Quality PC-Based oscilloscope?

[JoeyP]

resolution. On a normal scope you'd choose 1V/div where the signal would fill 6 (out of 8 ) divisions. And the people from Picoscope think they can get 12 bit from an 8 bit ADC and they are able to get 250MHz into a 1M Ohm 15pf input.

Tektronix was able to achieve a 250MHz BW into a 1M 20pF input all the way back in the 1970's (see Tek 475A), so 15pF is very reasonable. If you really understand scope inputs, then you recognize that the input impedance is not a simple capacitive load. They typically have series resisters that make them look much more like 50 ohms at high frequencies. Scopes with 1M inputs are typically equalized to achieve their specified bandwidth with a 50 ohm source, terminated with an external 50 ohm load at the input. Take a look at a cal procedure.

The people at Picoscope think they can do the same with standard 1:10 1M Ohm probes. Just look at their brochure. 15pf at 250MHz is around 42 Ohms so you may get lucky with the Picoscope and 50 Ohm terminators at the input but if you are into RF then I'd suggest equipment with 50 Ohm inputs. My rule of thumb is to go for 50 Ohm cabling and proper HF probes (active or passive divider) for measuring anything over 100MHz.

The topic here is about PC-based vs standalone scopes. Picoscope is not claiming anything that Agilent and Tek haven't been claiming for decades. Everyone recognizes that 10:1 1M probes have non-zero input capacitance, and that any capacitance will represent a load to a high speed signal. But it has nothing to do with whether the brand on the box is Tek or Picoscope. It's just a reality. Your "passive probe", if it is 10:1 would have an input impedance of 500 ohms, which would represent severe loading for many circuits. There are also active probes which reduce the tip capacitance to <1pF, but they're expensive and not everyone needs to spend that kind of money to get good results. People have been doing quite a lot with standard probes for many years.

[JoeyP]

If oversampling where a good option then we would all be using 1 bit AD/converters (comparators) and noise

The key factor which makes oversampling work is that the noise has to be enough to trigger at least 1 bit digitization changes. If your signal has +/-1mV noise and your minimum ADC resolution is 4mV (what it is for a 1V reference with 8-bit resolution), you won't get increased accuracy no matter how many samples you read.

In broad-band devices like scopes, there's rarely if ever an issue of not having enough noise. You can always sum extra noise at the ADC input to get the required level of dither, so even a perfectly clean signal could be oversampled and averaged. You can keep the resulting noise level low by summing in a predictable dither pattern, and then subtracting it back out of the digitized result.

If oversampling where a good option then we would all be using 1 bit AD/converters (comparators) and noise

To get a 1 GS/s 8-bit effective resolution using a 1-bit ADC and appropriate noise, you would need to sample at 65.536 Terra samples per second! With a 1 GS/s 1-bit ADC, you only get ~15.259 KS/s at 8-bits effective accuracy..

Exactly. That's one reason we don't see 1 bit converters used in scopes, but we do see this technique commonly used to increase resolution in scopes.

[nctnico]

resolution. On a normal scope you'd choose 1V/div where the signal would fill 6 (out of 8 ) divisions. And the people from Picoscope think they can get 12 bit from an 8 bit ADC and they are able to get 250MHz into a 1M Ohm 15pf input.

Of course they can...

Tektronix, Agilent and others have been getting 12bits from 8 bit converters for many years. Just over-sample and average. Input noise provides the needed dithering. That's how their "high res" mode works.

I put that to the test on my Tektronix scope and as expected it looks fine for a sine wave but a triangle or sawtooth get distorted due to inperfections of the noise and non-linearities in the ADC.

So did I. Images below for 8 bit normal sampling and 12 bit oversampling. No distortion noted. It just cleans up the noise as you'd expect for any form of averaging.

Hmm, you expect 12bit (that is 4096 steps) to show in an image about 350 pixels high? The whole extra bits in oversampling mean nothing when a scope's display hasn't substantially more pixels in height than the number of steps of the ADC. So Tek, Agilent, et all can safely claim their high resolution modes add several bits to an 8 bit ADC with about 7 ENOB without customers ever knowing the only thing they get is noise suppression. Just look what happens if I put my Tek scope to the test to see what it does with a signal about 3 LSB in height. The first image is the original signal, the second is the same signal attenuated 100 times and then enlarged 100 times. If the 10 bit high resolution mode would work as promised I'd expect a much better picture.

[JoeyP]

Hmm, you expect 12bit (that is 4096 steps) to show in an image about 350 pixels high?

No, I expect to see the distortion that you claim to exist in a practical application for which the scope was designed. I don't.

happens if I put my Tek scope to the test to see what it does with a signal about 3 LSB in height. The first image is the original signal, the second is the same signal attenuated 100 times and then enlarged 100 times.

So this is how you think people normally use scopes? A signal of 3LSBs p-p expanded 100x? Sounds pretty silly to me.

[tinhead]

If the 10 bit high resolution mode would work as promised I'd expect a much better picture.

nah, you forgot something. Change now to "Sample" mode and compare result to your second picture.
And now you can see that Hi-Res is working. If you fair enough, make screenshot of hat i'm talking about and post it here.

EDIT: i did for you, in sample mode as expected 100x times attenuated signal is not readable, in hi-res it is much better.

[helloworld922]

In broad-band devices like scopes, there's rarely if ever an issue of not having enough noise. You can always sum extra noise at the ADC input to get the required level of dither, so even a perfectly clean signal could be oversampled and averaged. You can keep the resulting noise level low by summing in a predictable dither pattern, and then subtracting it back out of the digitized result.

I was just pointing out some potential pitfalls with using over-sampling, and that the required "noise" level will need to be larger for a lower resolution ADC, or a larger reference voltage. Yes, you can compensate for this required gap by either adding artificial noise or amplifying the input signal so any natural noise can be picked up by the ADC (one of the reasons you want a signal's peak to peak to match as closely the peak to peak range of the ADC).

[nctnico]

Hmm, you expect 12bit (that is 4096 steps) to show in an image about 350 pixels high?

No, I expect to see the distortion that you claim to exist in a practical application for which the scope was designed. I don't.

happens if I put my Tek scope to the test to see what it does with a signal about 3 LSB in height. The first image is the original signal, the second is the same signal attenuated 100 times and then enlarged 100 times.

So this is how you think people normally use scopes? A signal of 3LSBs p-p expanded 100x? Sounds pretty silly to me.

Which means you agree with me that claiming to get more bits from an ADC is safe to do for a scope manufacturer because the whole claim is useless to begin with because the screen's resolution isn't high enough to show the errors.  Or you just miss the point entirely... see below.

@tinhead:
I did that but I don't have the screenshot (note the date on the other screenshots, I have posted these pictures before). What you get is a few dots from which you can clearly see the scope's ADC resolution. Hi-res mode does it's best to dream up a signal which could be there but without the actual information the scope's hi-res algorithm's guess is as good as mine. I don't need the scope to guess for me though, I want it to show me what it sees at its inputs. Ofcourse you'd crank the V/div down to get a good view at such a small signal instead of zooming in but that is besides the point that I'm trying to make here and that is that you cannot get more resolution from an ADC than what it has been designed for by just oversampling c.q. averaging.

[tinhead]

the point that I'm trying to make here and that is that you cannot get more resolution from an ADC than what it has been designed for by just oversampling c.q. averaging.

of course you can, you can see it even on yuor DSO. When you above the  hi-res limit the signal will be not better
(simply use 10x att. and 10x zoom for better view).

[tinhead]

I agree the chingchangchong probes sold for 20$ on fleabay hardly clamber above 100MHz .. But a real probe doesn't blink at 500MHz ... Then again, these real probes cost more than the base Rigol oscilloscope.... Buy 4 probes, get scope free 

oh yeah, there are some really funny probes on ebay. Sometimes it's simply question of compensation range.
Example below, Hantek P6200 cheap blah blah probe. Rated 200MHz, which is seems to be the truth on a Hantek
DSO (with a +2dBm peak ~150MHz). On a different scope (TDS 7xx series) it's simply useless :

Hantek P6200 200MHz x1/x10 Probe

  1MHz      0dBm
 25MHz   +1.2dBm
 50MHz    +2.0dBm
 75MHz  +2.8dBm
100MHz  +4.3dBm
125MHz  +5.4dBm
150MHz  +5.8dBm
175MHz  +6.4dBm
200MHz  -7.1dBm
225MHz  -7.2dBm
250MHz  -7.1dBm

this is the best result i was able to get. Sure, such probe costs 25USD, but that's already 25USD too much
(except someone need a cheap probe for 'DC').


Some other probes (as well tested on TDS7xx):

TESTEC TT-HF512 500MHz x10 Probe

  1MHz    0dBm
 50MHz   +0.2dBm
100MHz  +0.0dBm
150MHz  -0.6dBm
200MHz  +0.2dBm
250MHz  +1.2dBm
300MHz  +2.9dBm
350MHz  +2.3dBm
400MHz  +1.1dBm
450MHz  -0.8dBm
500MHz  -1.7dBm
550MHz  -3.0dBm



TEXAS 250/II 250MHz x10 Probe

  1MHz      0dBm
 50MHz   -0.2dBm
100MHz  +0.1dBm
150MHz  +1.3dBm
200MHz  +1.5dBm
250MHz  +2.3dBm
300MHz  +1.3dBm
350MHz  +0.2dBm
400MHz  -1.8dBm
450MHz  -3.1dBm
500MHz  -3.8dBm

TESTEC HF512 is about 130USD, so not a cheap one anymore. What not funny, Testec was not able to provide me
any detailed informations about frequency response nor impendance vs. frequency dependency. For an german
manufacturer a real pain in the ass. At least their probe have nice accessory kit included.

Texas costs 50USD, so it's bewteen 'ok, i will use it' and 'no, i will buy used TEK probe'

[alm]

Those responses look pretty terrible, do you have any data for a decent probe (eg. the matching Tek P6139a probe)? Resonance in the pass band is pretty bad, especially for the Hantek and Testec probes. I'm not so worried about behavior beyond the pass band, as in the case of the Texas probe, since you're supposed to use these with <= 250 MHz scopes that would attenuate these signals anyway.

It's very rare to see Bode plots for probes or scopes, so that Testec provide those doesn't surprise me. Impedance vs frequency is standard in any decent probe datasheet, however. I'm quite sure Testec used to produce the Hameg-branded passive probes; I would have expected them to be at least halfway decent.

500 MHz is the upper limit for passive high-Z probes. Probes beyond 200 MHz or so are often tuned to the impedance of a particular scope attenuator, which is why Tek made a whole bunch of ~300 MHz passive probes at some point (P6130, P6131, P6133, P6134, P6136, P6137, P6138, P6139), one for each series of scopes. The recommended probe for a 2465 was the P6131, for the 2465A the P6136 and for the 2465B the P6137. The probes would compensate for bumps in the scopes input impedance. By the time the TDS700 came along, they started standardizing front-ends so the P6139a could be used on all new 250+ MHz designs, but I wouldn't be surprised if a random commodity probe is a bad match for this front-end.

[nctnico]

the point that I'm trying to make here and that is that you cannot get more resolution from an ADC than what it has been designed for by just oversampling c.q. averaging.

of course you can, you can see it even on yuor DSO. When you above the  hi-res limit the signal will be not better
(simply use 10x att. and 10x zoom for better view).

Yes, but at those zoom levels the extra bits (and errors) added by the hi-res mode won't be visible. The whole point of having a hi-res mode would be to zoom in at a part of a signal and see the extremely small wiggles.

About the documents you linked: please read the conditions in which oversampling can work very carefully and you'll see I'm not saying anything else then the documents you linked. What I'm also saying is that in a scope those conditions are not met by design so the scope display may not show the actual signal. Again, oversampling sounds great in theory but in practical applications there are a lot of pitfalls.

[JoeyP]

Hmm, you expect 12bit (that is 4096 steps) to show in an image about 350 pixels high?

No, I expect to see the distortion that you claim to exist in a practical application for which the scope was designed. I don't.

happens if I put my Tek scope to the test to see what it does with a signal about 3 LSB in height. The first image is the original signal, the second is the same signal attenuated 100 times and then enlarged 100 times.

So this is how you think people normally use scopes? A signal of 3LSBs p-p expanded 100x? Sounds pretty silly to me.

Which means you agree with me that claiming to get more bits from an ADC is safe to do for a scope manufacturer because the whole claim is useless to begin with because the screen's resolution isn't high enough to show the errors. 

Agree with you? Hardly. Tinhead has it exactly right. His images tell the story, and if you can't tell by comparing the two images that oversampling has greatly improved the resolution, then there really must be something wrong with you. Perhaps you shouldn't be allowed near test equipment at all. You might get hurt!

By multiplying the result up 100X, you are displaying a small 10 bit signal with 16.6 bits resolution - and you're somehow surprised to see distortion? Despite the absurd conditions you've set up, you've actually proved the validity of the technique with your own image. You've just failed to analyze it correctly. I've taken the liberty of finishing the analysis for you:

I drew in black, a more or less ideal representation of the original triangle wave you captured. The height is 290 pixels, and you've stated its amplitude is 3LSBs of the 10 bit result. So each LSB of the 10 bit result is 96.7 pixels.

I then searched and marked the worst case deviation that I could find and drew it in blue. It is 36 pixels.

36 / 96.7 = 0.37, or a deviation (DNL) of just 0.37 LSB referred to the 10 bit result. That is a *very* respectable figure for an ADC. Hi-speed ADCs typically spec DNL on the order of 0.5 LSB or more.

So I'm sure that Tektronix will be relieved to hear that you've validated their product for them. I'll bet they were very worried about your opinion.

[tinhead]

Those responses look pretty terrible, do you have any data for a decent probe

yep, i do. And here is the less funny part, all my measurments are bad, the terminator i've used is broken. With different one
terminator and P6243 everything looks flat as in user manual - with the broken one even P6243 was 3dBm out at 400MHz.

I have to sleep, tomorrow i will provide corrected values

[nctnico]

@JoeyP:
You got the numbers wrong. The signal is about 3 LSB high from the ADC's perspective, not the 10 bit fictional resolution. On top of the 10 bit resolution there is also the scope's sin(x)/x interpolation to make a neat line. Here are two other pictures at lower zoom level settings:

[JoeyP]

@JoeyP:
You got the numbers wrong. The signal is about 3 LSB high from the ADC's perspective, not the 10 bit fictional resolution. On top of the 10 bit resolution there is also the scope's sin(x)/x interpolation to make a neat line. Here are two other pictures at lower zoom level settings:

Actually you got them wrong too. I should have ignored your numbers and done all of the math from the start. Your signal was 1.2Vpp in the first image. You attenuated it 100X for the second image, so it is 12mVpp. 256 x 0.012 / 1.6VFS = 1.92 LSB referred to the 8 bit converter. That's 7.68 LSB referred to the 10 bit result. So the actual DNL in the image is 0.95 LSB referred to the 10 bit result. Still monotonic and happens to be exactly the DNL spec for the TI/National ADC08200, a typical 200MHz 8 bit converter.

[tinhead]

Those responses look pretty terrible, do you have any data for a decent probe

yep, i do. And here is the less funny part, all my measurments are bad, the terminator i've used is broken. With different one
terminator and P6243 everything looks flat as in user manual - with the broken one even P6243 was 3dBm out at 400MHz.

I have to sleep, tomorrow i will provide corrected values

here we go, both TESTEC and TEXAS are definitely good enough

TESTEC TT-HF512 500MHz x10 Probe

  1MHz    0dBm
 50MHz   -0.1dBm
100MHz  -0.1dBm
150MHz  -0.2dBm
200MHz  +0.2dBm
250MHz  +0.2dBm
300MHz  +0.4dBm
350MHz  -0.3dBm
400MHz  -0.7dBm
450MHz  -1.7dBm
500MHz  -2.2dBm
550MHz  -2.8dBm



TEXAS 250/II 250MHz x10 Probe

  1MHz     0dBm
 50MHz  -0.3dBm
100MHz  -0.5dBm
150MHz  -0.1dBm
200MHz  -0.4dBm
250MHz  -0.1dBm
300MHz   0.0dBm
350MHz  +0.3dBm
400MHz   0.0dBm
450MHz  -0.5dBm
500MHz  -1.1dBm
550MHz  -2.4dBm


The Hantek probe, and i assume all the others almost identical probes on ebay (where only name and color seems to be random),
is still useless on Tek TDS7xx series. Probably with different trim cap insde it would work, but yeah, i don't give a s* on that.


Hantek P6200 200MHz x1/x10 Probe

  1MHz     0dBm
 25MHz     0dBm
 50MHz  +1.7dBm
 75MHz  +2.9dBm
100MHz  +3.8dBm
125MHz  +4.4dBm
150MHz  +4.5dBm
175MHz  +4.7dBm
200MHz  +4.9dBm

[nctnico]

@JoeyP:
You got the numbers wrong. The signal is about 3 LSB high from the ADC's perspective, not the 10 bit fictional resolution. On top of the 10 bit resolution there is also the scope's sin(x)/x interpolation to make a neat line. Here are two other pictures at lower zoom level settings:

Actually you got them wrong too. I should have ignored your numbers and done all of the math from the start. Your signal was 1.2Vpp in the first image. You attenuated it 100X for the second image, so it is 12mVpp. 256 x 0.012 / 1.6VFS = 1.92 LSB referred to the 8 bit converter. That's 7.68 LSB referred to the 10 bit result. So the actual DNL in the image is 0.95 LSB referred to the 10 bit result. Still monotonic and happens to be exactly the DNL spec for the TI/National ADC08200, a typical 200MHz 8 bit converter.

You can reason all you want but the last images I posted clearly show that the lack of noise causes the interpolation algorithm not to work properly. In case of an oscilloscope you actually want the input stages to be as low noise as possible which contradicts the noise requirement for oversampling. I'm sure that when I add enough noise to the input signal the hi-res mode will show a triangle waveform with straight lines. I'll leave that as an exercise for the reader though. I think I clearly demonstrated that hi-res modes on scopes can and will produce false readings.

[Wuerstchenhund]

I know most PC-based oscilloscopes are pretty much garbage.  But I'm wondering, does anyone make a pc-based scope with pro-level electronics on par with Tek/Agilent/Rigol, etc?   To give an example: If I can't find an adequate PC-based one, my next choice would probably be a Rigol DS2072)

There are lots of external 'scopes' (digitizers), which generally can be divided into two categories: 'low-end/crap' (with often pathetic specs) and good quality ones. Unfortunately, if you look at the latter and take out digitizer modules for VXI, PXI and other embedded systems there is not too much left, and what's left mostly consists of (very expensive) highend stuff like Agilent's DSO90008 series.

Since you mention Rigol I assume the DSO90008 is not what you're looking for.

I know what you're saying: Why not just get a regular DSO?   Well as I said I'm a programmer, so I spend most of my working life in front of 4 gigantic monitors and screaming fast development PCs.  And I'm much faster with a mouse than those buttons and knobs on a regular scope.  I'd love to have the quality capture abilities of a high-end scope without paying for the tiny scope display.

The problem is that you generally pay excessively for *not* having the tiny scope display. Most of the better external scopes/digitizers are for special applications and come with a price that reflects that. At the end of the day, the most economic way is to just buy a regular DSO which can be remote-controlled, ignore that it has local controls and a screen, and just use it with your PC as you would do with an USB scope.

[JoeyP]

You can reason all you want but the last images I posted clearly show that the lack of noise causes the interpolation algorithm not to work properly.

Your last images may "clearly show" -you- something, but you seem to be working very hard to contradict the obvious.

tinhead has posted a good example of oversampling/hires mode for a Tektronix scope here: https://www.eevblog.com/forum/reviews/quality-pc-based-oscilloscope/msg183740/#msg183740

I've posted an example from an Agilent scope below. The test signal was a 20mV p-p ramp. Each image was captured after a single-shot acquisition using real-time sample mode.

The "Signal x 1" image is the actual signal catpured at 128mV/div. I used the vernier to make the math more intuitive. With 8 divisions, full scale is 1024mV so the 20mV test signal is 2% of full scale. Note the indicated amplitude of 52mV due to internal noise that the scope has added to the test signal. If you magnify the image, you can see that each 8 bit LSB occupies 2 or 3 vertical pixels depending upon rounding error for each position.

The "Normal x 16" image shows the same captured signal zoomed vertically 16x (after the acquisition halted). The 8 bit resolution is quite visible, and as expected 255 x 52mV/1.024V FS yields 13 LSBs p-p.

The "Hires x 16" image shows the same signal captured with hires mode turned on, also at 128mV/div, and then zoomed vertically16x to 8mV/Div. No other changes were made to the scope or the signal. I used 16x because the scope has increased the converter's 8 bit hardwre resolution to a 12 bit oversampled result, so 4096/256=16.

The increase in resolution in the hires image speaks for itself. If you magnify the image, you can see that each LSB of the 12 bit result now occupies 2 or 3 pixels in height, exactly as expected. You can also see that hires mode has cleaned up much of the noise resulting in an accurate amplitude measurement of 20mV p-p. The result is 79 individual steps, which is very close to the theoretical 4095 x 20mV/1.024V FS = 80.

At the very least, the hires/oversampling feature cleans up noise and allows much lower level single-shot signals to be measured. And if you really need the extra resolution, it's there too.

So if nctnico wants to pretend that this feature doesn't work, he is certainly welcome to do so. But while disparaging an entire industry, or a particular manufacturer using a fabricated argument and "polite" language may be easy, it is not fair to those being disparaged nor to those who may be reading in an effort to learn more about a valuable feature they may desire in a scope. It needs to be challenged, I will be here to do so any time I see it happen.

[nctnico]

Let me repeat myself: a scope input has to be low noise to be usefull to begin with. The oversampling used in the high-res mode however requires a certain amount of noise to work. There is a clear contradiction here.  To put that to the test I've demonstrated that the required noise level can actually be insufficient in a normal situation and cause the oscilloscope to show a difference signal than there is in reality without any warning. BTW, for my test I've used a simple RC generator which can be bought for $50 through Ebay nowadays. It most certainly is not some super high end ultra low noise generator.

To put it simple: if there is not enough noise the oversampling simply will not work. You can't change the laws of physics. Oscilloscope makers clearly cut a corner here.

[tinhead]

To put it simple: if there is not enough noise the oversampling simply will not work.

of course you right here. I've made some other pictures - to finally end up the "discussion" -
and to show a nice Hi-Res trap:

Let's take a 1MHz sine wave (R&S SMX used for that), with amplitude high enough to not add
"any" trigger jitter (which could cange our results):



Tektronix TDS is using 2 vertical pixel (in 8bit mode) for one LSB. One DIV is 50pix high, zoomed 5 times
we have 16mV/40mVper DIV, so 0.4 x 10pix, this makes 4pixel in not zoomed which is equal to max. 2 LSB moise:



The zoomed sine wave looks of course not really clean:



With hi-res it looks better:



But here is the Hi-trap. I have enabled 20MHz BW filter (to have only max. 2LSB noise),
one can think "this will optimize the waveform", well no ... this kills H-Res, the waveform is less smooth:



as here where the full BW is enabled (sure, total amplitude deviation "seems" to be higher, but that't the
result of broader freq. specturm. The hi-res is giving smooth edges and fine steps between values, that's the deal):



The reason is of course the amount of noise, here we have 24mv/40mv per DIV, so 0.6 x 10pix, this makes 6 pixel
which is equal to 3 LSB of the waveform:



Therefore, properly used, hi-res make sense and it is possible.

You can't change the laws of physics. Oscilloscope makers clearly cut a corner here.

Well, normally there will be always some noise available, if not - in most case - it is enough to disable all BW filters,
then you will pickup enough noise to let Hi-res work. In case this is still not enough, apply some random ^^

[JoeyP]

nctnico then:

... And the people from Picoscope think they can get 12 bit from an 8 bit ADC and they are able to get 250MHz into a 1M Ohm 15pf input.

nctnico now:

To put it simple: if there is not enough noise the oversampling simply will not work. You can't change the laws of physics.

I'm glad to see that you now acknowledge that pico, or any other scope vendor can indeed achieve 12 bits resolution from an 8 bit converter through oversampling of the signal+noise typically present in broad band scopes.

You'll be relieved to know that neither I nor Agilent violated any laws of physics in capturing the images posted here:
https://www.eevblog.com/forum/reviews/quality-pc-based-oscilloscope/msg185551/#msg185551

No one is trying or claiming to violate any laws of physics. It is those very laws that explain why noise is proportional to bandwidth. In a broad band device like a scope, lack of noise is rarely if ever an issue. We should all be so lucky as to have scopes with such low noise levels that oversampling were not possible. I would make that trade in a heartbeat, but that option is not available. Scopes have noise. Manufacturers have wisely made good use of that noise to enable a very useful hires mode. But like many scope settings, if you try hard enough you'll find a way to get an invalid measurement.