Re: FeelElec New Arrival FY-6900 Signal Generator

[TurboTom]

In case of the DG800, I checked all the available options: In both square and pulse mode, regardless of selected frequency, the slope observed after 10ms looks identical, as shown in the screenshot above. Since both references (AWG and Scope) were free-running (i.e. not locked to an external reference), I guess what's observed is rather the phase noise of the two internal references over 10ms vs. each other than the phase jitter of the generated waveform. The sampling clock aof 250MHz can be clearly observed when looking at arbitrary waveforms with steep edges.

Attached you see a "square" wave slope (1ns timebase) of slightly below 10MHz (to eliminate whole-number factor vs. sampling frequency effects), sampled over about one minute with infinite persitance. Trigger point is one period of the square wave to the left.

Once again, just for comparison...

[rf-loop]

In case of the DG800, I checked all the available options: In both square and pulse mode, regardless of selected frequency, the slope observed after 10ms looks identical, as shown in the screenshot above. Since both references (AWG and Scope) were free-running (i.e. not locked to an external reference), I guess what's observed is rather the phase noise of the two internal references over 10ms vs. each other than the phase jitter of the generated waveform. The sampling clock aof 250MHz can be clearly observed when looking at arbitrary waveforms with steep edges.

Attached you see a "square" wave slope (1ns timebase) of slightly below 10MHz (to eliminate whole-number factor vs. sampling frequency effects), sampled over about one minute with infinite persitance. Trigger point is one period of the square wave to the left.

Once again, just for comparison...

In my previous msg I try carefully tell why it is better to use settings where things are so that we compare apples to apples.

This other user image (Owon scope) was 100Hz aka 10ms period, cycle to cycle jitter.

Now you show around 10MHz signal, previously around 1MHz. 

What I prevuiously told different frequencies different jitter.  Many times it is not not at all same to do it with 10MHz and after 10ms drelay or 100 Hz and there 10ms, and more better if avoid these golden frequencies and use some bit off like you just did with 10MHz.

Rigol DG900
Jitter (rms) 
Typical (1 Vpp)
≤5 MHz: 2 ppm of the period + 200 ps
>5 MHz: 200 ps

Same in Sig 1000X
Jitter (rms), Cycle to cycle
300 ps +
0.05 ppm of period

now 10ms period have 10 000 000 ns.

2ppm from it is 20ns.  This is why 100Hz can look extremely different.
And it is rms...  depending distribution peak value can not know without information about distribution. It is not always gaussian.

So how this 100Hz jitter looks like with DG800/900  compared to this @masterx81 test.

Arrgh...  I do not have scope now available...

[TurboTom]

@rf-loop --
of course, you are right. I attached a measurement with a square wave set to 99.999 280 Hz. The slope observed is one period after the triggering slope. I selected the amplitude to 12Vpp, high impedance but connected via a 50R BNC cable. The AWG input absorbs the reflected wave. If I set a higher amplitude on the AWG, rise time will be approximately doubled. It seems, the AWG firmware takes care of the finite slew rate of the output drivers in order to keep ringing and overshoot under control. No question, the FY6900 is capable of faster slopes and a (slightly) higher output amplitude (24Vpp vs. 20Vpp).

What's interesting to observe is that apparently the phase jitter of the output signal of the DG800 / 900 series only depends on the time delay between the trigger and the observed slope. It appears to perform way better than the spec, both the period-dependent figure as well as the absolute, "fast" figures.

Edit: As an additional observation, I added in the same screenshot a trace of the second channel, outputting a square wave of "arbitrary origin", i.e. without "controlled slope technology" (Rigol calls it "SiFi II"). The jitter originates from the 250MSa/s sampling frequency plus a little phase noise and what not. The rise time dropped from ~10ns to ~6ns, but overshoot and ringing are way worse. But even under these conditions, the jitter is still much better than spec.

[radiolistener]

Just tried to test PSG9080 with 10 ms delay after trigger, but unfortunately my oscilloscope SDS1102X doesn't support trigger delay more than 10000 divs time, for 5 ns/div the maximum delay is 50 uS.

Here is my test for PSG9080, taken with 1 meter RG58 with 50 Ohm termination (to avoid cable dependency).
Screenshots taken with iinfinite persists and 1 minute accumulation.

1) Square wave 100 Hz with 10 ms trigger delay and 1 us/div (minimum time per div for 10 ms delay on my oscillsocope)

2) Square wave 9'999'280 Hz with 99.8 ns trigger delay and 2 ns/div

3) Square wave 10'000'000 Hz with 20 us trigger delay and 2 ns/div

[rf-loop]

Just tried to test PSG9080 with 10 ms delay after trigger, but unfortunately my oscilloscope SDS1102X doesn't support trigger delay more than 10000 divs time, for 5 ns/div the maximum delay is 50 uS.

Here is my test for PSG9080, taken with 1 meter RG58 with 50 Ohm termination (to avoid cable dependency).
Screenshots taken with iinfinite persists and 1 minute accumulation.

1) Square wave 100 Hz with 10 ms trigger delay and 1 us/div (minimum time per div for 10 ms delay on my oscillsocope)

2) Square wave 9'999'280 Hz with 99.8 ns trigger delay and 2 ns/div

3) Square wave 10'000'000 Hz with 20 us trigger delay and 2 ns/div

Why you do not use 2nd time scale in 1st image. Or is it so that it do not work in this old X model.

[rf-loop]

@rf-loop --
of course, you are right. I attached a measurement with a square wave set to 99.999 280 Hz. The slope observed is one period after the triggering slope. I selected the amplitude to 12Vpp, high impedance but connected via a 50R BNC cable. The AWG input absorbs the reflected wave. If I set a higher amplitude on the AWG, rise time will be approximately doubled. It seems, the AWG firmware takes care of the finite slew rate of the output drivers in order to keep ringing and overshoot under control. No question, the FY6900 is capable of faster slopes and a (slightly) higher output amplitude (24Vpp vs. 20Vpp).

What's interesting to observe is that apparently the phase jitter of the output signal of the DG800 / 900 series only depends on the time delay between the trigger and the observed slope. It appears to perform way better than the spec, both the period-dependent figure as well as the absolute, "fast" figures.

Edit: As an additional observation, I added in the same screenshot a trace of the second channel, outputting a square wave of "arbitrary origin", i.e. without "controlled slope technology" (Rigol calls it "SiFi II"). The jitter originates from the 250MSa/s sampling frequency plus a little phase noise and what not. The rise time dropped from ~10ns to ~6ns, but overshoot and ringing are way worse. But even under these conditions, the jitter is still much better than spec.

Thanks. Perhaps this SiFi II  is something like Si have done long time ago... who knows.
So or so but in this single test looks like better than  data sheet worst case limits what perhaps are there but need find - if need.
I do not have here things for tests but here one old image about SDG1000X

Around in same ball park related to c-c jitter but in SQR  risetime small difference - naturally.


Your image for compare
btw, how this freq is under 10MHz and delay is under 100ns and both are based to same clock
ETA: oh... and now I hit scope is also stopped...hmmh



Old Si SDG1kX from some my old tests.

But then for every random readers. This is NOT for start any competition between these! Different things for different needs. All have some good and some bad things. What is best for one depends always hist personal needs and budget for hobby or work. Same for every piece of things. Kitchen machines, cars, test instruments and what ever, also clothes and food. There is not universal best and never come.

[radiolistener]

Why you do not use 2nd time scale in 1st image. Or is it so that it do not work in this old X model.

What is 2nd time scale? Do you mean sequence sampling?
I don't use it, because it works slower and it looks worse from my point of view and it's not clear for me how it's implemented

[Johnny B Good]

@radiolistener,

 You might have more luck with the 'scope's timebase zoom feature looking at this jitter. Just a thought.

John

[rf-loop]

@radiolistener,

 You might have more luck with the 'scope's timebase zoom feature looking at this jitter. Just a thought.

John

Yes as I say previously, using second time scale and because there is not full screen runtime zoom it can use splitted window zoom where other window have different time scale (based to same timebase!)

Of course in first place I can use also timebase zoom but with some reason I did not (And I do not anymore remember also how this Old obsolete X model works exactly in this case) ... we need slowly learn thinking timescale instead of timebase. In analog scopes world we did timebase change and dual timebase zoom and dual time base dual trig and delayed zoom and what ever we like but mostly not in common digital scopes. These are different things.  Yes also I use still mostly term "timebase" but slowly I need learn away this thinking because it is not true. We do not even have delayed or trig after delay second timebase zoom... we have only second time scale display, splitted window or whole window "zoom" like old Owon did in history, dual time scale zoom in "feature" what was laughed by lot of peoples.
 

[radiolistener]

You might have more luck with the 'scope's timebase zoom feature looking at this jitter. Just a thought.

Hm... yes, you're right. In Zoom mode aka dual time scale SDS1102X allows to see 10 ms delay.
I never used Zoom feature before and thought it's complete useless. But in such case it helps

See picture in attachment for 10 ms delay.
PSG9080 generator, square wave 6 Vpp, 100 Hz.
The same 1 meter RG58 coax cable and 50 Ohm termination.
Infinite persist mode with 1 minute accumulation.

[rf-loop]

You might have more luck with the 'scope's timebase zoom feature looking at this jitter. Just a thought.

Hm... yes, you're right. In Zoom mode aka dual time scale SDS1102X allows to see 10 ms delay.
I never used Zoom feature before and thought it's complete useless. But in such case it helps

See picture in attachment for 10 ms delay.
PSG9080 generator, square wave 6 Vpp, 100 Hz.
The same 1 meter RG58 coax cable and 50 Ohm termination.
Infinite persist mode with 1 minute accumulation.

Excellent. Thanks.

[TurboTom]

Hm... yes, you're right. In Zoom mode aka dual time scale SDS1102X allows to see 10 ms delay.
I never used Zoom feature before and thought it's complete useless. But in such case it helps

See picture in attachment for 10 ms delay.
PSG9080 generator, square wave 6 Vpp, 100 Hz.
The same 1 meter RG58 coax cable and 50 Ohm termination.
Infinite persist mode with 1 minute accumulation.

 

One thing to keep in mind when when doing high-resolution (timing) measurements after long delays with non-synchronized timebases: You're always observing the correlation of all the phase noise / jitter of both timebases involved, the generator's and the scope's. In order to characterize a single instrument, you would need the "counterpart" to be known and proven at least an order of magnitude better than the one to be characterized. Which is rearely the case with these (mostly) entry level instruments, unfortunately...

[radiolistener]

for comparison, here is ROJON TCXO 96 MHz with the same trigger delay setting.

TCXO output pin connected directly to SMA connector with almost zero line length.
SMA connector is connected directly to oscilloscope through SMA-BNC adapter.
Power supply chain: KORAD KA3005D 5.00V => LM2931 3.3V => tantal 100uF => ferrite bead => tantal 47uF => ceramic capacitors => TCXO

[Labrat101]

Hi Again ,
There seems that every one has put a lot of effort into this .
It would be more interesting if the results were taken from the actual output pins of the cyclone .
Ie the PSG9080 has a genuine cyclone chip . Were as the FY has a Clone Cyclone and on the new release
of the fy69 some one actually upload a photo of the chip . which has 'Feeltec Pro' stamped on it ..
 So there can't be any comparison between an Apple and a Lemon.
 If some one has Fy & another AWG with a genuine cyclone which to the spec's of the Cyclone does not have a Jitter problem
 due to it inbuilt patterned clock control circuit. The Clone version does not have this circuit .
 Also the Fy PCB has a layout mistakes that were not corrected and probably will never be done as the PCB would need a reprint.
which was corrected in many other models by other Makers . Of course this increased the cost .
 So All tests on other AWG's will never be able to duplicate this Error .
Therefore:. The 4ns Jitter will always be present on the clone version. over the full spectrum range .
 

   @ radiolistener
 That is a healthy drift that would be expected to be seen from a TCXO . shows the temp control is working

[radiolistener]

Ie the PSG9080 has a genuine cyclone chip

PSG9080 uses Xilinx Spartan 6 XC6SLX9 instead of Altera Cyclone IV

[Labrat101]

Ie the PSG9080 has a genuine cyclone chip

PSG9080 uses Xilinx Spartan 6 XC6SLX9 instead of Altera Cyclone IV

  Even Better     My Bad my error I meant to write genuine chip

[battlecoder]

Well you could all have done this fantastic analysis of the PSG9080 before 11.11. 
What I'm going to do now with the Uni-T I purchased? 

[masterx81]

Tried every trigger setting, but at 100hz and watching the rising edge 10ms after the trigger point the jitter is always the same. I've also doubt about the timebase of my scope, but by now i not have other ways to check it.
The psg9080 seem a quite promising device. I'm only a bit upset as i've spent a lot of time in this fy6900, and seem not so well spent 🤬

[Johnny B Good]

for comparison, here is ROJON TCXO 96 MHz with the same trigger delay setting.

TCXO output pin connected directly to SMA connector with almost zero line length.
SMA connector is connected directly to oscilloscope through SMA-BNC adapter.
Power supply chain: KORAD KA3005D 5.00V => LM2931 3.3V => tantal 100uF => ferrite bead => tantal 47uF => ceramic capacitors => TCXO

 Since that TCXO looks exactly like the JYEC 50MHz unit fitted to a "Hi-Fi clock board" pictured here:-

https://www.aliexpress.com/item/32962398016.html?spm=a2g0o.detail.1000013.5.56363e63yTUr7u&gps-id=pcDetailBottomMoreThisSeller&scm=1007.13339.169870.0&scm_id=1007.13339.169870.0&scm-url=1007.13339.169870.0&pvid=410d0e76-b02e-45aa-8020-b3a4f3048a97&_t=gps-id:pcDetailBottomMoreThisSeller,scm-url:1007.13339.169870.0,pvid:410d0e76-b02e-45aa-8020-b3a4f3048a97,tpp_buckets:668%230%23131923%2353_668%23808%234094%23330_668%23888%233325%234_668%234328%2319940%23905_668%232846%238112%23575_668%232717%237562%23480_668%231000022185%231000066059%230_668%233468%2315613%23482

 but purchased from an ebay seller two years ago for almost exactly the same price and being customisable to any end user's  frequency requirements and is at an even higher frequency of 96MHz, it's quite obviously a DDS type which is likely responsible for most of that jitter you can see. I suspect that if you select a less delayed zoomed segment to examine (say 1ms), the jitter will be reduced (possibly to just 10% of the 10ms delayed segment?).

 I'd bought the 50MHz TCXO clock module intending to salvage the JYEC oscillator to directly (using short wire stilts) replace the original 50MHz ten cent smd XO used on the FY6600's main board. Luckily, I discovered the blazing hot 55 to 60 degree C operating temperature of the smd XO in time to avoid taking the oscillator board apart and installed the whole module above the intake fan I'd installed into the base of the AWG to act as a deflector and stay within 2 deg of room temperature to avoid the rather wild temperature excursions it would have suffered If I'd gone ahead with my original plan.

 I recently tested that module (long since usurped by the current 10MHz OCXO upgrade with a 3N502 multiplier chip sat where the original smd XO chip had been) in order to quash the myth that the metal lidded smd 10MHz VCXO used in the SDG2000X series is a VCTCXO. It isn't - it's just a high quality VCXO allowing the user to re-calibrate it via the UI. The saving grace in this case being the presence of an external 10MHz reference socket. Unfortunately, that socket is bi-directional but I won't bore you with the details, suffice to say I hope Siglent will be interviewing job applicants for the position of chief UI developer sometime very soon.

 The biggest drawback with those TCXOs is the backlash in the trimmer adjustment. If that ROJON TXCO is anything like my JYEC one, it'll meet its 0.1ppm temperature stability spec with ease.   The backlash in the calibration trimmer makes achieving better than (a quite achievable) 25ppb accuracy a bit of a black art, involving as it does, skilful application of the trimming tool and a large dose of luck. It can be done, given enough patience and a stable calibration frequency reference source to adjust against but it's no easy task.

 I'm in the middle of quantifying the ionospheric induced timing errors in the GPSDO's output (some 4 to 6ns pk-pk phase modulations peaking around the 2 to 3mHz rate otherwise I would have set that 50MHz TCXO clock board up again to examine it using the zoomed timebase option (50ms or more delay) to see for myself (I'd only directly compared the exported internal 10MHz clock of the Siglent to my example of an actual TCXO so only saw  "jitter free" signals).

 I'll set that experiment up after I've finished my current test run with the GPSDO versus the sine wave output of the FY6600-60M locked to the RFS in order to let me dial in uHz offests to trim out the frequency drift between the GPSDO and the RFS, allowing me a more accurate view of these 4 to 6ns phase modulations imposed by the varying state of the ionosphere (the major source of timing errors with a single frequency GPS timing receiver such as the M8T I'm currently using - I can't really justify the 200 quid plus of a ZED9 right now ).

 Since the use of a cheap AWG is still on topic, I've attached a series of screenshots showing just how effective a tool these cheap AWGs (well my FY6600 at least) can be once the curse of the ten cent XO has been replaced by the blessing of an OCXO that can be locked to an external reference (the RFS in this case).

 The first five (note the date & time bottom RH) is a sequence started some 30 or 40 minutes after making a 40uHz adjustment to the FY6600 (representing the RFS, magenta trace). I'd noticed the frequency shift with the increase of room temperature from the central heating being turned on an hour earlier, corrected the offset and promptly forgot all about it. The last five were a similar sequence but I'd gotten distracted from making the initial screenshot some 42 minutes earlier by being called for tea.

 I'm triggering from CH2, the RFS since it's the GPSDO on CH1 that's doing the 4 to 6ns waltz. Also, it's worth bearing in mind the roughly half nanosecond's worth of jitter introduced by using the FY6600 as a frequency correction intermediary for the RFS (which will have some small amount of jitter of its own). I've been subtracting 0.5ns off the apparent pk-pk phase deviations to get a more accurate assessment.

 Incidentally, the SDG2000X series in spite of having an external clock reference socket, simply cannot be used for this type of measurement for lack of digits by which to set its output frequency (8 using the rotary encoder and digit cursor buttons or else 11 if using the keypad - 5 short of what you need for (the claimed) 1uHz resolution at or beyond 10MHz). Siglent have got a mountain to climb before they're even on a par with Feeltech's UI (which seems to have benefited from the absence of a keypad!). It seems, barring a catastrophic failure, that my much modded FY6600 has a lot more useful life left in it despite the acquisition of an SDG2042X

John

[rf-loop]

FY6900-60

I have looked two individual units. Here better one. Other one initial freq off is well over 30ppm.
This one is "only" 22ppm after it have warmed up. Its xtal reference stability is quite poor.
In my position, at this time until I am back in my homeland, I  do not have nearly any test equipment.
Here is FY start up drift and after then some random drifting around in normal room without notable temp changes in test period.

Note that in image scale 100 is  10 000 000Hz  and measuring receiver freq accuracy is far better than 0.05ppm and drift/jitter performance for this purpose so good that I can say as rolls royce company... "enough".  So this is not even visible in this scale.

For my use these FY6900 toys are totally in class "utterly useless".  UI and design is "nightmare" like some just out from school kid engineer designed and made in some car garage floor.
Example just one single detail. AM modulation is something what looks that designers have not any idea for what they are in generator and then example modulation depth is totally out of order. Yes there is wrong math. And UI. How a hell they think it can use in practice for anything. Example if you adjust modulating freq, whole modulation need stop. Then you set new Mod freq (other channel is modulating) after then you start MOD function again.  If you do not use external modulation source ... well two channel generator can do just one channel AM out. So can not produce even two channel independent AM. And when try adjust MOD frequency whole mod drop off before it can adjust.
And then, when adjust modulation depth it change also carrier level. What a hell. If set 100% mod carrier drops 6dB related to 0% mod depth. (even this I can accept if it then stay fixed...but shit, it do not) and if set other mod% it also move carrier level.  Then if look example AM side bands levels... they are totally off. They did not know what are sideband levels related to carrier when modulation depth is x%. What they have done in school... if they have been there.
Right levels are

100% -6dB
50%   -12dB
10%   -20dB
2%    -40dB

And as can see in image they are off. And error is not small. I did not take under 2% mod levels because perhaps not so important and also used spectrum have not enough traces for one image.

Whole gen is total nightmare in UI and in HW for any serious purpose even if it have many dingle dongle "features".


(FY6900-frequency-drift-after-start-and-out-of-specs-initial-offset-pub.png)


(FY6900-random-frequency-drifting-in-normal-roughly-constant-room-temp-pub.png)


(FY6900-AMmod-100-50-10-2-levels-pub.png)

[TurboTom]

...
 
And then, when adjust modulation depth it change also carrier level. What a hell. If set 100% mod carrier drops 6dB related to 0% mod depth. (even this I can accept if it then stay fixed...but shit, it do not) and if set other mod% it also move carrier level.

...

This is something that you will find with (virtually) all AWGs. When applying AM, in the AWGs, the maximum Vpp level is kept constant, i.e.

Vout(t) = A * sin(w_carrier*t) * (1 + m/2 * (sin(w_mod*t) - 1)) -- assuming both sinusoidal carrier and modulation signals and modulation depth m =[0..1]

The maximum peak amplitude will be just A. And this approach makes good sense in case of an AWG. You simply don't want to "overdrive" the attached DUT and prevent it from possible damage by supplying a voltage level that's beyond your preset limit.

You simply cannot expect the characteristics of an RF signal generator from an AWG, it's a different instrument, made for different applications, though some AWGs are getting close ... both price-wise as well as frequency-wise.

With your other points, I completely agree.

[masterx81]

In my unit i have a 0.9ppm tcxo. So the jitter came elsewhere...

[Labrat101]

Quote from: masterx81 on Today at 04:35:25 pm

In my unit i have a 0.9ppm tcxo. So the jitter came elsewhere...

I guess you meant..  So the jitter comes from elsewhere...
Sorry But   Yes .

[rf-loop]

...
 
And then, when adjust modulation depth it change also carrier level. What a hell. If set 100% mod carrier drops 6dB related to 0% mod depth. (even this I can accept if it then stay fixed...but shit, it do not) and if set other mod% it also move carrier level.

...

This is something that you will find with (virtually) all AWGs. When applying AM, in the AWGs, the maximum Vpp level is kept constant, i.e.

Vout(t) = A * sin(w_carrier*t) * (1 + m/2 * (sin(w_mod*t) - 1)) -- assuming both sinusoidal carrier and modulation signals and modulation depth m =[0..1]

The maximum peak amplitude will be just A. And this approach makes good sense in case of an AWG. You simply don't want to "overdrive" the attached DUT and prevent it from possible damage by supplying a voltage level that's beyond your preset limit.

You simply cannot expect the characteristics of an RF signal generator from an AWG, it's a different instrument, made for different applications, though some AWGs are getting close ... both price-wise as well as frequency-wise.

With your other points, I completely agree.

Yes for AWG I can accept also this principle that keep modulated total (sidebands and carrier sum) for peak level.
I know many AWG works like this.  Exept Siglent after they finally change it in some previous FW update.


But.

In FY6900 situation is total mess with this moving carrier level + same time wrong modulation %.
They move carrier when AM% change, it can see idea is perhaps just this "keep peak" principle. If it do ALONE this it can accept - just like example in Rigol DG800/900 or some  Siglent models. Not so nice but still acceptable and in some other kind of use just ok.

But then in FY6900 sidebands level related to true current carrier level is totally out of order. If user do not have things for measure modulation depth, using this % setting he get wrong true %.

When I look example DG800 sidebands level related to carrier (dBc) is - perfect.   Just tested DG using 0dBm out and AM from 100% down to  0.06%   sidebands -70dBc  and below this my measurement can not say sure due to noise (because I need use 50dB attenuator between generator and test system) but perhaps even under 0.06% still quite ok . With this DG800, it is easy to correct carrier level if it need keep in some specified level with higher modulation %. If need correct carrier level it is simple just correct level. And all adjustments can do without interrupting modulated signal (except some very tiny glitches)
It follow math just perfectly.

But with FY need some  table where can find what AM% give wanted AM%. If I need this... I feel it is better to throw it out from window before total burn out with it and do things with lot of better tool without even enormous price difference.

[rf-loop]

FY6900-60

I have looked two individual units. Here better one. Other one initial freq off is well over 30ppm.
This one is "only" 22ppm after it have warmed up. Its xtal reference stability is quite poor.
In my position, at this time until I am back in my homeland, I  do not have nearly any test equipment.
Here is FY start up drift and after then some random drifting around in normal room without notable temp changes in test period.

Note that in image scale 100 is  10 000 000Hz  and measuring receiver freq accuracy is far better than 0.05ppm and drift/jitter performance for this purpose so good that I can say as rolls royce company... "enough".  So this is not even visible in this scale.

For my use these FY6900 toys are totally in class "utterly useless".  UI and design is "nightmare" like some just out from school kid engineer designed and made in some car garage floor.
Example just one single detail. AM modulation is something what looks that designers have not any idea for what they are in generator and then example modulation depth is totally out of order. Yes there is wrong math. And UI. How a hell they think it can use in practice for anything. Example if you adjust modulating freq, whole modulation need stop. Then you set new Mod freq (other channel is modulating) after then you start MOD function again.  If you do not use external modulation source ... well two channel generator can do just one channel AM out. So can not produce even two channel independent AM. And when try adjust MOD frequency whole mod drop off before it can adjust.
And then, when adjust modulation depth it change also carrier level. What a hell. If set 100% mod carrier drops 6dB related to 0% mod depth. (even this I can accept if it then stay fixed...but shit, it do not) and if set other mod% it also move carrier level.  Then if look example AM side bands levels... they are totally off. They did not know what are sideband levels related to carrier when modulation depth is x%. What they have done in school... if they have been there.
Right levels are

100% -6dB
50%   -12dB
10%   -20dB
2%    -40dB

And as can see in image they are off. And error is not small. I did not take under 2% mod levels because perhaps not so important and also used spectrum have not enough traces for one image.

Whole gen is total nightmare in UI and in HW for any serious purpose even if it have many dingle dongle "features".


(FY6900-frequency-drift-after-start-and-out-of-specs-initial-offset-pub.png)


(FY6900-random-frequency-drifting-in-normal-roughly-constant-room-temp-pub.png)


(FY6900-AMmod-100-50-10-2-levels-pub.png)

After this FY6900 I look other generator. (DG800 "DG992")


I looked bit about its internal reference stability. Drift after startup and after it have roughly temperature equilibrium in room in its place. Place was also exactly same where FY6900 was when tested. Same stability in room temp etc.

DG is really and totally like from different universe.
Due to its reference much higher stability I use different scale. When small div was 0.1ppm in FY test, now in this test small div is 0.01ppm. (and test freq 100MHz) There is also other difference.The frequency offset scale is inverted so that trace go right it mean frequency move down and vice versa. Just opposite with FY test.

Then other thing. AM modulation.
As told I do not like when carrier levvel change depending AM modulation depth. Of course this problem is not meaningful with low modulation depths. But with higher modulation % it change more and need note. Finally with 100% modulation carrier drop of course 6dB. But in this generator sideband levels related to carrier true level are all ok. Need say - perfect. 
If you have 0dBm out and turn AM mod on and 100% just turn output level 6dB up then it is 0dBm. If use lower % look what is needed level correction if need keep carrier level. Mostly <10% do not need correction in practice.
Most important (least for me) is that AM modulation depth is correct and and also other modulations, here DG is very good and FY is just terrible.
Down to 0.02% AM mod what I can now here measure, it is  very accurate.
Internal reference is totally from different universe if compare to this FY6900.







DG drift from startup. 1h20m follow up.
NOTE SCALE. It is 10x magnified related to FY6900 test image!!!





DG800/900  AM modulation % accuracy. Better than good, it is perfect.
ETA: Some explanation about image. Modulation depths 0.02% 0.2% and 2% (blue, yellow, orange) are all done with same carrier frequency 7.1MHz.
There carrier level change is so small amount that this spectrum display resolution is not enough for shoe difference.
20% mod, carrier tuned bit lower freq, red trace. As can see carrier is bit dropped for keep sum level same. Finally 100% mod again bit lower freq just for better visible carrier level.

Sidenote. If someone wonder this base noise level...
Signal level in generator output is 0dBm.
Between instrument is 50dB attenuator and then display have same 50dB level offset. .With this arrangement used "measuring" receiver level accuracy-linearity is better.