4 GS/s Oscilloscope with averaging and VERY fast retrigger.

[ezalys]

Dear all,

I'm working on an experiment which measures the response of a system to a 2-10 us long pulse sequence. The response is very deep in the noise and requires a lot of averaging of the response signal. The system recovers instantaneously after the pulse sequence, and thus we would like to repeat the pulse sequence as soon as possible. I'm struggling to find a signal averaging oscilloscope that's able to respond faster than 1 microsecond (FastFlight-2 signal averaging oscilloscope, not available anymore) after the end of a trace. We can't miss triggers, as a missed trigger means missed data, missed averages, and therefore longer to collect enough averages to obtain meaningful data. Does anyone have an idea for how to average a million or so repetitive segments of analog signal without losing data?

Best,
E

[tautech]

I can't help thinking a modern deep memory DSO in slow timebase Roll mode and using History and a good trigger suite will provide what you're looking for.
How many channels do you need to use at the same time as this might have a negative effect on the sampling rate if 2 or more are used on the same ADC.

[nfmax]

The higher-specced Keysight MegaZoom 4 scopes, like the DSO-X 3104T or the 4000X series, advertise a waveform update rate of over 1 millison waveforms/sec, and claim to be the industry's fastest. I suggest you take a look, and/or call your local Keysight office.

[tautech]

I dunno, pick your poison, slow timebase setting and deep memory = zero rearm time or Xn wfm/s with some blind time.

[gf]

Does anyone have an idea for how to average a million or so repetitive segments of analog signal without losing data?

Is it really important not to lose any cycles if the signal is periodic anyway?
[ And if it is not periodic, what's the sense of averaging, then? ]

Do you need "real-time" display, or might it be an option to capture just one or multiple huge buffers of samples, and align and average them offline (e.g. with Matlab, Octave or whatsoever).

If the required sampling rate is lower than the highest sampling rate of the digitizer, "Hi-Res" acquisition mode may help additionally, in order to improve SNR.

[nfmax]

The higher-specced Keysight MegaZoom 4 scopes, like the DSO-X 3104T or the 4000X series, advertise a waveform update rate of over 1 millison waveforms/sec, and claim to be the industry's fastest. I suggest you take a look, and/or call your local Keysight office.

One minor problem - you are limited to 65536 averages in average acquisition mode. But maybe you could do 16 'blocks' of 65536 averages, with a longer gap between each block, using segmented memory mode, (or possibly you will have to retrieve each waveform buffer over the LAN, depending on how the average & segment acquisition modes interact). Then average each acquisition together separately.

I spent almost 20 years involved making systems that do just this sort of averaging: but we always used custom hardware, which progressed from a board full of logic chips & fast RAMs to a single, mid-range FPGA.

[nctnico]

Dear all,

I'm working on an experiment which measures the response of a system to a 2-10 us long pulse sequence. The response is very deep in the noise and requires a lot of averaging of the response signal. The system recovers instantaneously after the pulse sequence, and thus we would like to repeat the pulse sequence as soon as possible. I'm struggling to find a signal averaging oscilloscope that's able to respond faster than 1 microsecond (FastFlight-2 signal averaging oscilloscope, not available anymore) after the end of a trace. We can't miss triggers, as a missed trigger means missed data, missed averages, and therefore longer to collect enough averages to obtain meaningful data. Does anyone have an idea for how to average a million or so repetitive segments of analog signal without losing data?

I'd start with a 10 bit (or more) oscilloscope which can do fast segmented recording and then do the averaging on a PC. I think you probably find something in Lecroy's or R&S (RTA4000 for example) line-up. Doing the averaging on the oscilloscope itself is likely not going to work well due to not using enough bits to do the calculations to a level where you get meaningful data. But you can try; at least the RTA4000 has the ability to show an average of all recorded segments.

I don't think you will want to record a million segments though. At some point you start looking at the noise induced by the trigger and sampling system anyway.

[DaJMasta]

Why do this with just a scope?  What is your input frequency/bandwidth?  Rather than averaging like crazy to try and see the signal, pick a proper high dynamic range instrument or something with good out-of-band rejection and get a high resolution digitizer in there.  Either that or use a preamp and just look at the pulses with a normal scope.

[nfmax]

Dear all,

I'm working on an experiment which measures the response of a system to a 2-10 us long pulse sequence. The response is very deep in the noise and requires a lot of averaging of the response signal. The system recovers instantaneously after the pulse sequence, and thus we would like to repeat the pulse sequence as soon as possible. I'm struggling to find a signal averaging oscilloscope that's able to respond faster than 1 microsecond (FastFlight-2 signal averaging oscilloscope, not available anymore) after the end of a trace. We can't miss triggers, as a missed trigger means missed data, missed averages, and therefore longer to collect enough averages to obtain meaningful data. Does anyone have an idea for how to average a million or so repetitive segments of analog signal without losing data?

I'd start with a 10 bit (or more) oscilloscope which can do fast segmented recording and then do the averaging on a PC. I think you probably find something in Lecroy's or R&S (RTA4000 for example) line-up. Doing the averaging on the oscilloscope itself is likely not going to work well due to not using enough bits to do the calculations to a level where you get meaningful data. But you can try; at least the RTA4000 has the ability to show an average of all recorded segments.

I don't think you will want to record a million segments though. At some point you start looking at the noise induced by the trigger and sampling system anyway.

If the signal of interest is buried deep in the noise more ADC bits don't help, 8 is more than enough. Also, since the noise spreads the signal over many different bit transitions, the ADC effective DNL is much improved. Furthermore, I would sincerely hope that any decent scope has a wide enough acquisition memory to accommodate the maximum ADC sample summed over the maximum supported number of averages (24 bits, in the case of the Keysight scopes)!

The scope must be capable of doing the averaging itself, otherwise you are going to spend most of your time reading out the trace data for each pulse. Doing this once every 65536 pulses is much less of an issue.

Trigger uncertainty shouldn't be an issue, as you will be triggering from the circuit generating the stimulus, not the response. Sample jitter and clock drift also don't matter much, if the length of each response is fairly short: although the Keysight scopes I mention have high precision timebase anyway..

[ezalys]

DaJMasta - We use a preamp cooled to four kelvin already and the data is still this noisy. The 1/f knee is a real bear.
nctnico - We do see trigger noise and we average it out.

[DaJMasta]

That's probably about as much as you can expect out of a preamp then.... have you looked into PXI or similar high speed digitizers?  You can find models to 100MS/s and 14 or 16 bits for not too much money, though you need the chassis/controller (unless you go with a PCIe based one or something) - while they may not boast a high retrigger rate, if you have that much resolution, you may not need much averaging to be able to discern your signal.  If they're based off PCIe (or PXIe), then they may actually have enough bandwidth to run the sample rate you require without having to retrigger at all, just dumping to memory or a fast SSD.

Another option could be a spectrum analyzer with realtime analysis capability.  If you had 20MHz+ of instantaneous bandwidth, you could park the center frequency at like 10MHz and just directly digitize the input without sweeping, which gets you probably 14 bits of resolution with a much lower instrument noise floor.

Since your bandwidth requirement sounds low, slower, higher resolution digitizers are probably an option.  I assume you are already, but if not, bandpass filtering around your frequencies of interest could do a fair bit for system noise floor too!

[Marco]

500 kHz retriggering signal averaging, 4 GS/s sample rate :
https://www.engineersonline.nl/download/U1084AEN.pdf

[nfmax]

500 kHz retriggering signal averaging, 4 GS/s sample rate :
https://www.engineersonline.nl/download/U1084AEN.pdf

And built-in averaging firmware as well - nice! This is the sort of hardware averager design I've been involved with over the years.

[JDubU]

Don't know if this is helpful, but the new low end Keysight 1200x series advertises a 1uS trigger re-arm time in segmented memory mode.  Maximum number of segments is only 500 though.
The higher end Keysight InfiniiVision scopes have the same re-arm time but up to 1000 segments.

https://www.keysight.com/main/redirector.jspx?action=ref&nid=-32110.0.00&lc=eng&cc=US&ckey=3035376&cname=EDITORIAL&nfr=-32110.1286929

[ezalys]

Marco - Looks like Bruker uses these in their mass specs. Very similar requirements to what we do... very promising.

DaJMasta - The bandwidth requirement is very high. We need about 1 GHz of bandwidth going down to DC, hence the 4 GS/s in the title. We have a setup that works using the FastFlight-2 system, but it's no longer made.

[Someone]

Marco - Looks like Bruker uses these in their mass specs. Very similar requirements to what we do... very promising.

DaJMasta - The bandwidth requirement is very high. We need about 1 GHz of bandwidth going down to DC, hence the 4 GS/s in the title. We have a setup that works using the FastFlight-2 system, but it's no longer made.

Do be careful with the term "averaging", that U1084A-AVG linked accumulates all the acquisitions equally. On the other hand while many oscilloscopes have an "average" acquisition mode, its typically a weighted average (IIR type response) and not a true boxcar. Some Lecroy models offer both a pre-processing (IIR) and post processing (accumulation) averaging option.

That said a high waveform rate even with the less effective IIR averaging can still be more powerful for filtering noise in a given time. And can be combined with accumulating periodic captures of that.

[ezalys]

It's VITAL that this operate in accumulation and not IIR mode.

[DaJMasta]

If you really need 1GHz bandwidth (not what I expected from reading 2us+ long pulses) and this kind of retriggering... I think you're looking at a fairly high end scope, then.

Keysight's MSOS series (or better, the MSOS only specifies 300k+ waveform updates/sec)
Rohde & Schwarz's RTM/RTA/RTO 2000 series
Tektronix's MDO4 series or better (the 3 series specifies up to 280k waveform updates/sec, but the 4 series goes to 500k)

It sounds like LeCroy's HDO series can only capture that fast in sequence mode, but the WaveRunner HD 8000 series could do it - they specify up to 650k waveform updates a second with no mention of sequence mode only


Not sure if any of these meet every requirement, but they can update the screen frequently enough to likely meet your requirement, have at least 1GHz bandwidth options available, and are all high resolution (10bit+).  They're also all basically current-gen, so be prepared to shell out a bit (though it sounds like a research application, so this is probably less of an issue).  And with any case, get a demo to make sure.  For this price, it's important to check and be sure.

[Keysight DanielBogdanoff]

You also might look into the Keysight 6000 X-Series scope (D/MSOX6004A). It's the highest-end InfiniiVision scope, and goes from 1 GHz to 6 GHz (vs. 1.5 GHz max on the others). It uses the same front end as the MSOS mentioned below, which means the noise is quite low. But, it uses the same backend ASIC as the other InfiniiViision scopes(Megazoom IV). It does some fancier analysis stuff too, like color grading, jitter analysis, etc.

It goes up to 450,000 waveforms per second (aka triggers). The 3000T and 4000X are also great scopes, will get you double the update rate, and go up to 1.5 GHz.

We can definitely get you a demo if you'd like.

[Berni]

I think what you want instead is a deep memory scope.

For example the Keysight MSO9000 has 1Gpts of memory so at 4GS/s this gives you 250ms of continuous recording time on one channel. So if you have a event that is 10us long and can repeat it instantly then 25 000 of these events can be captured before the scope runs out of memory and stops. You get a 1GB sized file out of it that you then process with your own software to seek out the events of interest and process them. If you need help identifying where the pulses start you can use another channel on the scope to also capture a trigger signal while still getting 1GPts per capture.

But the ultimate device for your task would be a high speed ADC and a FPGA that can continuously sample and process the data in real time. So if you are up to doing FPGA development you might be able to get a existing ADC+FPGA board like for example the RedPitaya.

[nctnico]

500 kHz retriggering signal averaging, 4 GS/s sample rate :
https://www.engineersonline.nl/download/U1084AEN.pdf

And built-in averaging firmware as well - nice! This is the sort of hardware averager design I've been involved with over the years.

This Keysight  U1084A acquisition card seems like it has been designed with the purpose of the OP in mind. It could be a very good solution compared to an oscilloscope. One thing that worries me about using an oscilloscope is that maximum waveform rate of oscilloscopes is only valid for very specific (optimal) conditions which often don't even resemble a normal usage scenario. Averaging may slow it down and the only way to find out is to test the oscilloscope. It also seems that oscilloscope in general don't offer to average over 50k to 100k acquisitions.

[fcb]

Pico Technology 6407 or the newly released 6000 series.

[Wuerstchenhund]

I'm working on an experiment which measures the response of a system to a 2-10 us long pulse sequence. The response is very deep in the noise and requires a lot of averaging of the response signal. The system recovers instantaneously after the pulse sequence, and thus we would like to repeat the pulse sequence as soon as possible. I'm struggling to find a signal averaging oscilloscope that's able to respond faster than 1 microsecond (FastFlight-2 signal averaging oscilloscope, not available anymore) after the end of a trace. We can't miss triggers, as a missed trigger means missed data, missed averages, and therefore longer to collect enough averages to obtain meaningful data. Does anyone have an idea for how to average a million or so repetitive segments of analog signal without losing data?

So what is the purpose, i.e. is this to take a snapshot (i.e. a few cycles) or is this for continuous monitoring?

Scopes are great for the former but pretty bad for the latter.

If you just want to record a number of cycles then I'd suggest you get a scope with very large sample memory. Ignore waveform rates, even the fastest scope will be blind for around 90% of the time so you need a scope that can capture the required sequence in one go. You need to decide how many cycles you want and then multiply with 4GSa/s and then you got the sample memory size you need.

I understand that the BW requirement is 1GHz, so yes you want at least 4GSa/s.

As to which scopes, forget any of the embedded scopes like the DSO-X, they don't have the memory sizes or capabilities to do what you need. Instead, I would look at the Keysight Infiniium-S DSO-S104. The DSO-S has the advantage of letting you not only select memory size but also sample rate, and it can be optioned up with large memory (536Mpts at <5GSa/s in single-channel mode). I believe it also supports boxcar averaging, and if not it can optionally run MathLab scripts so you can do the averaging there.


However, if this is for continuous monitoring then I would just forget bench scopes and look into high speed digitizers, for example as PXI Express system. You can get the raw sample stream from your digitizer and feed it into Mathlab/Octave/SciLab/whatever to get your data.

[frogg]

I'm not sure I totally understand your question.

A Virtex-7 dev board could easily keep up with your sampling rate. How many bits ADC do you need?

The real metric you need to define is cross-sectional bandwidth.

The math is pretty simple and could be synthesized with a few blocks in MATLAB.

[Someone]

I'm not sure I totally understand your question.

A Virtex-7 dev board could easily keep up with your sampling rate. How many bits ADC do you need?

The real metric you need to define is cross-sectional bandwidth.

The math is pretty simple and could be synthesized with a few blocks in MATLAB.

Accumulating is easy in concept, but the OP is asking for an integrated product rather than building something. Scopes have the front end required and triggering performance (critical for the correlation to work) but insufficient offload/processing bandwidth to average as fast as they would like. The U1084A Marco linked to is one of the few products that has everything together in one unit/system.

It's VITAL that this operate in accumulation and not IIR mode.

Ignore scopes, fastest back-to-back accumulation is going to be something with user programmable logic/FPGA that sees the full sample rate. Pingpong buffers and offload while accumulating the next batch.