Does DSD measure better than PCM? That's the question I wanted to answer a while back.
I compared this in the digital domain, as this eliminates some of the complexities and noise of converting to analog and back. Using my own null-testing software I call DeltaWave, I compared DSD256 and PCM 24/96 of the same track by BlueCoast music. The track was by Jenner Fox, Twice as Lonely from album Buffalo.
Here's the spectrum comparison of the entire track @ 192KHz . DSD is pink, PCM is blue:
There is almost no difference up to about 30KHz, at which point DSD is starting to droop slightly compared to PCM. And of course, PCM shows no frequencies above 48KHz. The content above about 24KHz looks to me to be mostly noise pushed out of the audio band by noise shaping.
Here's the same comparison, but now limited to 32KHz using a low-pass filter:
Up to 28KHz, the measurements show that the two files compare very well:
- Difference (rms) = -90.41dB
- Correlated Null Depth=108.37dB
That's a pretty good match!
Here's the spectrum of the difference of the two files. Mostly under -120dB:
Phase is closely aligned up to about 20KHz. This plot shows phase difference for both tracks at each frequency bin in the FFT:
Of course, I can also do a similar comparison with 16/44.1KHz version of the same track. Here's the spectrum of the difference:
And the phase difference plot (diverging after about 22Khz, as expected). Interesting spike at about 4KHz, not sure what that's about.