Siglent SDS2000X HD, 12bit, 2GSa/s high definition oscilloscope serie.

[2N3055]

Looks like something with (whatever this means):

hw_adc_bits = 12
data_adc_bits = 16

All Siglent 12 bit scopes has 16 bit internal representation.

Everybody works on new products all the time.. 

[TopQuark]

Increase the bode plot speed of your scope by 10x with this one trick (kinda) 

I did an little experiment that puts the dynamic range of the SDS2000X HD into good use. I have been reading up on dynamic signal analysers (https://www.testunlimited.com/pdf/an/5988-6774EN.pdf), and saw them mention a DSA can be used to perform bode plotting at real time. I was quite intrigued given my grievances with the speed of Siglent's bode plot tool.

I am planning to build my own DSA, but thought I'd test the concept with the SDS2000X HD first, as the hardware is already there, and it is easy to hack together some code to test out the idea.

The idea behind the concept is quite simple, network analysis (bode plotting) basically involves injecting a known signal into a network under test, and see what is the response. In the case of the scope, a sig gen is used as the stimulus, one channel of the scope is used to measure the stimulus at the input of the network under test, and another channel used to measure what comes out of it.

The "traditional" way of bode plotting, as used by the Siglent scope's built in bode plot function, is to inject a sine wave at a specific frequency into the network and measure the magnitude and phase change of the sine wave at the output compared to the input.

The way DSA and this experiment works is, you inject a wideband noise source into the network to stimulate all frequency of measurement interest, capture one sample with the scope (e.g. 1s of data at 20MSPS), run a FFT on both the input reference signal and the output response signal, divide the complex output result by the complex input result to get the transfer function of the network under test, and compute the magnitude and phase of this transfer function.

The advantage of this method of bode plotting is, you only need to do one data capture with the scope, and the data contains all the information needed to compute a bode plot across a range of frequency, versus the traditional way where data capture(s) are needed at every single frequency point of interest. So in terms of sampling time needed, you can expect to see a couple order of magnitude in reduction.

There are also drawbacks to this method. The traditional way of bode plotting allows the scope / network analyser to adjust its gain at every frequency measurement point to fill up its dynamic range, and so you can easily get massive dynamic range even with a 8-bit ADC and lousy frontend. In this method, the maximum dynamic range of the bode plot is limited by the inherent dynamic range of the scope's frontend and ADC. 12-bits is barely enough in this case, and this is the reason why this method is usually only reserved for DSA with very high input dynamic range.

Nevertheless, I wrote a Python script that performs the DSA bode plot method and ran a quick test comparing both methods of bode plotting on the SDS2000X HD with a simple RC filter. The results are interesting, The DSA method took 30 seconds to give a bode plot, although half of the time is used to transfer the data from the scope to the PC, and half was used to crunch the FFT and plotting, only 1 second was used to capture the data (20Mpts, 20MSPS, 1 second). The Siglent's built in bode plot was used, (50 pt per decade, auto gain) and it took so long my ADHD brain forgot to time it around 8 minutes. Of course, the traditional bode plot is much higher fidelity with the output data, still more trustworthy IMO, and the Siglent bode plot is quite slow compared to competitors.

This post is not really about saying the DSA method is much better and the built in bode plot is bad or whatever. It is more of an academic study into different ways of bode plotting and to gain some insight for when I do build my DSA project. Hope you found this interesting.

[mawyatt]

Very interesting!!

Another way to do this is to use the time domain impulse function as input since it's Fourier Transform is just unity across all frequencies so no need to do an FFT on the input, then perform the FFT only on the output as required.

Best,

[2N3055]

Reason why it is implemented  this way is maximizing dynamic range and resilience to outside noise and signals.

[TopQuark]

Another way to do this is to use the time domain impulse function as input since it's Fourier Transform is just unity across all frequencies so no need to do an FFT on the input, then perform the FFT only on the output as required.

Good point, might give that a shot

Reason why it is implemented  this way is maximizing dynamic range and resilience to outside noise and signals.

In an ideal world both methods would be implemented in the scope. The scope does not need to transfer the data to an external PC (cut time by 50%), and FFT can be hardware accelerated like it already does (save most of the remaining 50%). I'd gladly trade the fixed gain mode in the bode plot menu for this approach, and leave the auto gain set mode for precision work.

[switchabl]

The same "single-shot" approach can be used with other stimulus signals that might give better results:
1) For best SNR/measurement time, we would like to get as much signal power into the DUT as we can but this is usually limited by the max. peak-to-peak voltage range of the generator or the linear range of the DUT and receiver. So a stimulus should have high RMS at a given level (i.e. it has low crest factor). White noise has high peaks so that is not good (in fact, ideal white noise basically would have infinite crest factor).

Similarly, a unit impulse is essentially the mathematical limit of compressing a given pulse energy into something with an infinitely high peak. So this is generally not something we should strive to approximate as a stimulus source and a big reason people usually do not try to measure impulse response directly. (Because of the theoretical importance of the impulse response it is still a good experiment if you are doing it for the educational value).

2) A nice bonus would be a stimulus that is periodic for our measurement time. Then we would not have to window our FFT and get zero spectral leakage for free.

The most common solution is a sine "chirp", i.e. a sine with continuously varying (instantaneous) frequency. That may sound suspiciously similar to the "stepped" approach but there are no steps and you don't stop for measurements either. You still record it all in one go and then apply FFT. It has the same crest factor as a fixed sine and covers a wide (adjustable!) spectral range.

I also want to briefly mention that maximum length sequences (MLS) could also be used. They are periodic, pseudo-random binary sequences and are quite interesting from a theoretical perspective (but a bit tricky to properly use in practice).

[mawyatt]

Another way to do this is to use the time domain impulse function as input since it's Fourier Transform is just unity across all frequencies so no need to do an FFT on the input, then perform the FFT only on the output as required.

Good point, might give that a shot

Reason why it is implemented  this way is maximizing dynamic range and resilience to outside noise and signals.

In an ideal world both methods would be implemented in the scope. The scope does not need to transfer the data to an external PC (cut time by 50%), and FFT can be hardware accelerated like it already does (save most of the remaining 50%). I'd gladly trade the fixed gain mode in the bode plot menu for this approach, and leave the auto gain set mode for precision work.

The impulse method also suffers from limited dynamic range, even more so than the broadband noise input method.

The method Siglent utilizes "seems" (don't know exactly how this is implemented) to employ a method similar to the classic Synchronous Sampling technique which effectively straps a narrow bandpass filter around the sampling to significantly reduce outside influence and "noise" effects.

Your DSA efforts sound interesting, and with the available chips could yield excellent results in the hands of a knowledgable designer like yourself

Best,

[mawyatt]

The same "single-shot" approach can be used with other stimulus signals that might give better results:
1) For best SNR/measurement time, we would like to get as much signal power into the DUT as we can but this is usually limited by the max. peak-to-peak voltage range of the generator or the linear range of the DUT and receiver. So a stimulus should have high RMS power at a given level (i.e. it has low crest factor). White noise has high peaks so that is not good (in fact, ideal white noise basically would have infinite crest factor).

Similarly, a unit impulse is essentially the mathematical limit of compressing a given pulse energy into something with an infinitely high peak. So this is generally not something we should strive to approximate as a stimulus source and a big reason people usually do not try to measure impulse response directly. (Because of the theoretical importance of the impulse response it is still a good experiment if you are doing it for the educational value).

2) A nice bonus would be a stimulus that is periodic for our measurement time. Then we would not have to window our FFT and get zero spectral leakage for free.

The most common solution is a sine "chirp", i.e. a sine with continuously varying (instantaneous) frequency. That may sound suspiciously similar to the "stepped" approach but there are no steps and you don't stop for measurements either. You still record it all in one go and then apply FFT. It has the same crest factor as a fixed sine and covers a wide (adjustable!) spectral range.

I also want to briefly mention that maximum length sequences (MLS) could also be used. They are periodic, pseudo-random binary sequences and are quite interesting from a theoretical perspective (but a bit tricky to properly use in practice).

Agree, the impulse method is limited in practical usage for the reasons mentioned, and getting as much signal power as practical (only limited by Signal source or/and DUT linearity limitations) into the DUT within the instantaneous measurement bandwidth in paramount in achieving high dynamic range. We actually developed long ago a simple device we called a "Pinger", this was a Diriac Doublet (+- impulses) that could be injected into many closed loop systems to quickly evaluate the system response. This worked well because most complex linear systems can be distilled into a classic 2nd order function approximation with decent fidelity. The Doublet was used because this introduced no DC term in the closed loop, whereas an impulse does introduce a DC term which might upset certain closed loop systems as they attempt to correct the injected error function.

The sine chirp method is a subset the little known Fourier Transform technique called the Chirped Z Transform, where sine and cosine signals are swept across the band of interest and multiplied against the input signal, then the results convolved. This method can be very effective as we utilized such back in ~1980 to create a handheld real time spectrum analyzer that ran on ~3 watts and had the equivalent computing of 1/3 Cray Super Computer doing the classic FFT approach. We developed custom chips for this including custom CCDs for the Sine and Cosine Convolvers, and this involved some very early Discrete Time Continuous Amplitude signal processing.

It's interesting you mention the MLS, these were utilized way back in early SPICE days as a means to simulate noise performance in the time domain, long before Cadence offered this SPICE feature. We created various files of MLS that were scaled in sampling time and amplitude and utilized by voltage and current sources. These sources were then placed in series and/or parallel with time domain noiseless components like R, L, C, and even non-linear Diodes, BJTs and MOS devices to allow thermal and shot noise effects to be studied in the time domain where we could investigate frequency translation and other non-linear effects.

Better stop as this is getting way off topic, maybe you and TopQuark might want to continue in another thread?

Best,

[tv84]

EEVBlog guru: "I have an idea!"

Mike: "Done that."

I love it. 

[2N3055]

Another way to do this is to use the time domain impulse function as input since it's Fourier Transform is just unity across all frequencies so no need to do an FFT on the input, then perform the FFT only on the output as required.

Good point, might give that a shot

Reason why it is implemented  this way is maximizing dynamic range and resilience to outside noise and signals.

In an ideal world both methods would be implemented in the scope. The scope does not need to transfer the data to an external PC (cut time by 50%), and FFT can be hardware accelerated like it already does (save most of the remaining 50%). I'd gladly trade the fixed gain mode in the bode plot menu for this approach, and leave the auto gain set mode for precision work.

The impulse method also suffers from limited dynamic range, even more so than the broadband noise input method.

The method Siglent utilizes "seems" (don't know exactly how this is implemented) to employ a method similar to the classic Synchronous Sampling technique which effectively straps a narrow bandpass filter around the sampling to significantly reduce outside influence and "noise" effects.

Your DSA efforts sound interesting, and with the available chips could yield excellent results in the hands of a knowledgable designer like yourself

Best,

What must be noted is that Siglent also uses adaptive stimulus level in addition to adaptive amplifier gain. That allows variable drive at different frequencies, which is something broadband noise and chirp cannot do. That is useful both with passive but also with active DUT.  Also with active DUT, single frequency method avoids IMD artefacts that can interfere with measurements at higher signal  levels.
Method used was chosen carefully to extract maximum dynamic range, accuracy and resilience to interference. That comes in hand when looking at switchers for instance. Or in not very EMC quite environment.
There are other methods, but what must be taken in consideration that all those alternative implementations are basically  a different FRA measurement module. They differ enough in implementation that only common factor is scope hardware. So manufacturer would need practically to make and test 3 different FRA modules... With manufacturing and all the process that goes with it is a bit more complicated to when I whip up an Octave program to crunch some numbers  in one afternoon.

But of course, improvements are always welcome even to already decent products.

[2N3055]

EEVBlog guru: "I have an idea!"

Mike: "Done that."

I love it. 

Yep, that's Mike!

[mawyatt]

You folks are funny 

"What must be noted is that Siglent also uses adaptive stimulus level in addition to adaptive amplifier gain. That allows variable drive at different frequencies, which is something broadband noise and chirp cannot do. That is useful both with passive but also with active DUT.  Also with active DUT, single frequency method avoids IMD artefacts that can interfere with measurements at higher signal  levels.
Method used was chosen carefully to extract maximum dynamic range, accuracy and resilience to interference. That comes in hand when looking at switchers for instance. Or in not very EMC quite environment."

Must say that this works quite well indeed, they get amazing performance from that 8 Bit ADC, more so than would be expected

Think this is what separates these Mid-Level DSOs, is what is "done" with the basic hardware that all seem to have and this points to the firmware and it's enabled features that Siglent seems to do a better job with in general.

Anyway, just an option!!

Best,

[Martin72]

Today I´ve tested my TM502A&AM503A system from tektronix, together with a probe from work.
How to get a current flowing thing....Ah...I still got my deskew fixture 
Good to have it here, not only for deskewing in this case...
And the signal looks on the HD really nice...

[Martin72]

Promo-weeks until 31.12., e.g. Batronix:

https://www.batronix.com/shop/oscilloscopes/Siglent-SDS2000X-hd.html

Including free decoder bundle and bandwith upgrade to the next level.

[tautech]

Promo-weeks until 31.12., e.g. Batronix:

https://www.batronix.com/shop/oscilloscopes/Siglent-SDS2000X-hd.html

Including free decoder bundle and bandwidth upgrade to the next level.

Do note the SDS5000X range is also included in this promo:
https://int.siglent.com/info/detail-62.html

[kripton2035]

don't know why I waited some time to buy my 2000hd, but I'm glad I did ...

[Martin72]

Me I always loose in such cases, as I´m unpatient when I want to have some...

[kripton2035]

whaou the decoder offer is worth EUR 2500 !
I don't need most of them, may be i2s one day but this is a nice offer.


edit: RAT! the logic probes are out of stock until 30-40 days...

[tautech]

whaou the decoder offer is worth EUR 2500 !
I don't need most of them, may be i2s one day but this is a nice offer.


edit: RAT! the logic probes are out of stock until 30-40 days...

SPL2016 is not part of the SDS2000X HD promotion. Did you plan to add it to a purchase ?

[kripton2035]

yes, planning to buy the scope (of course) the logic probe and the LA software
... although I expect to unlock the scope one day...

[Martin72]

I´m still hoping there will be a discount offer for the probes only someday.
Yes, there will be DIY alternatives, but I like to have the "original".
But not for this price.

[kripton2035]

2104X HD scope ordered... without the logic probes. will buy later if any discount .

[TopQuark]

Having some fun with writing apps for the scope:

Code: [Select]

/usr/bin/siglent/usr/mass_storage/U-disk1 # ./rusty_siglent_app
Hello, world!!!

This is Rust software written on my computer, compiled and placed on a usb stick, plugged into the scope, and excecuted natively on the scope.
This means there's a possibility to write custom apps to run on the scope, skipping the hassle of going through all the SCPI and data transfer over ethernet stuff. This is getting interesting... 

[kripton2035]

waiting for the rusty app that will dump the scope memory to the usb stick...

[TopQuark]

Wanted to look at the UART console again, took the opportunity to take a peek at the analog front end... Enjoy