[sdiy] Digital audio demo on Youtube

[rsdio at sounds.wa.com]

On Feb 18, 2014, at 11:49, Eric Brombaugh wrote:
> On 02/18/2014 12:33 PM, cs80 at therogoffs.com wrote:
>> I’ve read about the affects of ZOH using standard sample/hold DAC  
>> outputs but what’s the alternative?  In mathematical theory, the  
>> DAC output should be zero-width pulses whose area equal the sample  
>> value.  Do you have pointers to articles / part specs on DACs that  
>> try and approach this?
>
> Typically the solution is oversampling and interpolation. Inverse  
> sinc correction and a decent reconstruction filter can also be used  
> where oversampling in the digital domain isn't possible, but for  
> audio that's usually not a problem.
>
> Eric

I'm familiar with a number of DAC circuit implementations - binary- 
weighted resistor network or current source, R-2R ladder, PWM,  
thermometer coded "flash" converters, and hybrids - but none which do  
anything other than hold the output between samples.

Can you cite a reference part number for us which does not hold the  
voltage between samples?

It seems like an incredibly complex circuit would be needed to  
produce a voltage with instantaneous changes or undefined value in  
between samples.

I think it's a bit misleading to talk about oversampling, since  
you're still usually talking about a ZOH - just at a higher sample  
rate. If you convert a 48 kHz sample train to 192 kHz using  
oversampling, you still end up holding each sample at a constant  
voltage until the next sample period. True: from the point of view of  
an original 48 kHz sample stream, the samples are not held between.  
But from the point of view of the intermediate 192 kHz sample rate  
(or a 192 kHz sample stream taken directly from the analog input and  
preserved as an HD audio data stream), the conversion involves ZOH at  
that high rate.

The one exception is sigma-delta, where I suppose it's rather easy to  
design a desired slew rate into the output driver so that the voltage  
increases or decreases at a constant rate rather than holding at a  
constant. However, this is not without it's own distortion, because  
it's not a narrow impulse with perfect zero Volts in between samples.  
I suppose that PWM and/or flash converters could also be designed  
with slew rate limiting.

I realize that what you're talking about is sinc "correction" - but  
my understanding of that is that you're still designing a DAC circuit  
with ZOH, and then the reconstruction filter corrects for the  
frequency response imparted by the ZOH. It seems pretty difficult to  
avoid the ZOH rather than accept it and correct for it.

One of the advantages of ZOH is that it is the least distorted  
approximation of a finite pulse. If you disconnect the S&H during the  
unpredictable settling time of the DAC output, then you have a very  
rectangular output. Although the width is greater than zero, it's at  
least predictable and can be compensated. Anything like slew rate or  
triangular interpolation is not at all like an impulse because it's  
not a simple rectangle.

As Achim mentioned, there are return-to-zero DAC implementations, but  
those are not ideal. They have a non-zero impulse width, and thus  
affect the frequency response similarly to ZOH, although to a much  
lower degree.

Brian Willoughby
Sound Consulting