Delta-sigma noise shaping

You can make a decent d-s DAC inside an FPGA, and just RC lowpass an output pin. That works a lot better than PWM in a lot of situations. Good for stuff like tweaking DC offsets or trimming VCXOs.

John

Yes, in that sort of application it makes a lot of sense to do that. In a pulse height spectrometer application the delta modulation method isn't really an option.

I have looked pretty hard a couple of times. It is really hard to get a clear let alone quality explanation. Even the old patents leave much to be desired. Just the same:

1. "nyquist rate" vs actual sample rate appears to be 1/ bits. Telco use 56kbps F(n) ~8kHz useful BW under 4kHz (at about 8 bits). I suspect worst case is more like 1/(bits*log2(bits)).
2. I have not found any explanation clear enough to evaluate this. Plus the ease of including dynamic compression (like a-law or u-law) makes this a much harder question. The "feedback" is very non-traditional.
3. The noise shaping is one of the easier claims to verify. And if you do not mind a bit of transcoding works for both ADC and DAC.
4. I cannot do enough math to evaluate this one. But the limitation would one of properties of the technique rather than one of the capability of the comparators.

They have been used lots of places, including for video. I expect the main reason that they are not seen in instruments is that far too few can follow the math to prove that the result can be as good as a SAR.

No, i expect that it only has to have low jitter (for about one conversion time) not absolute edge accuracy.

More like consistency, rather than absolute amount. A rather different property.

I dl'd the datasheet for "closer" study. But your summary makes me wonder if it is worthwhile.

Could be any or all of: don't know how, don't have any knowledge of noise, never heard of such a thing, but the profit margin is too low, my boss told me to do it this way, isn't that part obsolete, and all of the other excuses we have frequently heard.

You wouldn't happen to have a schematic for one, would you? How about a full transistor level one from old Telco days (~1960)?

I suggest "translated out of band" as a better description.

There were lots of dual slope converters doing 7+ digits in the 1970's and early 1980's. That is about 22+ bits.

Really? I thought 4-1/2 was as high as it went. (Of course I was still a babe in arms then. :)

Cheers,

Phil Hobbs

Well, try finding an analog switch whose charge injection is stable to attocolombs over temperature.

John

Dielectric absorption is the killer.

John

I vaguely remember _glass_ capacitors ;-)

As I recall 4-1/2 digits was actually a piece-a-cake, early '70's.

...Jim Thompson

I read some. It is a little bit better than average. It is becoming clear, the only way to understand is to make one (or more) and plan with it some (them some).

Glass is mediocre for DA, as I recall. Polystyrene is pretty good. I suppose one could make a decent-capacitance air dielectric cap, with some cleverness.

I actually invented the dual-slope ADC when I was a kid, before Fairchild patented it. But I assumed I'd have to use relays, and that the timing/bounce would mess up the accuracy, so I never built one. Pity.

John

For the same reasons i would expect that the charge injection would only need to be stable per a few consecutive conversions. A differential rather than an absolute requirement.

Oops, play with them.

You make it sound so easy.

John

I'll find out by fiddling with one or more home built ones.