readme

CCITT G.711语音压缩算法

Dear Tony Robinson

Recently I retrieved the routines for the CCITT G.711 standard
from the ftp server at Cambridge University.

I noticed a remark "Thoughly tested by ajr and believed bug free.", 
and I am therefore e-mailing you since I have found some (minor) bugs in 
these routines. These are described below.

I am enclosing an updated version of these routines and a main program
for their inclusion. 

If you agree with my comments and updating you are very welcome to replace the
existing routines with these. 

I assume several people have applied these routines for conversion, so maybe 
it would be a good idea to notify the relevant news groups, but I will let 
you to decide on that.


With the very best wishes 


Borge Lindberg
 

Bugs in g.711 routines:
-----------------------

Tables for A-law /u-law conversion:
-----------------------------------
_u2a	/* u- to A-law conversions */:


	81,	82,	83,	84,	85,	86,	87,	88, 

should be

       	80,	82,	83,	84,	85,	86,	87,	88,



_a2u[128] /* A- to u-law conversions */:

	73,	74,	75,	76,	77,	78,	79,	79, 

should be


	73,	74,	75,	76,	77,	78,	79,	80,


The bugs found in the routines below, never occur on quantized 16 bit
data (data originating from a prior A- og u-law to linear conversion).

function linear2alaw:
----------------------
Generates wrong A-law value for linear input values in the range [-7 .. -1].

Segment numbers are found from full range values. This has to be changed
to 13-bit range values - otherwise a linear2alaw - alaw2linear - linear2alaw
sequence is not guaranteed to give identical A-law values.

function linear2ulaw:
----------------------
Generates wrong u-law values for linear input values having numeric
value above 32635 since clipping is omitted and results in something
like wrap-around.

Segment numbers are found from full range values. This has to be changed
to 14-bit range values - otherwise a linear2ulaw - ulaw2linear - linear2ulaw
sequence is not guaranteed to give identical u-law values.