Impressive Keysight 1000X series (EDUX1002G modded) SPI triggering rate

[TK]

I recently ran a test comparing the SPI triggering and decoding speed of Keysight EDUX1002G (modded to DSOX1102G) and Siglent SDS1104X-E

The details of the test was posted in a different thread:

https://www.eevblog.com/forum/testgear/everyday-bench-scope-a-3-way-race-edux1002a-gds-1054b-sds1202x-e/msg2412507/#msg2412507

Summary of test: STM32 nucleo board running @ 180MHz.  45MHz SPI clock.  The test outputs 50,000 full cycles of 8 frames from 1 byte to 8 bytes.  When a button is pressed, it sends a frame with 1 byte with value 0x3F.  Normally one of the 8 frames has byte value 0x37 in the same position.

Conclusion: Keysight was able to trigger 50,000 times per second when triggering on byte value 0x37.  Siglent SDS1104X-E triggered 3,000 times.  The difference is due to the fact that the Keysight is doing Serial triggering and decoding in the ASIC and the Siglent implemented it in software after the memory was sampled.

In initial tests, I used a GWInstek GDS1054B and Micsig TO1104 with similar results to the Siglent SDS1104X-E, all of them have serial triggering and decoding implemented in software.

[2N3055]

It would be interesting to see MSO5000 and RTB2000 doing this.

What code and Nucleo board you use?
I would like to try this on 3000T to see if mask is faster..

[TK]

I am using STM32F446 nucleo board.  I am using STM32CUBEMX and AC6 SystemWorkbench (Free IDE). 

You need to disable LD2 (on board LED) to be able to enable SPI1 to run at maximum of 45MHz.

I am getting an MSO5074 next week for the purpose of running the same test.

[JPortici]

I may try with the SDS5034X.
the 5034X has serial triggering done in the FPGA, decoding in software (I actually prefer this, i want to be able to add decoder AFTER. Also, if your decoder-in-asic picks up garbage because of wrong settings, or not high enough samplerate you have to throw away the acquision..) i believed the SDS1000X-E worked the same

[jemangedeslolos]

Hello,

Is it possible to share your System Workbench project  ?
I have severals Nucleo F446 and it will be nice to test on my own scope  : )

Thank you

[rf-loop]

I recently ran a test comparing the SPI triggering and decoding speed of Keysight EDUX1002G (modded to DSOX1102G) and Siglent SDS1104X-E

The details of the test was posted in a different thread:

https://www.eevblog.com/forum/testgear/everyday-bench-scope-a-3-way-race-edux1002a-gds-1054b-sds1202x-e/msg2412507/#msg2412507

Summary of test: STM32 nucleo board running @ 180MHz.  45MHz SPI clock.  The test outputs 50,000 full cycles of 8 frames from 1 byte to 8 bytes.  When a button is pressed, it sends a frame with 1 byte with value 0x3F.  Normally one of the 8 frames has byte value 0x37 in the same position.

Conclusion: Keysight was able to trigger 50,000 times per second when triggering on byte value 0x37.  Siglent SDS1104X-E triggered 3,000 times.  The difference is due to the fact that the Keysight is doing Serial triggering and decoding in the ASIC and the Siglent implemented it in software after the memory was sampled.

In initial tests, I used a GWInstek GDS1054B and Micsig TO1104 with similar results to the Siglent SDS1104X-E, all of them have serial triggering and decoding implemented in software.

Keysight is really lot of more fast to trig serial. This is natural due to very different design.
Of course also slow software system have still one advantage. It can decode offline. So or so, KS is lot of more fast and thats it.

But still I need ask did you all for get best speed using Siglent.
Why I ask:

I have not system what I can use for this speed of SPI or any other serial trigger and/or decode.
So  for this I can not say "anything".

But if look Siglent maximum wfm/s speed.  It is also limiting factor.

If it freerun as fast as it can... 

1 channel on.
Display mode dots
Delay 0
50ns/div
No measurements
Trigger Normal, edge 
Signal square (tested with 5MHz and 20MHz)

average is roughly 118kwfm/s
burst speed roughly 128kwfm/s  (between TFT refreshs, there is extra blind time every TFT refresh (40ms period) roughly 3.5ms)

1 channel on.
Display mode Vextors (Sinc or x)
Delay 0
50ns/div
No measurements
Trigger Normal, edge 
Signal square (tested with 5MHz and 20MHz)

average is roughly 20kwfm/s
burst speed roughly 22kwfm/s  (between TFT refreshs, there is extra blind time every TFT refresh (40ms period) roughly 3.5ms)

And now, same but 4 channels on. As need be with full SPI.

All 4 channels on.
Display mode dots
Delay 0
50ns/div
No measurements
Trigger Normal, edge 
Signal square (tested with 5MHz and 20MHz)

average is roughly 10.5kwfm/s
burst speed roughly 11.4kwfm/s  (between TFT refreshs, there is extra blind time every TFT refresh (40ms period) roughly 3.5ms)

All 4  channels on.
Display mode vectors (Sinc on (x is bit faster ))
Delay 0
50ns/div
No measurements
Trigger Normal, edge 
Signal square (tested with 5MHz and 20MHz)

average is roughly 4kwfm/s
burst speed roughly 4.5kwfm/s  (between TFT refreshs, there is extra blind time every TFT refresh (40ms period) roughly 3.5ms)

As can see even if decode or serial trigger do not affect this make limit, so even if Siglent works hard for make decode faster and tso also serial trigger it can not go very fast due to limits of maximum wfm/s except if they do not make very extremely radical changes.

But perhaps if need maximize Siglent speed as it is now, it is good to use display mode dots or least linear interpolation what also (in some cases) may give tiny amount more speed. But dots mode give much more. (If dots mode is selected no matter if there is  Sinc or x selected)
Normally for digital pulse streams linear interpolation is ok, specially when signal is fast related to samplerate so it do not generate "fake" overshoots. (gibbs ears)

I have no idea how much these different settings may affect for high speed serial trigger and decoding speed.

(Btw, decoding affect also maximum wfm/s speed.  When it capture between TFT refresh not so much but Decode need more time between TFT refresh. (exmple as told in this test, it was around 3.5ms  and if I just only turn decode on with some random settings and without serial signals etc, just only push Decode function on. Every TFT period this blind time cap (some processing) rise to around 20ms. What of course drop also average wfm/s speed.
(segmented acquisition do not have at all this processing time during acquisition but after sequence, then display processing time is - long - because it overlay all segments to TFT (before start next sequence if repeating is selected) also trig rearm is faster. Least with normal signals, example just with previous test (50ns/div and total 45526 segments) all channels on, 20MHz signal independent of display modes, etc... 320ksegment/s (kwfm)
and then one channel in use 487ksegment/s)

[tv84]

@TK,

Could you test with different SPI speeds to see when you get the 50,000 cycles in Siglent and the others?

BTW, I think the test you've done is extremely relevant and I find it very strange how, even doing things in software, the Siglent is only able to do 3.000 captures.

One thing is HW vs SW, the other is realtime vs post processing... 

[rf-loop]

....and I find it very strange how, even doing things in software, the Siglent is only able to do 3.000 captures.

As I told in previous message.
Even with normal edge trigger 4000 captures/second (aka 4kwfm/s)
(4 channels on, 50ns/div, display vectors, interpolation Sinc aka Sin(x)/x)

Even when it can maximum average 118kwfm/s with same 50ns/div and one channel on,  what is most fast for wfm/s speed and other settings so that maximum can reach but more channels on and not speed optimal settings speed is very different. This is not alone serial trigger/decode thing. 
(sequence acquisition is then totally different and there it is nearly 10 times faster than KS but this is totally other case.)

[tv84]

As I told in previous message.
Even with normal edge trigger 4000 captures/second (aka 4kwfm/s)
(4 channels on, 50ns/div, display vectors, interpolation Sinc aka Sin(x)/x)

Even when it can maximum average 118kwfm/s with same 50ns/div and one channel on,  what is most fast for wfm/s speed and other settings so that maximum can reach but more channels on and not speed optimal settings speed is very different. This is not alone serial trigger/decode thing. 
(sequence acquisition is then totally different and there it is nearly 10 times faster than KS but this is totally other case.)

I can't understand completely what you are trying to say... sorry. But I assume it's not easy to predict the value so some other measures/speed tests would be great.

Nonetheless, with all the factors the you seem to be indicating, I find the test a very real scenario and a great demo of  performance/architecture.

[TK]

@TK,

Could you test with different SPI speeds to see when you get the 50,000 cycles in Siglent and the others?

BTW, I think the test you've done is extremely relevant and I find it very strange how, even doing things in software, the Siglent is only able to do 3.000 captures.

One thing is HW vs SW, the other is realtime vs post processing... 

During the test I tried 1MHz and some other SPI speed and the result was similar, I will test later again but I doubt the SDS1104X-E will get any higher than 3,000 trigger updates per second.

[TK]

Hello,

Is it possible to share your System Workbench project  ?
I have severals Nucleo F446 and it will be nice to test on my own scope  : )

Thank you

I tried to clean it and compress, but it is still larger than the max 1MB allowed by the forum.  I will also change the frame output to be different to "Siglent" as it is not relevant to Siglent only anymore, and will find a way to share the project.

[rf-loop]

@TK,

Could you test with different SPI speeds to see when you get the 50,000 cycles in Siglent and the others?

BTW, I think the test you've done is extremely relevant and I find it very strange how, even doing things in software, the Siglent is only able to do 3.000 captures.

One thing is HW vs SW, the other is realtime vs post processing... 

During the test I tried 1MHz and some other SPI speed and the result was similar, I will test later again but I doubt the SDS1104X-E will get any higher than 3,000 trigger updates per second.

Yes, very possible.
But still I ask, did you use dots or vectors display because it is just there around in borderline affecting or not affecting with this t/div (50ns/div in your image) and all or 3 channels on. (and then if also decode is on, it drops more due to longer blind time cap, once in every TFT refresh 40ms period. I do not mean between acquisitions blind time what is other but independent thing in scope processing)

Even without serial trig and/or decode maximum wfm/s speed is as told previously. 
But as you can see maximum is different depending also display  etc settings.
With serial trigger of course it can not be faster than with normal edge trigger or without signal auto free run.
So, with this 3000/s  result.  With dots or vectors mode? If vectors mode was it Sinc or x?

[TK]

@TK,

Could you test with different SPI speeds to see when you get the 50,000 cycles in Siglent and the others?

BTW, I think the test you've done is extremely relevant and I find it very strange how, even doing things in software, the Siglent is only able to do 3.000 captures.

One thing is HW vs SW, the other is realtime vs post processing... 

During the test I tried 1MHz and some other SPI speed and the result was similar, I will test later again but I doubt the SDS1104X-E will get any higher than 3,000 trigger updates per second.

Yes, very possible.
But still I ask, did you use dots or vectors display because it is just there around in borderline affecting or not affecting with this t/div (50ns/div in your image) and all or 3 channels on. (and then if also decode is on, it drops more due to longer blind time cap, once in every TFT refresh 40ms period. I do not mean between acquisitions blind time what is other but independent thing in scope processing)

Even without serial trig and/or decode maximum wfm/s speed is as told previously. 
But as you can see maximum is different depending also display  etc settings.
With serial trigger of course it can not be faster than with normal edge trigger or without signal auto free run.
So, with this 3000/s  result.  With dots or vectors mode? If vectors mode was it Sinc or x?

On both scopes I used vector mode.  I remember trying decode on/off and the result was similar on the Siglent.  I will test dots mode later

[tv84]

BTW, why all others stick at the 3.000 mark?  Their characteristics are not so similar...

This is indeed very interesting/intriguing...

I'm trying to decide on what equipment to buy and now you've thrown this cold water bucket.... 

[TK]

BTW, why all others stick at the 3.000 mark?  Their characteristics are not so similar...

This is indeed very interesting/intriguing...

I'm trying to decide on what equipment to buy and now you've thrown this cold water bucket.... 

I am sorry, on the Micsig it is around 340... Probably I stopped measuring all of them when I was getting between 300 and 1-3K and I continued pushing the limit with the SDS1104X-E and Keysight and forgot about the others.

[TK]

Another interesting capture.  Keysight triggers correctly on 0x37 and decodes the SPI frame (0x37 0x45 0x45 0x56 0x42 0x4C 0x4F 0x4F = "7EEVBLOG"), even when the waveform is very distorted, almost beyond recognition due to having only 1Mpts sample memory.  It was captured with timebase @ 1ms, acquisition: 50MSa/s





Siglent SDS1104X-E capture at 1ms timebase, 500MSa/s, 7Mpts sample memory.  Waveform looks OK, but cannot decode... it runs out of decode memory

[Keysight DanielBogdanoff]

Another interesting capture.  Keysight triggers correctly on 0x37 and decodes the SPI frame (0x37 0x45 0x45 0x56 0x42 0x4C 0x4F 0x4F = "7EEVBLOG"), even when the waveform is very distorted, almost beyond recognition due to having only 1Mpts sample memory.  It was captured with timebase @ 1ms, acquisition: 50MSa/s

That's pretty interesting, I suspect it has something to do with the way the trigger system works. It's *possible* that the trigger's signal path does some edge cleanup with comparitors to get a good trigger, even if the capture is undersampled. I know that internally the ADCs sample at full rate all the time regardless of the time/div, so that could be it too. It would be interesting to see this same capture in peak detect mode and see what the bus looks like.

[rf-loop]

Can you open Siglent decode list when it is running (or both stop and running) (image where is 1ms/div) and tell how many messages it have totally decoded, as you can see in image left bottom there is decoded (MOSI) and light blue mostly tell, without errors. First over 1.5ms from beginning of acquisition.

It have many times told that Siglent have limits in decode result. I have not measured it using SPI but with UART and CORR/ETA: it is 3000 for both simultaneous decoders and IIC (Wire), it is 1000 messages (tested with IIC 32bytes messages and 1 - 32bytes, max decode result length is messages is 1000)

Then open also KS decode list when it is running (or both, run and stop)  (image where is 1ms/div) and tell how many messages it have really decoded in decode list. And what mean this bottom decode display (S1) color change and gap around 1ms from left. And why it show 350kHz.


Also, these other two images, you set KS for 500ns/div
then you set Sig 200ns/div

Also you explain that signal is crap... as can see in KS 500ns/div iomage. Yes it is. Compare this signal with Siglent 200ns/div display. Signal is not so crap.
If you compare, is it better that you keep even bottom basics same.
On the KS screen I think you do not even see this "waveform" what its trigger and decode system see. (due to its very extremely different circuits - is it analog side pathway trigger system? and is it so that same analog side pathway is used for hardware decoding circuits what also mean of course that it can not at all decode offline. When sofware system as Siglent, can do. Example decode from wfm history buffer collected from normal run or from segmented memory (what is nearly 10 times faster than KS)  But positive thing is live (run time) speed what software decode (Siglent) can not)

[Keysight DanielBogdanoff]

Trying to avoid a rep war here, sorry everyone! Regarding online/offline decoding, the Keysight InfiniiVision scopes decode whatever's on screen nearly instantaneously with the ASIC as discussed above. Even changing the time/div setting on a capture will cause the scope to re-decode the screen. This is part of the speed, it only processes what you can see & analyze and ignores what's off screen.

[rf-loop]

Trying to avoid a rep war here, sorry everyone! Regarding online/offline decoding, the Keysight InfiniiVision scopes decode whatever's on screen nearly instantaneously with the ASIC as discussed above. Even changing the time/div setting on a capture will cause the scope to re-decode the screen. This is part of the speed, it only processes what you can see & analyze and ignores what's off screen.

No rep war at all.  Just for this I also make question to TK.  Because some things he show looks bit weird. If you do not recognize these things, sorry.

And also it is well recognized KS is amazing fast.



But then other thing, Daniel..

I have heard many times (not tested myself) that KS can not offline decode.
This is now good place correct this "rumor" if it is true or not.

So, turn normal acquisition on. Trigger (example) edge.
So that this capture include, (example) simple UART signal (so that also most users can test it without needing nearly anything)
After then, stop scope and even if want disconnect signals.
Now there is acquisition - UART data stream.
Turn on Decode.
Now it is decoding offline (as I think what is offline). Just from memory without any live signal decoded until scope is stopped and after then asked fore decode it.

Also it can test using segments. Capture segments (without decoding or even wqithout serial trig)
After then look segments and start decode what is there.

Can it do it or not. Simple question yes or no.
If yes then we need shoot down all these wrong claims and rumors.
I have used Hewlett-Packard test instruments made over tens of years. I have (mostly) deeply respected things what they promise and how all have manufactured..  Today marketing materials and salespersons words looks bit different.

[rf-loop]

Another interesting capture.  Keysight triggers correctly on 0x37 and decodes the SPI frame (0x37 0x45 0x45 0x56 0x42 0x4C 0x4F 0x4F = "7EEVBLOG"), even when the waveform is very distorted, almost beyond recognition due to having only 1Mpts sample memory.  It was captured with timebase @ 1ms, acquisition: 50MSa/s

I believe it is not decoded from this waveform what you see on screen. There in display you see waveform what is produced and displayed using after ADC decimated samples. This useless junk wfm can not be what Decode process use if all you told about test signal is ok. It is somewhat easy to deduce. (and it can clearly also see if you look same signal in Siglent display. (I believe signal is same)
What is real input to Decode processing or serial trigger circuit in KS. Least it is not decimated sample stream. Is it using analog side pathway - comparator principle? I really do not know anymore...

[TK]

Can you open Siglent decode list when it is running (or both stop and running) (image where is 1ms/div) and tell how many messages it have totally decoded, as you can see in image left bottom there is decoded (MOSI) and light blue mostly tell, without errors. First over 1.5ms from beginning of acquisition.

It have many times told that Siglent have limits in decode result. I have not measured it using SPI but with UART and IIC (Wire), it is 1000 messages (tested with IIC 32bytes messages and 1 - 32bytes, max decode result length is messages is 1000)

3000 Bytes

Then open also KS decode list when it is running (or both, run and stop)  (image where is 1ms/div) and tell how many messages it have really decoded in decode list. And what mean this bottom decode display (S1) color change and gap around 1ms from left. And why it show 350kHz.

KS 1000X does not have decode list.  I think it shows 350KHz because it is trying to measure on CH1 that has a distorted waveform and it is not counting the actual 45MHz SPI clock signal

Also, these other two images, you set KS for 500ns/div
then you set Sig 200ns/div

Just trying to show that after some packets (which happens to be 3000), SDS1104X-E stops decoding.

[rf-loop]

Can you open Siglent decode list when it is running (or both stop and running) (image where is 1ms/div) and tell how many messages it have totally decoded, as you can see in image left bottom there is decoded (MOSI) and light blue mostly tell, without errors. First over 1.5ms from beginning of acquisition.

It have many times told that Siglent have limits in decode result. I have not measured it using SPI but with UART and IIC (Wire), it is 1000 messages (tested with IIC 32bytes messages and 1 - 32bytes, max decode result length is messages is 1000)

3000

Then open also KS decode list when it is running (or both, run and stop)  (image where is 1ms/div) and tell how many messages it have really decoded in decode list. And what mean this bottom decode display (S1) color change and gap around 1ms from left. And why it show 350kHz.

KS 1000X does not have decode list.  I think it shows 350KHz because it is trying to measure on CH1 that has a distorted waveform and it is not counting the actual 45MHz SPI clock signal

Also, these other two images, you set KS for 500ns/div
then you set Sig 200ns/div

Just trying to show that after some packets (which happens to be 3000), SDS1104X-E stops decoding.

You can show it better if you zoom to this position where is last decoded message and start undecoded part

Now it looks weird (specially alone this 200ns image because there was not any decoded packet and it looks like .. oh Siglent can not decode at all but then you show KS what have decoded. Nearly like some biased intrument owner except that I do not believe this is case at all.

But how you get decode result out from this image where KS t/div is 1ms.

Also you did not answer this question about KS bottom image decode display question. How many messages it decode in this image?
If there is not decode list with time position and data how you look decoding result at all with this setting. I mean,  what is purpose of Decode if result is unavailable.

If it is unavailable (difficult to believe that it can decode but can not show result - it do not feel weird if it is some unknown junk  but front panel label read Keysight (aka raped Hewlett-Packard) or if it need more exercise how to use these very different machines for get out decode result. As we also know different vehicles driving need different handling.


Ok. So with SPI it have limit 3000,  more than with IIC what have limit 1000. (32000 bytes)

[TK]

Can you open Siglent decode list when it is running (or both stop and running) (image where is 1ms/div) and tell how many messages it have totally decoded, as you can see in image left bottom there is decoded (MOSI) and light blue mostly tell, without errors. First over 1.5ms from beginning of acquisition.

It have many times told that Siglent have limits in decode result. I have not measured it using SPI but with UART and IIC (Wire), it is 1000 messages (tested with IIC 32bytes messages and 1 - 32bytes, max decode result length is messages is 1000)

3000

Then open also KS decode list when it is running (or both, run and stop)  (image where is 1ms/div) and tell how many messages it have really decoded in decode list. And what mean this bottom decode display (S1) color change and gap around 1ms from left. And why it show 350kHz.

KS 1000X does not have decode list.  I think it shows 350KHz because it is trying to measure on CH1 that has a distorted waveform and it is not counting the actual 45MHz SPI clock signal

Also, these other two images, you set KS for 500ns/div
then you set Sig 200ns/div

Just trying to show that after some packets (which happens to be 3000), SDS1104X-E stops decoding.

You can show it better if you zoom to this position where is last decoded message and start undecoded part

Now it looks weird (specially alone this 200ns image because there was not any decoded packet and it looks like .. oh Siglent can not decode at all but then you show KS what have decoded. Nearly like some biased intrument owner except that I do not believe this is case at all.

But how you get decode result out from this image where KS t/div is 1ms.

Also you did not answer this question about KS bottom image decode display question. How many messages it decode in this image?
If there is not decode list with time position and data how you look decoding result at all with this setting. I mean,  what is purpose of Decode if result is unavailable.

If it is unavailable (difficult to believe that it can decode but can not show result - it do not feel weird if it is some unknown junk  but front panel label read Keysight (aka raped Hewlett-Packard) or if it need more exercise how to use these very different machines for get out decode result. As we also know different vehicles driving need different handling.


Ok. So with SPI it have limit 3000,  more than with IIC what have limit 1000. (32000 bytes)

Isn't SDS1104X-E decoding after the capture?  Then why when capturing at 1ms timebase, it decodes 3000 packets, then when I change the timebase to where it triggered, it doesn't decode what is being shown in the screen?  Wasn't it supposed to decode whatever is in shown as waveform in the screen?  It is a bug or doesn't make any sense to me.

The color change in the KS decode information when showing at 1ms timebase is a visual effect.  You can move wherever in the 1ms capture and change timebase to 200us or whatever and it will show the correct decoded information, without need to do another capture.

So I am not sure which scope is doing post decode and which not...

[iMo]

Capturing SPI is easy. Rising clock edge takes the data. 45MHz SPI could be decoded by $1 CPLD (or a 10cent 74HC shift reg). No special processing is needed. It could be the KS is simply reading SPI data that way, while the others mess with an analysis of the sampled waveforms.