FPGA Central

Hi all,

Currently, I am studying on how to get Hi Fidelity effects with
ADPCM related compression methods. But it seems that the traditional
IMA ADPCM and MS ADPCM will not get the satisfied results.
Any one can give me some suggestions? 16 bit PCM to 5 bit PCM will
be accepted by me.

Thanks in advance.

Jessecw

Jessecw wrote:

> Hi all,
>
> Currently, I am studying on how to get Hi Fidelity effects with
> ADPCM related compression methods.

Why trying to reinvent mp3 ?

But it seems that the traditional
> IMA ADPCM and MS ADPCM will not get the satisfied results.

Of course it will not, unless your sample rate is 96+ kHz. For music at
44.1kHz, the maximum prediction gain is about 20dB. Consequently, for a
reasonably good fidelity you will need at least 10 bits per ADPCM sample.

> Any one can give me some suggestions? 16 bit PCM to 5 bit PCM will
> be accepted by me.

Consider multiband ADPCM. After making some refinements to it, you will
reinvent the MP3.


Vladimir Vassilevsky

DSP and Mixed Signal Design Consultant

http://www.abvolt.com

Jesse wrote:
> Hi all,
>
> Currently, I am studying on how to get Hi Fidelity effects with
> ADPCM related compression methods. But it seems that the traditional
> IMA ADPCM and MS ADPCM will not get the satisfied results.

ADPCM is just a simple form of (adaptive) LPC, with fixed prediction
coefficients. General, adaptive LPC can be used for anything between
lossless audio coding (eg.
http://www.nue.tu-berlin.de/wer/liebchen/lpac.html) down to veeery low
bit-rate audio coding (usually for speech, eg.
http://en.wikipedia.org/wiki/CELP).

Take your pick.

Regards,
Andor

"Vladimir Vassilevsky дµÀ£º
"
> Jessecw wrote:
>
> > Hi all,
> >
> > Currently, I am studying on how to get Hi Fidelity effects with
> > ADPCM related compression methods.
>
> Why trying to reinvent mp3 ?
>
> But it seems that the traditional
> > IMA ADPCM and MS ADPCM will not get the satisfied results.
>
> Of course it will not, unless your sample rate is 96+ kHz. For music at
> 44.1kHz, the maximum prediction gain is about 20dB. Consequently, for a
> reasonably good fidelity you will need at least 10 bits per ADPCM sample.
>
> > Any one can give me some suggestions? 16 bit PCM to 5 bit PCM will
> > be accepted by me.
>
> Consider multiband ADPCM. After making some refinements to it, you will
> reinvent the MP3.

Thanks a lot. Could you give me some URL links to multiband ADPCM?
You help will do a lot of help to me.


>
>
> Vladimir Vassilevsky
>
> DSP and Mixed Signal Design Consultant
>
> http://www.abvolt.com