[sdiy] Digital audio demo on Youtube

[rburnett at richieburnett.co.uk]

> So, going from time-discrete, infinitely short samples to a
> stair-stepped waveform means low pass filtering the signal, albeit
> that's a crude low-pass. But it makes the analogue reconstruction
> filtering easier, not more complicated.

You need to make up for the droop in the gain at the top of the audio 
pass-band (the inverse sinc filter that Eric B mentioned) and then 
implement a steep roll-off above 20kHz to attenuate the images due to 
the sampling process.

> Am I completely off?

No, not at all.  The ZOH is identical to convolving the sampling 
impulses with an impulse response that is a boxcar function of 1/Fs in 
duration.  It's just that the boxcar function of the ZOH isn't the only 
impulse response for the DAC reconstruction, and it's certainly not the 
best one for an audio converter.

For control purposes, the ZOH makes more sense because it's hard to 
interpolate smoothly from the current sample to the next sample when you 
don't necessarily know when the next control sample is going to come 
along.

-Richie,