[sdiy] Differential 12dB/oct filter

[Tom Wiltshire]

Hi Richie,

> I'm also working with the dsPIC audio DAC here too, so interested in what you're doing.  Are you finding it's output excessively noisy above 20kHz?

Well, I don't know about "excessively", but I'd certainly like less noise. I'm playing with using the 33FJ64GP802 for short delays. I've got audio input into the 12-bit ADC (at 32KHz sampling rate), and audio output from the 16-bit DAC at 64KHz, so it's currently a one-chip-wonder!
12-bit audio isn't too bad (after all, 12-bit/32KHz was an expensive studio rack mount delay processor in the middle 1980s!!) and the 16-bit output gives me headroom for processing.

It's funny, I haven't had such noise problems with it when doing oscillator designs, so I think perhaps some of my noise is coming from the ADC or the signal going into it. I should probably do everything I can to ensure that the input is smooth and clean before worrying about the output.

Can you recommend how I should generate a low noise 1.65V reference for the input biasing?

> You could try filtering each of the DAC+ and DAC- outputs from the dsPIC with first-order RC filters first before they are fed into the inputs of you "differential-amplifilter".  That would make a 2nd order filter.
> 
> Provided you design the first-order RC sections to take into account the loading due to the input impedance of your diff amp that follows you should get poles where you want them.  You won't be able to get complex poles though using this schemes, only real placement.  So, for instance you won't be able to make a perfect 2nd order Butterworth response, but you can probably get close enough.
> 
> The thing to watch out for is common-mode rejection.  If you do something to one of the balanced outputs from the DAC, you need to do exactly the same thing to the other output, otherwise common-mode noise coming out of the dsPIC won't cancel completely in the diff amp.  This has implications for resistance tolerance, and particularly capacitor tolerances!

Yes. Matching capacitors in such a filter might be the killer here. The whole point of the differential output is to kill the common-mode noise, so if I screw that up, I might as well go back to square one.

> If your ultimate aim is a single-ended audio output

It is. Since this is a delay processor, the signal comes in to a buffer, then splits. The wet path goes through an anti-aliasing filter to the dsPIC, then out of the dsPIC to the diff amp, then a post-DAC MFB filter, and then meets the dry signal at the output mixer. The dry path never gets digitised.

> I might be tempted to just follow the diff amp with a Sallen-Key or MFB filter though.  You have more control over the pole placements, no loading effects, less R's and C's, no common-mode degradation due to capacitor tolerances, but the extra cost of another op-amp :-(

Yes. I've done this as well. But it's only a 12dB filter, so it's not a huge effect.

Thanks,
Tom