EEVblog #1312 - Siglent Oscilloscopes CRIPPLING History Mode!

[EEVblog]

Siglent oscilloscopes have a crazy crippling automatic waveform History mode that is always on, you can't switch it off, and it gobbles up all your sweet juicy sample memory.
Why?

[tom66]

This is a design decision.

You can capture a longer timebase and zoom in. That's essentially the same as what you're asking for.  Siglent have just traded off memory for a history function. The default setting on most newer Rigol scopes is the same - when in 'Auto' mode there is almost no memory saved outside of the capture area.  To do what you are asking,  Siglent would have to capture ~100ms of data for every acquisition, which would seriously drop the acquisition frame rate, and customers might not understand that the memory depth setting affects this.

I did notice that the Siglent scope we had on loan did not have any auto memory depth setting which I found strange.

[Kleinstein]

It is a design decision to use the horizontal scale knob to set the length of the captured signal.
It is a little odd in the single mode - but not a big deal, as one can start slower first and than zoom in. So there is an easy work around.
A separate setting in run mode may make sense: one may need the display to manually judge the signal and than decide to capture and in this case may want the data well before that event.
The question is a little what does the memory depth setting do than ?
If used for the lenght of the segments one would need a finer granularity, not just decades.

[mroek]

This is something I never really noticed, but now that I know, it is kinda annoying that Siglent does it this way. I would prefer to be able to make this choice myself, and if one could actually turn off history mode and instead have the scope use all the memory for the single capture, that would provide the best of two worlds. History mode/segmented memory is a useful thing, but it should be up to the user how the memory is utilized.

The workaround using zoom mode is doable, but quite clumsy compared to just zooming out on whatever is on your screen. Anyways, thanks for the video, at least I learned something new.

[David Hess]

So now we have to start carefully distinguishing between memory depth and record length.  Marketing will of course advertise the former rather than the later because it will be larger.

How does this limitation make sense for performance when the same amount or even less of the data would be processed?  What requires more processing when a smaller part of the acquisition record is displayed and the sample rate did not change?

[tom66]

So now we have to start carefully distinguishing between memory depth and record length.  Marketing will of course advertise the former rather than the later because it will be larger.

How does this limitation make sense for performance when the same amount or even less of the data would be processed?  What requires more processing when a smaller part of the acquisition record is displayed and the sample rate did not change?

It has always been record length.

The Rigol DS1000Z, for instance, has 64MB of DDR(2/3) memory tied to the FPGA.  In normal acquisition, only 24Mpt of memory is available in any configuration: 1ch/2ch/4ch, with the memory divided between enabled channels. However, the full memory length is available when segmented memory is selected, in individual segments.

Why they made this choice is not understood by me,  but it might be because they want two buffers of memory available (plus scratchpad for trigger correction or other management), and alternately stream one or the other to the processor. Or, it might simply be a marketing decision, to not compete with their higher-end models and offer the full memory space in normal acquisition.

In the Siglent, the full record length is available by increasing the timebase, and Siglent have made the decision that the true acquisition length should always set the memory depth.  This appears to be a technical decision rather than a marketing decision, but doesn't seem to imply any hardware or software limitation. If you want to capture the full 200Mpt, simply zoom the instrument out, capture the data, and zoom in.  If the instrument was continuously capturing the 200Mpt samples, then it would need to acquire and potentially process all of those while displaying only 0.01% of the actual acquisition.  This would reduce the acquisition rate considerably.

I've noticed a "Tek Mode" buried in the utility/settings page of the SDS5000X;  perhaps that would change this behaviour?

[0xdeadbeef]

I tend to disagree with Dave here that it's a crippling and insane decision to not always capture the full memory. At least for lower sampling rates, it would take very long to fill the whole buffer. E.g. at 1MSa/s, it would need 200s to fill the buffer. I'd think nobody would accept a single shot mode that needed more than a second to display a signal after the trigger. On the Keysight scopes that people like to present as the gold standard, you usually only have less than 2MPts or so which kinda eliminates this problem due to a simple lack of deep memory.
Of course there would be cleverer ways to solve this by e.g. dynamically limiting the memory to ensure a relatively quickly (few hundred milliseconds) reaction in single trigger mode dependent on the sample rate. I'm sure that even then people would complain about the incomprehensible memory usage. So Siglent went the easy way. IMHO this is a reasonable decision and actually having a permanent history is a cool feature, isn't it?

[Fungus]

You can capture a longer timebase and zoom in. That's essentially the same as what you're asking for. 

No it isn't.

It's of no use whatsoever when the event only happens infrequently and you just managed to capture one and want to know what happened before/after.

[David Hess]

So now we have to start carefully distinguishing between memory depth and record length.  Marketing will of course advertise the former rather than the later because it will be larger.

It has always been record length.

I checked the Siglent marketing material before posting and it only refers to memory depth.

Why they made this choice is not understood by me,  but it might be because they want two buffers of memory available (plus scratchpad for trigger correction or other management), and alternately stream one or the other to the processor. Or, it might simply be a marketing decision, to not compete with their higher-end models and offer the full memory space in normal acquisition.

In the Siglent, the full record length is available by increasing the timebase, and Siglent have made the decision that the true acquisition length should always set the memory depth.  This appears to be a technical decision rather than a marketing decision, but doesn't seem to imply any hardware or software limitation. If you want to capture the full 200Mpt, simply zoom the instrument out, capture the data, and zoom in.  If the instrument was continuously capturing the 200Mpt samples, then it would need to acquire and potentially process all of those while displaying only 0.01% of the actual acquisition.  This would reduce the acquisition rate considerably.

Honestly I prefer that the acquisition record be the same size as or lightly larger than the viewed area unless I specify otherwise if it means a higher acquisition rate.  Maybe Dave will go back and see how the acquisition rate changes at different time/div settings when the record length is being limited.

[nctnico]

Honestly I prefer that the acquisition record be the same size as or lightly larger than the viewed area unless I specify otherwise if it means a higher acquisition rate.  Maybe Dave will go back and see how the acquisition rate changes at different time/div settings when the record length is being limited.

There is a long thread about this but the point is that for some use (for example embedded software debug & verification) it is very handy to have data which goes far beyond the screen (without needing to resort to zoom modes which are clumsy).  With memory length under full user control the user can make the decission between deep memory or fast waveform updates.

I tend to disagree with Dave here that it's a crippling and insane decision to not always capture the full memory. At least for lower sampling rates, it would take very long to fill the whole buffer. E.g. at 1MSa/s, it would need 200s to fill the buffer. I'd think nobody would accept a single shot mode that needed more than a second to display a signal after the trigger. On the Keysight scopes that people like to present as the gold standard, you usually only have less than 2MPts or so which kinda eliminates this problem due to a simple lack of deep memory.

But now you are optimising for update rate. In some case that doesn't even matter. Recently I used 5s/div in roll mode to measure something; for long time/div you tend to use roll mode anyway. If you need to trigger then you get set a shorter memory depth either manually or automatically (like R&S allows to do).

Of course there would be cleverer ways to solve this by e.g. dynamically limiting the memory to ensure a relatively quickly (few hundred milliseconds) reaction in single trigger mode dependent on the sample rate. I'm sure that even then people would complain about the incomprehensible memory usage. So Siglent went the easy way. IMHO this is a reasonable decision and actually having a permanent history is a cool feature, isn't it?

The way Siglent implemented it: no; it is half assed. For example R&S and Yokogawa allow to trade-off record length and history depth manually to have maximum control.

[tom66]

You can capture a longer timebase and zoom in. That's essentially the same as what you're asking for. 

No it isn't.

It's of no use whatsoever when the event only happens infrequently and you just managed to capture one and want to know what happened before/after.

I think you misunderstand ...

If you set a non-Siglent scope to 200Mpt depth and a timebase of, say, 1us/div across 10 divisions, it only displays 10us of data at 2GSa/s = 20kpt.   The full buffer is captured but this is equivalent to having the scope set to a timebase of 10ms/div and zoomed in. 

There is *no difference* in acquisition performance, trigger performance or oscilloscope results, to a scope set to a long memory depth and zoomed in, or the same long memory depth and zoomed out to fit the entire acquisition on the display.  They are functionally equivalent and capture exactly the same data.

In both cases you will capture your "glitch" that you want to observe.  It may be helpful to set the trigger up initially at a short timebase, and then zoom out to capture a wider span.

And indeed Siglent offers this with the "zoom" function ... they have just decided, that by default, the instrument always uses the highest update rate configuration.  Whether that is a sensible decision or not, I suppose, is down to the application and user.

[SMB784]

Of course there would be cleverer ways to solve this by e.g. dynamically limiting the memory to ensure a relatively quickly (few hundred milliseconds) reaction in single trigger mode dependent on the sample rate. I'm sure that even then people would complain about the incomprehensible memory usage. So Siglent went the easy way. IMHO this is a reasonable decision and actually having a permanent history is a cool feature, isn't it?

The way Siglent implemented it: no; it is half assed. For example R&S and Yokogawa allow to trade-off record length and history depth manually to have maximum control.

The question in my mind is: is this something that could be changed in software so that the trade-off could be added as a feature?  Or is this tied to hardware.

[TK]

The way Siglent implemented it is like you need to know beforehand what you want to capture and set the right timebase before.  I used the SDS2104X plus to capture SPI data, decoded information in screen was not enough and wanted to see more... You cannot.  You need to set a new timebase and capture again.  You might have lost your golden opportunity of capturing the fault.  You end up doing 3-5 captures until you get what you want to see.  Or like some member said, capture and zoom in... it is ridiculous.  In my new GW-Instek GDS1054B, I capture and then zoom out and in as I like to find what I want to see in detail.  This is why UI responsiveness is important.  Rigol can capture full memory, but scrolling is painful.  Keysight is ultra fast and GW-Instek is fast too.

[NiHaoMike]

I tend to disagree with Dave here that it's a crippling and insane decision to not always capture the full memory. At least for lower sampling rates, it would take very long to fill the whole buffer. E.g. at 1MSa/s, it would need 200s to fill the buffer.

Why would you slow down the ADC that much? At 1GHz sample rate, it would be a very reasonable 0.2s.

[bdunham7]

In my view this is related to the klunkiness of the  FFT function in that the record length is tied inextricably with the displayed data.  I would like to see more user control, just so that the scope behaves the way the user wants it to.

I can easily set up the scope to capture what I want it to, but it requires wasting half my screen on something that generally yields no visible information.  To be fair here, this allows me to see the waveform at just a few cycles, but it has captured a full 14M points and if I stop the scope, I can scroll through the capture for the rest of the day.  No real issue in capturing a long record while looking at the trigger point up close, but it does take an additional step.



Now that doesn't look to bad, but you do have to set it up beforehand and it cannot be configured as the default, and as I said, you've wasted half the screen.  Here everything is clear, but add in the FFT and some measurements and you have a mess.



I can't change the FFT to full screen without losing my zoomed-in waveform, which is what I typically really want to see instead of the unusable information on the top half. 

Just another 'feature' of an almost-pro tool. 

[nctnico]

Of course there would be cleverer ways to solve this by e.g. dynamically limiting the memory to ensure a relatively quickly (few hundred milliseconds) reaction in single trigger mode dependent on the sample rate. I'm sure that even then people would complain about the incomprehensible memory usage. So Siglent went the easy way. IMHO this is a reasonable decision and actually having a permanent history is a cool feature, isn't it?

The way Siglent implemented it: no; it is half assed. For example R&S and Yokogawa allow to trade-off record length and history depth manually to have maximum control.

The question in my mind is: is this something that could be changed in software so that the trade-off could be added as a feature?  Or is this tied to hardware.

I think this is likely a couple of hours of work for Siglent to change. All the functionality is there in the software; it just needs to be wired a bit differently. The changes come down to adding an auto / fixed memory setting and create a hidden zoom function which is set the memory length selected by the user.

[bdunham7]

I think this is likely a couple of hours of work for Siglent to change. All the functionality is there in the software; it just needs to be wired a bit differently. The changes come down to adding an auto / fixed memory setting and create a hidden zoom function which is set the memory length selected by the user.

It does seem simple.  The most basic step would be to allow the often-useless display of the full capture to be turned off.  The next would be to allow a configurable option of A) how it is now, where the 'MEM DEPTH' function determines how many triggered frames are stored in memory, or B) 'MEM DEPTH' controls how many points are stored in a single acquisition and if you choose a lower amount, then perhaps you get a certain amount of history captures, but with each one at that full specified record length. 

Then we could talk about allowing the MEM DEPTH to be set in increments other than decades and setting the record length directly.  Same for FFT.

[kevinpt]

I believe this is a carry over from the LeCroy 9300 series stretching back 30 years. Those scopes started out with a 68020 processor with support for an unprecedented level of math operations. These needed to be restricted to the displayed record to be functional on such limited hardware.

The 9300s and direct descendants have a similar scheme for configuring acquisition where you set a maximum record length and the scope computes the optimal sampling rate for the current timebase as well as a potentially shorter record length if it decides the maximum is too big for the screen. The whole idea at the time being to limit how much data you can choke the processor with so that the scope stays more responsive. For long acquisitions you were expected to set a slower timebase and use the zoom waveforms to inspect the results in detail. HP MegaZoom hadn't been invented yet.

It also works in conjunction with the segmented acquisition where you need to set a limit on the length of each segment. Once you've configured an acquisition with more segments than will fit in memory, it will drop the segment length to allow more to fit. These were some of the most advanced deep memory scopes at the time when the competition was happy to max out at 4k acquisition records. I don't know how the current modern scopes behave but if you drop the (max) record length way down, the sample rate may drop as well so that the screen can still be filled.

This is akin to Tek keeping the cursors locked to the screen on zoom and scroll to replicate their analog scope cursors.

[EEVblog]

Martin72 posted this table in another thread:



There is absolutely no excuse for not allowing the user to manually select any memory depth at all timebase settings.

[EEVblog]

I tend to disagree with Dave here that it's a crippling and insane decision to not always capture the full memory.

That's not what I was saying. I was saying that it's crippling and insane that they don't give you the option to do it.

[rambling@midnight]

I'm a newbie looking for my first scope which I'm saving up for ^_^. I was really set on the SDS1104X-E because of the great price, but this is really disturbing. Do we think Siglent will listen and fix this in a firmware update? (can it be fixed in firmware?)
Also can someone make a video showing how to use the dual display mode hack to get the sampling we want? (maybe demonstrating with trigger/single shot)

[mroek]

I'm a newbie looking for my first scope which I'm saving up for ^_^. I was really set on the SDS1104X-E because of the great price, but this is really disturbing. Do we think Siglent will listen and fix this in a firmware update? (can it be fixed in firmware?)
Also can someone make a video showing how to use the dual display mode hack to get the sampling we want?

It can most likely be "fixed" in firmware, but as others have mentioned, this is a design choice, so it is doubtful that they will change it.

No need for a video, I think. It is dead simple. Adjust the horizontal to whatever you want (you can see the number of samples captured on the screen), then press the horizontal knob to go into zoom mode. The original waveform is shown on top, and on the bottom you have a zoomed version of the sama data. If you now adjust the horizontal, you will adjust the zoomed one, and you can use the position knob to choose which part of the full waveform you want to see.

[tautech]

I'm a newbie looking for my first scope which I'm saving up for ^_^. I was really set on the SDS1104X-E because of the great price, but this is really disturbing. Do we think Siglent will listen and fix this in a firmware update? (can it be fixed in firmware?)
Also can someone make a video showing how to use the dual display mode hack to get the sampling we want? (maybe demonstrating with trigger/single shot)

Welcome to the forum.

Reply #14 shows you what you need to see with the first image in classic Zoom mode with the primary timebase setting adjusted to display max mem depth. The small sliver in the middle of the display represents the lower zoomed portion. Single or Stop the acquisition and you can pan through the upper primary timebase with the zoomed portion.
As SDS1104X-E is entry level the Zoom split is 50/50 whereas SDS2000X Plus models are ~75/25 split.
Very simple to use once you embrace it.

Edit
Added for simplicity.

[rambling@midnight]

Thanks guys. Maybe I will end up still getting the aforementioned model. I really dislike the visual layout of the Rigol's with the dual columns of buttons on either side and tiny on-screen icons, and the two shades of blue in the channels. Siglent looks really clean.

[tautech]

Thanks guys. Maybe I will end up still getting the aforementioned model. I really dislike the visual layout of the Rigol's with the dual columns of buttons on either side and tiny on-screen icons, and the two shades of blue in the channels. Siglent looks really clean.

You won't be disappointed.
The video that bought you here is making a mountain from a molehill when those 'smart' guys that really want to give it some real thought of the reason why Siglent has implemented memory management the way they have.................and that's just started a flame war.