Input on 1st digital oscllioscope purchase

[videobruce]

Even the cheap hobby ones have just as much if not more than the "lab grade" scopes.

If I didn't know better, thus spec sounds more as a marketing tool just like in consumer electronics where they try a different 'spin' each year to sell something. Example; "LED" TV vs LCD scam Samsung started a few years ago. 
I found a couple of white papers on memory depth I will take a look at. One is here, the other is from Agilent;
http://www.eetimes.com/document.asp?doc_id=1279463
.

Be aware that a modern 1GS/s scope almost certainly won't let you see a 500MHz signal. (Older digital scopes did; I once used a 1GHz scope to see sub-ns risetimes, but it only did 40MS/s). For digital signals, decide on the risetime of the signals that you wish to see.

This is what I mentioned in the 1st post regarding the sampling rate spec.

I have ruled out the 320x240 screens. Actually in this day, 640x480 is outdated AFAIC.

[videobruce]

There is also this with Rigol mentioned. I don't know if you guys agree with the response or not;
http://electronics.stackexchange.com/questions/76377/what-is-the-importance-of-memory-depth-when-selecting-a-digital-storage-oscillos

[Fungus]

I found a couple of white papers on memory depth I will take a look at. One is here, the other is from Agilent;
http://www.eetimes.com/document.asp?doc_id=1279463

Huh? "Sour grapes" springs to mind when reading that - marketing people trying to tell us why their 'scopes don't have as much memory as the competitor.

Isn't the best solution to have lots of memory and let the user choose a smaller buffer if he feels the 'scope is bogging down? (Like Rigol does...)

[nctnico]

There are a number of specs that are new to digital scopes I will have to read up on that I won't ask here unless I can't find sufficient data elsewhere.

There isn't much you can do wrong in terms of specs these days. Any scope with MB worth of sample memory is what you want, and 1GS/s sample rate...

Be aware that a modern 1GS/s scope almost certainly won't let you see a 500MHz signal. (Older digital scopes did; I once used a 1GHz scope to see sub-ns risetimes, but it only did 40MS/s)

For digital signals, decide on the risetime of the signals that you wish to see. That will determine the bandwidth of the analogue front end using the standard equation BW = 0.35/tr. For digital circuits 100MHz is probably the lowest you should use, since many circuits have 1ns risetimes and you may well want to see time differences of the order of a few ns.

Actually rise time doesn't affect seeing time differences. Even if the risetime of an oscilloscope is 5ns you'll still see a 1ns delay between two signals provided both channels have similar accuracy. In other words: the ability to see differences in timing between 2 channels has everything to do with how well the channels are matched. This should be somewhere in the specification of an oscilloscope.

Regarding memory: more is better as long as it doesn't slow the oscilloscope down and the memory is easy to scroll through.

[tggzzz]

There are a number of specs that are new to digital scopes I will have to read up on that I won't ask here unless I can't find sufficient data elsewhere.

There isn't much you can do wrong in terms of specs these days. Any scope with MB worth of sample memory is what you want, and 1GS/s sample rate...

Be aware that a modern 1GS/s scope almost certainly won't let you see a 500MHz signal. (Older digital scopes did; I once used a 1GHz scope to see sub-ns risetimes, but it only did 40MS/s)

For digital signals, decide on the risetime of the signals that you wish to see. That will determine the bandwidth of the analogue front end using the standard equation BW = 0.35/tr. For digital circuits 100MHz is probably the lowest you should use, since many circuits have 1ns risetimes and you may well want to see time differences of the order of a few ns.

Actually rise time doesn't affect seeing time differences. Even if the risetime of an oscilloscope is 5ns you'll still see a 1ns delay between two signals provided both channels have similar accuracy. In other words: the ability to see differences in timing between 2 channels has everything to do with how well the channels are matched. This should be somewhere in the specification of an oscilloscope.

True, but a faster risetime

• makes it easier to accurately discriminate the time at which a threshold voltage is crossed
• reduces the requirement for the ADCs non-linearities to be matched, and the channel'a analogue response to be matched

Thus, in this non-ideal world, a faster risetime is strongly positively correlated with discriminating smaller time intervals - even though there isn't a simple direct relationship.

[Precipice]

This should be somewhere in the specification of an oscilloscope.

And if you particularly care, you'll swap the inputs (and probes) to check...
There are too many ways to get errors, not to, and what's on the scope screen isn't always what's at the probe tip (or even at the nodes when the probes are off!)

[David Hess]

In rough order of importance, the things I look for include:

Peak Detection
Graded Intensity Display
High Acquisition Rate

I am less interested about a long record length but there are some applications which require it.

Some applications require 4 channels or logic analyser capability.

If the oscilloscope does not support equivalent time sampling, then I would be concerned about sampling errors in the form of non-linearity and jitter in the digitizer but this is not usually specified and nobody tests for it so there is not much to be done unless you have the opportunity to evaluate it yourself.  Thankfully like record length, in most applications it will not matter.

[tggzzz]

In rough order of importance, the things I look for include:
Peak Detection
Graded Intensity Display
High Acquisition Rate

Why do you regard peak detection as being pf paramount importance? What's the "use case" or application?

[videobruce]

I found a couple of white papers on memory depth I will take a look at. One is here, the other is from Agilent;
http://www.eetimes.com/document.asp?doc_id=1279463

Huh? "Sour grapes" springs to mind when reading that - marketing people trying to tell us why their 'scopes don't have as much memory as the competitor.
Isn't the best solution to have lots of memory and let the user choose a smaller buffer if he feels the 'scope is bogging down? (Like Rigol does...)

Can you comment further where you disagree what what was said? I understand there is a sales angle to any of those briefs, but it isn't all 'sales'.
.

Regarding memory: more is better as long as it doesn't slow the oscilloscope down

That was discussed in that paper.

[nctnico]

In rough order of importance, the things I look for include:
Peak Detection
Graded Intensity Display
High Acquisition Rate

Why do you regard peak detection as being pf paramount importance? What's the "use case" or application?

I don't touch a DSO with a stick if it doesn't have peak detect. Long memory or not without proper peak detect you'll miss narrow pulses for sure.

[David Hess]

In rough order of importance, the things I look for include:
Peak Detection
Graded Intensity Display
High Acquisition Rate

Why do you regard peak detection as being pf paramount importance? What's the "use case" or application?

This comes down to any application where the record length is insufficient to maintain the highest sampling rate at the current time/div setting although there are other advantages even if this is not an issue.  Peak detection also makes for an quick sanity check that aliasing is not occurring.

[tggzzz]

In rough order of importance, the things I look for include:
Peak Detection
Graded Intensity Display
High Acquisition Rate

Why do you regard peak detection as being pf paramount importance? What's the "use case" or application?

This comes down to any application where the record length is insufficient to maintain the highest sampling rate at the current time/div setting

Ah, so it is the "peak measured value within the displayed sample interval".

It isn't clear to me under which circumstances that non-linear algorithm is/isn't the most appropriate. I suspect that for many purposes the mean value might be a more appropriate algorithm, and that using the peak could have some unexpected and subtle side-effects.

I need to ponder, but I'm not sure I'll come to firm conclusions.

although there are other advantages even if this is not an issue.  Peak detection also makes for an quick sanity check that aliasing is not occurring.

With a conventional analogue scope, aliasing wouldn't be invisible, since the trace would just appear noisy or blurred. Strike one against naive digitisation.

Perhaps if measured samples (5.0, 4.9, 5.1, 5.0. 4.9, 5.1) are all within a single displayed sample, the displayed sample should be a 4.9V-5.1V smudge, rather than a single value 5.0V(mean) or 5.1V (peak) point.

[David Hess]

With a conventional analogue scope, aliasing wouldn't be invisible, since the trace would just appear noisy or blurred. Strike one against naive digitization.

CRTs have a "writing rate" which could in theory limit the ability to show fast glitches but I have never seen it occur.  This is one of the advantages of using a high bandwidth analog oscilloscope even in lower bandwidth applications; they have faster CRT writing rates which goes along with brighter and clearer CRTs.

Perhaps if measured samples (5.0, 4.9, 5.1, 5.0. 4.9, 5.1) are all within a single displayed sample, the displayed sample should be a 4.9V-5.1V smudge, rather than a single value 5.0V(mean) or 5.1V (peak) point.

Peak detect returns a pair of values for each sample point in the record so the record length is usually halved.  Some DSOs display a single point if all of the values are within a small range of each other which has the effect of removing noise.

I have not had an opportunity to do detailed tests on a modern DSO with a graded index display so I have no idea how they operate but I have seen them miss glitches without peak detect mode so something is different.  We had a discussion about what they should display if not simple index grading here months ago but it had more to do with accurately displaying noise rather than peaks.  There are all kinds of clever things which could be done which are decidedly non-analog like.

[tggzzz]

I have not had an opportunity to do detailed tests on a modern DSO with a graded index display so I have no idea how they operate but I have seen them miss glitches without peak detect mode so something is different.  We had a discussion about what they should display if not simple index grading here months ago but it had more to do with accurately displaying noise rather than peaks.  There are all kinds of clever things which could be done which are decidedly non-analog like.

This is a most interesting discussion, and I'm going to follow up that search term.

One of my current projects is to build a cheap, very fast scope which will have some decidely strange characteristics in order to accomodate fast+cheap simultaneously. It will necessarily have something akin to a graded index display even for fast waveforms. I am aware that I haven't considered how it would work with slower waveforms, and this will force me to concentrate my mind. OTOH I might decide that the sole raison d'etre for the scope is speed, and, in the short term, merely ignore the issue

Thanks

[nctnico]

Ah, so it is the "peak measured value within the displayed sample interval".

It isn't clear to me under which circumstances that non-linear algorithm is/isn't the most appropriate. I suspect that for many purposes the mean value might be a more appropriate algorithm, and that using the peak could have some unexpected and subtle side-effects.

The reason you need peak detect is that in many cases you want to see whether there is a pulse somewhere in the signal. If the signal is undersampled then you'll miss those peaks. An example is looking at vsync signals. These are very short compared to the period.
Just averaging the signal is asking for trouble because you basically miss all the HF information.

[tggzzz]

Ah, so it is the "peak measured value within the displayed sample interval".

It isn't clear to me under which circumstances that non-linear algorithm is/isn't the most appropriate. I suspect that for many purposes the mean value might be a more appropriate algorithm, and that using the peak could have some unexpected and subtle side-effects.

The reason you need peak detect is that in many cases you want to see whether there is a pulse somewhere in the signal. If the signal is undersampled then you'll miss those peaks. An example is looking at vsync signals. These are very short compared to the period.
Just averaging the signal is asking for trouble because you basically miss all the HF information.

Having done a little research, I now have a better understanding of what's being done - and yes you are correct, of course.

In particular, I found these pictures informative https://www.eevblog.com/forum/testgear/oscilloscope-specs-which-are-really-important/msg181637/#msg181637

[David Hess]

I have not had an opportunity to do detailed tests on a modern DSO with a graded index display so I have no idea how they operate but I have seen them miss glitches without peak detect mode so something is different.  We had a discussion about what they should display if not simple index grading here months ago but it had more to do with accurately displaying noise rather than peaks.  There are all kinds of clever things which could be done which are decidedly non-analog like.

This is a most interesting discussion, and I'm going to follow up that search term.

One of my current projects is to build a cheap, very fast scope which will have some decidely strange characteristics in order to accomodate fast+cheap simultaneously. It will necessarily have something akin to a graded index display even for fast waveforms. I am aware that I haven't considered how it would work with slower waveforms, and this will force me to concentrate my mind. OTOH I might decide that the sole raison d'etre for the scope is speed, and, in the short term, merely ignore the issue

The short discussion came about in connection which how an analog oscilloscope can be used to accurately measure the RMS value of wide bandwidth Gaussian noise which is usually a rather difficult thing to do requiring a specialized instrument like a sampling or thermal RMS voltmeter.  DSOs should be able to measure it directly using the same techniques that sampling RMS voltmeters use but some work better than others doing this.  What no DSO can do is duplicate how an analog oscilloscope does it and they work every time.  At least I have never seen or heard of one which can do it.



I doubt the above technique, tangential noise measurement, is feasible or even desirable in a DSO but it led to a short discussion about *how* a display which is not exclusively index graded could produce the same result.  (*) My thought was that such a display could also show detected peaks without compromising index grading.

For myself, all of my DSOs are old and slow and lack index graded displays but they do have peak detection and oddly enough very high display resolution which makes up somewhat for their other shortcomings.  "Retina displays" existed in DSOs about 30 years before Apple came up with the term.

Producing an index graded display record during decimation would preserve display quality and high waveform acquisition rates at the expense of record length.  I think Tektronix wrote some white papers discussing this and how their early "digital phosphor" oscilloscopes worked.  I wonder if this can be done inexpensive with an FPGA based design.  I know Rigol is not doing it this way.  This is in contrast to display generation from long record lengths so has little advantage as far as specmanship and marketing.

(*) Technically index grading is not required to do tangential noise measurement.  A halftone display produced via variable dot density like you would find in an analog sampling oscilloscope is sufficient.

[nctnico]

About a decade ago I embarked on designing my own USB oscilloscope. PC memory was already cheap so I opted to put 2GB of memory on it which provides enough storage for 1Gpt/s per channel. In that design I kept the sampling frequency constant and have the FPGA process the data (average + peak detect) into display data on the fly OR I could replay the stored waveforms from the FPGA to create the display data.