| Products > Test Equipment |
| Siglent SDS1x04X-E BodePlot II (SFRA) features and testings |
| << < (17/21) > >> |
| luciof:
Thanks tautech; I had already read the whole topic before, and specifically the post you pointed out. I also realize that I haven't made it entirely clear what I meant. To be clear, I'm quite happy with the scope; I'm not criticizing it, but trying to point out some details to improve it further. Regarding the skew and the related phase plot derating: 1- In my very basic setup, there is no significant skew related to cables and connections; in fact, using a single generator output and two "ADC coupled" scope channels (CH1 & CH2 or CH3 & CH4) the phase plot looks quite good (actually, the plots go haywire beyond 60 MHz, but this is because I clamped both the probe clips to a stub of wire stuck into the generator output and both ground clips clamped to the BNC ground shell. In previous tests with impedance-matched cables, terminations etc. this didn't happen, but anyway the plots were quite the same below 50 MHz, and it's enough for this evaluation). 2- The main issue arises using unrelated channels ( (CH1 or CH2) & (CH3 or CH4) ). In this case I consistently find a phase derating due to a constant delay between the scope channels; in my tests, the calculated delay is around 850-900 ps. 3- At first, one would think it's the skew between the scope channels, but it is unrealistically large and - in fact - the real skew is far smaller (about 100 ps or less, difficult to measure properly and usually negligible). Moreover, setting a deskew on the scope channels seems to make no difference 4- Anyway, in real measurement setups some significant time mismatch is always possible, and it would be very handy to be allowed to enter a compensation (delay value) parameter in Bode Plot II settings 5- In fact, when using two coupled generator outputs for bode plotting (as suggested somewhere to avoid power splitting fixtures), in previous impedance matched tests I found that the generator outputs were quite well matched in both frequency and amplitude, but there was a delay of about 280 ps between them. The only way to practically compensate this, I think, would be a specific compensation parameter (i.e., the delay parameter of point (4)). 6- Finally, I think that this kind of issue is usually underestimated because the main plot is the amplitude (modulo) plot, and often phase plot is discarded after a quick look; neverthless, in more involved works it can be very useful. That's all for now. Sorry for the numbered layout above: it helped me to better explain my points while struggling to write in English..., and thanks again for your attention. |
| Performa01:
Yes, the limited accuracy of phase measurements between the two separate ADCs have been discovered a long time ago, as can be demonstrated by the attached old screenshots from June 2019: SDS1104X-E_Y-t_Phase_Error SDS1104X-E_FRA_Phase_Error These demonstrate how four signals with phase differences of no more than 3.15° in the time domain can be measured as up to 40° in the Bode Plot. Since it is quite obvious that this is a channel skew issue, the fact that the channel Deskew parameter is ignored by the Bode Plot has been discovered back then as well and I’m surprised that we never got a fix. Maybe there are some restrictions in the HW architecture that prevent an easy solution. In any case, this does not make the Bode Plot useless, even at high frequencies. Just restrict measurements to a single ADC whenever you need accurate phase information at frequencies beyond a couple MHz. You lose the ability to plot three channels in parallel, but then again, we should never forget that competing products are limited to a single channel up to 25 MHz or so, hence a single channel up to 120 MHz should still be a major advantage. And there’s still the option of having three channels, even when phase information gets inaccurate above a couple MHz. Maybe sometimes we just can’t get everything in the cheapest package; at least for the more advanced HW-platforms, Siglent have addressed these issues. Reply #3585 in the SDS2000X Plus thread demonstrates just this: https://www.eevblog.com/forum/testgear/siglent-sds2000x-plus-coming/msg4370986/#msg4370986 I’m not keen on the idea of introducing a specific correction parameter in the Bode Plot. We want a correlation between time domain and frequency domain (at least in the higher class instruments we can watch the time domain during bode plot), so just using the Deskew parameter that is already there would be the obvious solution. I suspect this is the reason for the phase error in the SDS1004X-E in the first place; the normal y-t operation always needs some compensation for runtime differences anyway – and there it works. If Bode Plot would use these corrected data, there wouldn’t be any problems. For some reason it seems to use the uncorrected RAW data instead. I’ll try to clarify this. |
| luciof:
--- Quote from: Performa01 on August 21, 2022, 08:22:40 am ---Yes, the limited accuracy of phase measurements between the two separate ADCs have been discovered a long time ago, as can be demonstrated by the attached old screenshots from June 2019: SDS1104X-E_Y-t_Phase_Error SDS1104X-E_FRA_Phase_Error These demonstrate how four signals with phase differences of no more than 3.15° in the time domain can be measured as up to 40° in the Bode Plot. Since it is quite obvious that this is a channel skew issue, the fact that the channel Deskew parameter is ignored by the Bode Plot has been discovered back then as well and I’m surprised that we never got a fix. --- End quote --- The actual skew we see (3.15° @ 100 MHz) boils down to 97 ps, while the [more or less] 32° @ 100 MHz we see in the plot are about 890 ps, so the "time domain delay" and "bode plot delay" are definitely distinct; moreover, even if I change the deskew, the phase plots don't change anyway. --- Quote ---In any case, this does not make the Bode Plot useless, even at high frequencies. Just restrict measurements to a single ADC whenever you need accurate phase information at frequencies beyond a couple MHz. You lose the ability to plot three channels in parallel, but then again, we should never forget that competing products are limited to a single channel up to 25 MHz or so, hence a single channel up to 120 MHz should still be a major advantage. And there’s still the option of having three channels, even when phase information gets inaccurate above a couple MHz. Maybe sometimes we just can’t get everything in the cheapest package; at least for the more advanced HW-platforms, Siglent have addressed these issues. Reply #3585 in the SDS2000X Plus thread demonstrates just this: https://www.eevblog.com/forum/testgear/siglent-sds2000x-plus-coming/msg4370986/#msg4370986 --- End quote --- I agree, the system is very well usable within its limits, but I don't see this scope as a "cheap" one, rather a "low cost, high performance/cost ratio instrument". I think this should translate to the [reasonably] best exploitation of its hardware combined with a well honed UI. While this is still a cost, I would see it more as an investment, because a degree of sloppiness in a $500 device might be also expected in a $5000 one from the same firm [of course this is not the case, just to make the point :) ] --- Quote ---I’m not keen on the idea of introducing a specific correction parameter in the Bode Plot. We want a correlation between time domain and frequency domain (at least in the higher class instruments we can watch the time domain during bode plot), so just using the Deskew parameter that is already there would be the obvious solution. I suspect this is the reason for the phase error in the SDS1004X-E in the first place; the normal y-t operation always needs some compensation for runtime differences anyway – and there it works. If Bode Plot would use these corrected data, there wouldn’t be any problems. For some reason it seems to use the uncorrected RAW data instead. --- End quote --- Here I don't agree. Ideally, we would sample both input and output of the DUT in the exact same instant; this is obviously not possible, but if this timing error is reasonably constant, we can take it into account very easily when doing the math. For sure, this delay is not part of the results we need and - if well carachterized - can and should be cancelled out. Concluding (and reiterating): First, it's true that only in very rare occasions more than 2 channels are needed, and even then, when accurate and reliable plots beyond - say - 10 MHz are required, they'll need very accurate setups and probably something more than a generic scope and generator. But why give up something the hardware is well capable of doing, with only minor firmware adjustments? (And anyway, I see the abnormal and unmodifiable skew in bode plot more as a bug than a missing feature; if the skew was the actual one we see in time domain it might be good enough without compensation). Second, a "delay" parameter in the Bode Plot environment would take care very well of the issue we are talking of, and allow to compensate for non-instrument related delays. Two examples come to mind: 1) I have the following setup: generator --> 50 ohm cable --> DUT (with 50 ohm input and output Z) --> 50 ohm cable --> 50 ohm pass-thru terminator --> scope BNC the output channel is OK; now I need to also connect the DUT input to the scope; below - say - 100 MHz, I think the best way is to clip an active probe (only 1-2 pF loading) to the DUT input. But now we'll have a "large" delay (>5 ns for sure) which would make the phase plot practically unusable 2) Using an SDG2000X generator with coupled channels to make my connections simpler (one output to the DUT input, the other one to the scope input channel. In my case this adds up a further delay of about 300 ps (see picture) In both examples, a simple user parameter would cleanly solve the problem. To me, this is a fine instrument, and not "for its price", but I would better say "within its class and specifications". And I think it could be improved further with minor refinements. |
| rf-loop:
--- Quote from: Performa01 on August 21, 2022, 08:22:40 am ---Yes, the limited accuracy of phase measurements between the two separate ADCs have been discovered a long time ago, as can be demonstrated by the attached old screenshots from June 2019: SDS1104X-E_Y-t_Phase_Error SDS1104X-E_FRA_Phase_Error These demonstrate how four signals with phase differences of no more than 3.15° in the time domain can be measured as up to 40° in the Bode Plot. Since it is quite obvious that this is a channel skew issue, the fact that the channel Deskew parameter is ignored by the Bode Plot has been discovered back then as well and I’m surprised that we never got a fix. Maybe there are some restrictions in the HW architecture that prevent an easy solution. In any case, this does not make the Bode Plot useless, even at high frequencies. Just restrict measurements to a single ADC whenever you need accurate phase information at frequencies beyond a couple MHz. You lose the ability to plot three channels in parallel, but then again, we should never forget that competing products are limited to a single channel up to 25 MHz or so, hence a single channel up to 120 MHz should still be a major advantage. And there’s still the option of having three channels, even when phase information gets inaccurate above a couple MHz. Maybe sometimes we just can’t get everything in the cheapest package; at least for the more advanced HW-platforms, Siglent have addressed these issues. Reply #3585 in the SDS2000X Plus thread demonstrates just this: https://www.eevblog.com/forum/testgear/siglent-sds2000x-plus-coming/msg4370986/#msg4370986 I’m not keen on the idea of introducing a specific correction parameter in the Bode Plot. We want a correlation between time domain and frequency domain (at least in the higher class instruments we can watch the time domain during bode plot), so just using the Deskew parameter that is already there would be the obvious solution. I suspect this is the reason for the phase error in the SDS1004X-E in the first place; the normal y-t operation always needs some compensation for runtime differences anyway – and there it works. If Bode Plot would use these corrected data, there wouldn’t be any problems. For some reason it seems to use the uncorrected RAW data instead. I’ll try to clarify this. --- End quote --- Channels Skev adjustment works also in FRA (least in my old SDS1004X-E) but way Ch Skew adjust works is really poor. Have not tested now enough, but when I found my old handwritten piece of paper ("post it") note. I tried it just but due to other things very busy only very limitedly and quikly(=error warning!). My SDS FW is 6.1.37R8 Looks somehow (when FRA in use) that it is still so that when ADC's are interleaved (true ADC speed 1GSa/s) Skev adjust resolution is 1ns and when ADC's non interleaved mode (true ADC speed 500MSa/s) resolution is 2ns! what resolutions are really enormously too rough when we work higher frequencies (example over 1MHz. @1MHz 1ns responds 0.36 degree and @60MHz 21.6 degree what also was masured result here - naturally because time shift IS 1ns in this case.). It need remember what are sample intervals. This is only true time resolution! There is nothing below 1ns resolution (except fine intepolation between true raw samples) In quick and "dirty" test I use 1-60MHz sweep. FRA: Source generator Ch1. This was splitted with matched cable pair to oscilloscope ref CH1 and "dut out" CH3 If CH1 and CH3 alone active and all CH Settings all zero. Maximum phase error around 0.4 degree If I adjust then CH3 Skew nothing happen until Skev is 1ns. With this Skew result is roughly +22 degree error. But then FRA settings all same (just Ch1 and Ch3 etc): Oscilloscope all 4 channels in use. (2 and 4 not FRA use) Now ADC's work at 500MSa/s (I do not here talk anything about decimated samplerates, it do not matter here) CH's Skew all 0. Result at 60MHz. Phase error roughly -22 degree. Okay, then I think I adjust Ch3 Skew. I set it to 1.0ns Result: nothing happen, FRA displayed same roughly -22 degree phase at 60MHz. Next I set Ch3 Skew to 2.0ns. Result: Phase error at 60MHz is now roughly +22 degree. I can assume that if I can set Skev to 1.0ns it is quite same as originally when only two channels are in use and 1GSa/s But now 1ns set is ignored (it really looks like resolution is 2ns (sample interval without fine interpolation) When Skew only works every 1ns or 2ns, it's like a 22" spanner on a watchmaker's workbench. Of course if one do only 1kHz or 10kHz things... it is different case. Why there is this 1ns shift between ADC's interleaved and non interleaved mode. Even if Skew can not work with fine interpolation in FRA, still this 1ns BUG between ADC's interleaved/non interleaved mode is imho BUG. Naturally it can compensate example using 1ns delay. Example ~20cm RG223. Set of fixed delay lines is always good to have (adjustable delays/phase shifters even better - and expensive). It is bit surprise this BUG is still there! .... somehow as if old jobs are left unfinished when you jump into a new job and think "time to take care of problems". Time doesn't take care of them. These unfinished jobs still have to be taken care of fully, and that requires doing the right thing, not just pretending. After all, bosses often want their workers hands to wave a lot, but they forget that quality is more important than quantity. Insufficient quality, however, sometimes later pulls the rug from under your feet. |
| mawyatt:
Not knowing exactly how the instrument takes the sample and in what order if in interleave or sequential mode, so this is just a "What if" statement. If the samples of "Input" and "Output" are interchanged in time sequence for a single averaged "Sample Point" on the Bode Plot it seems the time skew would be removed in the averaging for a single Sample Point. Think the Bode Plot does use multiple samples to render a final Sample Point to improve Sample Point fidelity, so alternately swapping actual In/Out sample time sequence then averaging seems reasonable. Best, |
| Navigation |
| Message Index |
| Next page |
| Previous page |