01-17-2020, 08:22 AM
Un (court) extrait de mes échanges avec Bruno Putzeys à propos du DSD :
Like anyone else these days I treat DSD as simply one of the formats that music comes in, and I convert it to the format that I found most suitable for the design of the DAC stage. That format is one-sided PWM. It has all the advantages of true one bit (=no matching problems) coupled with all the advantages of multibit (much lower noise) AND a total absence of ISI. See the bottom half of the first picture above. One PWM cycle (3.125MHz) in the Tambaqui has 33 possible pulse lengths of which I’m using the middle 25 or so. That means I have about 5 bits of resolution, and the guarantee that the rising/falling transients have completely died down before another one hits. One edge repeats at completely constant intervals, the other edge is modulated by the signal. The wonderful thing of this arrangement is that any difference in rise/fall times can only cause a fixed DC shift with no signal dependence at all. This is why we’re quite happy to claim that the Tambaqui is the only completely ISI free converter in existence and it’s one of the major reasons why it sounds the way it does. The noise shaping is 7th order. It’s explained in a fair bit of detail in https://www.hypex.nl/img/upload/doc/an_w...al_PWM.pdf
Feeding the modulator is the upsampler. This is an asynchronous sample rate converter with a fixed 3.125MHz output rate. That was also completely our own design. Basically it works by first upsampling any incoming PCM data 4 times, and then fitting a 7th order polynomial over the resulting waveform. That results in a virtually infinitely upsampled representation (i.e. a continuous time signal) that is then sampled at the output rate. DSD is low-pass filtered prior to being interpolated, as one would do in any attempt to upsample it. The time base for the resampling stage is a ratio tracker of which we’re actually pretty proud but which would take the discussion a bit far.
En espérant que ce soit plus clair...
Like anyone else these days I treat DSD as simply one of the formats that music comes in, and I convert it to the format that I found most suitable for the design of the DAC stage. That format is one-sided PWM. It has all the advantages of true one bit (=no matching problems) coupled with all the advantages of multibit (much lower noise) AND a total absence of ISI. See the bottom half of the first picture above. One PWM cycle (3.125MHz) in the Tambaqui has 33 possible pulse lengths of which I’m using the middle 25 or so. That means I have about 5 bits of resolution, and the guarantee that the rising/falling transients have completely died down before another one hits. One edge repeats at completely constant intervals, the other edge is modulated by the signal. The wonderful thing of this arrangement is that any difference in rise/fall times can only cause a fixed DC shift with no signal dependence at all. This is why we’re quite happy to claim that the Tambaqui is the only completely ISI free converter in existence and it’s one of the major reasons why it sounds the way it does. The noise shaping is 7th order. It’s explained in a fair bit of detail in https://www.hypex.nl/img/upload/doc/an_w...al_PWM.pdf
Feeding the modulator is the upsampler. This is an asynchronous sample rate converter with a fixed 3.125MHz output rate. That was also completely our own design. Basically it works by first upsampling any incoming PCM data 4 times, and then fitting a 7th order polynomial over the resulting waveform. That results in a virtually infinitely upsampled representation (i.e. a continuous time signal) that is then sampled at the output rate. DSD is low-pass filtered prior to being interpolated, as one would do in any attempt to upsample it. The time base for the resampling stage is a ratio tracker of which we’re actually pretty proud but which would take the discussion a bit far.
En espérant que ce soit plus clair...