[time-nuts] A/D specs (was EFC tracking)
magnus at rubidium.dyndns.org
Mon Jun 28 19:47:02 UTC 2010
On 06/27/2010 08:38 AM, Hal Murray wrote:
> jimlux at earthlink.net said:
>> So, over the, say, 100 Hz and up range, they're probably pretty good.
> I like to think I'm pretty good at reading data sheets, but when it comes to
> modern A/D chips, I'm not so sure.
> 20 years ago, the specs were for DC and you hoped it did something sensible
> at high frequencies. Now the specs for many chips are for AC (typically they
> show a FFT plot) and you are in trouble if you want to use it for DC.
A typical audio signal chain wants to reject DC. When editing audio, a
shift in DC level causes a click. When changing the level it causes a
click. DC rejection is thus a necessary thing to get the wanted quality,
so it is not supprising. For audio DC is not part of the useful
spectrum. Low frequencies may however be useful. DC rejection on the
analog side is assumed, but it is also expected to exist in the digital
path to handle any remaining offset. The DC compensation is usually a
single-pole IIR filter with a fixed shift-down. The cost is very low in
the digital domain.
> Typical audio A/Ds are oversampling with a low pass filter implemented in
> DSP. This makes the anti-aliasing filter a lot simpler/cheaper.
> Has anybody tested modern audio A/Ds at very low frequencies?
I was about to consider such a test.
> Is there some simple way I should be thinking about this tangle?
> Has anybody tried adding a signal, say 1 KHz, and filtering it out? I'm
> fishing for something along the lines of the old chopper amplifiers - shift
> things to a mode that is known to work well and then filter out the junk you
> don't like.
If you chopper it to say 1 kHz and then de-chopper it and low-pass
filter it, then you will filter it out. Doing the filtering prior to the
de-choppering would require twice as large filter (in pole and zero
count - or their equivalent in tap counts) than the low-pass filter.
Regardless how the ADC is performed, it would need to be evaluated to
know the performance, accordance to the good-enough factor of the
More information about the time-nuts