How much accuracy do YOU really need?

[TomKatt]

I know that there are many folks here with technical skills WAY above mine, so just curious about how accurate you need your test gear to be...

It struck me that 99% of the time I just need 'ballpark' readings to confirm if something is operating within a known set of parameters.  My projects vary from small ucontroller (AVR/PI) digital circuits to audio / amp repair and finally computers, household appliances and automotive work.  Few instances require great accuracy.

It seems that repeatability is more important to many things I measure - is the bias on the left channel the same as the right (no matter the actual voltage measurement)?  Or adjust something for minimum or maximum deflection.  It's amazing how tolerant so many circuits appear to be - audio tube amps run on double digit tolerance parts and it never ceases to amaze me when Shango066 gets an old tube tv running that's been sitting in dried mud for 20 years simply by replacing a few caps and a burned resistor.

I am cognizant of the spec'd accuracy when I buy gear, but most of the time it's not in my top 5 criteria...  Most modern 'decent' gear seems "good enough".  There are obviously times when a few millivolts or ma will make or break something operating correctly, but I'm not sure how many times I based that on my readings rather than tweaking things and observing the results.

SO - How do YOU folks rate gear accuracy in your work?

edit -autocorrect corrections ;-)

[BradC]

I'm often looking at repeatable discrepancies, so it's not accuracy so much as precision I need.

[tom66]

For day-to-day usage of a multimeter?  0.1-0.2% accuracy is probably OK.  That eliminates the multimeter as a likely source of error in most systems I work on.  However as mentioned, I would agree that repeatability is far more important, so if that 0.2% error is consistent across all ranges (for e.g. voltage, it is a 'reference' error rather than a 'gain' error) is better than one where the individual ranges are scattered in accuracy.  It's also important it doesn't vary with battery state of charge (looking at you, "bottom of toolbox emergency multimeter") or too much over time and temperature.

[TomKatt]

For day-to-day usage of a multimeter?

I'm curious about all equipment.  I agree with your sentiments about the DMM's.  But even when I pull out the scope, I'm not sure how many times I relied on it's accuracy either...  Certainly I got by for many years using my ancient HP180A cro.

While I don't do much with RF, that's one field where I suspect true accuracy might be an important spec (then again - the Shango066 tv example.  Love the way he tests caps by playing phone audio through them and listening for the filtering)

[Bud]

You picked a poor example, the guy even never bothers to blow the 50 year dust off of the poor TVs before poking his crude tools in them.

[tom66]

Even a modern high performance scope probably has around 0.1 dB of gain error, which is very roughly around +/-2% voltage accuracy.  I have found, at least on cheaper scopes, the attenuation ranges are far less consistent than a multimeter.  For instance my Rigol DS1074Z jumps up about 2-3% in voltage measurement when the attenuator relay clicks in around 200mV/div.  Also you have to be aware most scopes are only 8-bit and so quantisation error can dominate any other error, especially on scopes where the lowest range (1mV/div or so) is usually a doubling of the last range rather than a true analog gain.

Since an oscilloscope is usually a tool to measure time-voltage signals the voltage accuracy is less critical than the time accuracy.  Time accuracy on an oscilloscope is sacrosanct, you cannot measure things like baud rate accurately if it is out by more than a few percent. Thankfully nearly every scope at least uses a basic TCXO for the master clock which obtains around +/-10~20ppm accuracy.  If you are measuring more precisely than that, I guess you can use the 10MHz ref input on many scopes. 

[TomKatt]

You picked a poor example, the guy even never bothers to blow the 50 year dust off of the poor TVs before poking his crude tools in them.

Come on - that's half the entertainment value!  And really goes to show how tolerant those circuits are and how little accuracy sometimes is needed.

At least he has a (beat up missing segments) Fluke 27/FM 

[mendip_discovery]

For the most part it's not needed as the conditions and spec of the parts you are measuring are so big for it not to be an issue. But it is a rabbit hole and you find yourself looking for better kit just because you are trying to figure out the small issues.

Also we aware that if you try to measure a 1% resistor with a 0.5% spec meter. Are you able to say that the resistor is in spec or not?

I have wanted to check my vehicle battery for a parasitic drain. I tried a clamp meter with banana leads to see if that would give me an idea of where those Ah are running off too. But the readings vary so much that I can't use any of the data. Even trying it to see how the battery is charging isnt much use. So I am now looking at other methods to get reliable and repeatable results.

[tggzzz]

I know that there are many folks here with technical skills WAY above mine, so just curious about how accurate you need your test gear to be...

It struck me that 99% of the time I just need 'ballpark' readings to confirm if something is operating within a known set of parameters.  My projects vary from small ucontroller (AVR/PI) digital circuits to audio / amp repair and finally computers, household appliances and automotive work.  Few instances require great accuracy.

It seems that repeatability is more important to many things I measure ...

I am cognizant of the spec'd accuracy when I buy gear, but most of the time it's not in my top 5 criteria...  Most modern 'decent' gear seems "good enough".  There are obviously times when a few millivolts or ma will make or break something operating correctly, but I'm not sure how many times I based that on my readings rather than tweaking things and observing the results.

Good to see someone thinking about what they are doing Too often that fails to happen A good engineer will always understand the minimum required, and how to achieve that with imperfect components and test equipment.

I'll expand "repeatability" to "repeatability over time".

I once worked for a company that made an instrument that measured coaxial cable impedance. The accuracy was 0.1dB, but the customers didn't care about that since they had other equipment that was more accurate. They were interested in its resolution and stability over 1 week, which was 0.001dB. Why? They were interested how the cable's impedance changed with temperature, so they stuck the sample in an oven. A big oven, since the submarine telecoms cable had a rather large minimum bend radius. It took a week to raise/lower the temperature.

I'm sure you are aware of

[shapirus]

I'll expand "repeatability" to "repeatability over time".

Plus resolution, to observe slowly changing values.

[TomKatt]

I'll expand "repeatability" to "repeatability over time".

Plus resolution, to observe slowly changing values.

One reason I keep a few old analog lab meters kicking around.  The lcd bargraphs have their place, but are no substitute for a moving coil.

These used to be dirt cheap on eBay when schools started upgrading to 'modern' equipment.

They're probably not 'accurate', but they are 'repeatable over time' 

[madires]

accuracy <-> precision <-> resolution

... and what you really need depends on the work you're doing. For tinkering with Arduinos any inexpensive DMM might be sufficient, while for calibrating an 8.5 digit bench DMM in a metrology lab you would need something more suitable.

[TimFox]

Even a modern high performance scope probably has around a dB of gain error, which is very roughly around +/-2% voltage accuracy.  I have found, at least on cheaper scopes, the attenuation ranges are far less consistent than a multimeter.  For instance my Rigol DS1074Z jumps up about 2-3% in voltage measurement when the attenuator relay clicks in around 200mV/div.  Also you have to be aware most scopes are only 8-bit and so quantisation error can dominate any other error, especially on scopes where the lowest range (1mV/div or so) is usually a doubling of the last range rather than a true analog gain.

Since an oscilloscope is usually a tool to measure time-voltage signals the voltage accuracy is less critical than the time accuracy.  Time accuracy on an oscilloscope is sacrosanct, you cannot measure things like baud rate accurately if it is out by more than a few percent. Thankfully nearly every scope at least uses a basic TCXO for the master clock which obtains around +/-10~20ppm accuracy.  If you are measuring more precisely than that, I guess you can use the 10MHz ref input on many scopes.

A 1 dB gain error translates into approximately 12% voltage error:  20 x log₁₀(1.12) = 0.98 dB.

[tom66]

A 1 dB gain error translates into approximately 12% voltage error:  20 x log₁₀(1.12) = 0.98 dB.

D'oh.  I've probably slipped a decimal point a place somewhere in my memory, I didn't actually do the maths, but recall some Rigol datasheet that specified both.  Thanks for the correction

[Someone]

A 1 dB gain error translates into approximately 12% voltage error:  20 x log₁₀(1.12) = 0.98 dB.

D'oh.  I've probably slipped a decimal point a place somewhere in my memory, I didn't actually do the maths, but recall some Rigol datasheet that specified both.  Thanks for the correction

The DC accuracy is often in the 2% ballpark, with AC accuracy worse. So everyone is correct!

[artag]

There's somebody here who's signature consists of a series of crossed-out remarks of the form

4.5 digits 6.5 digits 7.5 digits are enough for anyone.

Always gives me a tickle. Sometimes the precision isn't important but achieving more is fun in itself.

[pqass]

I confess that I have a touch of gear acquisition syndrome; 3 bench meters in 3 years (so far). Although, my limit is 6.5 digits since 7.5 digits is right out of my $$$ range.

But I didn't really appreciate the low end (microvolts, milliohms) until I was able to do a 4-wire measurement.  It enabled me to precisely cut NiChrome shunts for my e-loads.  I couldn't do this with my hand-held [4000 count] meters.  Then it dawned on me that there is an important world of applications between 1 ohm and 2 milliohm (8" of 14AWG).  Also, having microvolt measurement capability allowed me to monitor thermocouples directly!

My latest meter, an HP3456A can do 6.5 digits but the features, moreso, just blew me away.  It doesn't directly offer current measurement but that doesn't matter because you can provide your own shunt, measure its resistance, then enter that into the "Y" register. The meter can then calculate (X-Z)/Y where X=the measured voltage drop and Z=0!  And there are many other great features like: offset compensation, choosing PLC, stats, error calc. (binning), temperature (via NTC),  etc.

[mumoto]

I worked as a metrologist for nearly two decades.

With my current work,  all I can say is "not a lot" in the general electrical sphere.

I understand a good deal of the science of measurement, but sometimes (ha! I'm no long working in the field) good enough is good enough 

I could start a long scientific and philosophical explanation, but it's horses for courses.   

[exe]

10k counts is enough for everything I do, but for some reason I have 6.5digit dmm...

[vk6zgo]

For day-to-day usage of a multimeter?

I'm curious about all equipment.  I agree with your sentiments about the DMM's.  But even when I pull out the scope, I'm not sure how many times I relied on it's accuracy either...  Certainly I got by for many years using my ancient HP180A cro.

While I don't do much with RF, that's one field where I suspect true accuracy might be an important spec (then again - the Shango066 tv example.  Love the way he tests caps by playing phone audio through them and listening for the filtering)

Nah!---------RF is normally specified in dB.

You'd usually be happy if you could read better than 0.5 dB on a Spectrum Analyser, & that is just under 11%.
Really "scrunching up your eyes" you might be able to "guess" 0.25dB, which is a tad under 6%.

This can be "a trap for young players".

On one occasion, we had a test where we had to input +16dBm of audio into the (600 ohm) L & R inputs of a TV stereo encoder.
"Easy peasy" we thought, the audio signal generator has a low Z (30 ohm) output, so we can just "parallel 'em up".
The error would only be a fraction of a dB, so "it's cool for cats".

Sadly, that was not the case, as the Tx spec was 50% deviation, which showed up with this setup as 47%!!

What looks like nothing with dB, soon mounts up to something quite a lot more concerning.

At another job, we had some UHF transmitters, which as is common, consisted of a bunch of individual smaller amplifiers paralleled up
via diplexers, complete with unbalanced loads.
Unfortunately, we found ourselves having multiple amplifier failures.

It turned out that the "rent an EEs" back in the dear old PRC had been fairly "laid back" both in the ratings of the unbalanced loads & the frequency response/ amplitude matching of the amplifiers.

"Meh!---- They're within a coupla dB" didn't quite hack it, the unbalance loads blew up, followed by the associated amplifiers, ending up in a mess.

In the end, I had to go through all the smaller amplifiers with an SA & tracking generator, getting their response at the operating frequency within 0.5 dB to avoid killing the unbalance loads.
I was unimpressed at having to push chip capacitors around on PCBs to get exactly the right response, something I hadn't done for around 30 years!

[David Hess]

The old solution for more accuracy and precision than your instrument can return is to make a null or slideback measurement using a precision accurate DC reference source.

For instance the Tektronix 7A13 differential comparator allows an analog oscilloscope to make voltage measurements to 4 digits of resolution and 0.1% accuracy.

[ebastler]

accuracy <-> precision <-> resolution

Umm... what do you mean by the "<->" symbol?

[Gregg]

Slide rules are another example of how much accuracy the old timers really needed.

[tggzzz]

Slide rules are another example of how much accuracy the old timers really needed.

12/24m long scales in some cases, giving 4 to 5 digits, e.g. https://en.wikipedia.org/wiki/Fuller_calculator

[David Hess]

It really depends on the circuit and application.  With precision linear circuits, it is possible for an error of greater than 1 count out of 100,000, or even 1 million, to indicate an error which requires fixing.  These are the same kinds of circuits where it may be mandatory to have a voltmeter with gigaohm+ input resistance for useful measurements.

More than once I have diagnosed the existence of a problem in a precision analog-to-digital converter because the least significant digit flickered more than 1 count when it should not have.  I have also done the same a couple times now with digital counters.