Question re: Effective Number of Bits vs. the resolution of an ADC

Status
Not open for further replies.

Constantin

Well-known member
So one of the ADCs I've been using is a MCP3911, which microchip calls an analog front end, which is specialized for power measurements. One of the interesting aspects of this device is that it advertises 24bit resolution with no missing codes yet, the the highest effective number of bits is about 16 per the data sheet (depends on how you use it). Per Wikipedia, "ENOB specifies the resolution of an ideal ADC circuit that would have the same resolution as the circuit under consideration". My interpretation is that if you manage everything right re: the PCB, layout, components, etc., that the best you will be able to do with this ADC is 16 bits.

The MCP3911 communicates via SPI and the Voltage and Current readings are transferred in 8 bit increments. Thus, the device offers 3x8bit transfers per power register to read out a 24-bit number or a truncated/rounded/etc/ 16-bit number via 2-8bit transfers. It seems weird that a manufacturer would promote downloading 24-bit data packets for a ADC that at best can achieve 16 ENOB.

Why would one even offer 24-bit readouts if the ADC can't be relied on to produce accurate data at that resolution? (at least if I understand the meaning of ENOB). For the follow-on device (MCP3910) they even offer 32-bit register readouts so you can get a signed 24-bit value. Seems like a fools errand to me, or am I missing something?
 
My experience is that at about 12 bits and more, the analog circuit design is very challenging in terms of noise, linearity, etc.
As to precision versus resolution - that's a common paradox, and much is written.
Some systems have a signal source such that statistical processing can be done - and especially so if the noise is Gaussian or other known a priori type.

Some use a sample and hold (analog) ahead of the A/D, to deal with non-recurrent signals.

It's a profession unto itself.
 
In most of these 24-bit systems, the lower bits are effectively just noise. No system has enough accuracy (e.g. the voltage reference) to measure accurately to 24 bits. However in some applications you don't need precise measurement, but need to be able to detect small changes --e.g. in an audio system, you don't really care if the volume is 1 % higher (accuracy), but you would (might !) care if the signal had 1 % noise (or distortion or errors). In those types of applications, additional resolution beyond the ENOB may be somewhat useful.

Conversely, using a digital voltmeter, you might care about absolute accuracy if you need to measure a voltage precisely -- you won't easily find a DVM better than 0.05 % accuracy -- this represents about 12 bits ( ~3 digits); yet a meter with 4 digits could be useful to indicate if a signal is rising or falling -- you can see small changes, but know that the reading itself if not precise.
 
My interpretation is that if you manage everything right re: the PCB, layout, components, etc., that the best you will be able to do with this ADC is 16 bits.

Yes, and sometimes even that is optimistic.

Always remember, datasheets are written by marketing departments. Their one and only purpose is to sell the part. It may not look like a sales pitch, but it most certainly is. They will often manipulate the specs or use creative ways to specify key parameters to make the product appear as desirable as possible. Shortcomings or limitations are often not mentioned at all, or glossed over, in hopes you'll already be too far down the path of using the part before you discover any issues.

Then again, nearly all the semiconductor manufacturers are working with roughly the same technology, so competitive parts of similar vintage and pricing will often have similar limitations, and similarly misleading datasheets.
 
Well in addition to the English majors hired to supply 'creative fiction' by the marketing department, you have the translation of techno-speak to English from the engineers, and these days, you have the translation of Japanese and lately Chinese techno-speak to English techno-speak to English datasheet-speak, so there are endless opportunities for subtle points being lost. And then there are the errata, which always reminds me of functionary that follows a politician around, and says to the media what the president, senator, congresscritter, mayor, alderman, etc. meant to say was ....
 
You guys are amazing - thanks! As a non-EE, it sometimes takes some double-checking the reason I see things in datasheets that baffle me. Thank you again.
 
Status
Not open for further replies.
Back
Top