PCM up-sampling with the TI/BB PCM5122 DAC chip

Hi Roon Up-sampling experts out here

Are any of you able to have a quick look at the TI / Burr Brown PCM5122 DAC chip datasheet linked below and advise which Roon upsampling method of:

“max PCM sample rate” or “max factor of 2” is best to use:

Your help would be greatly appreciated


Cheers, Sean

Does Table 8 suggest “factor of 2” should be the best Roon up-sampling method to use for this particular Dac chip? @brian

Based on that data sheet, I think I would want to run that DAC at 384kHz, assuming the driver + implementation around the chip supports that frequency.

Based on Table 8, I can see that 384kHz content is not going to go through any of the oversampling filters in that data sheet–all of which have significantly worse technical parameters than ours.

“Power of 2” is not really about the DAC–it is for people who believe that non-integer-ratio conversions are worse than integer-ratio conversions. This is not actually true for Roon’s sample rate converters–which use the same approach either way–but some people prefer to avoid them nonetheless, and that’s why the setting is there.

Thanks Brian ! I always wondered what the obsession was with “factor of 2” vs “factor of 1.09 = 192/176” for example or factor of 2.18 = 192/88

As in what the integer had to do with anything but thanks for mentioning there’s no advantage in Roon with the integer method.

Is that something you could elaborate upon?

Specifically, I have been using the “High-Attenuation x* Interpolation Filter,” which I assume is the equivalent of your “Precise, Linear Phase” filter. How do the respective technical parameters of these filters compare?

FWIW, the Hifiberry DAC+ Pro inplementation only admits sample rates up to 352.8kHz. So I have not been able to try your filter at the full 384kHz.

The TI filter looks like a general-purpose brickwall filter.

Our precise filter is not a brickwall. It’s more tuned towards listening use cases. It’s slightly leakier (IIRC -60dB @ Fs, where as this one is -100dB). The stopband rejection is much deeper (IIRC, -180dB vs ~-100dB). Fewer cycles of ringing. I can’t quite tell from the chart, but we also start the dropoff at a point designed to make ours an apodising filter–not sure if TI is doing that or not.

It’s a lot more expensive to implement ours, in terms of work done per sample. We were thinking about listening use cases when designing it and had a large processing time budget since we are on general purpose CPUs. TI are putting a common utility filter into a commodity DAC chip. Both filters are going to serve the function of changing sample rate, but the design process was pretty different, and I wouldn’t necessarily think of them as directly comparable.

Leakier, but still linear phase?

It might be nice to have a bit of detail on your various filters in the KB, as the names are perhaps not as descriptive as they might be.

Linear phase and leaky are independent variables. Our filter is about -60dB at Fs/2. So not very leaky, but it’s not fully stopped at that point.

So are “apodizing” and “minimum phase”. But in audio-reconstruction-filter-land, they are often taken to be synonymous. Hence my desire for clarification.

I had sort of assumed that “Precise linear phase” meant a “Chord-like” linear phase brick wall. Thanks for clearing up that misconception.

FWIW, I found these measurements of Roon’s PCM upsampling filters.

The “Precise Linear Phase” filter is, indeed, not as steep as the PM5122 “High-Attenuation” in my HifiBerry DAC+ Pro, but it’s considerably steeper than the AK4493 “Sharp Roll-off” filter in my Topping E30.

I wish folks could come to some sort of Gentleman’s Agreement about a naming their reconstruction filters. Or, failing that, just publish graphs like the above.