Forums

SCALING OF LPC COEFFICIENTS

Started by mgencarellli September 30, 2002
I have written a LPC10 encoder which produces the reflection
cofficients,gain, pitch parameters etc using a 6711. These values are
in floating point notation. What I want to know is how are these
values are converted to a integer representation, for transmission,
considering that for LPC10,these parameters are represented as
anything from 3 to 7 bits depending on the parameter (eg. gain is 7
bits, refection coefficients 3 to 4 bits). Is it a simple scaling of
the integer range for each parameter?
Maurizio Gencarelli
Electronics Engineer
Royal Australian Air Force



Maurizio-

> I have written a LPC10 encoder which produces the reflection
> cofficients,gain, pitch parameters etc using a 6711. These values are
> in floating point notation. What I want to know is how are these
> values are converted to a integer representation, for transmission,
> considering that for LPC10,these parameters are represented as
> anything from 3 to 7 bits depending on the parameter (eg. gain is 7
> bits, refection coefficients 3 to 4 bits). Is it a simple scaling of
> the integer range for each parameter?

For LPC10 I'm not sure, but for newer coders it's normally a combination of
linear or
log quantization followed by additional coding, such as vector-quantization.
For
example, log10 of pitch might be quantized to 0-127 (7 bits), or LP coefficients
might be quantized to -32768 to -32767 (Q15) then combined into a vector to
access
codebook indexes.

Reflection coefficients are 3-4 bits linear-quantized in LPC10? Yikes, no
wonder it
sounds so bad.

Jeff Brower
DSP sw/hw engineer
Signalogic