PCM to DSD upsampling

[MLGrado]

A question for those of you in the know.

In PCM to DSD upsampling, whether in software or in a sigma-delta DAC chip, a reconstruction fitler is typically used for PCM.

For instance, in the case of Jriver Media Center's PCM to DSD conversion, measurements show it uses a typical linear phase filter, which has not an insignificant amount of ringing in the impulse response.

Why is a reconstruction filter needed at all? When we are going from 44 to 192 khz sampling to a sample rate of a few million??? It isn't like aliasing is going to be an issue. Furthermore, any aliasing would be totally masked in the ultrasonic noise anyway.

Any one have an idea why Jriver and maybe others use typical PCM oversampling/reconstruction before oversampling to 64fs and greater??

Just seems unnecessary to me..

Andrew

 

[Ken Newton]

...Why is a reconstruction filter needed at all? When we are going from 44 to 192 khz sampling to a sample rate of a few million??? It isn't like aliasing is going to be an issue. Furthermore, any aliasing would be totally masked in the ultrasonic noise anyway...

So long as the spectrum of the sampled waveform is self-limited to less than the Nyquist frequency, an filter to prevent baseband (the desired audio band) aliasing is not required during ADC sampling. However, low-pass (reconstruction) filtering is still desirable at the DAC side during playback to remove the ultra-sonic image bands which otherwise repeat at integer multiples of the native sample rate. While these images are not directly audible (being ultrasonic), they may cause subsequent amplifier intermodulation, and/or RF input misbehavior. In addition, some suggest that, left unfiltered, these ultrasonic images may responsible for effects such as DAC noise floor modulation, and increased jitter artifacts.

It seems that your real objection to the digital anti-alias and anti-image filters may have to do with their strong impulse response ringing. Objectively, this strong filter ringing is what correctly reconstructs the original band-limited waveform from only a series of periodic samples. Subjectively, I used to suspect the ringing responsible for digital's typically strident sound character. However, my own experiments in this area have led me to conclude that the strong filter ringing, whether pre or post, is not the source of digital's subjective ills that I had once suspected it was.

 

[MLGrado]

So long as the spectrum of the sampled waveform is self-limited to less than the Nyquist frequency, an filter to prevent baseband (the desired audio band) aliasing is not required during ADC sampling. However, low-pass (reconstruction) filtering is still desirable at the DAC side during playback to remove the ultra-sonic image bands which otherwise repeat at integer multiples of the native sample rate. While these images are not directly audible (being ultrasonic), they may cause subsequent amplifier intermodulation, and/or RF input misbehavior. In addition, some suggest that, left unfiltered, these ultrasonic images may responsible for effects such as DAC noise floor modulation, and increased jitter artifacts.

It seems that your real objection to the digital anti-alias and anti-image filters may have to do with their strong impulse response ringing. Objectively, this strong filter ringing is what correctly reconstructs the original band-limited waveform from only a series of periodic samples. Subjectively, I used to suspect the ringing responsible for digital's typically strident sound character. However, my own experiments in this area have led me to conclude that the strong filter ringing, whether pre or post, is not the source of digital's subjective ills that I had once suspected it was.

Yes, and thanks for the answer.

You are correct about my concern about filtering. I know reconstruction filters are used in the conversion from PCM to DSD (Jriver for example). DSD is full of ultrasonic noise already, so I was wondering why we needed to use 'steep' filters to cut aliasing noise that would surely be 'lost' in the ultrasonic noise of DSD.