Siglent SDG1020 Arbitrary FunGen- First Impressions and a possible Signal Glitch

[BlueLaser]

I purchased the Siglent SDG1020 just yesterday and thought I would offer up some thoughts and bring up a possible concern. 

First, the unit seems to be built quite nicely.  Good shielding and solid chassis, bright screen that is easily view-able and not washed out.  etc.  User interface is managable but could definitely be improved with a firmware update.  There are many settings that I would like to see be toggled using the rotary encoder rather than having to press the soft key at the side of the screen multiple times.  The voltage accuracy seems very good as well.  I particularly like the DC output option which seems very close to spot on at voltages above about 10mV.  The DC output is fully adjustable from 1mV upto 10V.  I measured it with a Uni-T UT61E    and the 1-5 mV range was off by upto 0.4mV.  After that, the values all the way through the full range were almost always +/- 1mV.  Nice! 

So naturally, I am interested in this unit for testing of devices which requires non-DC signals so I hooked it up to my (also) new DS2072 and found a strange glitch.  I had it set to 4V pk2pk Square wave and At 100kHz on Channel 1 only, there were a number of glitches which begin occurring about -50ns before trigger and continue past about +40ns.  They seem pretty significant and are not present at all frequencies.  I have attached pictures to get others' thoughts.  One minor other thought is I found the jitter between channel 1 and 2 to be pretty significant (and above the listed spec of 6ns) but I assume the spec sheet was referring to the jitter of the channel with respect to itself.  They must have completely different synthesizers as they do not seem correlated.  Either way, kind of strange and may cause problems for some measurements.  I think when I get a second, I will compare it to the TTL out to figure out what is going on.  I wanted to know what others thought and it this is just an issue with my unit.  I just emailed Siglent and am hoping to hear back from them soon.  All the best. 

Update
A number of SDG1000 Series units seem to suffer from a glitching/jitter phenomenon when viewed carefully along the rising or falling edge of the SQUARE wave function.  The issue is most visible in the low frequency area <1MHz.  This issue only seems to affect the square wave as it stems from hysteresis on the ADCMP562 Comparator located near the fan on the main board.  A solution recommended by Siglent to me was to replace R82 and R87 (located very close to the ventilation fan on the board) with 20k ohm well matched 0603 resistors.  The existing resistors seem to be 40k in some units and are VERY closely matched (Mine were <0.1% tolerance) so when replacing, I manually binned the 5% resistors I had to ensure similar response on each channel.  It should be noted that I only experienced improvement in the upper frequency range >2-4MHz after this repair.  Please observe strict ESD protection rules if you are attempting this repair.  For vastly improved jitter and no glitching, the pulse function should be used instead setting the duty cycle to 50%.  The pulse function is limited to 5MHz, however. 

[saturation]

I don't own this unit, but at only 100kHz it abnormal; waveforms should be pristine in appearance and stability, as this FG is rated to 20MHz.  Beyond a flaw, and if one isn't so worried about voiding a warranty, I'd tear it down and check the construction quality to make sure its not something easily user fixable caused by poor assembly.

[BlueLaser]

Hmm, yeah.  This is what I suspected.  The glitches are pretty significant.  Let me see what the manufacturer says and what the seller will do?  As long as I am keeping the unit, I would definitely be willing to crack it open and see if there is soldering error that could be resulting in the error.  That said, I don't want to do anything until I know they're not going to ask for it back. 

Regarding firmware, from what I understand, you must also contact the manufacturer directly?  Does anyone know if there are any hardware differences between the different models of the SDG1000 series?  (Is it just software crippled for sub 50MHz models?)

[pauln]

I have this exact same AWG.  FWIW I have duplicated your settings into my old Owon DSO and do not get the glitches shown in your pictures.

In case you are not aware there is a known issue when using the 1020 with dbm units - the output jumps dramatically at certain points.  Sorry I cant remember the exact points.  I have written several emails to Siglent, starting last August 2012 asking when it will be sorted and have only received 1 response saying "wait" it is under test.  The sudden output increase when using dbm units could cause a problem or 2 - so be aware.

Paul 

[Mark]

I have this exact same AWG.  FWIW I have duplicated your settings into my old Owon DSO and do not get the glitches shown in your pictures.

Same here with the same equipment as you.  No glitches with the OP's settings.   

[BlueLaser]

Thanks Mark and Pauln.  I did want to point out that the glitches do not occur on every trace.  They are probably about 1 out of 500 or so traces with a glitch but luckily I can keep my ds2072 on infinite persistence and the 50,000 wfms/s helps too!     I'll wait to see if Siglent tells me anything.  Otherwise, I guess I'll leave it up to the sales channel to see if I can get an exchange or get fully or partially compensated. 

[andy1]

I have the SDG1010 so not exactly the same model but with this frequency there shouldn't be that much difference, no glitch for me either when capturing for ~30min with infinite persistence with my hantek DSO so I think with even the much lower capture rate than the nice Rigol I should have seen something.

[rf-loop]

First you can look these tests (specially about jitter)

In these some tests have used SDG1010 and  1025.
In most of things they are same.


Look there jitter image 6. Made with two channels and ~93kHz square.

Then look image 4 and then image 7 with lower frequancies.

Also image 7 is made for jitter test so it is not good for look inside corner where signal rising edge start. I will test and take later some images with better settings for look inside this corner. (I have something allready but I can not use these - becouse images have something (done in third party lab) where I need follow NDA)

There are (of course) lot of  "glitches" due to natural jitter and filtering. typically  they stay "typically" below ~10% of signal level and around jitter p-p time before rising edge start (are not equally spaced throughout the time domain, but the emphasis on close to the rising edge (statistical distribution)). 
For OP, please read also Agilent paper about this type of DDS generators principle.  This is simple DDS generator and jitter is natural with this principle.

If you have real glitches what levels are nearly 50% (as shown in your one image) from signal level something is badly wrong. Something is wrong in your measurement OR something is wrong in your Siglent SDG OR both. Also you can set your oscilloscope trigger to level where it do not meet these "glitches".  And for jitter, look next rising edge from  triggering rising edge.


After I have enough time for do some tests with Siglent SDG I will show lot of better images about this corner before square rising edge in this image.  This image here is not for this purpose, it is made for characterize amount of typical peak to peak jitter.
(it is typically as manufacturer specified). Filtering is always compromise but still it perhaps can do better without hop to next price group. Some amount of this corner impurity need accept. Just as we need also accept corners wobbling aliasing in oscilloscopes and many other natural things. If <0.5k$ machine is not enough good it can always buy 5k$ or 50k$ unit. But also if you get more accurate test equipments you find still (but more small) "errors" there. It is one important part of experience and knowledge to characterize what is fail and what is normal.

Also of course there is channel to channel jitter becouse both channels signal have jitter and now if you synchronize to other channel and look how other channel have jitter related to other channel - it is basic fundamental in this type of generator principle.
If need generator what do not have this this type of jitter then this generator type is wrong for your use. Also there are some golden frequancies where do not need "hop" for example adjust frequency or period (remember sampling rate).


If there is really this level of glitches what your image show then your generator have some fail. Simply talking - then your generator is broken.





----
(Yes I'm also aghp (well known seller as "brand" in my country and specially in some group of peoples)
but now IF I find something what these companies do not want publish  (data/hack/modification or other this kind of  things) for Owon/Siglent I can not at all publish these becouse  I'm both companies accepted reseller and service - if this is unethical - sorry, I keep these in my head.  ; ) 

But original "accident" happend after some old  customer and friend ask me why I'm rf-loop in EEVblog becouse I'm "aghp"... after 3 minute "thinking" it was ready... ok: now I'm aghp, satisfied now? What rf-loop or aghp have talk here: there is not any kind of  discrepancy.

I did not destroy rf-loop account becouse I think it may be useful in future some situations IF there come data or something what seller can not publish without loose deal with manufacturer.  ; ) )  Then also sometimes becouse I have several computers (some have same IP behind NAT and some in other locations and even some in China sometimes) Sometimes have happend accident that I open PC and go to EEVblog and I have not looked login (automatic) and just start write without thinkin who I'm becouse I'm myself always.


All know outside of equipment are not secrets.. and same for inside... all can measure and test these. But inside there may be things where can use some given information about repair service work.

Of course I can still publish errors and all information what is normal from seller who do not sell only "nice boxes". 

- rf-loop  (aka and "official": aghp or aghp55)

[rf-loop]

Here some images made with HP
Siglent SDG1025 connected to HP's 50ohm input. (it is 500MHz old work horse)
SDG software version is: 1.01.01.23   and HW version 0f-00-00-19-23

Images have explanation.
But look carefully scales and remember signal frequency.
Remember: SDG samplerate is 125MSa/s. One sample every 8ns.

It is good to note what happend if think reverse. Take oscilloscope what have example 50MHz analog bw and it have 125MSa/s sampling rate. Then try look sguare with example 6ns rising time. What you see (if all is ok). You see corners wobbling.  Something littlebit like this 100kHz rising.

But then... what happend 10kHz. I do not know.
Level is under 10% mostly but  sometimes higher but NOT nerly 50%.
I do not know what this OP measuring method may affect or what is going on in Rigol "image makeup" process. 

My first suspect is that OP's SDG is somehow broken.  These glitches are lot of too high.
Is it possible analyze real sampled data points so that Rigol signal makeup process do not "destroy" real data.

In these HP images there is raw real data without any "show nice image" manipulation.





NOTE images vertical and horizontal scale and signal under test dimensions.
100kHz signal period is 10000ns!  and 10kHz signal 100000ns!  and signal level is 3.5Vp-p but on the scope screen is only 800mV  (100mV/div)
For images 4 and 5.  remeber that SDG ample perior is 8ns!

[BlueLaser]

RF-loop, Thank you so much for all the insight.  You made me worried about my new DS2072 so I decided to check it with my DS1102e and sure enough, the glitch is there.  BADLY!  There is DEFINITELY something wrong with my particular unit.  I understand the nature of DDS will introduce glitches based on the true internal sampling rate.  The problem I am having is that they are orders of magnitude larger than the jitter specification.  Please see the included pictures.  For these, I decided to turn ON sync in the utility menu and can now see that channel 2 also exhibits some (but not as severe) glitching.  I mean, this is 4us of glitches!!!     (Not ns or ps) That was on a 1kHz square wave.

Okay, so besides that, I was wondering if someone could chime in on the impedance switching.  When I switch form High-Z to 50ohms, the voltage amplitude seems to double.  Is this a normal response or yet another problem?  Thanks again everyone!   

[PA0PBZ]

Okay, so besides that, I was wondering if someone could chime in on the impedance switching.  When I switch form High-Z to 50ohms, the voltage amplitude seems to double.  Is this a normal response or yet another problem?  Thanks again everyone!   

When set to 50 Ohms you should terminate 'the other side' with 50 Ohms, I guess you will find that the amplitude will be correct again 

[BlueLaser]

lol, I guess I slept through too much of my electrical engineering classes!    When I get back to the bench, I'll try matching the impedance.  I'm 99% sure that's what I was missing.  Thanks. 

[rf-loop]

Glitches before rising edge, there is some but they need be quite low level! It can not at all accept if these glitches level is 30% or even 50% from signal pp. 10-15% is maybe normal for this kind of unit.

Jitter is totally different case.

Most easy way look jitter is (just as Agilent do).

Trig to rising edge. Go to look NEXT rising edge. Zoom in. Wait enough time. Finally you find peak to peak value for jitter. It is highly related to square wave frequency (period).

100kHz square long time jitter here in image. This data have collected long time leaving scope with infinite prsistence alone. Trigger point is 10us (one cycle) before (rising edge)

If some manufacturer tell (as someones do) jitter rms value.. or typical value...  nice numbers and they do not lie..

(and for Owon owners here... all these can also look with Owon)

[rf-loop]

I have this exact same AWG.  FWIW I have duplicated your settings into my old Owon DSO and do not get the glitches shown in your pictures.

Same here with the same equipment as you.  No glitches with the OP's settings.   

With Owon, of course  it can see (but with 100kHz square just same as with HP images before.)

Here image (not best possible due to lack of time to do it) how Owon see these "glitches" using SDG1025 with 4Vp-p square and 10kHz frequency.

Note that image show only around 900mV area from 4V rising edge.
And bottom line (after last falling edge is around 50000ns. Glitches are inside 120ns before rising edge and most them are inside 30ns.

Nice to see also that Rigol have learn something from Owon front end and also it seems  they have take  exactly same  RuiFeng ADC. But of course Rigol have developed nice and totally different product.

[Wuerstchenhund]

I, too, have a Siglent SDG1020 AWG (HW 02-00-00-19-24, SW 1.01.01.23), and because of this thread I did some tests yesterday. And low and behold, my unit does show the same glitches at 100kHz 4Vp-p square wave as BlueLaser's one. I never noticed the glitches because I only use the unit for sinus and pulse signals, and occasionally for square waves in the MHz range.

However, I am not so sure that this is a hardware problem. I did some few more tests with other signal forms (pulses, sinus), and lower and higher frequencies. What I found was that the glitches only show up in square wave mode, they don't however come up in other modes (i.e. pulse mode with 50% duty cycle), and that it gets worse if the frequency gets lower. But at 1Mhz or more the signal is perfectly stable.

Ch 2 doesn't show any glitches in square wave mode.

This leads me to believe that the issue may as well be a firmware bug, which perhaps only shows up on specific hardware variants.

BlueLaser, Pauln and Mark, maybe you can post your HW and SW versions?

[rf-loop]

I, too, have a Siglent SDG1020 AWG (HW 02-00-00-19-24, SW 1.01.01.23), and because of this thread I did some tests yesterday. And low and behold, my unit does show the same glitches at 100kHz 4Vp-p square wave as BlueLaser's one. I never noticed the glitches because I only use the unit for sinus and pulse signals, and occasionally for square waves in the MHz range.

However, I am not so sure that this is a hardware problem. I did some few more tests with other signal forms (pulses, sinus), and lower and higher frequencies. What I found was that the glitches only show up in square wave mode, they don't however come up in other modes (i.e. pulse mode with 50% duty cycle), and that it gets worse if the frequency gets lower. But at 1Mhz or more the signal is perfectly stable.

Ch 2 doesn't show any glitches in square wave mode.

This leads me to believe that the issue may as well be a firmware bug, which perhaps only shows up on specific hardware variants.

BlueLaser, Pauln and Mark, maybe you can post your HW and SW versions?

Please can you show image about these glitches. Trig to rising edge around 70% of full level.
Use (example) 100kHz symmetric square (not pulse)
Look just before rising edge... something example 30 - 100ns area before rising.
Use 4Vpp set in generator (50ohm)  and  50ohm input with oscilloscope.
What is maximum glitch level? (do not use any "Sinx-like"  cosmetics in scope. Real sampled points.

How about 3Vpp

Then use 400mVpp signal from SDG all other same. Max glitches level?
(this level is "direct" in output stages)

Can you confirm with just equal settings (note CH1 and CH2 output limits) that CH2 do not have glitches but CH1 have.

Here is table about level ranges inside HW. (for these there are set of attenuators and also fixed 20dB amplifier in HW)  They are used combination what also depends about offset setting range (<=3V and >3V)

[Wuerstchenhund]

Please can you show image about these glitches. Trig to rising edge around 70% of full level.
Use (example) 100kHz symmetric square (not pulse)
Look just before rising edge... something example 30 - 100ns area before rising.
Use 4Vpp set in generator (50ohm)  and  50ohm input with oscilloscope.
What is maximum glitch level? (do not use any "Sinx-like"  cosmetics in scope. Real sampled points.

Getting images may be difficult as the only scope I have at the moment is a HP 54510A 2Ch 250MHz 1GSa/s DSO for which I don't have any screen capture utility (Agilent's IntuiLink software doesn't work with this scope), and my 54542A which can capture screen shots is currently out of order due to a PSU problem.

But I guess I could try to make some old-fashioned photos.

I'll have a look when I am back home tonight.

Update: Sorry, something more important came along so I had't time to play around with the AWG. If it's still of interest I can do that within the next days.

[rf-loop]

After very small HW modification in analog circuit.
Changed one component value.

Measurement just all as my earlier hp image 1.
(also it can see that hp scope trig is more stable after this mod. (look rising edge)

20minute time with infinite peristence as also before.

Perhaps now better?

I have not tested all functions fully but looks like not any bad efect in other signal modes and features.

But this test show clearly where in circuit this problem is hiding....  (partially or fully) 

Images before test mod and after test mod


Note: Signal top level is 3.5V  and here is only bottom corner  before rising.
Now glitches are inside 70ns before rise and under 150mV (what can find with this oscilloscope using 20minute)
Before around 140ns area and 400mV  (with used around 10kHz square.)

[BlueLaser]

My unit is versioned as follows:
Boot Strap Number: 1
Software Version:  1.01.01.23R1
HardWare Version:  02-00-00-19-24
Model:  SDG1020

RF-loop, what change did you make.  I can't see your image of the circuit you're referring to. 

I find this interesting because Siglent got back to my seller and said the problem may require changing R87 on the board to a 20kohm resistor.  I am really not sure what this will do because I was then advised in the next email to NOT open the case as it may complicate the return/exchange process.  I'm currently in the process of either getting replacement unit or board from the seller.  And yes, I seem to only see these glitches when in square wave mode.  I do still see them at higher frequencies (>1MHz) but they are much less pronounced and do not go back as far in time.  It guess this does show signs of a firmware issue but I wonder if it has to do with the square generator or filter.  If someone has one that is out of warranty or are okay opening, what is the current value is R87 anyway? Thanks a bunch guys. 

[rf-loop]

Nice to hear that someone else have also find solution.

Yes resistors R87 (for CH1) and R82 (for CH2)
can change.
Original resistance may be different in different manufacture versions.
Original may be normally 40k (40.2k) as example in image.

And it is too high in my opinion.

Change to 20k may help enough but this hp oscilloscope image was from test with around 9k.

My opinion is that some value between 7k - 33k is ok for most of uses. (perhaps this 20k is compromize with less glitches but minimum other efects.   I have not tested what possible things it affects in some other things.. phase accuracy etc.. Lack of time to do full tests.

(but remember, my test was with 10kHz, not 100kHz as original post here  and also I have looked 100Hz and really lot of better than without change. As we know example 1MHz there is very small amount of glitches. More low freq and more wide "time window" for glitches before rising edge. This is natural with this circuit principle.)

This image show original, NOT modified!

[nixxon]

I did some tests on my SDG1020 with newest firmware 1.01.01.23 this evening with my Rigol DS1102E.

Square wave, 50% duty cycle, 100 kHz, 50 ohm terminated inside SDG1020 and at scope BNC input (T-connector).
Amplitude 3 Vpp, as CH2 only goes up to 3 Vpp.

"Exposure" time: approx. 13 minutes for both runs (SDG1020 Ch1 and Ch2).

When I saved the screenshot from output Ch1, the triggering was not stopped on the scope and there was a bad connection between the 50ohm termination and T-connector at the scope end. Thus the amplitude increased a bit and got captured on the screenshot. I have crossed out the wrong part (in red color).

All measurements done in basement "lab" @ 10 degrees Celcius. Equipment was not warmed up at all, other than during the tests that took an hour or so in total.

Rigol settings:
Analog Ch  State   Scale    Position   Coupling  BW Limit  Invert
CH1        On      500mV/   0.00uV     DC        Off       Off

Analog Ch  Impedance   Probe
CH1        1M Ohm      1X

Time    Time Ref    Main Scale    Delay
Main    Center      5.000ns/      0.000000s

Trigger  Source      Slope    Mode      Coupling     Level    Holdoff
Edge     CH1         Rising   Auto      DC            460mV   500ns

Acquisition    Sampling    Memory Depth    Sample Rate
Normal         Realtime    Long Memory     500.0MSa   

[nixxon]

Some DC measurements of the SDG1020, using my Fluke 87 (and 83-5)

DC out ------> Reading Fluke 87
+1 ------> 1.006
-1 ------> -0.999
+2 ------> 2.000
-2 ------> -1.981
+5 ------> 5.00
-5 ------> -4.96

Some mV readings:
DC out ------> Reading Fluke 87
0.000 ------> 0.001 (+0.001V)
0.001 ------> 0.002
0.002 ------> 0.003
(etc. (+0.001V) all the way up to:)
0.133 ------> 0.134
0.134 ------> 0.135
0.135 ------> 0.137 (+0.002V)
0.136 ------> 0.138
(etc (+0.002) then some bigger steps
0.200 ------> 0.202
0.300 ------> 0.303 (+0.003V)
0.400 ------> 0.403
0.500 ------> 0.503
0.600 ------> 0.604
0.700 ------> 0.703
0.800 ------> 0.804 (+0.004V)
0.900 ------> 0.905 (+0.005V)
1.000 ------> 1.006 (+0.006V)

Pretty accurate DC output and within specs: 1% +3mV

EDIT: See enclosed table for %-errors
EDIT 2: Corrected wrong numbers in table

[nixxon]

I also enclose a 25 minute sample from Ch1 output @ 3.5Vpp (same settings as above)

I'll enclose the scope settings (25minp.txt) for this screenshot in case I missed any details:

Analog Ch  State   Scale    Position   Coupling  BW Limit  Invert
CH1        On      500mV/   0.00uV     DC        Off       Off

Analog Ch  Impedance   Probe
CH1        1M Ohm      1X

Time    Time Ref    Main Scale    Delay
Main    Center      5.000ns/      0.000000s

Trigger  Source      Slope    Mode      Coupling     Level    Holdoff
Edge     CH1         Rising   Auto      DC            1.00V   500ns

Acquisition    Sampling    Memory Depth    Sample Rate
Normal         Realtime    Long Memory     500.0MSa 

[rf-loop]

There are more glitches with lower frequency.
You can look example 10kHz and 1kHz.
Also note that example in my images there is 100mV/div but signal from SDG is 3.5V.
Only looked bottom corner beforer signal rising edge (for better visibility)
Why there is glitches more if use more low frequency.
Answer is simple but it need pen and paper to explain whole "theory and practice".
But it can understand as know that squarewave is produced from sinewave.
More slow wave there is more time where input to comparator is just very very near to treshold before input crosses the threshold. Now, in practice and in theory all signals have more or less noise. There is not "ideal" sinewave. Also we know how sinewave is produced. With 14bit DAC where some lowest bits are also "noisy".  Comparator whole environment is "noisy", also littlebit itself comparator Analog Devices  ADCMP562 (Dual High Speed PECL Comparator).

COMPARATOR HYSTERESIS
The addition of hysteresis to a comparator is often useful in a
noisy environment, or where it is not desirable for the
comparator to toggle between states when the input signal is at
the switching threshold. The transfer function for a comparator
with hysteresis is shown in Figure 20. If the input voltage
approaches the threshold from the negative direction, the
comparator switches from a 0 to a 1 when the input crosses
+VH/2. The new switching threshold becomes ?VH/2. The
comparator remains in a 1 state until the threshold ?VH/2 is
crossed, coming from the positive direction. In this manner,
noise centered on 0 V input does not cause the comparator to
switch states unless it exceeds the region bounded by ±VH/2.
Positive feedback from the output to the input is often used to
produce hysteresis in a comparator (Figure 24). The major
problem with this approach is that the amount of hysteresis
varies with the output logic levels, resulting in a hysteresis that
is not symmetrical around zero.
In the ADCMP562, hysteresis is generated through the
programmable hysteresis pin. A resistor from the HYS pin to
GND creates a current into the part that is used to generate
hysteresis. Hysteresis generated in this manner is independent
of output swing and is symmetrical around the trip point

Resistor what need change is just for set this hysteresis.
It is also possible that optimum value is different in some units.

If want do lot of more then also internal sinewave used for square may need more filtering if want really good. But it need think compromize. Also it is possible that comparator enviroment (powers and GNDs need cleaning etc etc) This is cheap generator. Not high-end generator. If want high-end generator it need build totally all agen...or buy different generator (with different price)

EDit/Add
I get beter smd resistor values set for this purpose and later I will do some more and better tests.

[nixxon]

I will do some tests with the recommended setup:
Signal generator: 4Vpp, Square wave, 50% duty cycle, 1 kHz or 10 kHz (or maybe 12.345678 kHz)
Rigol: Horizontal scale 20 ns/div, Vertical scale 100 mV/div (& place trace to the right and bottom of screen)

Maybe I will capture more glitches if I set Memory Depth to "Normal" (16kpts@1GS/s) rather than "Long memory" (1Mpts@500MS/s). I dont't believe I need that MegaZoom-like feature at all. (If I scroll sideways, the persist-image disappears) I think reducing memory depth also may reduce the blind time and increase the waveform capture (or update) rate. Or maybe this makes no difference when measuring low frequencies.

If I capture the same signal with both Memory Depth settings (Normal and Long Memory), can anyone predict how the 2 screenshots will differ from each other?