Siglent SDS800X HD Review & Demonstration Thread

[2N3055]

Performa01
I'm not impressed by the HF response step size the first 15ns; there is much undefined over and undershoot. The same behavior can also be seen on the SDS1000HD. Almost as if they use the same front-end design.

Can you also make a test on a 100Hz square wave input signal ? Some recent Siglent scopes also have problems with this (not showing a real square)

Thank you

Why would they not share input design, when their specifications for inputs are pretty much the same...? Of course, they would only share 1MΩ input path design.


Are you referring to images with top part of signal zoomed in 20X ?
(Which by the way is something not many scopes can do, not to mention in this class.)

Your definition  of recent is very different from mine.  Problems with compensation were in scopes from 2017..  that is going on to 7 years now... there were 3 generations of new products released. Just saying...

Picture SDS824X_HD_PR_1GSa is NOT zoomed in, and clearly shows strange strange aberrations, almost there is a reflected wave interfering with one of the internals...
A TDR or VNA  would shed some light on this to see what is going on. (at least on a SDS1000X HD this is possible, but outside the scope of this thread)

There is nothing wrong with sharing a good design, I did not imply this is a bad thing, just an observation, so no defense from your side is needed here.

As for the LF compensation in the front-end, Siglent does not have a good track record for this, specially on the low end scopes. I consider the SDS800 as a low end scope give the fact it has only a 7''  screen hence my curiosity for the LF distortion on this model.

Thanks Performa01 and RF-loop for the tests

Thank you for explaining.  I wasn't defending, just wanted to know what exactly you mean.

But I think "As for the LF compensation in the front-end, Siglent does not have a good track record for this, specially on the low end scopes" is a bit over the top.
There were 2 entry level scopes, 7 years ago, then new, sharing input design that had error in manufacturing on one batch.

Shall I start with all the major problems Keysight, Tek, R&S had in their equipment over the years.
From outright bad designs to major manufacturing problems. Does that makes them "they don't have good reputation making T&M equipment"
Of course not.

He who works makes mistakes... that is just a fact.
In a perfect world there would be none.

In real world, as long as problem is recognized and taken care off, it is all we can ask for.

[Performa01]

There were 2 entry level scopes, 7 years ago, then new, sharing input design that had error in manufacturing on one batch.

Even that statement is not correct.

The SDS1202X-E, introduced to the public in Autumn 2016, had missing compensation capacitors in the frontend in the very first production batch.

Somewhere I've read claims that the SDS1104/1204X-E, introduced in early 2018, also was affected - but this is Chinese Whispers at best. I got my SDS1104X-E several months before its market start and it did not have any issues. You'll also not be able to find any complaints from users about such issues.

[2N3055]

There were 2 entry level scopes, 7 years ago, then new, sharing input design that had error in manufacturing on one batch.

Even that statement is not correct.

The SDS1202X-E, introduced to the public in Autumn 2016, had missing compensation capacitors in the frontend in the very first production batch.

Somewhere I've read claims that the SDS1104/1204X-E, introduced in early 2018, also was affected - but this is Chinese Whispers at best. I got my SDS1104X-E several months before its market start and it did not have any issues. You'll also not be able to find any complaints from users about such issues.

I stand corrected then. Thank you for clarification.

[Performa01]

The following content has been added:

Reply #7
https://www.eevblog.com/forum/testgear/sds800x-hd-review-demonstration-thread/msg5293762/#msg5293762

1.   Digital Channels

The opening posting has been updated accordingly.

[ebastler]

Many thanks for being very clear on the capabilities as well as limitations! I also appreciate the older photos of the probes (in the 1104X-E thread, had not come across these yet) -- they give a good impression what to expect.

I am pleasantly surprised that in your screenshot, the digital channels hardly ever show the +- 1 ns skew stated in the datasheet. The DHO900 logic analyser is specified for +- 5 ns skew between digital channels, typical (!). Although I don't know whether it actually exhibits these rather large deviations.

[Performa01]

The following content has been added:

Reply #7
https://www.eevblog.com/forum/testgear/sds800x-hd-review-demonstration-thread/msg5293762/#msg5293762

1.   Digital Channels: fast pulse measurement added

Reply #8
https://www.eevblog.com/forum/testgear/sds800x-hd-review-demonstration-thread/msg5293765/#msg5293765

1.   Measurements 2 (Trend/Track)
2.   FFT Dynamic range

The opening posting has been updated accordingly.

[Performa01]

The following content has been added:

Reply #9
https://www.eevblog.com/forum/testgear/sds800x-hd-review-demonstration-thread/msg5293771/#msg5293771

1.   True Vertical Sensitivity
2.   Peak Detect
3.   Bode Plot at a glance

The opening posting has been updated accordingly.

[Martin72]

Thank you for your incredible efforts....    

[Performa01]

Bode Plot Example

Here is my standard test for Bode Plot: a simple 455kHz IF filter, consisting of a Kyocera KBF-455R-20A ceramic 6 element filter with two resonant 2nd order L-matching networks for the 50/1500 ohm impedance transformation at both the input and output.


SDS824X HD_Bode_50_S21_IF455kHz

A complex structure like this has a somewhat chaotic phase response, especially in the passband and the transitions into the stopband. The amplitude shows nice steep transitions into the stopband, yet there are some unwanted resonances as is typical for this type of filters. The important fact is that we need high frequency resolution to capture all the fine details. Furthermore, this test demonstrates at least 90 dB dynamic range.

The data table has been adjusted to the nominal center frequency of the filter, which should be 455 kHz. A vertical cursor marks the selected frequency in the plot and from the table we can see that insertion loss of this filter is some 3.66 dB. The frequency with the lowest attenuation of 3.57 dB is 458 kHz though. This “data cursor” is independent of the manual cursors that are available in the Display menu.

I’ve set up some specific measurements: UF (Upper Frequency), LF (Lower Frequency) and BW (Band Width) to characterize the most important properties of the filter. We can see at a glance that the bandwidth is precisely 21.4 kHz.

For those bothered by the complex phase plot, there is always the option to disable any trace we like:


SDS824X HD_Bode_50_S21_IF455kHz_NP

[tszaboo]

Few questions: Is that 500 steps to measure the Bode plot? What did you use to make the signal, USB to a desktop AWG or the S-BUS AWG?

[Performa01]

Few questions: Is that 500 steps to measure the Bode plot? What did you use to make the signal, USB to a desktop AWG or the S-BUS AWG?

It is 501 points Bode Plot with an external SDG6052X connected via USB.

For better dynamic I've used the SDG6052X in Tracking mode, so I could connct the reference channel and the DUT directly to the AWG and didn't need a power splitter, which would have caused 6 dB signal loss.

[mawyatt]

Few questions: Is that 500 steps to measure the Bode plot? What did you use to make the signal, USB to a desktop AWG or the S-BUS AWG?

It is 501 points Bode Plot with an external SDG6052X connected via USB.

For better dynamic I've used the SDG6052X in Tracking mode, so I could connct the reference channel and the DUT directly to the AWG and didn't need a power splitter, which would have caused 6 dB signal loss.

Good idea using the AWG tracking mode instead of a splitter when you have a 50 ohm system

Best,

[rf-loop]

BodePlot. It is also good to look for performance limits. So what can't be done with it. Here are a couple of examples of where performance becomes a limiting factor. In one, the dynamics of the measurement is limiting, and in the other, the frequency resolution.

First, an IF filter built using the "Bob Peace" style. (there is cascaded 2 very cheap Murata filters and some "this and that" for reduce cross over etc.)
In addition to the form factor of the filter, the stopband is extremely important. This doesn't even come close to see it. Here it is necessary to reach at least somewhere -120----130dB below the passband and this can not at all.
The steepness of the edge is also such that it would be good to have more frequency resolution. Of course, the edges could be run separately with a narrower Span if need go more deep.


455kHz IF filter wide sweep
Green trace is BP noise level. (just empty channel) There can see jump in noise level perhaps due to overlapping between two different RBW when RBW change, just there is one point where BodePlot "receiver" bandwidth (RBW) is changing.


Same filter, BP Span now 50kHz

It is here clear that this BodePlot can not at all see this filter stop band charasteristics. For this, there need be lot of more dynamic range down from 0dBm level.

Steep edges can analyze separately if need more frequency resolution. (50kHz Span minimun step is 100Hz what is not enough for very steep and narrow filters, but this filter is not so steep)



Then two xtal resonators. The frequency resolution ends in both. That approximately 84kHz resonator should be driven with perhaps a 0.1Hz step sweep. Because the minimum Span is 500Hz and the maximum number of samples 501 limit in this case limit is 1Hz.


Here bit over 84kHz resonator.
Now Span is 500Hz and 1Hz steps. (501 pts). It is clear that this 1Hz step is far too big for this if need carefully look series and parrel resonances.
It need least 10 times more resolution.

Notice that the phase information in the table is just random noise. It's a pity that in scalar measurements that part cannot be closed. So user have to ignore them only in his own head. (In this case there is not reference signal at all.)



And same here in next.

Cheap around 2.4MHz resonator. Sweep resolution is not well enough but here (if any reason) if need look series and/or parallel resonances mode more detailed, here can do it running these separately using more narrow span. Limit is 500Hz Span and 501pts what give 1Hz resolution.




In the ad, of course, a real careless salesman could write that the resolution is up to 275204396 pts/decade, which is not false if we are precise. On the other hand, it is in all cases at least 70/decade when full sweep 10Hz - 120MHz.







Finally, a picture of what it can do quite easily.


This is very simple cascaded HiPass CRCR (green trace) and LoPass RCRC filter for get BandPass (yellow trace). Traces are BandPass, HighPass and one trace (pink) is after first CR in HiPass.
As can see there is visible also this capacitor resonace.
Green trace signal use 1x probe. (this is reason for higher freq attenuating)

But overall it still amazing how much this can do.

It's good to know what can do. It's also good to know what can not.

[Performa01]

SPI Speed Test

We could ask the question: how fast an SPI data stream can we decode with a 200 MHz DSO? Huch much oversampling do we need?

For some DSOs it is said that they need a fair bit of oversampling for proper decoding. Yet the answer is, a proper implementation doesn’t require much in this regard, hence it should be perfectly adequate to have a bandwidth three times the SPI clock frequency. The sample rate on the other hand should be irrelevant, as long as the Nyquist criterion for the required bandwidth is fulfilled. That makes for more than 6 times the SPI clock frequency.

The SDS824X HD true bandwidth is limited to 200 MHz in full channel mode and its sample rate is 500 MSa/s. According to the hypothesis stated above, this bandwidth would allow a max. SPI clock of 66 MHz and the sample rate is sufficient for that. The only way to know how good this works is to try it out…

I felt adventurous, hence didn’t bother with 66 MHz, but tried 100 MHz right away:


SDS824X_HD_SPI_100Mbps

The above screenshot shows a bidirectional (full duplex) 100 Mbps SPI data stream with a message length of 11 bytes. No special trigger has been used, just falling edge trigger on the /CS line.

The decoding works without issues, but how can we prove that the results are correct? Here’s a simple test: I’ve replaced the MOSI signal with a phase synchronized copy of the 100 MHz clock signal, phase shifted 90° such that it is always sampled at its low state and then shifted by another 180° so that it is always sampled at its high state. This way we’re putting the maximum stress on the acquisition chain, whereas it would be an easy task if we had just static signals for MISO/MOSI.


SDS824X_HD_SPI_100Mbps_0


SDS824X_HD_SPI_100Mbps_1

The expected results of all 0x00 for the first test and all 0xFF for the second one has been achieved, proving that there is a good chance to decode a 100 Mbps SPI data stream correctly with only twice the bandwidth and five times the sample rate.

If we trigger on the rising edge of the clock, we can produce an eye diagram which clearly shows that there is plenty of margin for correct decoding. Even the MISO signal, while slightly delayed, causes no problem in this scenario.


SDS824X_HD_SPI_100Mbps_Eye

The fast pulses look very soft on the 200 MHz scope and the clock signal is a pure sine now, since we can only capture the 2nd harmonic, but not the 3rd.

Compare this to the same clock and MOSI signals displayed at ten times the bandwidth of the SDS6204X HD:


SDS6204_Pro_H12_SPI_100Mbps_Eye

So, could we go up to even 200 Mbps maybe?


SDS824X_HD_SPI_200Mbps_Eye

This is quite revealing. There might be a small chance to get it working with a fair share of luck, yet this certainly isn’t a robust solution. The transitions are way too slow to give any reasonable error margin for the decoder.

[gf]

SDS824X_HD_SPI_200Mbps_Eye

This is quite revealing. There might be a small chance to get it working with a fair share of luck, yet this certainly isn’t a robust solution. The transitions are way too slow to give any reasonable error margin for the decoder.

The eye diagram is still very nice and has fully open eyes. If MISO/MOSI are sampled at the zero-crossing of the clock, there is still plenty of error margin, although there is a small time offset between the 4 signals. I think the potential problem is rather that the decoder is possibly working directly with the original (asynchronous) 500MSa/s samples. With prior interpolation/upsampling, I think decoding should still work fine. Can the decoder run on an interpolated math trace?

[Performa01]

The eye diagram is still very nice and has fully open eyes. If MISO/MOSI are sampled at the zero-crossing of the clock, there is still plenty of error margin, although there is a small time offset between the 4 signals. I think the potential problem is rather that the decoder is possibly working directly with the original (asynchronous) 500MSa/s samples. With prior interpolation/upsampling, I think decoding should still work fine. Can the decoder run on an interpolated math trace?

Yes, the eye is wide open, but timing is extremely critical. The slightest phase shift throws off the decoder and during my tests, I could never get it completely stable.

I do not know whether the decoders work on the raw sample data, but I guess not on the other hand I do know that there is no reconstruction/interpolation for slower time bases, where the record length approaches or exceeds the screen width. It might rather also have to do with the digital 200 MHz bandwidth limiter, that is activated whenever more than two channels are active and helps to suppress aliasing artifacts in the region from 250 to ~290 MHz.

Decoders cannot work on anything other than true input channels. Depending on the operation, Math traces might get too slow to act as reliable source for the decoders, and even more importantly they are not suitable as a trigger source. The serial triggers are implemented in hardware and operate on a separate trigger data stream, which is never decimated but always either sin(x)/x reconstructed or linearly interpolated, according to the user settings.


The 200 Mbps SPI challenge

While we cannot decode a SPI data stream in 4-channel mode, it appears to be no problem if we stick to just two channels (I’ve switched to 32 bit words by now):


SDS824X_HD_SPI_200Mbps_2Ch

Here’s the obligate 0 and 1 test:


SDS824X_HD_SPI_200Mbps_2Ch_0


SDS824X_HD_SPI_200Mbps_2Ch_1

… and the eye diagram:


SDS824X_HD_SPI_200Mbps_2Ch_Eye

The eye-diagram looks way better now, as the bandwidth is 245 MHz in this configuration and there is no digital filter. The sample rate is twice as high, so your guess is as good as mine as to which of these factors actually makes all the difference.

EDIT: my guess now is that it's actually the sample rate.

[mawyatt]

BodePlot. It is also good to look for performance limits. So what can't be done with it. Here are a couple of examples of where performance becomes a limiting factor. In one, the dynamics of the measurement is limiting, and in the other, the frequency resolution.

First, an IF filter built using the "Bob Peace" style. (there is cascaded 2 very cheap Murata filters and some "this and that" for reduce cross over etc.)
In addition to the form factor of the filter, the stopband is extremely important. This doesn't even come close to see it. Here it is necessary to reach at least somewhere -120----130dB below the passband and this can not at all.
The steepness of the edge is also such that it would be good to have more frequency resolution. Of course, the edges could be run separately with a narrower Span if need go more deep.


455kHz IF filter wide sweep
Green trace is BP noise level. (just empty channel) There can see jump in noise level perhaps due to overlapping between two different RBW when RBW change, just there is one point where BodePlot "receiver" bandwidth (RBW) is changing.


Same filter, BP Span now 50kHz

It is here clear that this BodePlot can not at all see this filter stop band charasteristics. For this, there need be lot of more dynamic range down from 0dBm level.

Steep edges can analyze separately if need more frequency resolution. (50kHz Span minimun step is 100Hz what is not enough for very steep and narrow filters, but this filter is not so steep)



Then two xtal resonators. The frequency resolution ends in both. That approximately 84kHz resonator should be driven with perhaps a 0.1Hz step sweep. Because the minimum Span is 500Hz and the maximum number of samples 501 limit in this case limit is 1Hz.


Here bit over 84kHz resonator.
Now Span is 500Hz and 1Hz steps. (501 pts). It is clear that this 1Hz step is far too big for this if need carefully look series and parrel resonances.
It need least 10 times more resolution.

Notice that the phase information in the table is just random noise. It's a pity that in scalar measurements that part cannot be closed. So user have to ignore them only in his own head. (In this case there is not reference signal at all.)



And same here in next.

Cheap around 2.4MHz resonator. Sweep resolution is not well enough but here (if any reason) if need look series and/or parallel resonances mode more detailed, here can do it running these separately using more narrow span. Limit is 500Hz Span and 501pts what give 1Hz resolution.




In the ad, of course, a real careless salesman could write that the resolution is up to 275204396 pts/decade, which is not false if we are precise. On the other hand, it is in all cases at least 70/decade when full sweep 10Hz - 120MHz.







Finally, a picture of what it can do quite easily.


This is very simple cascaded HiPass CRCR (green trace) and LoPass RCRC filter for get BandPass (yellow trace). Traces are BandPass, HighPass and one trace (pink) is after first CR in HiPass.
As can see there is visible also this capacitor resonace.
Green trace signal use 1x probe. (this is reason for higher freq attenuating)

But overall it still amazing how much this can do.

It's good to know what can do. It's also good to know what can not.

My god man, it's just a DSO after all

Evidently (we don't have one.....yet!!), a pretty damn good one at that

Best,

[Performa01]

Mask Test

Consistent with the previous SPI speed test, I want to demonstrate the usefulness of the full speed mask test, as it’s a perfect tool for automatic eye diagram monitoring.

Let’s get back to the 100 Mbps eye diagram of the previous test and set up a mask for it. For this, there is an integrated mask editor; we cannot just use the automatic mask creation here, because we don’t want to define a particular waveform with certain tolerances, but rather some forbidden area, so that the “eye” stays wide open.


SDS824X HD_Mask_Editor

I didn’t spend much time creating a perfect mask; this one has just been cobbled together quite quickly. Furthermore, the eye most likely doesn’t need to be that wide open in practical applications, yet this is just for demonstration’s sake.

Now we let the mask test run. With the clean signals produced in a near ideal lab environment, we could have run that test for days without any failure, so I’ve added 224 mVrms noise with 500 MHz bandwidth to the data signal, hoping that I’ll get a mask violation eventually:


SDS824X HD_Mask_Test_run

The mask test is implemented in hardware, so we are getting a high number of passes within a short time. The screenshot has been taken a few seconds after the test started and up to this point, no mask violation has occurred. Yet it only takes a little while longer and a total of 411262 test runs until the first violation occurs.


SDS824X HD_Mask_Test_fail

As can be seen, the mask test was set up in a way that it would beep and stop on the first violation. Alternatively, we can also get an automatic screenshot every time the violation occurs.

This is another small difference to the SDS2000X HD: there we can also have the option “Failure to History”, which means that all mask violations (and only these) are stored in the history, so we can have the mask test run overnight and then analyze all the mask violations in peace later.

The info block in the display tells us that the failure rate in this test scenario was less than 0.001%. Of course, it can be seen quite clearly that this violation wouldn’t have prevented the SPI decoder to deliver correct results. It’s all a matter of setting up the mask appropriately…

[Performa01]

Zoom Expectations

When using a 12-bit DSO, some people tend to get enthusiastic and expect miracles. Maybe we all should come back to earth and ask ourselves what we can realistically expect.

There’s the zoom, where the SDS1000X HD has been shown to exhibit some Sinc artefacts at extreme zoom settings. But shouldn’t we first of all get a feeling what a realistic zoom factor is?

Consider a signal that has to be viewed at a vertical gain of 1 V/div in order to fit on the screen:


SDS824X HD_PR_H50ns_Stop

Now we want to zoom in, e.g. 5x:


SDS824X HD_PR_H50ns_Stop_Zoom5

We already see a bit of noise creeping in. Apart from that, we should determine how much zoom is feasible, before we try to zoom in any further:

Just like the SDS800X HD, an SDS1000X HD will have 480 LSB (aka codes) per division. Consequently, 20x zoom is about the sensible limit, because then we get 24 LSB per division – and with this, there would still be a chance to see something meaningful. 20x zoom means 1 V / 20 = 50 mV/div:


SDS824X HD_PR_H50ns_Stop_Zoom20

Now the noise is stronger and it already gets hard to spot any details. Yet we can take it to the extreme and try 100x zoom:


SDS824X HD_PR_H50ns_Stop_Zoom100

We are now at 4.8 codes per division and what we see is just (mostly granular) noise – all this has nothing to do with the real signal anymore. Most obvious when viewing it in Dots mode.


SDS824X HD_PR_H50ns_Stop_Zoom100_Dots

We can take it one step further and zoom in horizontally as well (while still in Vectors mode), so that it becomes less obvious that we’re actually looking at pure noise:


SDS824X HD_PR_H2ns_Stop_Zoom100

As can be seen, the SDS800X HD doesn’t show any Sinc artifacts, yet the much more important insight should be that it is completely irrelevant what a DSO shows at such extreme zoom levels, where we see nothing but noise anyway.

Of course, we could also do it the correct way. Whenever we need to use some extreme zoom, then we also need a means to

a)   Increase the vertical resolution
b)   Reduce the noise

The tool of choice for repetitive signals is Averaging, because it increases the vertical resolution, suppresses any modulation (hence also noise) and doesn’t affect the bandwidth.

Even at the extreme 100x zoom setting we can get the following:


SDS824X HD_PR_H50ns_Stop_Zoom100_AVg1024

We leave the input channel at its original gain of 1 V/div, but set the final time base while still in Run mode and set up a math trace with averaging, which can be displayed at any vertical scale, i.e. also 10 mV/div for a 100x zoom. When the desired number of averages has been processed, we can stop the acquisition, yet this is not required, as the input channel is left unchanged anyway. In this example it is 1024 averages shown at 10 mV/div, hence a 100x zoom again.

Now compare this with the previous screenshots. This one now has much less curves and kinks than even the pointless capture at 2 ns/div before, proving that it’s just nonsense (=noise) what we get at such zoom levels without proper averaging.

Of course, 1024 averages are quite a lot. In theory, it would enhance the resolution by 10 bits, making for a total of 22 bits. The current platform doesn’t support sample data and digital signal processing results at a higher resolution than 16 bits, so this is what we get as soon as we use 16x or higher averaging. The ENOB on the other hand benefits a lot more from this, even though it’s limited to 16 bits as well. But it starts at just 8.4 bit according to the data sheet, and 1024x averaging will increase this by 5 bits for a total of ~13.4 bits.

[ebastler]

There’s the zoom, where the SDS1000X HD has been shown to exhibit some Sinc artefacts at extreme zoom settings. But shouldn’t we first of all get a feeling what a realistic zoom factor is?

Thank you for following up on the discussion in the SDS1000/3000X HD thread. But I think you missed a key point of that discussion.

Even without any zoom (in your first screenshot), the effective vertical resolution of the trace display is cut in half! There are always pairs of identical pixels in the vertical direction, resulting in a trace that is fatter than necessary and has lower resolution -- despite the fact that there must be plenty of vertical resolution to spare in the actual sample data. See the magnified view below.

Why does the scope do that? The Math trace in your final screenshot shows proper resolution of individual pixels.

[Performa01]

Thank you for following up on the discussion in the SDS1000/3000X HD thread. But I think you missed a key point of that discussion.

Even without any zoom (in your first screenshot), the effective vertical resolution of the trace display is cut in half! There are always pairs of identical pixels in the vertical direction, resulting in a trace that is fatter than necessary and has lower resolution -- despite the fact that there must be plenty of vertical resolution to spare in the actual sample data.

Just because I didn't address that point yet, it doesn't mean that I "missed the point".

Regarding the pixel rendering, I've already looked into this and prepared a posting, until it occured to me that this product is not even released yet and I should rather report any issues, rather than publishing them. Because that's my personal approach for public reviews prior to a product release: I publish the content as I've planned it and minor bugs that I notice along the way will be reported, but not explicitely published. In case of severe bugs that prevent me from doing what I have planned, I would refuse to publish anything at all and just say the product is not yet ready for review.

To cut a long story short, the "point that I missed" has long been reported when I posted the "Zoom Expectations" article.

Of course the Sinc artifacts in the SDS1000X HD are a bug, and this has been reported quite a while ago (not by me, as I don't have an SDS1000X HD), but there has been Chinese New Year silence in the past two weeks and then again, just like the SDS800X HD, the SDS1000X HD is not even released outside of China yet.

[ebastler]

Just because I didn't address that point yet, it doesn't mean that I "missed the point". [...] To cut a long story short, the "point that I missed" has long been reported when I posted the "Zoom Expectations" article.

Apologies if I caused offense, I did not mean to. It just appeared to me that you had posted this in direct response to the current discussion in the SDS1000/3000X HD thread, and I wanted to point out that the vertical screen resolution issue discussed there is in fact not related to the Zoom function.

I think Siglent would be doing themselves a big favor if they can change the trace rendering to make use of the full vertical screen resolution. In the SDS1000/3000X HD thread, this started as a complaint about the "poor screen" used in the Siglents, in comparison with the Rigol DHOs -- but it turns out that the slightly lower physical resolution only plays a very minor role in this.

Edit: Thanks for the explanation in the other thread. So this is not going to be improved, due to hardware (FPGA block RAM) constraints in the budget HD scopes. Which is not a big deal for the SDS800X HD with its smaller screen, but a pity for the SDS1000X HD, in my opinion.

[Performa01]

Serial Decoders

This is just a quick test to demonstrate some serial decoders. For convenience sake (simple connections to the test board without having to bother with probes), I’ve used the digital channels of the SLA1016.

The I2C decoder has already been used in the course of the Digital Channels test in Reply #7, so I’ll not show it again here:

https://www.eevblog.com/forum/testgear/sds800x-hd-review-demonstration-thread/msg5293762/#msg5293762

Here’s some SPI data stream with simple edge trigger on /CS:


SDS824X HD_Digital_SPI

This is some UART data stream in 8N1 format with serial UART trigger on the start condition of a byte:


SDS824X HD_Digital_UART

Finally, a CAN telegram with serial CAN trigger on the start condition.


SDS824X HD_Digital_CAN

[Performa01]

Cursors
Cursors

Not everyone might be familiar with the various concepts of cursor measurements, so here comes a brief explanation together with some examples.

There are three kinds of cursors: Manual, Tracking and Measure.

In Manual mode, the x and y cursor pairs can be moved freely and used like calipers to measure distances in both axes, even at the same time if so desired.


SDS824X HD_Cursors_Manual

In Track mode, the x-cursor pair can be moved freely, whereas the y-cursor pair will always track the selected signal trace. Of course, this makes most kind of cursor measurements much easier and also more precise.


SDS824X HD_Cursors_Track

Finally, in Measure mode the cursors are fully automatic and just visualize the points of the signal trace that are used by a certain automatic measurement. Here are some examples:

Rise Time


SDS824X HD_Cursors_Measure_Rise

For the 10-90% rise time measurement, the measure cursors show the corresponding positions for the 10% and 90% threshold on the time axis.

Rising Edge Overshoot


SDS824X HD_Cursors_Measure_ROV

For the rising edge overshoot, the measure cursors show the amplitude difference between top and positive peak.

Positive Pulse Width


SDS824X HD_Cursors_Measure_Width

For the pulse width (according to the thresholds configured in the Measurement Config, default is 50-50%), the measure cursors show the corresponding positions on the time axis.

[Performa01]

System Performance with SP5050A Probe

Many folks worry about the adequacy of the supplied probes.

There are few situations where it would be appropriate to use passive 10x high impedance probes at a test node within a circuit carrying >100 MHz signals. With a tip capacitance of 10 pF the impedance at 100 MHz is just 160 ohms and forms a low-pass filter together with all not extremely low source resistances. This might still be okay for low impedance nodes like the outputs of line drivers, but certainly not anywhere else.

Apart from the fact, that at higher frequencies alternative probing solutions are required, the previous test of the good old 100 MHz PP510 has shown that they do not limit the system bandwidth, but even extend it to ~274 MHz and the rise time is quite adequate at 2 ns, see section “Probe Bandwidth” in Reply #5:

https://www.eevblog.com/forum/testgear/sds800x-hd-review-demonstration-thread/msg5293756/#msg5293756

If you wonder what you could gain with a much better (and much more expensive) probe – here’s a test with Siglent’s top model, the 500 MHz SP5050A.

Just to give you an idea, I’ve once tested the quite similar SP3050A, which is also rated 500 MHz, as always using the industry standard test with 25 ohm source impedance (2nd screenshot in Reply #90):

https://www.eevblog.com/forum/testgear/at-last-siglent_s-sds5054x-touchscreen/msg2165488/#msg2165488

On a 1 GHz SDS5104X, the system bandwidth was 1.096 GHz with this “500 MHz” probe. Of course, this is useless in practice as it would only work on a terminated 50 ohm port with the supplied coax-adapter – and in that case we could dispend with the probe and just use a direct coax connection instead.

Now here’s the system frequency response with SP5050A up to 500 MHz:


SDS824X_HD_Probe_PP5050A_FR

It can be seen, that this 500 MHz probe extends the system bandwidth to ~291 MHz (274 MHz with PP510 and 244 MHz with direct coax connection).

Of course, the probe has been properly LF-compensated prior to the measurements:


SDS824X_HD_Probe_PP5050A_PR_1kHz

The transition times are now in the 1.7-1.8 ns ballpark, hence very similar to the direct coax connection.

The ultimate test for proper HF-compensation is done with a fast (1 ns) risetime 1 MHz square wave.


SDS824X_HD_Probe_PP5050A_PR_1MHz_Zoom

It can be seen, that the SP5050A is a pretty good match for this scope because the initial overshoot is about the same level as the top of the pulse.

Verdict: yes, the SP5050A outperforms a PP510 in just about every regard. It’s a very nice probe overall and it sure demonstrates the bounds of possibilities with the SDS824X HD. There still isn’t a huge difference after all.