Why are oscilloscopes so inaccurate?

[drummerdimitri]

I realized how inaccurate my scope was when doing some DC measurements of a battery compared to my 6.5 digit DMM so I thought the unit needed calibration until I read the datasheet which shows a 2% voltage accuracy which is worse than the cheapest DMM on the market today.

Why is this the case? Even Keysight's Infiniium UXR scopes which can cost up to 1M $ have the same accuracy as my cheap Rigol.

[capt bullshot]

Because one uses a cheap DMM to measure a battery.

[Rerouter]

I would say that in general they are more concerned with AC performance, however if your DC is so far off, run a self cal on the scope, in most cases this should put you well under 1% for DC offset and for most should also reduce any gain errors, however I will warn that some scopes seem to only calibrate there AC coupled mode or there DC coupled mode, and do not store offsets for both (there are small offsets between these 2 on most front ends)

The other one is , in most scopes your only dealing with an 8 bit ADC, now you can use the channel offset in a lot of cases to squeeze out extra precision, but in the end most scope measurements are relative, with repeatability being more valuable than absolute accuracy.

If you want a faster, high precision ADC, that gets into purpose built test gear like keithley DAQ's, as to why the 1 million dollor scopes DC specs are poor, they are not intended to be used in that application, they needed the best bandwidth / sample rate and AC performance, how well it measured a slow DC signal was very much an afterthought.

[tggzzz]

I realized how inaccurate my scope was when doing some DC measurements of a battery compared to my 6.5 digit DMM so I thought the unit needed calibration until I read the datasheet which shows a 2% voltage accuracy which is worse than the cheapest DMM on the market today.

Why is this the case?

You might as well ask why an oxy-acetylene torch is hotter than a soldering iron. Different tools for different jobs, each with different necessary strengths and allowable weaknesses.

Choose the right tool for the job, work around the limitations of available tools.

[mikerj]

Why is my DMM useless for showing high frequency time domain behaviour? 

[Circlotron]

in most scopes your only dealing with an 8 bit ADC, now you can use the channel offset in a lot of cases to squeeze out extra precision,

So does that mean that if you shift the trace to the bottom of the screen and only use the upward portion of the trace, you are only getting 7-bit resolution? 

[wraper]

Because even your 6.5 digit DMM is no longer 6.5 digit if you set faster sampling speed. There is trade-off between accuracy and speed. You cannot easily get both, especially at reasonable price. Probing also is an issue with high resolution, that's why multimeters set sampling speed as number of power line cycles (NPLC), to minimize power line AC interference impact on your measurements.

NPLC is number of power line cycles.  DC Voltage, DC Current, and Resistance measurement resolution, accuracy is reduced by power line induced AC noise.  Using NPLC of 1 or greater increases AC noise integration time, and increases measurement resolution and accuracy, however the trade-off is slower measurement rates.  For highest measurement accuracy NPLC of 100 is recommended.

[soldar]

How on earth do you measure a voltage on a scope screen with better than 1% accuracy? Does it involve a microscope?

A 'scope is not meant for this kind of measurement, it is meant to visualize complex waves. Why would you use it to measure a DC voltage with any accuracy?

A lawnmower and a nail clipper are both cutting tools but I would not use a lawnmower to trim my fingernails nor would I attempt to cut the grass using a nail clipper. Different tools for different purposes.

[Rerouter]

technically you would be measuring that voltage with the statistics menu built in to most DSO's,

[Siwastaja]

Optimization of bandwidth * voltage accuracy * price, for specific purposes they are supposed to serve.

The most generic catch-all instrument would have massive BW, massive number of channels, with massively good voltaga gain and offset accuracy - but it would also be massively big, consume massive amount of power, and cost a massive sum of money.

Out of these features, DC accuracy is easiest to sacrifice, since it's the least needed in typical use cases. Multimeters are great and abundant.

The ADCs on scopes are almost always 8-bit, and the DC offset stability and gain accuracy specs of the analog frontend are not the #1 or #2 items in the optimization, even for mid-range (think $10k-ish) products.

Note that product pricing is not a linear equation. If a scope BOM cost is $1000, you can probably sell the scope for $3000, but if the BOM cost is $2000, you can't sell it for $6000, because now at $6000 you would have much fewer than half the number of customers, so less total money in, and the NRE costs need to be calculated again per projected unit sales. Sometimes this ends up to totally ridiculous sums of money, and sometimes the result is that the market doesn't want the product.

[Siwastaja]

How on earth do you measure a voltage on a scope screen with better than 1% accuracy? Does it involve a microscope?

It involves reading out the number from the screen. You know these digital storage oscillosscopes that really penetrated the market in the 1990's.

[bd139]

Scopes are shit at everything they do. They just do a lot of shit things at the same time which turns out is a lot more useful than things that do one thing really well.

[magic]

The analog frontend could probably be fitted with a DC servo, frontends of cheap DSOs already use separate paths for high and low frequencies.

Attenuators could be measured and calibrated digitally during self-calibration.

Thermal stability of variable gain amplifier ICs and ADCs remains a question.

Some scopes are so bad that they could use improvements even with zero hardware cost. My Hantek doesn't calibrate gain errors in various ranges in self-cal, they just don't match each other no matter what.

[ogden]

A lawnmower and a nail clipper are both cutting tools but I would not use a lawnmower to trim my fingernails

And on that bombshell we end this thread 

[2N3055]

They are not. I have a scope that has 0,25% guaranteed accuracy. It is 16 bit.

General purpose scopes are 8 bit. Their resolution is roughly 0,4%.  That is smallest voltage step they are capable of. There is also DNL and INL in A/D converter, meaning, each step is not exactly 0,4%...

To keep specs under control, best they can promise is 1 -1,5% (some manufacturers do), with everything being ideal. Which is not.
Also it is really hard to make front end that has bandwidth from DC to 1GHz with max 1-2% total amplitude error.


6,5 Digit DMMs have approximately 23-24 bit resolution. And rest of meter specialized and optimized for just that.

[Gyro]

in most scopes your only dealing with an 8 bit ADC, now you can use the channel offset in a lot of cases to squeeze out extra precision,

So does that mean that if you shift the trace to the bottom of the screen and only use the upward portion of the trace, you are only getting 7-bit resolution? 

No, it's actually much worse than that...

1. Unless your signal is just the right amplitude to fill the entire input range of the ADC (ie, just before clipping) then it won't be giving you the full 8 bits anyway.

and

2. Most DSO LCD screens don't have sufficient vertical pixels to display the full 256 levels coming from the ADC, especially when you have multiple channels displayed plus the general crap of annunciators, measurements etc. clogging up the margins. Check your LCD resolution and see how many bits of resolution  it's actually displaying on a typical waveform - you will probably be surprised.

EG. The Rigol 1054Z has 480 vertical screen pixels (480x800). Now split that between 4 traces and the surrounding crud and you've probably got less than 100 pixels per channel, ie. under 7 bits.

[wraper]

2. Most DSO LCD screens don't have sufficient vertical pixels to display the full 256 levels coming from the ADC, especially when you have multiple channels displayed plus the general crap of annunciators, measurements etc. clogging up the margins. Check your LCD resolution and see how many bits of resolution  it's actually displaying on a typical waveform - you will probably be surprised.

EG. The Rigol 1054Z has 480 vertical screen pixels (480x800). Now split that between 4 traces and the surrounding crud and you've probably got less than 100 pixels per channel, ie. under 7 bits.

Not that you can accurately see actual voltage from the trace even if LCD resolution is high. But all modern scopes have measurement functions so you can see precise number in digits.

[Gyro]

Not that you can accurately see actual voltage from the trace even if LCD resolution is high. But all modern scopes have measurement functions so you can see precise number in digits.

True, though you do have to remember the aforementioned 8 bit ADC limitation in your definition of precise.

[drummerdimitri]

Lots of good info on here thanks for the clarification. 

[SiliconWizard]

Why is my DMM useless for showing high frequency time domain behaviour? 

Hehe, indeed.

The question was worth asking though - that's a very common "trap" for many beginners and even fresh engineers... I think it's already a good sign that the OP actually *noticed* the inaccuracy and asked the question. Many wouldn't even have cared.

[David Hess]

The design constraints needed to make wideband amplifiers with a response down to DC are not compatible with DC precision.  This is also why you cannot use an oscilloscope to directly measure settling time to high precision; instead, a rate of change or null measurement must be made.

Further with an 8 bit instrument, once you throw away 1 bit for linearity, that is only 1 part in 128 so about 2% when 1 bit is sign.

Higher resolution oscilloscopes and digitizers exist but for high accuracy, they have to compromise in other areas.

Sampling oscilloscopes avoid these problems by omitting signal conditioning before sampling so they can achieve much higher resolution at high frequencies.

How on earth do you measure a voltage on a scope screen with better than 1% accuracy? Does it involve a microscope?

You use an oscilloscope that includes a differential comparator.

The analog frontend could probably be fitted with a DC servo, frontends of cheap DSOs already use separate paths for high and low frequencies.

The problem with a DC servo is that in the event of overload, it ruins the overload recovery time.  This is also a problem with two path designs unless special care is taken.  There is a lot to be said for the performance and simplicity of the old stacked JFET input buffer.

Thermal stability of variable gain amplifier ICs and ADCs remains a question.

Thermal balance is a big problem and especially so in integrated circuits.  Even worse, the methods used to combat it in integrated circuits cannot be used at high frequencies because of parasitic coupling.  Essentially the offset varies with operating point and time.  Very high performance instruments compensate for this in the digital domain.

A more obvious and less complex way to control FET mismatch-induced offset would utilize a matched dual monolithic FET. Readers are invited to speculate on why this approach has unacceptable high frequency error. - Jim Williams, Linear Technology application note 65, page 72.

[Alex Eisenhut]

You use an oscilloscope that includes a differential comparator.

Sounds like a use case for a vintage analog scope...

http://w140.com/tekwiki/wiki/1A5

[DaJMasta]

It's been mentioned, but different instruments for different applications.  It's a limitation of the practical design of the frontend and sampling system (low resolution, response down to DC and up being hard to keep linear, etc.), but it's also not something that's needed for the kinds of things scopes are usually known for.

The scope's strength in measurement is time - its triggering and sample rate have to be very low jitter and tightly controlled to give you that nice smooth sinewave on the screen, but in essence, it's not intended to be a measurement tool - a scope is for visualization.  A DMM, a power meter, a frequency counter... all of these are explicitly for measuring whereas the primary function of an oscilloscope is to show detail on a signal as a visualization tool, the measurement of a signal's parameters is sort of secondary addition (take a look at all-analog scopes, measurement was never a big priority until digitization started making it easier).

[coppercone2]

look at bandwidth. Measurement usually gets ugly when you widen bandwidth.

You have something that is widely considered perfectly linear (visible "DC", that is where you can track change with your eyeballs) and AC, in which every parameter follows 1/F tendancies. It's like looking at everything through a stroboscope (trying to visualize a waveform of 1MHz on a scope).

[SiliconWizard]

Scopes are shit at everything they do. They just do a lot of shit things at the same time which turns out is a lot more useful than things that do one thing really well.

Same can be said about most people.