor is 1E6 (10 MHz / 10 Hz), so that's what you would want to scale your phase data by in order for the data to make sense with the 10 MHz (1E7) nominal frequency.  (Be sure to enter 0.1 Hz as your sample interval, of course, rather than 1 Hz.)

[tkamiya]

I have scale factor at MINUS -6, such as 1E-6.  Is that what you meant? (you said 1E6)

Input frequency is 1E6 Hz
Sampling interval is 0.1second

[thinkfat]

I'm afraid the unwrapping code in TimeLab is too primitive. It heavily depends on "full_cycle_period_secs" being very, very accurate. If it isn't, you get a jump in phase at every wrap.

I don't know exactly what goes into the calculation of this variable, but I can import the original corby1.txt properly if I play with the "input frequency" in the import dialog. If I set it to 10.257e6, the file is imported properly. Resulting ADEV attached.

I'd be very suprised if there was such inconsistency in time intervals measured with the TAPR TICC. As I said, it's a timestamping counter and by design cannot be affected by start/stop ordering problems occuring near phase wrap. If there was an inconsistency, it'd have to be in the numeric library. Not likely for something as simple as a subtraction.

[KE5FX]

I have scale factor at MINUS -6, such as 1E-6.  Is that what you meant? (you said 1E6)

Input frequency is 1E6 Hz
Sampling interval is 0.1second

No, your input frequency would be 10E6 (10 MHz).  Since it's 10 Hz in reality, you would enter 1E-6 to scale the readings back by 1000000:1.

I'm afraid the unwrapping code in TimeLab is too primitive. It heavily depends on "full_cycle_period_secs" being very, very accurate. If it isn't, you get a jump in phase at every wrap.

Yes and no.  The unwrapping algorithm is meant to operate on the count, not the period.  If I'm doing a conventional STOP-START measurement between two 10 MHz sources, such that the TI readings can never exceed 100 ns, then a difference between successive readings that exceeds 50 ns indicates that a rollover occurred.  In that case, the reading is still correct modulo 100 ns.  The "phase unwrapper" is simply implementing a carry or borrow operation, adding more MSDs to the count.  It was intended to handle this case, and that's about it.  For the intended use case it's well below the noise floor of most counters. 

If I use a 5370B to compare two sources that are close to 10 MHz, for example, the unwrapper works well, or at least well enough.  I don't get a 2-ns error at the nominal 100-ns wrap points until I specify an input frequency that's 200 kHz off, i.e., 10.2 or 9.8 MHz.  That makes sense, considering that 1/10.2E6 or 1/9.8E6 is roughly 98 or 102 ns. 

So, thinking about it this way, it's not surprising that wrap errors that ordinarily don't show up in a straightforward TI measurement are suddenly a problem with a DMTD.  After all the whole idea is to magnify any phase-difference errors to measure them on a counter that normally doesn't have the required resolution.  What to do about it is a different question, of course.   An exercise for the proverbial student.

I'd be very suprised if there was such inconsistency in time intervals measured with the TAPR TICC. As I said, it's a timestamping counter and by design cannot be affected by start/stop ordering problems occuring near phase wrap. If there was an inconsistency, it'd have to be in the numeric library. Not likely for something as simple as a subtraction.

Yep, agreed.

[tkamiya]

It was late.  I meant 10E6 for input frequency.  Thanks again for staying up for me last night.

I learned also, unwrapping is an actively researched subject.  I found two PhD thesis on this.  In areas like geo-spatal imaging, MRI, and radar, all involves unwrapping.  I was reading it's easy to do at first but when we consider all the corner cases and what-ifs of real implementation, it suddenly get complicated. 

One thing I've noted.  I may have a reflection problem somewhere between counter input and DMTD output.  Every now and then, LED on A input flickers one too many time as B does - which should not happen.  It is possible that signal is reflecting and higher instantaneous voltage may be triggering A port every now and then.

I really don't care what counter - as long as it works.  I'm pretty happy to have it working on HP5335.  Once I have a solid foundation, I can start to improve.

John,
TimeLab is not open sourced, correct?  I'd love to be able to look at the actual code.

[tkamiya]

Retried TICC again with the same setup.

Wrapping issue still exists.  It does not unwrap. 

[thinkfat]

Retried TICC again with the same setup.

Wrapping issue still exists.  It does not unwrap.

I'm afraid the unwrapping code in TimeLab is too primitive. It heavily depends on "full_cycle_period_secs" being very, very accurate. If it isn't, you get a jump in phase at every wrap.

I don't know exactly what goes into the calculation of this variable, but I can import the original corby1.txt properly if I play with the "input frequency" in the import dialog. If I set it to 10.257e6, the file is imported properly. Resulting ADEV attached.

I'd be very suprised if there was such inconsistency in time intervals measured with the TAPR TICC. As I said, it's a timestamping counter and by design cannot be affected by start/stop ordering problems occuring near phase wrap. If there was an inconsistency, it'd have to be in the numeric library. Not likely for something as simple as a subtraction.

Try it. It takes a while to figure out the correct value, but it can be done. Just modify "Input Frequency" on import, try a few times and watch the "frequency difference" display, eventually the spikes will disappear.

[tkamiya]

I'll have to manually perform unwrapping to understand this.  What is odd is, last night's experiment with HP5335 had few wraps.  It worked correctly.  Original and unwrapped screen showed respective versions.  Frequency display was devoid of peaks.  This morning, I noticed original and unwrapped graphs are identical showing UNWRAPPED graphs.   Since I wasn't paying attention to yesterday's work, I don't know when that changed.  Possibly when new acquisition was started this morning.

There is something going on that I am missing entirely.  It may not be the calculation.  I'll repeat the experiment and see if I can recreate it.

[notfaded1]

Question do these looks close enough to try?  I'm going to do it soon anyhow but I spent a LOT of time getting the Offset Oscillator stable at the right stability over decent amount of time 8 hours.  You can see my work in the last three plots working on getting stable Offset Osc.  I hope I'm close enough get some decent data.

Regards,

Bill

[tkamiya]

What are you doing?  Noise floor test?

[notfaded1]

No these are measurements of the three Oscillators directly against the 5061B over 8 hours each and the offset 3 times while tuning it.  I'm pretty excited to see them through the DMTD next.  I was trying to get them all stable over decent amount of time without drifting all over.  This took me many days and many 8 hour sessions to finally get it where setting the wheels and letting it stabilize then running it for 8 hours would be consistent and close to frequency I'm looking for.

Bill

[tkamiya]

They don't look right.  Adev curb is rarely straight like that.  You should have an initial value, go lower, keep at certain level for a while before going up.  Why are intervals 2.5 seconds?  If offset is 1Hz, you should be able to read every second. 

I'm doing 10Hz offset and reading every 1/10 second. 

You have HP5061?  You have lots of neat toys.  Best oscillator I have is an old and worn out Cesium.  It still locks but how well it's working, I have no ways to know.

[notfaded1]

They don't look right.  Adev curb is rarely straight like that.  You should have an initial value, go lower, keep at certain level for a while before going up.  Why are intervals 2.5 seconds?  If offset is 1Hz, you should be able to read every second. 

I'm doing 10Hz offset and reading every 1/10 second. 

You have HP5061?  You have lots of neat toys.  Best oscillator I have is an old and worn out Cesium.  It still locks but how well it's working, I have no ways to know.

On the 53132A you can't get 12 digits of resolution without the longer sample interval... it'll cut off the digits of precision.  Believe me if I don't let them settle first after making voltage control changes they will curve a LOT more at first.  I've spent a lot of time and patience working on these.  You can see how the #2 unit after adjusting at first it wanted to curve back up right away.  It took me weeks to get them to this point.  The nice thing about the 1050A's are the wheels for adjusting the control voltage.  The wheels adjust like this:

Switch #
1   4 x 10^-12
2   4 x 10^-11
3   4 x 10^-10
4   4 x 10^ -9
5   4 x 10^ -8

Bill

[tkamiya]

Right.  But you don't need that many digits.  If you do 1 second gate time, you'll get as good as it'll ever do.  I don't know what the optimal value is, but this is DMTD, so 12th digit, if you have it, represents like 10E-18 or some ridiculous value. 

Are you reading frequency or are you reading time interval?  You should be setting it to have port 1 to start, and port 2 to stop. 

[notfaded1]

This isn't through the DMTD yet.  It's just the 3 LO's against the 5061B so I wanted the 12 digits to see how close I was getting.  The tweaking I was doing in the last two digits.  I'm hoping to then run these 3 through the DMTD next with the offset and two others here.

Bill

[tkamiya]

Oh, I misunderstood.  Apologies.

[edpalmer42]

I don't know exactly what goes into the calculation of this variable, but I can import the original corby1.txt properly if I play with the "input frequency" in the import dialog. If I set it to 10.257e6, the file is imported properly. Resulting ADEV attached.

And the tolerance on that value is quite tight.  10.250e6 is too little and 10.260e6 is too much.  Both result in spikes.

I've seen this issue in the past.  It tends to show up when the number of data points per wrap is low.  When the number of points is high, everything works.  I suspected it was caused by Timelab having difficulty determining exactly when the wrap points occurred.

I never considered it to be a 'bug' but rather a 'limitation'.  As the quality of my equipment and measurements increased the issue disappeared and I haven't seen it for years.

[thinkfat]

I don't know exactly what goes into the calculation of this variable, but I can import the original corby1.txt properly if I play with the "input frequency" in the import dialog. If I set it to 10.257e6, the file is imported properly. Resulting ADEV attached.

And the tolerance on that value is quite tight.  10.250e6 is too little and 10.260e6 is too much.  Both result in spikes.

I've seen this issue in the past.  It tends to show up when the number of data points per wrap is low.  When the number of points is high, everything works.  I suspected it was caused by Timelab having difficulty determining exactly when the wrap points occurred.

I never considered it to be a 'bug' but rather a 'limitation'.  As the quality of my equipment and measurements increased the issue disappeared and I haven't seen it for years.

Naturally, if you have few datapoints per cycle you have few samples around the wrapping point. This makes it more difficult to pinpoint it. But what I see when importing corby1.txt is a different issue, the wrap is detected quite well but the cycle period is waaay off. I guess TL doesn't even attempt to find the proper cycle period, it just uses the input frequency from the import dialog.

It could be done, though. One method could be to find the wrapping point, then differentiate over some 10 samples around it and adjust the cycle period by searching for minimum standard deviation. I've used this method for finding an optimal value for the ominous "time dilation" parameter in the TAPR TICC configuration.

[tkamiya]

That's what John said.  One thing about TICC data is that it takes two rows of data to complete the transition.  Appears unwarpping code gets confused about that.  It takes one point to unwrap it and leave the other one in place.  That becomes the deep null.  I can try to find the cycle but the problem is, this is not usable for measuring.  I'm going to use HP5335A that just works.  I may try to make DIY counter as well.  I want my whole setup in one package for ease of use. 

How's yours coming along?

[tkamiya]

I'm kind of wondering about DMTD.

Now we have Corby's DMTD and some of us have it working, what can we do to move this forward?  Improve noise floor?  Make low pass filter adjustable?  What determines DMTD's performance, anyway?  I personally have a plan to make this a "one box" solution.  Package DMTD, TICC, and an adjustable hetrodyne local source.  All I need in addition is reference source and a computer.  For the local source, I am going to use 11801-6011 and have a switchable EFC control that are preset to 1Hz, 5Hz, 10Hz, and external.  For switching, mercury wetted relays will be switch controlled.  I am also wondering if I can pack a small PC and CRT into discarded oscilloscope case and make it truly a one box solution.  I have a few TINY version of Lenovo PC, and small LCD displays.

Ideas anyone?

[jpb]

I'm kind of wondering about DMTD.

Now we have Corby's DMTD and some of us have it working, what can we do to move this forward?  Improve noise floor?  Make low pass filter adjustable?  What determines DMTD's performance, anyway?  I personally have a plan to make this a "one box" solution.  Package DMTD, TICC, and an adjustable hetrodyne local source.  All I need in addition is reference source and a computer.  For the local source, I am going to use 11801-6011 and have a switchable EFC control that are preset to 1Hz, 5Hz, 10Hz, and external.  For switching, mercury wetted relays will be switch controlled.  I am also wondering if I can pack a small PC and CRT into discarded oscilloscope case and make it truly a one box solution.  I have a few TINY version of Lenovo PC, and small LCD displays.

Ideas anyone?

Just a thought - but before you build it all into a one-box solution, think about whether you want to do three-cornered-hat measurements. You may want to build 3 DMT boards and a three way splitter for the offset oscillator into the same case - plus 3 TICCs. Obviously this puts the costs up considerably but it potentially allows you to get around the need for a very good reference oscillator or at least allows you to order oscillators in terms of ADEV.

[thinkfat]

That's what John said.  One thing about TICC data is that it takes two rows of data to complete the transition.  Appears unwarpping code gets confused about that.  It takes one point to unwrap it and leave the other one in place.  That becomes the deep null.  I can try to find the cycle but the problem is, this is not usable for measuring.  I'm going to use HP5335A that just works.  I may try to make DIY counter as well.  I want my whole setup in one package for ease of use. 

How's yours coming along?

Well I've received the PCBs and parts a while ago but I'm stuck with other stuff. I'm completing a LEA-M8T breakout board first, then populating and testing the DMTD board will come next. I've meanwhile bought a Bodnar GPS reference, hoping I'll be able to use it as the offset oscillator. I've also built a passive splitter, hoping that I'll be able to use that for testing the noise floor.

Unfortunately the Bodnar GPS is not flexible enough to provide offset as well as signal source for the noise floor test so I'll have to think about something else.

[tkamiya]

I'm not sure if GPSDO is a right choice for reference, as frequency gets yanked up and down periodically.  I'm using OCXO.  It will drift but at least it's a smooth transition.  In short term, it's more stable than GPSDO.  I am now using 11801 tuned 10Hz offset.  It easily moved that much and more.

Yes, I've considered 3 cornered hat version.  My problem is slightly different signals causing beat signal by leakage.  So I don't want to put 3 different DUT into one case until I solve that issue.

As to my data issue, I'm actually having two issues.  My DUT is 11801, REF is HP105B (both 10MHz) aligned pretty well to minimize wraps, and transfer oscillator is also 11801 tuned 10Hz higher.  This annomallies correlate precisely in timing.

Top graph is a frequency difference.  Notice the flat top mountain? 
Second graph is wrapped phase data.  If you look closely, there is an inflection point in negative slope, and of course the wrapping. 

I've drew a line through them to make it clear. 

Also, if you are to imagine those parts aren't there, rest of the data is pretty much contiguous.  Frequency stays steady, goes up, and comes back down to the same or similar point.  So phase wrap is causing some kind of annomaly and every wrapping resets it.  I'm sure this is a interference issue.  The shift is about 0.001Hz to 0.002Hz.  The exact location of this inflection point is consistent during a measurement but exact location changes when new oscillator is brought in. 

This change actually makes about 1 x 10^-11 worse in adev graph.  So it's fairly significant.
(0.001Hz is 1x10^10)  I need to solve this.

[tkamiya]

Today's experiment:

Knowing something is not handling wrapped data well, I worked to eliminate the wrap in the first place.  That meant for measurement period, 1 hour, it must NOT wrap!

Using oscilloscope, I aligned DUT and reference PRECISELY.  Currently, they are drifting with respect to each other in order of 1x10^5 second per second.  At this level, adjustment of HP105B fine adjustment is such that it's not much of a turn but a very very light touch.  At this level, visual observation is reaching its limit.  I switched over to HP5335 and tried to null it completely.  Unfortunately, it wasn't possible as resolution of the knob is limited.  (wire wound multi-turn pot)

Per TimeLab phase difference graph, it moves 1 x 10^9 second per every 300 minutes.  It is drifting up then down, so i'd imagine this is a temperature shift in the room.  Both DUT, and the transfer oscillators are in individual large tupper ware box with covers on them.  I've let them sit overnight to reach thermal equilibrium.

My HP10811 board uses 723 as a pre-regulator to feed the oscillator, then 78L05 to further drop down the voltage.  It is fed to 25 turn pot.  I'm afraid tempco isn't good enough at this fine level.  But this is a circuit HP uses on its on gear.  So I assume what I will get is a typical of this oscillator.  I may have to use a hyper accurate voltage reference and super low tempco multiturn pot next.  They are very expensive!

Here's an Adev and Original phase difference at 15 minutes mark.  Cooling cycle of my A/C just started, so room temp will shift by 1 degree F shortly.

I'm going to commit this and will update at 30 minutes mark.

[tkamiya]

At 24 minutes mark, I am definitely seeing the effects of temperature change in the room.  A/C is still cooling at this moment.  Phase has shifted in plus direction by 1.5 x 10^-9 so far.  At 26 minutes mark, I hear A/C is ramping down.  Trajectory of phase graph has reversed.  Now it is on up and down, but overall unchanged. 

At 28 minutes point, A/C has ramped down further.  Phase is definitely on downward trend.  Moving rather fast.  -0.5 x 10^9 per minute.

Here's phase and Adev graph at this point.

Will be back shortly.

[tkamiya]

At 33 minute mark, phase has shifted down by 3 units in negative direction.  (3 x 10^-9)  Still no wrapping!  Adev is pretty much steady at 2.5x10^-12.  I'm impressed!  Now A/C has stopped.  Overall, that whole event resulted in reduction of Adev by less than one unit at 10^-12.  Looks like a small hill....   

At 37 minute mark, worse Adev point is evident at 3.55x10^-12.  Likely, aging might be the biggest factor at this point.  39 minutes.  Phase difference is rapidly declining, and it has now passed starting figure in opposite direction.  Phase difference is now essentially zero.  (at the start of this measurement, it was +2.5x10^-9)

At 42 minute mark.  There was a HUGE disturbance.  Frequency difference was well within 1 x 10^-11 before.  The jump measures in BOTH direction 4 x 10^10.  I have nothing to explain this change.  On Phase difference, I see a staircase effect.  This is a wrap.  Looking at a counter, it is showing 99 x 10^3.  It may have briefly touched a wrapping point and went back and forth few times. 

Here is a graph from 45 minutes mark.
 

[tkamiya]

After the jump previously mentioned (wrapping event), whole Adev curb went up by 0.5x10^-12.  Not a big deal but it did have an impact.  WHOLE graph went up.

Counter's physical display now reads 98.xxxxx 10^-3 and declining.  So close to wrapping point, there is a risk for another wrap.  At 55 minutes mark, there is no significant event.  Wrapping resulted in phase difference increasing by 1 x 10-7 and it is staying there.

Here's a set of graph at one hour mark.

This run is actually set for 3 hours.  I will report back then and upload a TIM file as well.

[FriedLogic]

I'm kind of wondering about DMTD.

Now we have Corby's DMTD and some of us have it working, what can we do to move this forward?  Improve noise floor?  Make low pass filter adjustable?  What determines DMTD's performance, anyway?  I personally have a plan to make this a "one box" solution.  Package DMTD, TICC, and an adjustable hetrodyne local source.  All I need in addition is reference source and a computer.  For the local source, I am going to use 11801-6011 and have a switchable EFC control that are preset to 1Hz, 5Hz, 10Hz, and external.  For switching, mercury wetted relays will be switch controlled.  I am also wondering if I can pack a small PC and CRT into discarded oscilloscope case and make it truly a one box solution.  I have a few TINY version of Lenovo PC, and small LCD displays.

Ideas anyone?

Just a thought - but before you build it all into a one-box solution, think about whether you want to do three-cornered-hat measurements. You may want to build 3 DMT boards and a three way splitter for the offset oscillator into the same case - plus 3 TICCs. Obviously this puts the costs up considerably but it potentially allows you to get around the need for a very good reference oscillator or at least allows you to order oscillators in terms of ADEV.

  I was thinking about trying a microcontroller dev board as a cheap way of measuring multiple channels. Some have quite a few input capture pins that link to 32 bit timers, and use internal clocks in the 100MHz region. Add a stable 1PPS and clock drift could also be corrected for.

  The Propeller Mini 32150 that Corby mentioned might be able to that sort of thing too, but I'm not at all familiar with it.
...edit: I see that it can be made to something similar, and multple channels as well!
https://www.eevblog.com/forum/metrology/a-simple-time-interval-counter-for-dmtd-work/msg3396377/#msg3396377
http://leapsecond.com/prop/index.htm#protic

[tkamiya]

Here is my 3 hour results.

Wrapping is not periodical and its impact is not that much on Adev curb.  I've noticed when wrapping happens early in first go around, it's unrecoverable.  When it's later, eventually, it settles down to a reasonable amount.

[thinkfat]

There's still something odd going on around the wrap points. See how the noise level increases directly after the wrap? It eventually goes back to where it was, but effectively you can throw away this measurement. It's now showing what your oscillator is doing, instead contains a lot of stuff that comes out of the measurement system.

PS: if you look back at the curves I posted based on your TICC test run, there is no such effect after a wrap. The data is clean once you manage to import it properly. The data from your HP5335A isn't unwrapped properly, either, and it seems there's something really odd going on in a certain time interval range. It looks like a measurement error, loss of resolution or something similar happening for small time intervals. If I were you, I'd ditch the HP, go back to the TICC and spend the extra time when importing the data.

[tkamiya]

Since last night, I changed one testing parameter.  I've set DUT and REF very VERY close to each other in frequency.  My goal was to do a 30 minutes run without a wrap.  I almost made it.  At 26 minutes, it wrapped.  Then wrapping frequency increased.

What this low difference in frequency did was, when it does cross, a little noise can cause it to roll back, then another noise can roll it again.  The slope of change is very shallow.  The graph is showing such.  So what this means is, I need to eliminate the drift even more, start half way between the phase wrap point.  But that kind of defeats the purpose of measuring.

What I don't understand is this stair case effect after the wrap.  If I eliminate this graph (which is frequency difference), it does not form a contiguous line.  AND if you look little after, it goes back to the original Y axis and continues on.  It's like frequency shifting. But it only happens after wrapping for short duration and goes back.  I can't imagine what is causing this.  It's really hard to see but there IS an inflection point in phase graph indicating this "going back to the same Y" phenomenon.

Still there is SOMETHING going on. 

I WONDER....  if noise is rather high and frequency is changing rather highly, it's going back and forth between zero point BETWEEN the collection point?  So count increases and decreases but we won't see it.  Then when the noise level is far enough, it steady increases?

That means trigger point is too low.  No.  DMTD is a comparator system isn't it?  So it will swing logic 0 to logic 1.  Comparatorr gain is too high?  Dampening factor is too low?  Low pass cut off is odd?  Did I put a wrong part somewhere?

I'll have enough parts to make two more in few days.  I wonder if I should...

[FriedLogic]

I'll have enough parts to make two more in few days.  I wonder if I should...

  If you have only made one and are having strange problems, making another one could be useful. If they're both the same, that's fine as it just means that you now have 2 of them to test the rest of your setup even faster.
  I find that when it comes to development, it's useful have 2 boards (or whatever) available as it can make it easier to find problems, as well as seeing the effects of any attempted improvements.

  I've got one made up which seems to be working well, but I am getting oscillations on the noise floor test at 50s-800s which I'm still working to eliminate.

[thinkfat]

Randomness is in the nature of noise. You will always find those oscillations, and you'll find them more frequently the closer your DUT and reference are together. There's nothing you can do about that. Even if you eliminate all measurement noise (which is impossible), there will of course still be the phase noise (or jitter) from the oscillators.

If your system setup or analysis software cannot deal with time interval wraps, tough luck. Anyway, IMHO a DMTD system is not meant to be run for days. There are better ways to analyze long term stability of clocks. Your 53132A using a Rb standard as a reference will do a better job at this.

[FriedLogic]

  I don't think it's random, although whether it's removable or not is another question. I've attached a couple of plots from one run.
  I've not used anything with this sort of single shot resolution before, so it's a new game for me to chase down the source of a picosecond or so of ripple.
  At the moment I'm just using a multi output linear supply, so improving the filtering there is an obvious thing with all the switching supplies built into everything these days.

  It does get more challenging to use them for long runs, but a 53132A is not inexpensive, and neither is a TICC - especially outside of the US. Something based on a few TDC7200s or maybe a GPX2 would work for multiple channels.

[FriedLogic]

  Is there any reason not to use an LT1012 here? I just ask because there seems to be a bit of a shortage of the LT1008 at the moment.

[5065AGuru]

Taka,

Is IMG_4824.jpg a phase plot?

Also which portion is at minimum count and which at maximum?

My guess is that it wrapped down from a low count into the max count region.

If that's the case the noise from the offset oscillator will increase and this would explain why "the noise level increases directly after the wrap"

I think I mentioned once before but I select the input jacks of the mixer so that as the DUT ages it causes the phase numbers to decrease.

Then I adjust the DUT until the counts are low and rising slowly.

Then take your data. Your phase plot will show a rising curve that is slowing down as aging progresses.

It will eventually turn around and start dropping to the wrap point.

This will trace a parabola as shown in one of my posts.

This also allows the maximum amount of time prior to wrapping.

It also has the advantage of usually staying in the lower count region longer.

Cheers,

Corby

[tkamiya]

Corby,

Yes, that is a phase plot.  Please see IMG_4797.jpg to see the relationship between phase and frequency, and in regards to timing of wrap and slope change.

You are exactly right.  The frequency had a downward trend, and wrapped backwards.  In other words, it went from MIN to MAX.  That was my misguided attempt to zero the slope.  I completely forgot it will wrap both ways....  I blame few drinks for that.  I completely understand your logic on why the change in noise level.  What I have problem with, is the fact that change are abrupt.  Of course phase change artifact is abrupt, but change in slope happens; then it traces the original slope.  We are talking about changes in 0.0001Hz, so it may have something to do with resolution.

I am not discounting the possibility of outside influence either. 

I'm starting to think wrapping to be a major issue in successful DMTD measurement.  Also random noise near MIN and MAX can cause wrap to happen multiple times in succession. 

Right now, I can't put my head around this but how about lowering resolution, so it  has wider range between MIN and MAX?  I'll sleep on this idea. 

[tkamiya]

FriedLogic,

I don't think it's random either.  I see this graph as of random noise modulated by a lower frequency sine wave.

[tkamiya]

Corby,

Having OCXO as a standard, definite long term drift exists as well as short and mid term.  Obviously, this setup cannot be used to measure Cesium for days.  In using DMTD method, what is a reasonable expectation?  I was quite happy to have 25 minutes run.  With practice, I can probably do a bit better.

[Bruce G]

Surely an LTC6957-1 could be used as a more readily available input device?
At least its Phase noise is known and is lower than that of the LTC695-3/4 variants at low offset frequencies.
It also allows some tuning of its input stage bandwidth to better suit the input signal amplitude and fequency.

[jpb]

Surely an LTC6957-1 could be used as a more readily available input device?
At least its Phase noise is known and is lower than that of the LTC695-3/4 variants at low offset frequencies.
It also allows some tuning of its input stage bandwidth to better suit the input signal amplitude and frequency.

I agree - in fact I suggested this back on page 13 of this thread though without going into the reasons you've covered. I suppose the main argument against is it is quite an expensive.
I use it in my experimentation as you can get it on a handy (if somewhat expensive) demo board.

[5065AGuru]

Taka,

I do all my runs at 5Mhz with a 1Hz offset.

This gives you twice as much time (200ns vs 100ns between wraps)

However you will still have wraps on longer plots.

Yes if the change is slow enough you can get several steps at the wrap.

This is not a problem as you can eliminate them also by the remove step function.

Say my long plot has 3 wraps and I want to use the entire plot.

I would use single step to remove one step at a time and then have the complete plot with no wraps.

I will post an example here soon of how to properly use the single step that way as it's not straightforward until you are familiar with it.

Cheers,

Corby

[5065AGuru]

Taka,

I posted the unwrapping stuff in the DMTD tutorial post.

Cheers,

Corby

[tkamiya]

Thank you!

[Bruce G]

The built in hysteresis of the MC1650/1 results in AM to PM conversion which degrades the PN performance compared to a hysteresis free comparator when AM noise is present.
Has the effect of AM noise been considered and/or measured?

[Bruce G]

I used the demo board to measure the PN at 10MHz down to an offset of 0.01Hz of the LTC6957-4 and found that it was a little lower than that plotted on the datasheet.

[5065AGuru]

Bruce,

As noted in the beginning posts:

 "This was only to scratch an itch of mine to see how the Efratom TS105 stages would work as a standalone Dual Mixer.
I was not trying to build "modern"!
The 1650 or 1651 were purchased on ebay for reasonable prices.
I also wanted something that could be easily assembled as Sometimes that scares people off."

So I did not consider AM noise effects.

I have tested it against 4 other dual mixers of different designs and MANY Quartz and Rubidium oscillators and other than a bit higher noise floor at the lowest Tau it has performed well.

Well enough that I consider the itch scratched!

Cheers,

Corby

[tkamiya]

I made a simple serial read only terminal using Arduino.  All it does is read from TTL out of Propeller board and display it on 16x2 LCD.  The code is ugly as I needed to avoid certain things to keep up with 115K baud output.  Use of strings slowed down processing so much it wasn't even keeping up with 9600 baud.  I've used 16x2 I2C LCD display and Arduino Nano board.  Connection is just SCL, SDA, VCC, GND, and serial RX.  It only uses one line.


// ------------------------------------------------------------
// Serial monitor
// ------------------------------------------------------------
//  serialMonitor v1.0
//
// reads from serial input and prints on LCD
// this code is speed optimized.  Do not use strings or or statement
// as it will not keep up with 115K serial input
//
#include <Wire.h>
#include <LiquidCrystal_I2C.h>                          // I2C LCD library
LiquidCrystal_I2C lcd = LiquidCrystal_I2C(0x27, 16, 2); // LCD type is 16x2

void setup()
{
  Wire.begin();
  Serial.begin(115200);
  lcd.init();
  lcd.backlight();
  lcd.clear();
  lcd.print("Starting DMTD...");
}

void loop()
{
  if (Serial.available())
  {
    lcd.clear();
    while (Serial.available() > 0)
    {
      char inputChar = Serial.read();
      if (inputChar != '\n')
      {
          if (inputChar != '\r')
          {
               lcd.write(inputChar);
          }
      }
    }
  }
}

[nctnico]

I think there is a major error (on my part) in acquisition screen.

The setting is DMTD 10MHz + 10Hz, 10 readings per second.
Data type is time interval.
Input frequency is 10E-6
Scale factor is 10E-7

Late to the game... If it is a Keysight counter then there can be a problem when using time interval measurements. On the Keysight 53230A for example the interval loses accuracy when it becomes negatives (wraps around). If you measure the interval (from edge to edge) from channel 1 to 2 then 2 should always come after 1. When measuring 1PPS signals using a Keysight 53230A I make sure that the result can never become 'negative' because measurements over nearly a 1 second period do not have picosecond accuracy.

Edit: checked some results; in case of a wraparound on the 53230A the error can be in the order of 200ns.

[tkamiya]

Since then, I dumped HP51132A.  It drops a count when it's about ready to wrap around.  I'm using Cory's DMTD counter described elsewhere.

[tkamiya]

During this long thread, 3 port and 4 port DMTD setup was suggested.  I'm very curious.....

Mr. Riley's famous small DTMD is a combination of 2 port versions.  What was suggested in this thread, if I wasn't mistaken, are true 3 and 4 channel where each channel gets hetrodyned via offset oscillator.  From here, I'm a bit foggy.  How do compare the result?  4 port version will end up 6 combinations thus we'll have 6 different TI results.  Now what??

Would someone who tried these share what you found out? 

[jpb]

A three port  would allow three different DUTs to be compared in a three-cornered hat way giving the individual Allan deviations (if it all worked out) or at least allowing them to be ordered to get the best.
This assumes the noise is not correlated so the measured Allan variance (square of the Allan deviation) is the sum of the squares of the two separate Allan deviations.

Four ports  would allow the reference oscillator itself to be measured against itself which would give an indication of the noise introduced by the system.

[artag]

I got a few of the PCBs made to experiment with : more than I shall need.
Happy to send them out (I've got 6 or so) for free to UK readers.
PM me.

[Mrt12]

Hallo all

I have not participated in this thread for a while and just checked today what I have missed in the meantime ;-)
there are some interesting results here. But I wonder whether someone has also tried the ECL gates, like MC100EL16 or so, instead of the MC1650?
I am stil having the phase fluctuations on my design, and I still wonder whether the ECL gates are the cause of these problems or the mixers or the mixer filters.

The ECL gates should have a 0.7ps Jitter, so I am not sure which noise floor I could expect from them.

[tkamiya]

Is anyone still playing with this?  I have been sitting on mine for quite some time.  I still can't get my head wrapped around roll-over situations.  Then a thought occurred to me....  Roll-overs are nothing less than reaching counter's highest value and restarting from zero.  Then why can't I post process this, and every time it rolls over, add the highest value and create non-rolled over value.  Keep doing it.....  Then I can have my data without roll over.  Arduino is probably capable of handling this since only 2 data comes out every second. 

If I wanted to fill a missing data, I think I can post process this and fill-in-the-gap automatically as well, like someone did in head. 

I just wanted to know if anyone made significant headway in this project.

[jpb]

Is anyone still playing with this?  I have been sitting on mine for quite some time.  I still can't get my head wrapped around roll-over situations.  Then a thought occurred to me....  Roll-overs are nothing less than reaching counter's highest value and restarting from zero.  Then why can't I post process this, and every time it rolls over, add the highest value and create non-rolled over value.  Keep doing it.....  Then I can have my data without roll over.  Arduino is probably capable of handling this since only 2 data comes out every second. 

If I wanted to fill a missing data, I think I can post process this and fill-in-the-gap automatically as well, like someone did in head. 

I just wanted to know if anyone made significant headway in this project.

I've not really started yet. I have populated most of a board but then found I was a capacitor short and need to order one but I've had no time lately (I only have a chance to do it at weekends and family commitments have meant I've had no time) but I hope to get back onto it soon.
Roll over is, as you say, just the phase essentially going through two pi so, assuming random phase shifts are much less than this, it should be possible to post-process. I guess you also need to account for negative as well as positive phase at times but it should all be possible to do in post processing - it might be easiest just to do it on a pc once the file is saved.

[tkamiya]

I didn't think of it as >2pi but you are right.  I was thinking more in terms of delta(T) that if t(2) - t(1) is significantly different and going in opposite direction, add t(1) to t(last).  I'll have to rethink this, I guess.  The final algorithm might end up with complicated scheme....

I have had my time standards turned off for quite a while to save on electricity (as I wasn't using them).  I turned them all back on again yesterday.  I'm going to have to wait a months or so to establish equilibrium again.

[capt bullshot]

I've done some frequency difference measurement experiments some time ago, including a DMTD contraption. It's got nothing to do with the original posters DMTD board here, I've started my experiments independently of this thread.

Just for the records, here's a link to what I've done: http://wunderkis.de/tnfb/index.html

Anyway, that above link is just FYI, as I may refer to it later.

I've noticed some of you were experimenting and thinking about different input stages for the DMTD. As a follow up to my previous experiments, I've built a rather simple "modular" DMTD testing system. It consists of the input stage, the mixer stage and the output (comparator) stage. As TIC I'm using the MCU part of this: http://wunderkis.de/tnfb/fdif2.html . Using todays microcontrollers, it's a rather simple job to make up a dual channel reciprocal frequency counter with a resolution of 12.5ns (as in 80MHz timer clock frequency). So I don't have any issues with logging and phase wrapping using this setup.

For the input stage, I've used "no input stage at all" (sinusoidal signal directly connected to the diode mixer inputs) and some "fast" comparators (MAX9012) that were by chance available in my stash of components.



Next is the mixer stage, I've used a Mini-Circuits diode mixer and a mixer made up using a simple XOR logic gate.




Final stage is the IF amplifier / comparator, two variants: a rather simple one and a cascade of limiter amplifiers. Component values were somewhat optimized while testing.




Attached schematic of these modules: https://www.eevblog.com/forum/metrology/dmtd-board/?action=dlattach;attach=1323681

As my intention was to measure the noise floor of different combinations of these modules, I've used the 10MHz and 100Hz offset frequency output part of this (same link as the above TIC): http://wunderkis.de/tnfb/fdif2.html  I've used a simple transformer / hybrid based power divider to split the 10MHz into both inputs.

All measurements were done at 100Hz offset frequency, reduced by aggregation to 1Hz sample rate by the TIC. As both 10MHz signals and 10MHz + 100Hz offset reference were supplied from a common source (the 10MHz OCXO), I assume this setup basically measures the noise floor of the different circuit configurations.
 
Here's the results:



All results are within the 5e-14 to 1e-13 (Allan deviation @ 1sec) ballpark, with the "best" result provided by using "no input stage" and the multi stage comparator. Otherwise, using an XOR logic gate instead of the diode mixer gives rather similar results, slightly better with the simple output comparator.

I hope it's useful and I didn't make any gross mistakes

[tkamiya]

Wow, great work!

Where did you get that interesting perf board?  I've never seen anything like that before.

[capt bullshot]

Nearly everyone who sees them, asks where I get them

- at some rare occasions I find a partially used one in the scrap box @work
- otherwise one can order them from Mouser: "BusBoards Prototype Systems", Type "SP3T-50x50-G".

[tkamiya]

This is the neatest thing I've seen!  Perfect for RF experimentation!  Thanks for the information.

Where do you work?   

[Gerhard_dk4xp]

Perhaps we should have JLCPCB make a few, with 4 pads between the vias
so that SO-8 fit in between.  :-)

I have made quite a catalogue of stamps: inv / n.i opamps, diverse regulators,
ecl / CML gates / FFs, oscillators.. Everything SMD, bottom side is GND always,
so I can solder them directly on a copper-clad board.

[tkamiya]

I was actually thinking similarly. 

Pads on top side, ground plane on bottom, no plating on vias, and two rows of connected pads at top side and bottom side.  Those connected pads will be a power bus.  What I'm not sure is, how small I can actually go on pad sizes.  I used to use 0.1" spacing single sided perf board all the time, and after few solder/unsolder sessions, they start to peel off.  It was smart of the designer of aforementioned product to use square.  They offer more adhesive area.

[thinkfat]

I was actually thinking similarly. 

Pads on top side, ground plane on bottom, no plating on vias, and two rows of connected pads at top side and bottom side.  Those connected pads will be a power bus.  What I'm not sure is, how small I can actually go on pad sizes.  I used to use 0.1" spacing single sided perf board all the time, and after few solder/unsolder sessions, they start to peel off.  It was smart of the designer of aforementioned product to use square.  They offer more adhesive area.

And they are easier to solder, too, because it's way easier to connect the heat source to the pad.

[capt bullshot]

Here's a result of the DMTD I made, test setup is:

Internal reference synchronized to the PPS output of a simple GPS module (cheap standard ublox neo-6)
Input 1: True Position GPSDO
Input 2: One of my homebrew radio clocks, synchronized to PPS output of a random cheap GPS module (not the same as for the reference)

Internal reference and homebrew radio clock both use one of the cheap OCXOs that I bought from an ebay seller, see datasheet below.

As the microcontroller based TIC has two ratiometric frequency counters and a time difference measurement, it can continously log both differences of the inputs vs. the reference and the input difference in one run. Data are logged as frequency difference, not phase. I do import the frequency difference data into TimeLab using these settings:



ADEV plot by TimeLab  ADEV values are well above the measured noise floor, looks plausible as long as I didn't make any systematic mistake. Maybe someone with more experience in this kind of measurement can have a look at the plots and evaluate the plausibility of my measurements.



This is the recorded raw frequency difference, showing an arbitrary excerpt of the logged time series data to get some visual idea of how these oscillators behave.
Yellow trace is PPS disciplined internal reference vs. the GPS disciplined radio clock
Green trace is internal vs. TruePosition GPSDO
Blue is radio clock vs. GPSDO

[capt bullshot]

Another result:

Ref and Input 2 as above, Input 1 connected to the 10MHz output of the HP53131A with Opt. 10 (HS Oscillator). So this is a free running oscillator measured vs. the PPS disciplined internal reference of my contraption:



So the 10811-60160 (which is the part number of the internal 53131A HS reference) shows better short term stability than the chinese OCXO in my radio clock.

[capt bullshot]

Last but not least in this small row of measurements:

Now I've switched the internal reference from PPS disciplined to free running. MCU firmware basically keeps the EFC input constantly at the last value after switching off the PPS source, removing instability and noise introduced by the disciplining control loop. Compared against the HP53131A internal reference, both now were warmed up for about 36h. The 53131 relies on its native thermal properties, while I've put some "highly sophisticated" thermal shielding around my DMTD (just put it into a cardboard box, and added some thermal mass above the box. Room temperature was low, but halfway constant while the observation period.

So it's a comparison of two free running oscillators (blue trace) vs. the GPS disciplined radio clock OCXO (other two traces). HP53131A internal 10811 and DMTD contraption internal chinese OCXO apparently win. Wonder if I should increase the radio clock / internal reference PLL time constant? It's not clearly defined, as I've implemented some kind of auto on the PLL loop gain, starting with a rather fast loop and then slowing down, depending on PPS to Ref. phase deviation.

[jpb]

I've finally completed one of the two boards (kindly given to me by artag who had a surplus). I've recently switched to lead-free from lead solder and had terrible trouble getting temperature right and getting it to flow (especially on the surface mount parts) and also I was rather doubtful about the parts I'd had to acquire on ebay so I was very pleased when it just worked!

I need to change my software that talks to my counter and sort out a proper offset oscillator but as a preliminary experiment I tried using my Agilent 33522A.

If I had one channel as 10MHz split and the other as 1Hz offset then I got a difference of exactly 1 Hz (not surprisingly) and the ADEV at 1sec is around 2 to 2.5 x 10^-13 (see screenshot). The counter and the DDS are both connected to one of my two Star4+ GPSDO.

I thought this was a little unfair as a test so I split the output of my other Star4+ GPSDO and used the DDS at 9.999999MHz but I get a frequency difference of around 1.6 to 1.8Hz!!

If I measure the frequency of my Star4+ against the other one the difference is way down in the mHz (see screen shot) so this is a little puzzling. To get 1Hz I had to reduce the frequency on the DDS (again puzzling as I'd expect to have to increase it??).

Anyway, the ADEV at 1sec is then around 4 x 10^-13 so not too bad.

So using DDS (even a good one) seems to amplify frequency difference from < mHz to more like 0.6 Hz - something for me to investigate.

The ADEV figures make sense if the correlation eliminates one half the overall noise (or something like that - slightly arm-waving at this stage, I've not had time to do a proper analysis - just an initial play).

[ch_scr]

I think there is a problem with the input termination on the MC100EL16D like it is shown above.
To have the 50 in between the two inputs will overload the V_BB pin, datasheet says to limit current on that pin to +-0.5mA.
I found this page showing how it is usually done.
They show this picture for differential 50 input:

And add that: "Either input in Fig. 2A can also be driven by a single-ended AC coupled sine or square wave signal,
because such a signal would swing symmetrically with respect to the non-driven input which is now biased at V_TT."
I think V_BB could be used, but input coupling would have to be designed different to achieve high enough impedance.
Edit:
Maybe this could work?

[kpjamro]

Hi - really late to the party I guess. 

I have skimmed through several hundred postings and I am a bit cross-eyed.

I would like to ask for a summary/reprise of the DMDT project please. This may help others just tuning in.

Is the design as originally posted considered to work OK?  Are any changes recommended? Parts substitutions?

If it is largely OK as-is, then, what are the key files needed? 
I have found:

"ts105v19_D.pcb" which I can open with ExpressPCB and presumably use to order boards
"dmtd.png" schematic - is this available as a .SCH file? Is it current?
"STUFFINGTOPSDMV1.jpg" top view with components
"IMG_20201015_094408.jpg" bottom components
I found  BOM in a posting - is there a project file for Digikey or Mouser?

Thanks very much.

Ken

[tkamiya]

I and few others made this DMDT by Corby.  It definitely works as advertised.  As far as I know, there hasn't been unadvertised changes beyond what is posted here.  There is also another thread for easy-to-make-and-cheap period counter that is perfect for this project. 

As to what is the latest, I don't have file names, but I suggest reading thread backwards and find out which is the latest.  Known changes are change of capacitors on back of the board for temperature co-efficient, and addition of 50 ohm register to the output.  You can tell the latest circuit because said register is hand written in.

This creation inspired others to try and make variety of DMDT.

I buy from Digi-key.  BOM didn't work for me, so I ordered everything by hand selecting.  There are two hard to find new IC.  Those came from eBay.  DBM is of course Mini-Circuits.  I used the same PCB vendor Corby used.  I recall sending them a ZIP file for traces, drills, and (there is no silk screen). 

[kpjamro]

Who is Bert? and where can I find his boards?

[kpjamro]

Thanks for the reply. I did make one of those propeller board timers.

I searched the entire forum and ts105v19_D seems to be the latest PCB.  It has all the outputs in parallel instead of the 220 ohm resistor arrangement discussed and shown in a Kicad schematic by ch_scr (but without PCB files).  The PCB would need some significant rearrangement to accommodate those resistors.

But - perhaps there is no more recent PCB file available.  And I could not find a marked-up schematic. The only schematic of this is a PNG screen print taken from the ExpressSCH.

I guess I should try a DM to Corby - I hope he does not mind.  This discussion format has no facility for managing files...

Regards

[FriedLogic]

   There is also this post from chuckb if you just want the gerbers for any PCB manufacturer to make.
https://www.eevblog.com/forum/metrology/dmtd-board/msg3278036/#msg3278036
This is the one that I made and it works well.

   There is often no need for a fast low impedance output, particularly if it's just a short link to a counter in the same case.
   I use opto isolators for the input to the Propeller counter, which reduces noise and ground loop issues. I did not have any fast ones when I was setting it up, so added an HC132 in case the counter was not too happy with the 3.5us or so rise time. It also inverted the signal again so it was the same as it originally was.

[kpjamro]

Thanks

 ... I figured out how to look at those Gerber files in Kicad Gerber viewer.  It seems that the default file extensions the program offers do not include any of the important ones, for some reason.  I guess this is the same PCB pattern as the v19D PCB. Anyway, thanks for that info.

One thing I find strange is that in input devices are in sockets.  If this is where sub-nanosecond timing variability is important, I would think you would want to avoid sockets there.

This beastie needs 4 different voltages.  For a standalone unit I was wondering if I could just buy one PSU that covered all 4 and I found the Meanwell RQ-65B which can.   I don't know if purity of the supply voltages is an issue where you need to supply it from some low-noise dc/dc regulator chips or if the Meanwell would work.  I saw a schematic by one poster that showed what seemed to be separate regulators for every stage of this circuit. No PCB was included with that however.

I am getting close :-)

[Echo88]

Are 13dBm-mixers acceptable for the pcb from Corby or should one stay with the 7dBm ones?

[5065AGuru]

Echo88,

+13dbm mixers are OK but you may have to drive them at close to that level for the best results.

One think to be careful about. The mixers used must have isolated RF and LO inputs as you are driving them differentially!

Some mixers have one side of the RF and LO connected to the case, these wont work.

Cheers,

Corby

[Mrt12]

Hi together,
has anybody made some further experiments with the ECL input stages?
I just started to think again about this DMTD topic and I still wonder why I had this strange jitter issue with my ECL gates.

[Mrt12]

Hallo all,
today I performed a very simple experiment and I believe I found the culprit why my DMTD is so unstable.

First of all, I changed the 50 Ohm input resistor for the VBB pin of the ECL gates, as suggested in an earlier post. This brought additional stability.
Then I ran some tests, during which I left the room and closed the door. After a while I went back in, and noticed that there was a sudden jump in the TI data as soon as I opened the door!
The effect was reproducible. Obviously, opening and closing the door produced a tiny air blast, that changed the temperature of the ECL gates, altering their phase.
I am annoyed that I didn't find out this earlier. The effect is reproducible when I touch the ECL chips.

I modified my entire input stage to the mixer with flying wires, and voila, now I have the desired stability of better than 1e-12 at 1 second. At least when I do this test with 2 OCXOs as signal sources at the input.
Why couldn't I find this earlier? 
and what surprises me even more is that the board from corby doesn't seem to have this effect - why doesn't the ECL comparator change its phase when the temperature changes?
anyways, it seems like the temperature coefficient of the input stage is crucial. The MC100EL16 ECL gate doesn't seem to be stable enough for this task.
Can anybody confirm this?