Why most DSOs are 8bit?

[TNb]

I just noticed that many oscilloscopes, even expensive ones, actually use 8 bit ADC. Now there are some models with 12 bit ADC, but these are kinda new, before 2013 most of DSOs were 8 bits. This is kinda surprising, because the price difference between 8 bit and 12 bit ADC chips is basically 0 when you talking about 3-5 grand scope, but these 4 bits provide so much resolution, basically giving you 3840 additional points on vertical scale(2^12 - 2^8), or in other words 16 times(!) better vertical resolution.

Is there reason why they are made that way? I suspect maybe with lower resolution you can sample stuff faster, but nah... sacrificing so much vertical resolution seems so not right.

[danadak]

Largely due to sample rate, the 8 bitters capable of much higher sampling rate
then 12, 14. If you look at scope specs, bandwidth, Nyquist, you can see the
tradeoffs.

As time progresses you will see higher resolution for the same sampling rate,
its simply getting process device to device to behave.

And not all 12 bitters have better ENOB than 8     http://www.tek.com/blog/not-so-high-high-resolution



Regards, Dana.

[max_torque]

I suspect maybe with lower resolution you can sample stuff faster, but nah... sacrificing so much vertical resolution seems so not right.

If you look at how an Analogue to Digital Convertor functions, it should become obvious that resolution takes time!

Broadly (and this the very broad begineers ADC explanation!) an ADC, is a comparitor.  It has an architecture that roughly speaking has the voltage to be measured on one input, and then it switches in a certain proportion of a reference voltage, and checks is REF > IN ??
Because the voltage comparison takes a finite time to do, the more comparisons you do, the longer it takes.  Of course, with a higher number of comparison voltages, you can quantise the input signal to a high resolution.

In reality, a scope uses a fancy analogue front end with different gain/attenuation settings to help minimise the downsides of having a low resolution ADC.
For example, if you wanted to measure say a 10V signal, with an 8bit ADC, the maximum resolution is 10/1024  (~9.8mV).  So that fixes your resolution, so if you were to put say a 100mV input voltage into the same circuit, you wouldn't be able to see the difference between say 98 and 102 mV.   However, switch in some gain, and you also increase your effective resolution.  A gain of say 10, means you can now resolve 0.98mV on signal up to 1V.

And this is what a scope does.  As you change the vertical scaling, from say 5mV/div to 50mV/div, you are not just changing the input range, but also the input resolution.  This mostly works, because in general, the lower the voltage, more more we care about high resolution.  ie a circuit that is fed by 1000V probably doesn't care if it gets 990 or 1010V, (+-1%)it still works fine.  But a circuit that is processing 10mV would certianly care about +-10V......

Having said all that, as ADC improve, and get faster and faster, modern DSO's are starting to have simpler and simpler (ie cheaper!) front ends, and use the greater resolution in the ADC to get sufficient capability.

[rollatorwieltje]

What do you mean by no price difference? A quick search on Mouser; 1GS/s 8 bit ADC is maybe €50 (HMCAD1511), the cheapest 12 bit (ADC12D500RFIUT/NOPB) about €360?

Also the required memory bandwidth will increase significantly. Dealing with 12 bits is also awkward. Do you store them in 16 bit containers? Now you're wasting space. Do you pack the data? Now you need to dick around with alignment when performing stuff like math operations*.


_{* I don't know about FPGA/DSP stuff, maybe they can handle this without a problem?}

[Kleinstein]

At the high speed of DSOs, there is quite a price difference from an 8 bit to an 12 Bit ADC. For the fast converters (like 1 GHz) there is also considerable noise, that limits the useful resolution. Due to the high bandwidth the SNR is not much better than an 8 Bit ADC anyway, as the input is usually high impedance and this relatively high noise.

Even with an 8 Bit ADC they get higher resolution by oversampling if slower sampling rates are used. So the main advantage of a higher resolution ADC is not less noise, but better linearity.

An other point is that until a few years, the screen resolution was not that much better than 8 Bits in the vertical. So with only 480 vertical lines and some of them lost to the grid / axis, there is not even 9 Bit resolution on the screen.

[vealmike]

And not all 12 bitters have better ENOB than 8     http://www.tek.com/blog/not-so-high-high-resolution

This.
Take a look at the R&S 'scopes. They look expensive for their bandwidth, but the effective resolution and noise floor is excellent.
Yer gets what yer pays for.

[JanJansen]

I,m waiting with buying a scope, i want at least 16 bit.

[Kleinstein]

If you need a 16 Bit scope - use a sound-card based cheap solution. Even the 48 kHz sampling rate it is hard to get true 16 Bit resolution one an 1 M input. Though at the low speed one could get away without a proper probe and this way get lower noise.

[CatalinaWOW]

I think that in addition to the technical difficulty of getting higher resolution at high speed there is a demand side part of the answer.

DSOs historically were a replacement for analog scopes.  Which are largely used to find shapes of signals and do crude measurements.  Eight bits is enough to get very similar displays to an analog scope on what was a typical 4 to 5 inch oscilloscope screen.  So why push for higher resolution?

That is changing as display sizes are growing and DSOs are being used more and more as a front end data capture system, but a large part of the market is still just looking for the shape and comparative timings of the wiggles.

[helius]

Back in the day, this was one of LeCroy's advertised features: both the availability of 10-bit ADC in selected models, as well as software averaging with up to 11 bits for low-bandwidth signals, when using 8-bit ADC.

[Andreas]

I,m waiting with buying a scope, i want at least 16 bit.

Why wait.

there are 16 or even 24 bit scopes available.

Mine has up to 20 Bit (with oversampling).

With best regards

Andreas

[David Hess]

My oldest DSO  (Tektronix 7854) is 10 bits, has a 1024x1024 260x325dpi "retina" display, a bandwidth of 400 MHz, supports ETS up to 100GS/s, and was first produced in 1981 so I am getting a kick out of these replies.

There are a couple of reasons that DSO resolutions are typically not larger than 8 bits:

1. Higher resolution digitizers are either more expensive or sacrifice sampling rate and/or bandwidth.
2. 8-bit data paths are standard.  Using more than 8 bits will typically mean using a 16 bit data path. (1)
3. This also means that going from 8-bits to 10-bits halves the record length if byte wide memory is used or doubles the amount of memory required.  It also doubles the bandwidth requirement between the acquisition and display record which may impact the number of acquisitions per second.  Note however that good 8-bit DSOs already use 16-bit records for display and calculation purposes.
4. Noise at high bandwidths, especially for high impedance inputs, is higher than an 8 bit digitizer will support anyway.
5. It is very difficult to build a wideband vertical amplifier with a transient response and linearity better than 8 bits.

Ultimately typical applications for digital storage oscilloscopes do not require more than 8-bits of resolution.  More specialized instruments are higher resolution at a considerable cost premium but in some respects are just as limited because of other considerations.

(1) This was one of the major reasons the IBM PC was built with the 8-bit external bus 8088 instead of the 16-bit only 68000.  The 68008 did not come out until later.

[danadak]

I also have a 7854. Its sample rate of 500 Khz limits it for non periodic signal
work for sure. Its memory severely limited for capturing one off events. But
still a very useful scope. Especially plugin options.

Regards, Dana.
-

[David Hess]

I also have a 7854. Its sample rate of 500 Khz limits it for non periodic signal
work for sure. Its memory severely limited for capturing one off events. But
still a very useful scope. Especially plugin options.

Regards, Dana.

The 7854 operates as a literal digital storage sampling oscilloscope which actually makes its 500kHz sampling rate pretty high; sampling oscilloscopes did not achieve that until much later although they had bandwidths an order of magnitude higher and were often higher resolution.  If you work out the numbers, it is still effectively a 100 GS/s ETS instrument.  The interesting thing is that Tektronix made it 10 bits in all respects instead of 8 bits but that is because 10 bits was needed for a reasonable record length...

The 7854's record length in normal acquisition mode is limited by how it uses the sweep ramp to generate the horizontal position and has nothing to do with memory.  If its ADC was 12 bits, then the record length would be four times as long.  If you have the 7B87 timebase installed, then you can make single shot captures up to the real time sampling rate but since the memory addressing was designed for 10 bits in normal acquisition mode, it is still limited to a 1k record length.

A slightly more modern version of how the 7854 works can be found in the 2252 which only has hardcopy capability; it cannot even display stored records on its CRT and like the 7854, its digitizer is greater than 8 bits.

I ran some tests a couple years ago on my 7854 and the 10 bit resolution it provides can indeed reveal waveform features which are invisible on an 8 bit DSO.  They even showed up when using sampling plug-ins which points to why higher than 8 bit resolution is not wasted on true sampling oscilloscopes.

[Zero999]

Also the required memory bandwidth will increase significantly. Dealing with 12 bits is also awkward. Do you store them in 16 bit containers? Now you're wasting space. Do you pack the data? Now you need to dick around with alignment when performing stuff like math operations*.


_{* I don't know about FPGA/DSP stuff, maybe they can handle this without a problem?}

As you said at the bottom of your post, the unusual number of bits isn't a problem with an FPGA.

[mikeselectricstuff]

Simple - 8 bits are enough for most things that scopes are used for. Any more gets expensive, and the extra cost is not justifiable for most users.

[rx8pilot]

Simple - 8 bits are enough for most things that scopes are used for. Any more gets expensive, and the extra cost is not justifiable for most users.

It is not just the cost of the A/D convrter - the improved performance required of the analog front end is an expensive endeavor. It serves no purpose to digitize noise a the lower levels, so you really need a sweet front end to do any good with 12 bits. The LeCroy 12bit scopes are cool, but the sample rate is relatively low.

[tooki]

At least in the case of cheaper scopes, consider also the output resolution (the display). Like many of us here, I've got a Rigol 1054Z, whose display is 800*480px. By the time you subtract all the user interface, you're left with a precisely 600*400px viewport for data. A naked eye looking at a live signal isn't gonna see the difference between 256 levels scaled up to 400px, compared to 1024 levels scaled down to 400px if it were 10-bit, for example. You'd be adding a great deal of cost for almost no noticeable improvement in performance.

I mean, tell me if I'm wrong, folks, but my understanding is that the primary purpose of the scope is to see change over time (i.e. the shape of waveforms), and that for precise measurements, you use a separate meter for the value you're measuring.

[Kleinstein]

Most DSOs allow for a zoom in to a part of the curve - so the screen resolution is not such a hard limit. At least today they would offer higher resolution screens if the rest would give a better resolution.

I think the main point is the SNR: The typical 1:10 passive probes give quite some noise background. Except for high signal amplitudes, there is just no need for an ADC with less noise than 8 Bits at 1 GHz. If the Bandwidth is reduced and oversampling is used, the effective resolution will also go up. So they already get more than 8 Bit resolution for the lower speed settings.

[David Hess]

At least in the case of cheaper scopes, consider also the output resolution (the display). Like many of us here, I've got a Rigol 1054Z, whose display is 800*480px. By the time you subtract all the user interface, you're left with a precisely 600*400px viewport for data. A naked eye looking at a live signal isn't gonna see the difference between 256 levels scaled up to 400px, compared to 1024 levels scaled down to 400px if it were 10-bit, for example. You'd be adding a great deal of cost for almost no noticeable improvement in performance.

I mean, tell me if I'm wrong, folks, but my understanding is that the primary purpose of the scope is to see change over time (i.e. the shape of waveforms), and that for precise measurements, you use a separate meter for the value you're measuring.

There is some truth to this but the Rigol is a particularly bad example and misleading because it works directly with its 8 bit display record.  On a better DSO, the 8 bit acquisition record is stored using 16 bits for calculations and measurements which allows taking advantage of averaging and high resolution mode.  On the Rigol, these modes only reduce noise instead of also increasing resolution.

Most DSOs allow for a zoom in to a part of the curve - so the screen resolution is not such a hard limit. At least today they would offer higher resolution screens if the rest would give a better resolution.

In the past many or most DSOs had much higher screen resolution than digitizer resolution.  An extreme example of this is the old Tektronix 2232 and 2440 series which have 10 bits of vertical display resolution but only 8 bit digitizers.  They are calibrated to use 100 vertical display points per division (1) so over 8 vertical divisions, that is 800 points; the extra 200 points (actually 224 points) are used to overscan by 1 division (1.12 divisions) above the top and bottom of the graticule which can be seen by looking closely at the masking at the top and bottom of the CRT.  The extra range makes sure that clipping is beyond the screen boundaries.

So 255 digitizer counts over 800 vertical display points?  No, it is worse than that.  The 8 bit acquisition record is displayed across the whole 10 bit display record, which also prevents aliasing, so under normal circumstances, each digitizer count is 4 display counts or 25 digitizer points per division.  So why bother with a 10 bit display?  In averaging mode and when interpolation is used, the extra resolution is displayed and wow does it look smooth.  The readout is very nice looking also.  The extra resolution of the display makes up somewhat for the lack of index grading.

I think the main point is the SNR: The typical 1:10 passive probes give quite some noise background. Except for high signal amplitudes, there is just no need for an ADC with less noise than 8 Bits at 1 GHz. If the Bandwidth is reduced and oversampling is used, the effective resolution will also go up. So they already get more than 8 Bit resolution for the lower speed settings.

The SNR loss with a 10x passive probe is insignificant compared to the noise picked up at the high frequencies these probes support.  If the ground lead is not used and a coaxial connection or ground spring attachment is made, then the noise level approaches that of a 1x passive probe once the bandwidth is taken into account.

(1) Tiny 1cm divisions so 254dpi.  Bite me Apple.

[JanJansen]

I,m waiting with buying a scope, i want at least 16 bit.

there are 16 or even 24 bit scopes available.
Mine has up to 20 Bit (with oversampling).

Under 1000 euro not available currently.
Any suggestions ?

[Andreas]

Hello,

really?

https://www.reichelt.de/USB-Oszilloskope/PS-5242A/3/index.html?ACTION=3&GROUPID=5905&ARTICLE=146209&OFFSET=16&SID=28OFEwzawQARwAADdDEx42d8c6b274f003fee3df82c8cfbb3a86e&LANGUAGE=EN

below 1000 EUR if you can live with the bandwidth/memory.
I would go at least for the next higher model.

With best regards

Andreas

[Wollvieh]

One additional aspect may have been the accuracy of the input. Oscilloscopes aren't precision instruments, so with for example 1% error that is at a 256 resolution just the last one or two bits. Unlike blownup pictures from modern compact cameras with 20 megapixels behind a 1:5.6 crap lense which delivers maybe 4 megapixels of optical information.

The Philips PM 3320 had 10 bits of resolution. As far as is remember it has not a "real" adressable memory but an analog CMOS daisy chain like shift register per channel.

[KE5FX]

Short answer: people look at bits, but they buy MS/s.