Greg Simmons wrote:
<< This is a common misconception. Current Sample Rate Conversion
technique does not directly divide higher rates to the target rate,
i.e. 88.2k to 44.1k is not a case of simply dividing by two.
>>
<Right. There is no advantage in retaining an integer relation between samp=
ling rates.>
Hi Greg,
I beg to differ here. As someone who has already programmed (interpolation-=
based) sample rate conversion algorithms for both DSP chips and PC software=
, I see no reason why there should be any advantage to up-sample the signal=
to a higher intermediate sample rate in one just wants to convert the samp=
le rate from 88.2 to 44.1 kHz.
<At the core of this is the Nyquist frequency. A recording at 44.1ks/s ('ks=
/s' =3D kilo samples per second, to make a distinction from 'kHz' for the a=
udio signal) cannot contain audio signals above 22.05kHz (the Nyquist frequ=
ency for 44.1ks/s), or it will cause aliasing. A recording made at 88.2ks/s=
will contain audio signals up to 44.1kHz, so the sample rate conversion pr=
ocess MUST include a low pass filter to remove everything above 22.05kHz *b=
efore* converting to 44.1ks/s to prevent aliasing.>
Yes, this is true, but the required low-pass filter would be very straightf=
orward and can just operate on the original sample stream of 88.1 kHz. Any =
intermediate up-sampling (to 7.056 MHz) would unnecessarily increase the am=
ount of data that needs to be processed, which would take much longer to pr=
ocess.
<The mathematical process of filtering negates any theoretical benefits tha=
t an integer-related multiple sampling rate might have had.>
No, this should not negate the benefits as there would be no additional int=
erpolation or up-sampling required.
<Upsampling is a very common technique these days for many kinds of process=
ing, including the filtering mentioned above. As I understand it [and corre=
ct me if I'm wrong], it produces a cleaner result because it distributes th=
e error noise through a wider bandwidth, therefore less error noise within =
the bandwidth that we're interested in (nominally 20Hz to 20kHz).>
I believe that there is no benefit at least in simple low-pass filtering fo=
r sample rate decimation.
However, up-sampling can be indeed improving the sound quality when it come=
s to playing back the sounds through a D/A converter. So, you might hear th=
e same improvements when you executed the sample rate conversion without in=
termediate up-sampling and instead activated the up-sampling for the playba=
ck procedure only.
Regards,
Raimund
Avisoft Bioacoustics
http://www.avisoft.com
|