10 Mhz reference frequency distribution (Sine vs Square)

[(In)Sanity]

Ok,  this should be easy for everyone to understand.  I have a Nortel/Trimble GPSDO with a 10 Mhz 1v p-p sine wave output.   I need that to be around 5v p-p to feed off to SA, counter,  etc.   The question is do I keep it sine and just amplify it or convert it to a TTL logic level square wave ?    Which do you feel will have less phase noise ?   Logic kind of says maybe the square wave with a fast rise time would have less phase noise,  just not convinced and want some feedback.

Sine or square,  which is better for a reference distribution system ?

Thanks,

Jeff

[ConKbot]

Been working on this a bit at work myself. Check out the LTC6957, the datasheet, appnotes and related documents.  From the datasheet:

The logic families discussed and illustrated to this point [LVPECL, CML, 50 ohm single ended, LVDS] are generally a better choice for routing and distributing low phase-noise reference/clock signals than is CMOS logic. All of the logic types shown so far are well suited for use with low impedance terminations. Most of the time there is a differential signal when using LVPECL or CML, and LVDS always has a differential signal. Differential signals provide lots of margin for error when it comes to picking up noise and interference that can corrupt a reference clock.  CMOS on the other hand cannot drive 50? loads, is usually routed single-ended, and by its nature is coupled to the potentially noisy supply voltage half the time. The LTC6957-3/LTC6957-4 provide CMOS outputs, so it may seem surprising to read herein that CMOS is a poor choice for low phase noise applications. However, these devices should prove useful for designers that recognize the challenges and limitations of using CMOS signals for low phase noise applications. See the CMOS Outputs of
theLTC6957-3/LTC6957-4 section for further information

Also, check your SA input requirements, Ive seen lots with 0dbm, 10 dbm, or 1vpp requirements (and 50 ohm too, obviously) but I cant recall a SA that would take a TTL signal, though its not impossible. Also check your phase noise specs in the SA datasheet, to see if you can make sure that the signal youre feeding it actually has better phase noise than the internal one.    A frequency counter may take a 5V input, but obviously its only a frequency reference, and its all averaged over the gate time, so jitter wont be as critical.

If you do need 5v amplifying it may be a better approach, (even if youre clipping it some, as long as your amp is well behaved with the clipped output), as just feeding it into a inverter/buffer would skew the duty cycle if the threshold isnt exactly at the midpoint of the sine wave, and it would also convert any AM on the sine wave into jitter/PM on the output. 

There is an app note on on the 6957 for a 10 MHz clock input protection circuit, which is obviously relevant to this application.  They are using a balun to double the voltage ( and slew rate) of the sine into the differential input of the IC, which also helps reduce added phase noise.  Though I'd definitely appreciate any input from others with experience on this too.

[(In)Sanity]

Been working on this a bit at work myself. Check out the LTC6957, the datasheet, appnotes and related documents.  From the datasheet:

The logic families discussed and illustrated to this point [LVPECL, CML, 50 ohm single ended, LVDS] are generally a better choice for routing and distributing low phase-noise reference/clock signals than is CMOS logic. All of the logic types shown so far are well suited for use with low impedance terminations. Most of the time there is a differential signal when using LVPECL or CML, and LVDS always has a differential signal. Differential signals provide lots of margin for error when it comes to picking up noise and interference that can corrupt a reference clock.  CMOS on the other hand cannot drive 50? loads, is usually routed single-ended, and by its nature is coupled to the potentially noisy supply voltage half the time. The LTC6957-3/LTC6957-4 provide CMOS outputs, so it may seem surprising to read herein that CMOS is a poor choice for low phase noise applications. However, these devices should prove useful for designers that recognize the challenges and limitations of using CMOS signals for low phase noise applications. See the CMOS Outputs of
theLTC6957-3/LTC6957-4 section for further information

Also, check your SA input requirements, Ive seen lots with 0dbm, 10 dbm, or 1vpp requirements (and 50 ohm too, obviously) but I cant recall a SA that would take a TTL signal, though its not impossible. Also check your phase noise specs in the SA datasheet, to see if you can make sure that the signal youre feeding it actually has better phase noise than the internal one.    A frequency counter may take a 5V input, but obviously its only a frequency reference, and its all averaged over the gate time, so jitter wont be as critical.

If you do need 5v amplifying it may be a better approach, (even if youre clipping it some, as long as your amp is well behaved with the clipped output), as just feeding it into a inverter/buffer would skew the duty cycle if the threshold isnt exactly at the midpoint of the sine wave, and it would also convert any AM on the sine wave into jitter/PM on the output. 

There is an app note on on the 6957 for a 10 MHz clock input protection circuit, which is obviously relevant to this application.  They are using a balun to double the voltage ( and slew rate) of the sine into the differential input of the IC, which also helps reduce added phase noise.  Though I'd definitely appreciate any input from others with experience on this too.

Thanks for the feedback,  the SA in question is a Rigol DSA815-TG.  I know the phase noise can be a real concern as you pointed out with the SA.   I can cal the SA's 10 Mhz reference often against the GPSDO however I would rather just always have that reference working for me.   The SA wants 0 to +10 dbm  so 2v p-p max,  the 1v p-p I have will work,  not tried it yet.   I think I'll just go with the amplifier option as most counters will work with 2v p-p as well.   

My biggest concern was jitter which you have directed me in the right direction.   I tried to Google the subject...with no useful results.   

Some of the cheaper counters use oddball references which I figured out can just be dealt with using a simple PLL.   The phase noise of course goes up,  but cheap counter,  cheap reference clock. 

Jeff

[pa3bca]

The DSA815-TG only needs a few 100mv RMS. My 815 locks on a reference way below 0 dBm..
So 0 dBm is a nice level.
And: No way should you use a 10 MHz square for clock distribution in your lab! In my experience the harmonics will leak like crazy if you have a few instruments connected. Nice peaks on your 815 every 10 MHz till a few 100 MHz's.
I know, I tried. I also have a DIY 10 MHz GPSDO, and I have carefully lowpass-filtered the square wave. (3 or 5 pole filter, must check). Then a buffer amplifier and a 4 way splitter into 4 BNC's. harmonics > 40 dB down 

[XFDDesign]

There is certainly nothing wrong with using ridiculously-fast-edged squarewaves in a source-master distribution system. I have a commercial product which does just that, with <800ps rise/fall times. The likely difference is, each output goes to only one instrument and is either terminated _by_ the instrument, or it's externally terminated entering into the instrument. Sine-sources are more friendly to the erratic standards of various test-gear brands' Reference Input requirements simply on a reflection concern. But consider, these instruments are usually taking these signals to set their clocks to.

My suggestion, is to use your 10MHz source as a reference to a phase-locked VCXO. You can get high quality, low-jitter VXCOs from most major distributors. When you do this, you're locking a square-wave oscillator to your sine source, and skip all of the asymmetries or duty-cycle headaches. From your, now locked, VCXO source, you can drive the logic gates of your choice for distribution, but keep in mind transmission line effects. If you want sinusoidal outputs, receive your source, band-pass filter it with a home-grown 10.000MHz crystal filter, and then feed it to an array of output buffer amplifiers like the AD8000 configured in a gain-of-2 form. Put a 49.9 ohm resistor between the output of the amplifier and the transmission line.

[(In)Sanity]

Hmm,  we seem to have a bit of a conflict in a few responses.   I plan on running the signal around with times LMR-200 which I have a supply of.   So I'm still tossed up on the square vs sine,  I guess I could just run sine and square it on the far ends of the cable either phase locked or buffered.   I kind of like the phase lock idea as it would produce a bit less jitter if done right,  do it wrong and of course it'll just make things worse. 

Thanks for the feedback,

Jeff

[jpb]

I'm no expert, but the argument that I've heard for using sine in the actual distribution is that you avoid a lot of RF pollution from higher harmonics in a square wave. Also I guess if there is any dispersion in the transmission lines (unlikely to be significant at these frequencies) then square waves will be distorted while hopefully a pure sine wave will stay that way.

[pa3bca]

There is certainly nothing wrong with using ridiculously-fast-edged squarewaves in a source-master distribution system. I have a commercial product which does just that, with <800ps rise/fall times. The likely difference is, each output goes to only one instrument and is either terminated _by_ the instrument, or it's externally terminated entering into the instrument.

With a fairly fast-edged square wave as output I did see a lot of harmonics on my DSA815. Nowhere near 0 dBm of course, but when looking at very small signals near the noisefloor they were very bothersome. Possibly my coax was leaking, or the terminations in the DSA815 or the DG1032 signal generator were not 50 Ohms, or these devices themselves were leaking, who knows.
And I also did hear the harmonics on my HF transceiver. So I had to filter the output, now it is a nice sine wave.
I could not discern any difference in measured phase noise between using the internal reference of the 815 and the external GPSDO. But then again I find that locking the DSA815 to the GPSDO standard has limited value. The improved frequency accuracy with a RBW of 10 Hz ? which I use seldom because it is dead slow..

[TSL]

Having already done this, I can give you a few ideas.

If you're working with RF and low signals then sine.

If your only working with digital then you can probably do either.

For RF you want to distribute sine - especially if you're locking an SA, but before you do that you need to know if the phase noise of your GPSDO is better than your SA otherwise you're going to make your SA worse.

Likewise for signal generators - check the documentation that came with your instruments they'll often specify what sort of external reference they need.

Its handy to keep the the references of each instrument isolated to prevent ground loops or reflecting any DC bias onto the other instruments. To do this you can buy a kit from TAPR thus...

https://www.tapr.org/kits_tadd-1.html

I use one of those to feed my instruments.

I also use a high bandwidth video distribution amplifier to feed my ham radios, microwave and VLF transverters since they are less demanding than instrumentation.

The only item that is not locked to my GPSDO ( A thunderbolt ) is my HP8566B SA since the reference in it has a phase noise lower than the Thunderbolt. If I ever get a PRS10 Rubidium unit I'll lock that to the GPSDO and use its output to lock the SA since those things have a phase noise way lower than the Thunderbolts.

cheers

Tim

[XFDDesign]

With a fairly fast-edged square wave as output I did see a lot of harmonics on my DSA815. Nowhere near 0 dBm of course, but when looking at very small signals near the noisefloor they were very bothersome. Possibly my coax was leaking, or the terminations in the DSA815 or the DG1032 signal generator were not 50 Ohms, or these devices themselves were leaking, who knows.
And I also did hear the harmonics on my HF transceiver. So I had to filter the output, now it is a nice sine wave.
I could not discern any difference in measured phase noise between using the internal reference of the 815 and the external GPSDO. But then again I find that locking the DSA815 to the GPSDO standard has limited value. The improved frequency accuracy with a RBW of 10 Hz ? which I use seldom because it is dead slow..

What was your configuration for the DSA815? What was the level you measured your spurs at? I can try to replicate it.
Cranking up the sensitivity settings on mine, without anything attached to the input, this signal:


Gives me this plot:


Even for RF, the answer will be "it depends." Just for giggles, I set my HF receiver to 10.000000MHz and all I get is WWV. For this box, everything is running factory made RG-58 BNC cables and everything is terminated, except for the other 9 unused outputs from the Rubidium master/distribution box. For reference, when I stuff the signal into the '815 directly, I get this:

[(In)Sanity]

It's really starting to sound like in my particular case I'm better off just running the GPSDO to the freq counter and using that to calibrate the references of all the other gear on a routine basis.   I was only contemplating using it with all the gear because I have it and figured it would give me piece of mind.   I question that now.   I'll play around with the sine wave feeds and see if the before and after look any different in terms of noise floor,  spurs,  etc.  The counter as someone pointed out is more immune to phase noise ,  well depending on the design.   If it's the type that measures phase offset to get the sub count readings then it'll have problems as well.   I'm still shopping for a good counter,  of course I'm also trying to keep it cheap...hmm.  Guess it'll depend on if that Christmas bonus ever shows up. 

Thanks,

Jeff

[pa3bca]

What was your configuration for the DSA815? What was the level you measured your spurs at? I can try to replicate it.
Cranking up the sensitivity settings on mine, without anything attached to the input, this signal:
<snip>
Even for RF, the answer will be "it depends." Just for giggles, I set my HF receiver to 10.000000MHz and all I get is WWV. For this box, everything is running factory made RG-58 BNC cables and everything is terminated, except for the other 9 unused outputs from the Rubidium master/distribution box. For reference, when I stuff the signal into the '815 directly, I get this:
<snip>

Did you connect the input of the SA to anything? It looks suspiciously flat. So any leakage is not getting in via the external reference input of the DSA815. When I connected my DSA to devices (DUT's) in my "lab", with the DSA815 and the signal generator (DG1032Z) locked to the GPSDO I certainly saw the harmonics on the SA. Possible below -90 dBm (can't remember exactly). So either radiation from not-so-good RG58 cables, radiation leaking through the siggen, or..

Unfortunately I cannot easily reproduce my earlier measurements.

[tggzzz]

Ok,  this should be easy for everyone to understand.  I have a Nortel/Trimble GPSDO with a 10 Mhz 1v p-p sine wave output.   I need that to be around 5v p-p to feed off to SA, counter,  etc.   The question is do I keep it sine and just amplify it or convert it to a TTL logic level square wave ?    Which do you feel will have less phase noise ?   Logic kind of says maybe the square wave with a fast rise time would have less phase noise,  just not convinced and want some feedback.

Sine or square,  which is better for a reference distribution system ?

Think in basic terms.

Any extra circuit between source and destination will introduce some thermal noise, i.e. jitter i.e. phase noise, and possibly interference. The only exceptions to that are if

• the extra circuit has a high degree of narrowband filtering (e.g. a good PLL), or
• interference can be introduced between the source and destination, and the new circuit reduces the effects of the interference (e.g. a differential digital clock)

[(In)Sanity]

Ok,  this should be easy for everyone to understand.  I have a Nortel/Trimble GPSDO with a 10 Mhz 1v p-p sine wave output.   I need that to be around 5v p-p to feed off to SA, counter,  etc.   The question is do I keep it sine and just amplify it or convert it to a TTL logic level square wave ?    Which do you feel will have less phase noise ?   Logic kind of says maybe the square wave with a fast rise time would have less phase noise,  just not convinced and want some feedback.

Sine or square,  which is better for a reference distribution system ?

Think in basic terms.

Any extra circuit between source and destination will introduce some thermal noise, i.e. jitter i.e. phase noise, and possibly interference. The only exceptions to that are if

• the extra circuit has a high degree of narrowband filtering (e.g. a good PLL), or
• interference can be introduced between the source and destination, and the new circuit reduces the effects of the interference (e.g. a differential digital clock)

This is all really good feedback for me,  from your post and those before.   I've always been more of a Volt Nut then a Time Nut,  however I kind of want to join that camp as well.   I'm going to just go with calibrating the gear from the GPSDO for now and feed the sine wave from it to a counter.   Still need to buy the counter.   I've already done the zero drift technique with two channels of the scope which works out pretty well.  My office/lab stays pretty temp stable,  so overall not bad.

Jeff

[SeanB]

Probably the best is to use a sine wave and distribute using a regular video amplifier converted to 50R operation, and then at the equipment end use a crystal oscillator locked with the reference. That way you have a stable 10MHz and the low drift of the reference as well, and if there is no reference then it simply free runs at the crystal short term drift rate. All you need is a lock indicator which can simply be a detection of the cable delivering the reference signal voltage.

[(In)Sanity]

Probably the best is to use a sine wave and distribute using a regular video amplifier converted to 50R operation, and then at the equipment end use a crystal oscillator locked with the reference. That way you have a stable 10MHz and the low drift of the reference as well, and if there is no reference then it simply free runs at the crystal short term drift rate. All you need is a lock indicator which can simply be a detection of the cable delivering the reference signal voltage.

The short term (24 hour) drift on the Nortel/Trimble GPSDO without a lock is better then I actually need,   it doesn't unlock all that often either.  I should get a Rubidium standard as well just to compare it to.  Doesn't look like a Christmas bonus this year so my counter choices will be more limited.   I don't really need one for proper calibration however it would be nice.   I do have one on the 815 and one on the 2465BDM as well as the DS2000A series.   None of them however do sub hz resolution.    I'll just keep doing the zero drift method for now as it gets me in the sub ppm range without much effort.  Without a counter I have no reason to keep the GPSDO on 24/7.   If I want to calibrate something I can have it locked in half an hour or so,  depending on various factors. 

Jeff