switched ADPCM

Started by Jessecw December 14, 2006
Hi All,

I am preparing to develop an real time audio compression codec for high
quality audio signals. Due to the heavy calculation burden, I will not
adopt MP3 as my compression algorithm.

I get from google search engine that there is an algorithm called
switched ADPCM that can code high quality audio signals.

I would like to hear from some one that what switch ADPCM is or get
some useful web URL for it.

Any comments are welcomed.

Best Regards,
Jesse


Jessecw wrote:

> Hi All, > > I am preparing to develop an real time audio compression codec for high > quality audio signals.
What is "high quality" ?
> Due to the heavy calculation burden, I will not > adopt MP3 as my compression algorithm.
Poor you.
> I get from google search engine that there is an algorithm called > switched ADPCM that can code high quality audio signals. > > I would like to hear from some one that what switch ADPCM is or get > some useful web URL for it.
There is no silver bullet, and there can't be. The prediction gain is going to be about 20dB at 44.1kHz at the best. To have a reasonably good quality, you have to make up for another 60dB at least, and this is 10 bits per sample. A companding and Huffman may allow you to squeeze another couple of bits. Eight bits per sample, and here we go. If you insist on the lower bitrate, then split your bandwidth into several subbands and encode each subband with ADPCM with the different number of bits. Vladimir Vassilevsky DSP and Mixed Signal Design Consultant http://www.abvolt.com
Thanks a lot.

Then how about this method?

Encoder:
                    (Interpolation)
48kHz PCM -------------------->96kHz PCM (Note1) ------> For prediction
       |
       |-----------------------------> Encoding the difference with the
prediction value--------->



Decoder:

ADPCM Code------->Inverse Quantizer------>+--------------------------->
                                                                  ^
                                                                  |

|--------Predictor with 96kHz PCM, interpolated


Thanks in advance.

Jesse


"Vladimir Vassilevsky =D0=B4=B5=C0=A3=BA
"
> Jessecw wrote: > > > Hi All, > > > > I am preparing to develop an real time audio compression codec for high > > quality audio signals. > > What is "high quality" ? > > > Due to the heavy calculation burden, I will not > > adopt MP3 as my compression algorithm. > > Poor you. > > > I get from google search engine that there is an algorithm called > > switched ADPCM that can code high quality audio signals. > > > > I would like to hear from some one that what switch ADPCM is or get > > some useful web URL for it. > > There is no silver bullet, and there can't be. > > The prediction gain is going to be about 20dB at 44.1kHz at the best. To > have a reasonably good quality, you have to make up for another 60dB at > least, and this is 10 bits per sample. A companding and Huffman may > allow you to squeeze another couple of bits. Eight bits per sample, and > here we go. > > If you insist on the lower bitrate, then split your bandwidth into > several subbands and encode each subband with ADPCM with the different > number of bits. > > Vladimir Vassilevsky >=20 > DSP and Mixed Signal Design Consultant >=20 > http://www.abvolt.com
Do not top post, please!


Jessecw wrote:

> Thanks a lot. >=20 > Then how about this method?
There is no such thing as the perpetual motion. You can either have 10 bits per ADPCM sample at 48kHz or 5 bits per sample at 96kHz. The net bit rate is the same. VLV
>=20 > Encoder: > (Interpolation) > 48kHz PCM -------------------->96kHz PCM (Note1) ------> For prediction=
> | > |-----------------------------> Encoding the difference with the=
> prediction value---------> >=20 >=20 >=20 > Decoder: >=20 > ADPCM Code------->Inverse Quantizer------>+--------------------------->=
> ^ > | >=20 > |--------Predictor with 96kHz PCM, interpolated >=20 >=20 > Thanks in advance. >=20 > Jesse >=20 >=20 > "Vladimir Vassilevsky =D0=B4=B5=C0=A3=BA > " >=20 >>Jessecw wrote: >> >> >>>Hi All, >>> >>>I am preparing to develop an real time audio compression codec for hig=
h
>>>quality audio signals. >> >>What is "high quality" ? >> >> >>>Due to the heavy calculation burden, I will not >>>adopt MP3 as my compression algorithm. >> >>Poor you. >> >> >>>I get from google search engine that there is an algorithm called >>>switched ADPCM that can code high quality audio signals. >>> >>>I would like to hear from some one that what switch ADPCM is or get >>>some useful web URL for it. >> >>There is no silver bullet, and there can't be. >> >>The prediction gain is going to be about 20dB at 44.1kHz at the best. T=
o
>>have a reasonably good quality, you have to make up for another 60dB at=
>>least, and this is 10 bits per sample. A companding and Huffman may >>allow you to squeeze another couple of bits. Eight bits per sample, and=
>>here we go. >> >>If you insist on the lower bitrate, then split your bandwidth into >>several subbands and encode each subband with ADPCM with the different >>number of bits. >> >>Vladimir Vassilevsky >> >>DSP and Mixed Signal Design Consultant >> >>http://www.abvolt.com >=20 >=20