Tweaking DSD to PCM

Anyone have thoughts regarding the settings for DSD to PCM filter and DSD to PCM gain boost? What is the effect of these, and under what circumstances would you want to adjust them?

I have a couple comments:

Consider changing the gain boost. +3db instead of +6db. +6db can make some non compliant DSD content clip.

DON’T turn the filter off completely as it could possibly cause speaker damage.

Gain Boost

Most DSD content was represented in the PCM domain at an earlier point in the production process. In some cases, this is because the mixing/mastering workstation used PCM internally, and in other cases old PCM masters are converted to DSD. When it’s time to create an SACD master or DSD file, the PCM stream is transformed into a DSD stream using a technique called Sigma-Delta modulation.

One of the design tradeoffs in a Sigma-Delta is the management of mathematical instability. S-D modulators utilize feedback (sometimes feed-forward) loops that can become unstable if the input signal is too loud.

One of the most common practical measures used to prevent instability is limiting the amplitude of the input signal. In fact, the Scarlett Book (SACD) specifications require that the input signal be limited to -6 dBFS (meaning, 6dB less than the full scale dynamic range of the DXD signal).

This has a side effect: if you were to play a CD and SACD side by side on the same transport, assuming the transport does nothing to compensate for the -6dB adjustment that happened during authoring, the SACD would come out 6dB quieter.

Roon’s default is to reverse that -6dB adjustment when converting DSD to PCM. This should compensate for loudness differences introduced during the authoring process, and allows DSD content to take advantage of the full dynamic range of the resulting PCM samples.

This is great for all content that was properly adjusted during the authoring process. Unfortunately, there is content in the world that was not authored properly. If you hear clipping with some of your content, you may want to turn the 6dB adjustment down to 3dB or 0dB and see if it gets better.

Filter selection:

Most of the energy in a DSD signal is comprised of high frequency noise (mostly above 30kHz). This noise is an implicit consequence of encoding the signal into a 1-bit stream (meaning: the file isn’t wasting bits representing it, rather it is inherent to the DSD representation). this requires some changes in how we think about the signals.

DSD DACs are designed to do the “right thing” with DSD and output a reasonable analog signal that is not filled with high frequency noise.

When a DSD signal is down-sampled to PCM, some of the high-frequency noise is retained in the PCM representation. While this PCM representation mathematically accurate, it is not a nice thing to send to a PCM DAC or to your analog gear.

In many cases, the DAC or some downstream component will filter out the noise anyways. In other cases, SQ might be compromised or audible artifacts could result. In other cases, noise might reach your amp/speakers which aren’t meant to deal with that kind of noise. In extreme cases hardware damage could result.

Which brings us to the filter settings.

These settings allow you to apply a slow-roll-off IIR low-pass filter to the signal. They are gentle filters, and have very nice impulse response characteristics so as to avoid compromising the DSD signal. These filters attenuate that high frequency noise.

The 24kHz filter results in frequency response characteristics that are nearly guaranteed not to offend your PCM gear, but this involves placing a filter fairly near to the limits of human frequency perception. It’s a conservative/safe setting.

30kHz is (in my opinion) the best compromise, it’s the default in Roon, and what I use day-to-day. This setting nicely removes nearly all of the noise spectrum in DSD, while leaving some space between stuff we can hear and the filter itself.

50kHz preserves significantly more noise (especially during DSD64 playback), but is even less likely than the 30kHz filter to muck with frequency response in an audible way. This doesn’t mean it’s “better”. It’s possible that with some gear, this filter will sound worse than 30k depending on how the gear reacts to the HF noise.

Running unfiltered is NOT RECOMMENDED unless you are 100% sure that there is another low-pass filter in your setup that will serve the same purpose.

4 Likes

Thank you, Jeff.

Wonderful technical write up. Thank you, Brian.

I hope this (and many of Brian’s other excellent technical posts) are getting captured and put into the knowledge base.

Yup, good call.

Sorry to quibble but I think this statement is technically true - but a bit misleading.

I’ve got about 100 DSD albums, and I don’t think any of them are derived directly from DXD. The only label I know of that is recording and making DSD like this on a regular basis is 2L. AFAIK, there is very little actual recording to DXD. I don’t know of anyone recording any music other than classical on a regular basis to either DSD or DXD - there are of course a few exceptions (the BlueCoast label), but close to none in the popular music styles (rock, pop, country or hip hop).

What is true that to create DSD, in production lower level PCM is often upsampled to DXD levels and then converted to DSD post production, as production tasks are difficult to do in DSD and few studios have the equipment to do any post recording production directly in DSD. (BTW, almost all chip based DACs also do this PCM>DXD> DSD conversion internally, on their way to outputting analog. Even if they claim “no upsampling”). Some music that is actually recorded in DSD is converted to DXD (or parts are converted to DXD) for production, and then re-converted to DSD in order to sell a DSD file or make an SACD.

Lots of DSD is derived from PCM at much lower levels, even 16/44.1. DSD of legacy material is sometimes a conversion directly from analog tape. There are a few small labels recording in DSD: Classical, a little jazz, some world and folk musics. Some of them actually record in an analog/DSD hybrid and do almost no post recording production, and then sell “real” DSD.

I didn’t mean to suggest that most content is recorded in DXD–that is indeed rare. The DXD is usually coming out of mixing/mastering/remastering processes. And DXD a horrible marketing word–we should just call it PCM in a technical discussion.

For the purpose of this discussion, it doesn’t matter what the sample rate or bit depth of the PCM content was. The only purpose of bringing that up was to provide some motivating background for why SACD was formalized with a -6dB offset back in 1999.

I’ve replaced that paragraph. Better?

Most DSD content was represented in the PCM domain at an earlier point in the production process. In some cases, this is because the mixing/mastering workstation used PCM internally, and in other cases old PCM masters are converted to DSD. When it’s time to create an SACD master or DSD file, the PCM stream is transformed into a DSD stream using a technique called Sigma-Delta modulation.