Re: [Alsa-user] Digital bit perfect ouptut with ALSA

[Bill U. <un...@ph...>]

On Sun, 21 Dec 2008, Jamie Lokier wrote:

> Sergei Steshenko wrote:
>> On Sun, 21 Dec 2008 03:34:50 +0000
>> Jamie Lokier <ja...@sh...> wrote:
>>
>>
>>> Even 2x upsampling introduces distortion.
>>
>> No, it doesn't.
>>
>> It can be implemeted as a pure linear transform with constant coefficients,
>> thus it would be a simple FIR LPF.
>
> It depends what you are calling distortion.  An upsampling FIR has
> energy in the alias band between input-Nyquist and output-Nyquist,
> because of the finite number of taps.  To avoid those you need a sinc
> filter, which takes infinite taps; you can only approximate it.  Since
> you can't perfectly filter the alias band with analogue filtering
> after the DAC, and you can't filter it digitally due to finite taps,
> that is audio energy you don't want.  Thus the FIR output has a form
> of distortion from the desired continuous waveform, which has no
> energy in the alias band.

The question is not whether it has energy there, but what is the amplitude of
that and is it in the audible spectrum. It matters not a damn if  that
energy is 90dB below the signal level  or is at 45KHz.  And 24 bit
quantization error ( being at 135dB below the full signal-- ie completely and
utterly inaudible unless you happen to play your music as amplitudes where
your speakers create shock waves in the air-- in which case you are in far
more trouble from death than to have to worry about distorted hearing. And
even if you survive, you cannot hear anything anymore.

>
> (There is also quantisation error distortion - you can't implement the
> linear transform perfectly - but let's not go there.)
>
> Non-integer ratio upsampling is not a simple linear function, as you
> rightly say (it's a polyphase FIR LPF), and _can_ introduce more

Sure its a linear function.

> undesired energy all over the place, depending on the number of taps
> and filter design.
>
> But the point is that the increased output sample rate allows
> potentially better LPFs both in the polyphase FIR and in the analogue
> domain after the DAC, compared with the fixed 2x rate, which more than
> compensate at reducing alias band energy.

The question is what you mean by better. If it is at amplitudes or frequencies
which are inaudible, then it is irrelevant anyway.

>
> -- Jamie