DS2072A Vmax

[aroby]

I just bought one of the $20 voltage references from eBay.  I checked out my DMMs and they are pretty close to what the voltage reference puts out.  But if I connect the reference to my DS2072A and look at Vmax, it's about 0.1v out on all of the reference voltages.

Now, I'm new to oscilloscopes, but my understanding is that they are basically voltmeters with a time dimension, so I assume that this difference should not to be expected?

Anthony

[Mandelbrot]

To my knowledge, oscilloscopes measure voltage digitally using an 8-bit ADC. 8-bits only leaves room for 256 discrete voltage measurements, so there can be a large room for error. If an 8-bit ADC is measuring voltage from zero to five volts, the most accurate it can be is (5-0)/2^8 = 0.01953125V or about 20mV.

Also, it is important to keep in mind that wires can act as antennas and all sorts of interesting electromagnetic field magic happens that can also alter what your scope reads by a few millivolts.

Hopefully someone here will give a better explanation than me...

EDIT: I forgot that the 256 levels are spread across all vertical divisions, not just 1 division, so the error is even larger than I stated above.

[Rigby]

how are you hooking it up?

[Wim13]

it is even worse, on the scale of 10 Volts/dev on the DSO,

Full scale is then 80 Volts, over 8 divs, at 8 bits ( 256 steps) where 200 are visible...

That gives 80/200 as smallest change, thats is 0,4 Volts resolution... plus the jitter of the ADC..

Ofcourse on 1 Volt per div, it is 8/200 is 40 mV max resolution, plus jitter

[aroby]

how are you hooking it up?

Just with one of the probes that came with the scope.

Anthony

[aroby]

To my knowledge, oscilloscopes measure voltage digitally using an 8-bit ADC. 8-bits only leaves room for 256 discrete voltage measurements, so there can be a large room for error. If an 8-bit ADC is measuring voltage from zero to five volts, the most accurate it can be is (5-0)/2^8 = 0.01953125V or about 20mV.

I guess that makes sense, but I thought the whole point of an oscilloscope was to have a more precise measurement instrument.  Or is it that scopes are more used because they show the shape of what is happening over time, and that the accuracy of measurement of voltage in the shape is less important?

Anthony

[tautech]

An oscilloscopes primary function is to display a waveform in the time domain.
All other uses/accuracy are as per manufacturers spec.
Voltage measurements are indicative only not absolute.

[Hideki]

High resolution timing but sloppy amplitude measurement is common. It's hard to be precise in both axes at the same time, so the tradeoff for modern scopes is that the sample rate has gone up-up and up (Rigol is now at 2 gigasamples/s max) while the amplitude resolution is stuck at 8 bits. It's just enough to show a reasonable waveform shape on the screen, but not enough to make precise voltage measurements.

Some scopes use more than 8 bits, but they tend to be more specialized tools (like some of the picoscopes). The sample rate typically drops when you want more bits of resolution.

[aroby]

High resolution timing but sloppy amplitude measurement is common. It's hard to be precise in both axes at the same time, so the tradeoff for modern scopes is that the sample rate has gone up-up and up (Rigol is now at 2 gigasamples/s max) while the amplitude resolution is stuck at 8 bits. It's just enough to show a reasonable waveform shape on the screen, but not enough to make precise voltage measurements.

Thanks for the insights. This is my first oscilloscope, so I'm not sure what exactly to expect in terms of accuracy.

[alexwhittemore]

The answer that "it just sucks in the voltage axis" isn't quite accurate enough for me, so I decided to explore exactly what's going on.

A scope's vertical accuracy is pretty opaque, since generally you're not looking for super-precise vertical measurements - that much is definitely true.

For example, let's take a reference DC measurement and analyze the uncertainties. I've set my bench supply to 2V, measured by my let's-assume-accurate 4.5 digit meter to be 1.9997V

(See first screenshot, which I guess I can't embed)

In order to get as accurate a measurement as possible, I zoom in on the scope pretty far - the 10mV/div range. I could have gone farther in, but scrolling the position knob endlessly is a huge pain, since it scrolls about 1div/turn or so, which is to say 10mV/turn or 200 turns to set the full 2V offset. Anyway. Full scale, I've got 4 positive and 4 negative vertical divisions, so 8. Assuming the ADC limits at those edges for maximum resolution displayed, that's 8divs*10mV = 80mV, and 80mV/256adc steps = .31mV/ADC step. [aside, I think it actually doesn't, because the trigger can happen at ±5 divisions from center, and in fact, the voltage per division setting isn't even restricted to these coarse values]

Of course, because I'm measuring a full scale of 80mV range, I have to offset by a lot to see this signal at 2V.

Pause for a sec. Here's the datasheet for my Rigol DS1104z:
http://beyondmeasure.rigoltech.com/acton/attachment/1579/f-0317/1/-/-/-/-/DS1000Z%20Data%20Sheet.pdf
Page 5 has the relevant information.

On the 10mV range (in fact, all ranges 500mV or lower) the maximum offset is ±2V, which means I can offset the signal down 2V to put it on the centerline of the scope. Here's the first bit of interesting inaccuracy from the datasheet: DC Offset Accuracy is ±0.1 div, ±2mV and ±1% of offset setting. For my settings, that's ±1mv ±2mV ±20mV, for ±23mV total. So we can already ignore ADC graduation, as it's 2 orders lower in amplitude down here.

Now, we know we're going to measure our signal ±23mV, and assuming my meter is spot on, what we actually measure is 2.004V on a 1.9997v signal, or 4.3mV out. That's actually pretty good!

And we didn't even account for DC Gain Accuracy of ±4% full scale lower than 10mV or 3% full scale >10mV. Not that we had to, because we tried to center the ADC on our signal exactly via offset.


Let's take a different measurement, something with no offset. I've set the PSU to 1v, measured at .9962 actual. I can easily see this on-screen with the vertical scale at 500mV/div

(SCREENSHOT 2)

Now we're measuring 1.030V on the scope, both per the cursor and avg measurement. That's a bit worse than before, 33.8mV error. Okay, let's shoot this one. You'd think the offset error would be lower, given that we've got offset set to zero volts (surely, it just shorts the offset input to ground, right?! Maybe not!). But assuming it comes int play, and we do have to assume that, we've got ±0.1*500mV/div ±2mV ±1% of zero = 52mV. Well, that right there is larger than our mesasured error, so assuming the offset circuit isn't magically perfect at zero offset (which it may or may not be), we're within spec!

But now we've also got to account for DC gain accuracy, because we're not centered on the zero of the ADC! Our signal is above 10mV, so we're at 3% of full scale error. Now, I don't know what full scale for the ADC is right now, but it could well be ±5 divisions = 5V. 3% of 5V is 150mV, so if that's right, we're WAY better than worst case! Let's be conservative and assume the datasheet meant 3% of our INPUT signal. 3% of .9962V is about 30mV. Ignoring offset error, that's still pretty darn close to our actual measurement error.

Now, OP said the measured DS2072A was out by 100mV, which is significantly higher than what I've measured my scope to be off by. But again, depending on how the measurement was set up, 100mV of offset error isn't crazy if volts/div is relatively high, nor is it insane for gain error.

All of these parameters (except the DC Offset accuracy, which doesn't depend on input signal) will improve with frequency too, though I don't find specs for them in the datasheet. Making components accurate at both DC and frequency is hard, considering DC is infinitely low frequency. As an example, you'll never find a traditional sweep spectrum analyzer that goes all the way down to DC. Keeping things linear across an infinite number of decades (.1Hz, .01Hz, .001Hz...) is basically impossible, and ultra low frequencies aren't generally useful to pay attention to anyway! If you were going to be measuring those, you'd use a multimeter

Conclusion: Despite being, on an absolute scale, terrible resolution, 8 bits is actually high enough that ADC step size isn't really the problem here. However, because scopes aren't really designed for operation at DC, gain error at DC is QUITE high, and if you're using it (maybe even if you aren't? I'd love to know in detail how the input offset circuit works exactly), offset error is pretty high too.

[David Hess]

Attenuating probes have their own accuracy specification which can degrade the system gain accuracy further.

The digitizer is 8 bits but its input range is *larger* than the display to accommodate some overload and position control range.  I do not know about Rigol oscilloscopes specifically but 25 counts per division is typical allowing an extra division above and below the screen.  Cheap DSOs may use "digital gain" in place of some input attenuator settings which will lower the resolution further.

Temperature changes may alter the offset and gain as well so some oscilloscopes include some form of signal path compensation or calibration which should be executed before making measurements that need to be the most accurate.

All of that adds up to an accuracy and resolution at least 10 times worse than a good 3.5 digit voltmeter.  Since 3.5 digits is 2000 counts, ideal accuracy will be about 0.1% which oddly enough is the specification for my bench 3.5 digit multimeters.  An oscilloscope making a measurement with 200 counts should be 1% or worse and there are plenty of 3.5 digit multimeters only good to 0.5% making 5% total error almost reasonable for an instrument designed for high speed digitizing.