Are 8 bits really enough ?

[T3sl4co1l]

Well, how much do you need?

A typical DSO is only a few hundred pixels tall on the screen, so you can't *see* anything finer than about 8 bits (more like 9 bits these days, i.e. 512 pixels height, or even 10 bits for fancier or PC-based scopes).  If you need to zoom in without taking additional readings, you might need more, but you run into problems with dynamic range vs. offset, and bandwidth capability.

I can't see the detail, but won't the onscreen measurements benefit from higher bit depth?

Not really: most of the ones I've seen, display about three digits more than they represent, anyway!  It would be very nice if they put a sigma on those as well, sadly most don't (with I think Lecroy being one of the few who do, unless it's simply standard only on the embedded-PC-based pricey ones).

A simple measure of accuracy is how many digits are dancing around in a given steady state display.

Errors arise from the signal itself (voltage and phase noise), trigger noise (and delay if shifted from the trigger point), quantization noise (including time quantization*) and thresholding.

*Which depends on if it's a "deep memory" scope, or if it's reading by various methods: decimated to screen resolution, averaged to screen resolution (Hi-Res), histogram (DPO or heat map style); and how the measuring algorithm operates on that data (does it operate on what's on screen, or on that segment of the buffer, or the total buffer?).

From what little I know of the usual measuring algorithms, many of them have threshold limitations: for example, to measure "top" of a square wave, how do you determine which points are in the "top"?  What if you try measuring this on a wave that's decidedly not square: like a sine, or a switching waveform with ringing, or white noise?  There should be some measure of confidence corresponding to that measurement, which is relayed to the user.  But this is rarely provided.

(More broadly, signal and image processing needs to take these things into account.  It's amusing when you have a picture of a fire hydrant identified as "dog" by some AI; it's less amusing when your digital communications channel is fading out and it makes bad assumptions about what it thinks the signal is, rather than reducing bandwidth to a more confident level.)

In any case, a hard threshold is the worst possible signal processing method, and has noise equal to any given sample.  Ideally, you'd like the measurement to have 1/sqrt(N) noise, for N samples in the measurement span (whether it's piled on top as a histogram or stretched out in a very long buffer).

And yes, when you have megs of buffer, random errors can be very well refined indeed (1e6 points --> 1e3 reduction in noise!), leaving you with only the systematic errors (like the nonuniformity of bit sizes in the ADC).  If the measurement algorithm harnesses this resource, then it can be as accurate as the systematic limitations; my concern is that most will not go to this length, and worst of all, you have no way to know because they rarely tell you the uncertainty.

Tim

[Fungus]

I can't see the detail, but won't the onscreen measurements benefit from higher bit depth?

By how much? Half a percent?

I'm not sure how important the accuracy of the on-screen measurements are in an oscilloscope. Opinions will obviously differ (especially around here) but "a couple of percent" seems fine to me.

I wouldn't pay a lot more to go from (eg.) 2% to 0.5% and I see no reason at all for the 0.1% accuracy I demand from a multimeter. I'm much more interested in what's going on generally, looking at the shape of the waves, the rising edges, etc.

YMMV.

[Fungus]

A simple measure of accuracy is how many digits are dancing around in a given steady state display.

My DS1054Z dances around quite a lot in 8-bit mode. In 12-bit mode it's rock steady.

(on the calibration signal)

[coppice]

A simple measure of accuracy is how many digits are dancing around in a given steady state display.

My DS1054Z dances around quite a lot in 8-bit mode. In 12-bit mode it's rock steady.

(on the calibration signal)

That's all about the front end noise. You don't need a 12 bit ADC. You need an 8 bit front end, to match the 8 bit ADC.

[theoldwizard1]

Since you're asking, I'm guessing you're not using the 'scope for high-precision or commercial work.

An 8-bit ADC will give you nice traces on the screen.

Many have correctly stated the issue of why increase the "accuracy" when the display is the limiting factor and this is an excellent point.

But as the on-board computing capability continues to increase (look at Red Pitaya) as does display technology technology (look at the latest iPads) and 8 bit ADC is going to look like the engine out of an old Ford Pinto in a new Mustang (both 2.3L, although the later is turbocharger).

I was extremely happy to see Rigol's latest 'scopes move to a larger screen.  What is practical ? 8" - 10" ??  There are several battery operated, automotive directed 'scopes with 8" screen on the market these days (think really thick tablet with a heavy rubber case, all ridiculously over priced).

If you expect to use the 'scope as a waveform-capture device for later processing (for example, estimating spectral information through DFT), you may find that an 8-bit ADC is only giving you 6-7 bits of resolution between stepped gain/attenuation and quantization noise. You're not going to be seeing signals -40 dBc with that capture.

And that is where I am going.  That "fancy" post processing software is become more like "standard equipment" on these new 'scopes (and a market Red Pitaya and SmartScope are hoping to tap).  Or at least, a potentially profitable option (the manufacturer only pays for software once !)

All that said, I'm very happy with my Rigol 1054 for hobby use, especially for its price.

The new "low end" 'scopes are AMAZING bang for the buck !


FYI, My thoughts were, "How much would it cost to go from an 8 bit to a 10 bit 'scope ?  Maybe even a 12 bit 'scope "  14 bit is crazy for this marketplace.

[Fungus]

FYI, My thoughts were, "How much would it cost to go from an 8 bit to a 10 bit 'scope ?  Maybe even a 12 bit 'scope "  14 bit is crazy for this marketplace.

Probably quite a lot. Everything downstream from the ADC would have to get wider. The CPU would probably have to switch to 16-bit processing instead of 8-bit. Do they make 10-bit (or 12-bit) RAM chips? If not you're going to have to use multiple chips instead of the neat 8-bit wide RAM chips they're using at the moment. It all adds up to more money.

[theoldwizard1]

Probably quite a lot. Everything downstream from the ADC would have to get wider. The CPU would probably have to switch to 16-bit processing instead of 8-bit.

You are out of touch with reality !  The Red Pitaya has multiple processors, one of them is 64 bits wide !

Do they make 10-bit (or 12-bit) RAM chips? If not you're going to have to use multiple chips instead of the neat 8-bit wide RAM chips they're using at the moment. It all adds up to more money.

Of course they do and they have for years !

You could easily make one using part of one of the FPGAs, and some external "instrument grade" OP amps and laser trimmed resistors.  I suspect no one has done this because to calibrate a "discrete" design to less than 1mV and KEEP IT IN CALIBRATION while running, is to too difficult.

[Howardlong]

Probably quite a lot. Everything downstream from the ADC would have to get wider. The CPU would probably have to switch to 16-bit processing instead of 8-bit.

You are out of touch with reality !  The Red Pitaya has multiple processors, one of them is 64 bits wide !

It's just a shame the Red Pitaya has yet to show itself as anything other than a toy in the TE arena due to the extremely rudimentary software and 100MSa/s sampling rate.

The Red Pitaya has some niche marketsnd uses, I know, I am using them in very low volume OEM devices, but a piece of test equipment it ain't.

Do they make 10-bit (or 12-bit) RAM chips? If not you're going to have to use multiple chips instead of the neat 8-bit wide RAM chips they're using at the moment. It all adds up to more money.

Of course they do and they have for years !

You could easily make one using part of one of the FPGAs, and some external "instrument grade" OP amps and laser trimmed resistors.  I suspect no one has done this because to calibrate a "discrete" design to less than 1mV and KEEP IT IN CALIBRATION while running, is to too difficult.

Your idea of easy and mine are somewhat different I fear. How about doing it and educating us all how easy it is? 

[Andreas]

FYI, My thoughts were, "How much would it cost to go from an 8 bit to a 10 bit 'scope ?  Maybe even a 12 bit 'scope "  14 bit is crazy for this marketplace.

And how about 16 Bits. (20 Bits with oversampling ?).

https://www.picotech.com/oscilloscope/5000/flexible-resolution-oscilloscope

If you ever worked with a >= 12 Bit scope you never want to go back.
And with a PC screen you can at least use 10 bits at a glance and can easily zoom in at details of complex waveforms.

With best regards

Andreas

[theoldwizard1]

If you ever worked with a >= 12 Bit scope you never want to go back.

I guess that is where I am coming from !  "Back in my day" (>10+yo) the few DSO scopes we had were >= 12 bits, but I don't recall the sample rate.

[Kleinstein]

With the small screen found on normal DSOs there is little need for more than 8 Bits resolution. Especially as you get a little more resolution due to oversampling if the sampling rate is lower.  So users usually prefer the faster 8 Bit ADC over a 10 Bit ADC with lower speed. Often it is also more the input amplifier or the probe that limits the noise and not so much the ADC. So it is not only about having a better ADC and more memory (not a big deal any more) a higher resolution scope should also have a lower noise input stage.

A lower noise input stage could make the scope more useful - even with only 8 Bit resolution.

[Wuerstchenhund]

FYI, My thoughts were, "How much would it cost to go from an 8 bit to a 10 bit 'scope ?  Maybe even a 12 bit 'scope "  14 bit is crazy for this marketplace.

And how about 16 Bits. (20 Bits with oversampling ?).

https://www.picotech.com/oscilloscope/5000/flexible-resolution-oscilloscope

But look at the specs:


    Flexible Hardware Resolution, from 8 to 16 bits
    200 MHz analog bandwidth
    1 GS/s sampling at 8 bit resolution
    500 MS/s sampling at 12 bit resolution
    62.5 MS/s sampling at 16 bit resolution
    512 MS buffer memory
    100,000 waveforms per second
    200 MS/s Arbitrary Waveform Generator
    Serial decoding and mask testing as standard
    Windows, Linux and Mac software"


I'm not sure how useful a scope is that at 16bit is limited to a 30MHz sine wave (and much lower frequencies for non-sine waveforms). It's also not exactly cheap, at £699 for the 2ch 60MHz variant and £1695 for the 200MHz 4ch version with AWG.

PicoScopes are great for a few specific tasks but they make for very poor general purpose scopes. Even our engineer that love them wouldn't want them as their only scope.

[rf-loop]

With the small screen found on normal DSOs there is little need for more than 8 Bits resolution. Especially as you get a little more resolution due to oversampling if the sampling rate is lower.  So users usually prefer the faster 8 Bit ADC over a 10 Bit ADC with lower speed. Often it is also more the input amplifier or the probe that limits the noise and not so much the ADC. So it is not only about having a better ADC and more memory (not a big deal any more) a higher resolution scope should also have a lower noise input stage.

A lower noise input stage could make the scope more useful - even with only 8 Bit resolution.

(Entry level)
Owon XDS3000(A) have 1GSa 8bit ADC and 500MSa/s 12bit ADC.
Both are enough for full 100MHz BW.  If look some scope what have "high resolution" up to 12bit  useable frequency response is very low.

Screen resolution (800x600)  is not enough for 12bit but of course it is useable when scope is topped and do vertical zoom for details. Also of course full resolution is available in various file formats.

There is somewhere thread where it was looked with bit more details/tests.

[Fungus]

A lower noise input stage could make the scope more useful - even with only 8 Bit resolution.

That's expensive to do in hardware.

In software you can do averaging to get the same result.

The DS1054Z (for example) has an 8-bit averaging mode to reduce noise and a "12-bit" oversampling mode to give pseudo-higher resolution.

[rf-loop]

A lower noise input stage could make the scope more useful - even with only 8 Bit resolution.

That's expensive to do in hardware.

In software you can do averaging to get the same result.

The DS1054Z (for example) has an 8-bit averaging mode to reduce noise and a "12-bit" oversampling mode to give pseudo-higher resolution.

And frequency response with 12bit high resolution is?  Is it 2MHz or what? 
Btw, do you know what settings need use it really is 12bit least "in theory".  I have not seen any real data about it.

Can you get 12bit resolution data out from scope, example as .CSV file or is is just only for "nice image"
Example some Tektronix have normally 8bit data available but with software baased high resolution it give out double bytes so that this improved accuracy (done using box car averaging) is also available.

[nctnico]

FYI, My thoughts were, "How much would it cost to go from an 8 bit to a 10 bit 'scope ?  Maybe even a 12 bit 'scope "  14 bit is crazy for this marketplace.

And how about 16 Bits. (20 Bits with oversampling ?).

https://www.picotech.com/oscilloscope/5000/flexible-resolution-oscilloscope

But look at the specs:


    Flexible Hardware Resolution, from 8 to 16 bits
    200 MHz analog bandwidth
    1 GS/s sampling at 8 bit resolution
    500 MS/s sampling at 12 bit resolution
    62.5 MS/s sampling at 16 bit resolution
    512 MS buffer memory
    100,000 waveforms per second
    200 MS/s Arbitrary Waveform Generator
    Serial decoding and mask testing as standard
    Windows, Linux and Mac software"


I'm not sure how useful a scope is that at 16bit is limited to a 30MHz sine wave (and much lower frequencies for non-sine waveforms). It's also not exactly cheap, at £699 for the 2ch 60MHz variant and £1695 for the 200MHz 4ch version with AWG.

I don't buy their explaination. As usual Picoscope pulls 16 bits out of their arses. Generally you can't get 16 bits from an 8 bit ADC. Non-linearity will kill you long before you get to 16 bits not to mention clock jitter. Also there is the noise requirement to be able to upsample at all; basically the lowest 8 bits of the 16 bit result must be swamped in noise to be able to increase resolution. If you have a very steady, low noise DC level you won't be able to digitize that with 16 bit accuracy using an 8 bit converter because the result will always be stuck at a constant value from the ADC. Oh and please note Picoscope uses the word resolution which is something completely different than accuracy.