Noise and Resolution (App Note)

What is resolution?

Resolution in this context refers to the conversion of an analog voltage to a digital value in a computer (and vice versa). A computer is a digital machine and thus stores a number as a series of ones and zeroes. If you are storing a digital 2-bit number you can store 4 different values: 00, 01, 10, or 11. Now, say you have a device which converts an analog voltage between 0 and 10 volts into a 2-bit digital value for storage in a computer. This device will give digital values as follows:

So in this example, the 2-bit digital value can represent 4 different numbers, and the voltage input range of 0 to 10 volts is divided into 4 pieces giving a voltage resolution of 2.5 volts per bit. A 3-bit digital value can represent 8 (2^3) different numbers. A 12-bit digital value can represent 4096 (2^12) different numbers. A 16-bit digital value can represent 65536 (2^16) different numbers.
It might occur to you at this point that a digital input could be thought of as a 1-bit analog to digital converter. Low voltages give a 0 and high voltages give a 1.

In the case of the LabJack U12, a single-ended analog input has a voltage range of -10 volts to +10 volts (20 volt total span) and returns a 12-bit value. This gives a voltage resolution of 20/4096 or 0.00488 volts per bit (4.88 mV/bit).

What does it mean to say a device is 12-bit, 16-bit, or 24-bit?

When you see analog input DAQ devices from various manufacturers called 12-bit, 16-bit, or 24-bit, it generally just means they have an ADC (analog to digital converter) that returns that many bits.  When an ADC chip returns 16 bits, it is probably better than a 12-bit converter, but not always.  The simple fact that a converter returns 16-bits says little about the quality of those bits.

It is hard to simply state "the resolution" of a given device. What we like to do, is provide actual measured data that tells you the resolution of a device including typical inherent noise.

If you look at a device called "24-bit" just because it has a converter that returns 24-bits of data per sample, you will find that it typically provides 20 bits effective or 18 bits noise-free (like the UE9-Pro).  Devices such as the U6-Pro and T7-Pro provide some of the best performance around from a 24-bit ADC, and they provide around 22 bits effective resolution or 20 bits noise-free resolution. We sometimes mention that these devices have a 24-bit ADC (as that is what people look and search for), but we try not to call them "24-bit" in favor of providing a typical effective resolution figure.

Another interesting thing about your typical 24-bit sigma-delta converter is that you can look at them as only having a 1-bit ADC inside, but with timing and math they can produce 24-bit readings:
http://www.maxim-ic.com/appnotes.cfm/appnote_number/1870/

Hardware with a 24-bit ADC

T8: 21-bit Effective Res

U6-Pro: 22-bit Effective Res

T7-Pro: 22-bit Effective Res

UE9-Pro: 20-bit Effective Res

Hardware with a 16-bit ADC or less

U6: 19-bit Effective Res

T7: 19-bit Effective Res

UE9: 16-bit Effective Res

T4: 12-bit Effective Res

U3: 12-bit Effective Res

U12: 12-bit Effective Res

Inherent Noise Level of the LabJack

Analog to digital converters (ADCs) have an inherent noise level, and the support circuitry can add to that noise level.

From Appendix A of the U3 User's Guide, the typical peak-to-peak noise on an analog input is +/- 1 count, which is excellent.  If using a single-ended low-voltage channel, there are 4096 counts across a span of about 2.4 volts, so the voltage resolution is about 600 uV/count.  If using a high-voltage channel, there are 4096 counts
across a span of about 20 volts, so the voltage resolution is about 5 mV/count.

Appendix B of the U6 or UE9 User's Guide provide extensive tables for typical noise/resolution with those devices.

If you see more noise than you expect, start by looking a known stable voltage and look at the noise.  Usually, GND is a good way to go for this, except for a single-ened low-voltage U3 channel where 0 volts could be just below the lower rail and a 1.5 battery would be a better choice.

Why does my DMM reading look more stable?  Perhaps it is just showing fewer digits.  Sounds silly, but sometimes that is the explanation.  Usually, though, it has to do with the fact that the DMM is giving you the average or RMS over some time period (perhaps 0.5 seconds), whereas the LabJack is giving you data points acquired over some number of microseconds.  If you use the LabJack to acquire lots of points over the same time period you can mimic the DMM behavior.