BG7TBL FA1 and FA2 frequency analyzers

[EV]

Here is one test more. EXT-REF is ZYT GPSDO (10 MHz) and input signal 10 MHz is from Trimble Thunderbolt GPS.

- Gate time 10 s
- Number of samples 7953
- So test time is over 21 hours
- Avg is again 0.0246 mHz high

So last two digits are there for nothing! To get them accurate the gate time must be increased.

[edigi]

What puzzles me is how this translates to ADEV measurements. The Agilent 53131A has a one-shot resolution of 500psecs, my counter - the FCA3100 has a one-shot resolution of 50psecs and this translates pretty closely to an ADEV noise floor of 5 x 10-11 at 1sec (which all makes sense).
The results of loop back tests on the FA-2 give a noise floor (apparently) of around 2.6 x 10-12 at 1 sec which implies an accuracy or one-shot resolution of less than 3 psecs which is ten times better than the top of the range Agilent counter (20psecs).
This doesn't make much sense to me.

You cannot translate between the one-shot resolution and the ADEV value. Statistical counters make several thousands of time stamps of the signal within the measurement period in order to improve resolution beyond what is possible with a one-shot measurement.
Thus the real resolution will be much better than what is possible with a one-shot measurement (assuming that the signal measured has the needed amount of cycles).

[jpb]

Here is one test more. EXT-REF is ZYT GPSDO (10 MHz) and input signal 10 MHz is from Trimble Thunderbolt GPS.

- Gate time 10 s
- Number of samples 7953
- So test time is over 21 hours
- Avg is again 0.0246 mHz high

So last two digits are there for nothing! To get them accurate the gate time must be increased.

I guess that this is the drawback of the measurements not being back-to-back.
My counter (FCA3100 = Pendulum 95?) can do back-to-back measurements so that a high measurement will be cancelled out by a low measurement (if one period is measured too long then the next will be too short by a corresponding amount).
The FA2 measures frequency and then there is a gap before the next gate so that all the measurements can be high and it seems that this is the case, though not by very much.

The FA2 still is very good value for money, a counter < £100 compared to counters that cost £1500 or more but it is useful to know its limitations.

[jpb]

What puzzles me is how this translates to ADEV measurements. The Agilent 53131A has a one-shot resolution of 500psecs, my counter - the FCA3100 has a one-shot resolution of 50psecs and this translates pretty closely to an ADEV noise floor of 5 x 10-11 at 1sec (which all makes sense).
The results of loop back tests on the FA-2 give a noise floor (apparently) of around 2.6 x 10-12 at 1 sec which implies an accuracy or one-shot resolution of less than 3 psecs which is ten times better than the top of the range Agilent counter (20psecs).
This doesn't make much sense to me.

You cannot translate between the one-shot resolution and the ADEV value. Statistical counters make several thousands of time stamps of the signal within the measurement period in order to improve resolution beyond what is possible with a one-shot measurement.
Thus the real resolution will be much better than what is possible with a one-shot measurement (assuming that the signal measured has the needed amount of cycles).

My experience with my counter is that there is a pretty good relationship between the two.
It does do time-stamping and regression frequency measurement as well as back-to-back and I've done a lot of experimentation but even extending measurements to over a million seconds and using over-lapping ADEV does not affect the magnitude, though it does make for much smoother curves.

Measuring say 10MHz it will measure (or count 10^7) cycles in a second but the end points are only accurate to 50psecs (or 33psec resolution). The regression measurement should make the frequency a bit more accurate using time stamping but my practical experience is it doesn't improve it hugely.

Perhaps I should have said resolution rather than one shot resolution (though I think the two are closely related).
This page on leapsecond.com shows the general effect:
http://www.leapsecond.com/pages/adev-avg/

Averaging statistics will reduce the standard deviation (as sqrt(N)) and give a more accurate average value but ADEV statistics don't seem to work the same way.

[EV]

I guess that this is the drawback of the measurements not being back-to-back.
..
The FA2 still is very good value for money, a counter < £100 compared to counters that cost £1500 or more but it is useful to know its limitations.

Yes it is very useful device for the money and it has no fan. I must take power plug off if I don't want to listen Agilent's fan.

I am interested if someone can do same tests. Is this error same and always in the high side in other devices?

[edigi]

Here is one test more. EXT-REF is ZYT GPSDO (10 MHz) and input signal 10 MHz is from Trimble Thunderbolt GPS.

Have you tried reversing your setup? (Using as reference what you use as input currently, and using as input your current reference; just in case...)

[EV]

Have you tried reversing your setup? (Using as reference what you use as input currently, and using as input your current reference; just in case...)

I think it has no influence to the results. Look at Reply #147. There is same 10 MHz signal connected to EXT-REF and CH1.

[edigi]

There is same 10 MHz signal connected to EXT-REF and CH1.

I must have the golden copy of FA2 then (the one that I had to fix the relay switch  ) because doing the same test overnight it's precise even in the last digit.

[EV]

I must have the golden copy of FA2 then (the one that I had to fix the relay switch  ) because doing the same test overnight it's precise even in the last digit.

Congratulations! How have you connected the same signal to EXT-REF and CH1? I am using distribution amplifier.

If I connect the signal from REF output to CH1 I get Avg error +0.005 mHz with gate time 10 s and N=60. Maybe the error decreases with more samples.
This is much less than 0.023 mHz which I got using distribution amplifier.

[jpb]

I must have the golden copy of FA2 then (the one that I had to fix the relay switch  ) because doing the same test overnight it's precise even in the last digit.

Congratulations! How have you connected the same signal to EXT-REF and CH1? I am using distribution amplifier.

If I connect the signal from REF output to CH1 I get Avg error +0.005 mHz with gate time 10 s and N=60. Maybe the error decreases with more samples.
This is much less than 0.023 mHz which I got using distribution amplifier.

I have noticed a similar phenomenon with ADEV measurements on my counter.

I have two Oscilloquartz Star 4+ GPSDOs.

If I split the output from one and feed it to the Ref in and the input port of the counter then I find the ADEV at 1 sec is around 10^-10 or 100psecs whereas the counter is 50psecs.
If I do a loop back from the ref out to the input then I get an ADEV corresponding to 50psecs (in round numbers).

The interesting thing is, that when I measure the second GPSDO against the first the ADEV at 1 second is 5 x 10^-11 (50psecs) i.e. better than the single GPSDO when split!

My explanation (which is just a guess) is that the split case is too accurate so that it always measures one lsb too high (say) and then one lsb too low and averaging doesn't make any difference as there is no dithering because of noise. While when the second is measured there is some noise which allows the averaging to remove the quantization error. The loopback also has no noise to dither but is also got no delays etc so it doesn't matter.

[edigi]

How have you connected the same signal to EXT-REF and CH1? I am using distribution amplifier.

If I connect the signal from REF output to CH1 I get Avg error +0.005 mHz with gate time 10 s and N=60. Maybe the error decreases with more samples.
This is much less than 0.023 mHz which I got using distribution amplifier.

I just used an SMA T adapter (for the FA2 converted with BNC adapter to REF input and CH1 input). As the REF input is not 50 Ohm (but some high impedance) this means even some reflections...
As GPSDO aims for better accuracy and sacrifices noise for that (the GPS disciplining means small pushes and pulls and the controlling DAC may not have enough bits + the control wire may pick up noise) it's not a good source for this (this is why I have not used GPSDO output).
The error definitely decreases with more samples. It may actually require a huge amount of samples to go entirely to zero in some cases (or it may never happen, I don't know). I've made a new (and longer) run and what is in the attached picture is an example of that.
As the measurements results are assumed to be random (although this may not be true, could be cross checked with the distribution of measurement results, it may even worth especially in your case) in case of too few samples a short bad series can bias the average. In longer case it should not happen. The 0.005 mHz error with just N=60 is not a terrible result at all (although in my case it's probably in the range of 2-3 microHz in the worst case). Especially if you use a lesser stability source like GPSDO.
If you have very high expectations, probably you're better off with a branded device that specifies in much details what you can expect from it.
E.g. FA2 does not have the resolution for low frequency signals at all.

Update: Fixed soft photo

[FransW]

"If you have very high expectations, probably you're better off with a branded device that specifies in much details what you can expect from it.'

As far as I am aware, the forum members are busy creating the specifications within the limits of their knowledge, experience and used equipment.
So far I have seen nothing from the designer(s) but for a few relatively simple explanations.
No circuit info, no design specs, no detailed description(s), no test reports or reviews from cailbration labs and/or international standards institutes, etc.

Enjoy,  Frans

It is however a nice piece of equipment.
Time-Nuttery is a very limited subject for most of us.

[jpb]

The manual does have some vague specs as to accuracy and digits/sec on different ranges but given that the whole thing costs rather less than a calibration on an Agilent counter say it is not that surprising.

It would be nice though to have some sort of clue as to the theory of operation/circuit so we could understand its limitations or at least to satisfy our curiosity.

[mino-fm]

It would be nice though to have some sort of clue as to the theory of operation/circuit so we could understand its limitations or at least to satisfy our curiosity.

Reading the datasheet you get 8 digits/s for 1 Hz signals. I suppose it is a reciprocal counter with Fref between 100 - 200 MHz. User 'hgl' observed that Fin <= 0,6 Hz reduces the update rate to 10 s and at < 0,3 Hz display shows 0 Hz.
For higher input frequencies as 1 MHz linear regession promises three more digits, so 11 digits/s are possible. Because of CPLD's max. frequency of 200 MHz it could be possible to achieve >= 12 digits/s at this input frequency.
So far theory.

I tried to implement this theory using STM32H7xx with Fref = 240 MHz. Reciprocal counter function gives stable 8 digits/s. With higher sample rate up to 1 MS/s and linear regression over 1E6 samples (time stamps) I hoped to reach higher resolution: no succes at all! Maybe some day I'll find out what's going wrong or someone shows me the right way.
STM32F7xx and TDC7200 give 10 digits/s even at low frequencies. Decreasing update rate to 10 s - 100 s resolution will be 11 - 12 digits. This is a practicable compromise without 'magic', if you like to test other frequencies but 10.00000000000 MHz ;-)

[FransW]

There is a difference between the average, the mean, the expectation value and the "true" value.
Partly because they belong to different worlds: the classical mechanical world and the quantum mechanical world.

Resolution is not strongly connected. You can spend a lifetime (or more ...) to enhance the resolution, but this will not bring you closer tot the "true" value.

We have to accept that measurement accuracy has its limits. Primarily related to the purpose of these measurements.

Frans

[jpb]

It would be nice though to have some sort of clue as to the theory of operation/circuit so we could understand its limitations or at least to satisfy our curiosity.

Reading the datasheet you get 8 digits/s for 1 Hz signals. I suppose it is a reciprocal counter with Fref between 100 - 200 MHz. User 'hgl' observed that Fin <= 0,6 Hz reduces the update rate to 10 s and at < 0,3 Hz display shows 0 Hz.
For higher input frequencies as 1 MHz linear regession promises three more digits, so 11 digits/s are possible. Because of CPLD's max. frequency of 200 MHz it could be possible to achieve >= 12 digits/s at this input frequency.
So far theory.

I tried to implement this theory using STM32H7xx with Fref = 240 MHz. Reciprocal counter function gives stable 8 digits/s. With higher sample rate up to 1 MS/s and linear regression over 1E6 samples (time stamps) I hoped to reach higher resolution: no succes at all! Maybe some day I'll find out what's going wrong or someone shows me the right way.
STM32F7xx and TDC7200 give 10 digits/s even at low frequencies. Decreasing update rate to 10 s - 100 s resolution will be 11 - 12 digits. This is a practicable compromise without 'magic', if you like to test other frequencies but 10.00000000000 MHz ;-)

I think that the approach may not be the conventional one. For a reciprocal counter of conventional sort take the Hameg HM8123 - this has a 400MHz clock yet is restricted to 10 nsec or 9 digits/sec. My own counter, an FCA3100 which is based on a Pendulum 95(? I think) has interpolation with 50 psecs resolution and does do 12 digits a second but the ADEV noise floor is around 5x10^-11 (as I'd expect from 50 psec) whilst the FA2 seems to manage ten times better at around 3 x 10^-12.
Additionally the FA2 comes up with messages apparently (I don't have one, but this is what has been reported on this forum) indicating a loss of lock in a PLL - this may be lock to the reference of course but I think it is more than that, I think it may be how the FA2 works. That is the FA2 may have a digital PLL and locks its own reference to the DUT by adjusting R and N values (if it is an integer PLL) and then calculates the frequency for the R and N values giving lock.

Riley's system uses an approach like this:
https://www.wriley.com/A%20DDS%20Clock%20Measurement%20Module.pdf

[edigi]

I tried to implement this theory using STM32H7xx with Fref = 240 MHz. Reciprocal counter function gives stable 8 digits/s. With higher sample rate up to 1 MS/s and linear regression over 1E6 samples (time stamps) I hoped to reach higher resolution: no succes at all! Maybe some day I'll find out what's going wrong or someone shows me the right way.
STM32F7xx and TDC7200 give 10 digits/s even at low frequencies. Decreasing update rate to 10 s - 100 s resolution will be 11 - 12 digits. This is a practicable compromise without 'magic', if you like to test other frequencies but 10.00000000000 MHz ;-)

I also use TDC7200 in my own counter for the interpolating part.
Page 3 of http://n1.taur.dk/permanent/frequencymeasurement.pdf shows one way how resolution can be improved.

The resolution of FA2 is pretty impressive as it can show even small frequency changes happening much less than a 100sec (probably some kind of delay line TDC is used like shown in the code example https://cas.tudelft.nl/fpga_tdc/TDC_basic.html).

In practice however as even most of the low phase noise OCXOs have too high phase noise to fulfill the stability requirement of the 12 digit/s measurement multi-seconds averaging is needed...

[FriedLogic]

My own counter, an FCA3100 which is based on a Pendulum 95(? I think) has interpolation with 50 psecs resolution and does do 12 digits a second but the ADEV noise floor is around 5x10^-11 (as I'd expect from 50 psec) whilst the FA2 seems to manage ten times better at around 3 x 10^-12.

But do we know that the FA2 does not do any averaging or other processing? If it does, it's more comparable the averaging mode on the FCA3100 which manages a much better 12 digits/sec.

My FA2 has arrived, so when I get it up and running I'm thinking of measuring the weighting across the sample period, unless someone has already done that.

[FransW]

Did you see the upgraded version yet?
https://www.aliexpress.com/i/4000120075382.html

CH1 Frequency Accuracy (?, probably resolution):
- 0.1s Gate Time: 0.001Hz@10M
- 1s Gate Time: 0.0001Hz@10M
- 10s Gate Time: 0.00001Hz@10M
             
- CH1 Test Speed: 11 bits/second@10MHz
- CH1 Power Test Range: -50DBM to +20DBM, 1M to 550M@50 OHM
- CH1 Power Test Accuracy: 0.5DB@10MHz
- CH1 Connector Type: BNC

[jpb]

Thanks for the heads-up.
The upgraded version looks exactly the same on the outside as the non-upgraded version so if you bought one it would be tricky to know which you'd got without a frequency source above 6GHz to try it out with. (My frequency synthesizer only goes up to 3GHz.)

Edit: I notice that it has the date 20190622 (22nd June 2019) after the BG7TBL which the older model doesn't seem to have.

[EV]

I ordered also this new version from eBay by make offer method. 

[edigi]

In the upgraded version probably only the prescaler has been replaced because the rest of the specification seems to be the same.
As it's on a separate PCB, it's actually pretty easy to replace it.

[FransW]

Please keep us informed and when possible with internal pictures (see edigi's post).

Note:
BG7TBL's RF contributions over the past years are highly appreciated.
He (?) makes it possible to with limited means do measurements that would require
more expensive tools when not  made available by him.
I refer to GPSDO's, attenuators, filters, pre-scalers (1000:1 f.e.), noise generators etc.
Personally I appreciate the resolution of the FA-2. Digits!

[EV]

Please keep us informed and when possible with internal pictures (see edigi's post).

It will take about one month before I get it this new version.

[mino-fm]

For a reciprocal counter of conventional sort take the Hameg HM8123 - this has a 400MHz clock yet is restricted to 10 nsec or 9 digits/sec. My own counter, an FCA3100 which is based on a Pendulum 95(? I think) has interpolation with 50 psecs resolution and does do 12 digits a second.

To test the basic resolution of reciprocal counters you better use Fin = 1 Hz. So no statistics can embellish the results.
For HM8123 the datasheet shows resolution of one LSD = (1,25 x 10e-8s x frequency) / measurement time. At 1 Hz you can expect 8 digits/s and internal Fref should be 80 MHz I guess.
I don't know FCA3100 but an absolute resolution of 50 ps will lead to 10 digits/s @ 1 Hz.

I also use TDC7200 in my own counter for the interpolating part.
Page 3 of http://n1.taur.dk/permanent/frequencymeasurement.pdf shows one way how resolution can be improved.

You can give a practical example for regression? I know this document but I failed to get any improvement using a lot of timestamps.