Computer Audio Asylum: RE: Question about DSP and Filtering by John Swenson

I have been thinking hard about how to reply to this, it's a complex subject.

I think I'm going to go with the abreviated version and try and keep it short. That means there is going to be a lot left out unfortunately, but hopefully it will cast some light on how I'm thinking about this.

It all goes back to a bunch of experiments I did many years ago with regards to digital filters in DAC chips and NOS DACs etc. I won't bore you with the details, but the conclusion was that done right NOS significantly improves the sound IN SOME WAYS, and makes it worse in others. The sound is richer, more alive, conveys more of the emotional impact from the performers, BUT it sounds "dirty" around the edges, particularly in the high frequencies. Some people think it's a good compromise, others do not. The dirtiness is obviously the results of the high frequency aliasing caused by the unfiltered output. But why does getting rid of aliasing get rid of the "richness, emotional impact etc"?

After a lot of experimentation I came to the conclusion it was the digital filters themselves. Every single one of the hardware filters I looked at in DAC chips (and external ones as well, such as the DF1704) were compromised in some way. My supposition is that in order to cut costs in the chips the designers cut corners in the implemention. They are required to meet certain numbers in the spec sheet, and they can't do it properly and still stay on budget, so they cheat and play tricks in order to get good numbers.

In this forum there have been a lot of statements along the lines of "Shannon says that the filter will accurately reproduce the original waveform", but many are saying it doesn't sound that way, and others keep on saying the theorum is correct, I think this is reason for the dichotomy, the actual hardware implementations in most cases are NOT properly implementing the filter.

I have tested this hypothesis in two ways: creating my own digital filter in an FPGA and using software to do the filtering on the file. In both cases I have used two different DACs that I have built myself. One uses 1704 DAC chips (which do not have a filter) and the other uses a 1794 which does have a digital filter, but it can be turned off. Both use very low jitter local clocks, very low noise power supplies etc. For the 1704 I can feed the data directly (NOS), from a FPGA digital filter, or through a DF1704. For the 1794 DAC I can use either its internal filter, or bypass the internal filter and send direct, or use the FPGA filter.

The results of all this was that with both DACs using either the internal filter or the DF1704 produced very clean sound but it was lacking richness, aliveness etc. Stright NOS in both cases gave the richness, aliveness, but dirty sound. Using either the FPGA filter or filtering in software gave the best of both worlds, it still had the richness and aliveness, but was clean. This sound difference is NOT subtle it is almost starteling. Everyone who has heard this invariabley says something along the lines of "now THIS is what it is supposed to sound like".

I tried several programs to do the filtering in software, and they all did similar things. I wan't using any special audiophile programs, just the normal resampling algorithms in programs such as SOX etc. Yes you can hear slight differences between them, but they all sounded way better than straight NOS or filtering in the DAC chip.

Now all that was the explanation for my comments about filtering and Tony's player. To get the best sound I don't want to use filtering in the DAC chip, that leaves implementing my own filter in an FPGA or resampling in software external to the player. Doing it in the FPGA takes a lot of hardware resources and fast clocks which flies in the face of the concept "absolutely as little going on as possible", so the best bet would be resampling in software and sending the higher sample rate files to the player. The higher sample rate files take more memory accesses, it will be interesting to see if that is any better than the extra stuff going on in the FPGA for a hardware filter. My guess is that the esternal software filtering will sound slightly better.

Whew, that was long winded, but that was about as concise as I could get and still cover the material.

John S.