Comparing ADC noise parameters

Question

I'm looking at two ADC parts the ADS1230 and the ADS1231; the 1231 is an almost pin-compatible upgrade (if you don't need the ×64 gain setting) which provides 24-bit resolution and some more noise filtering.

However looking at the noise figure section in the datasheet there's something suspicious:

These are the spec for the 20-bit part:

the newer 24 bit part only supports gain 128 so they merged the table but the criteria are the same:

Now, these converter are designed for slow signals (mostly load cells but of course they could be useful for force/pressure bridges to); the recommended data rate is in fact 10sps (80sps has less powerline noise rejection due to the sampling notches), so while ENOB is a good merit figure (from dynamic tests at full range) I'd guess that the noise free bits could be a better indication.

So, looking at the 10sps, G=128 with 5V excitation we have: 20-bit: ENOB=19.8 noise-free bits=17.5 24-bit: ENOB=20.1 noise-free bits=17.4

…the 20 bit converter seems actually slightly better performing than the 24 bit one. Looking at the noise column is probably due to the noise shape (the 20 bit has more RMS noise but the pkpk is lower). At 3V the situation is similar.

At 80sps the 20 bit converter is even better.

So why should someone pick the 24 bit part for a better resolution (which would be probably swamped by noise)?

The price difference is significant but in fact the 20 bit converter costs more since it's an older part and probably TI tries to discourage it. Or simply the dual gain amplifier costs more… whatever (IIRC the resolution came from the digital filter and digital logic probably is cheaper than analog circuitry).

Is it maybe in fact a cheaper part hidden behind a bigger resolution? (i.e. a marketing ploy)

Answer 1

The number of bits on the label of a delta-sigma converter is merely the number of bits in the digital counter and digital filter (not shown).

Since bits are cheap, I assume it costs basically nothing to go from 20 to 24 bits, just a tiny bit of logic. Since everyone is marketing 24 bit ADCs these days, why not? Even the audiophile DACs are 32 bit now.

It's not entirely marketing though. In the case of a DAC, one could argue that having 32 bits at the input of the modulator means less headache for the DSP programmer, who doesn't have to program a nice dithering algo to chop off the DSP outputs into 16 or 24 bits, which is nice.

On the ADC side, having a few bits of noise at the end make it easier to average values to get more precision if needed. It makes the noise more uniform, less quantized into steps, which is also nice.

Now, performance depends entirely on the analog circuits before the counter. So, I suspect a die shrink, to save on cost. That would explain the slightly higher noise and the simpler analog conditioning.