A user seeks assistance adapting a 13-bit A-law compression table for a 16-bit FPGA implementation as required by LTE standards. The discussion focuses on the correct mathematical treatment of signed integers during the companding process.
The solution involves converting two's complement values to sign-magnitude format and specifically inverting the bits for negative ranges to maintain the correct trend.
- A-law compression requires converting 2's complement data into sign plus magnitude before processing.
- Standard 13-bit compression tables can be adapted to 16-bit by shifting bit indices toward the Most Significant Bit (MSB).
- For negative input values, bits must be inverted in the resulting table to achieve the correct compression trend.
- While other compression algorithms like Huffman or LZW exist, A-law is specifically mandated in certain radio-to-Ethernet standards like LTE.
Hi all,
Referring to this doc on A law compression (table 1):
http://www.young-engineering.com/docs/YoungEnginee...
The table is meant for 13 bits signed to 8 bits signed.
In my case I have to compress 16 bits to 8 bits and when I apply the above table I get the positive range correct but the negative range reduced by half or so. Can anybody help explain how to implement the compression using such table or what could be missing.
Thanks
Kaz
Hi Kaz,
Did you use exactly that table to compress the 15 bits + sign into 7 bits + sign? Or did you make a new table?
Can we see the code?
Regards,
Josy
Hi Josy,
Thanks for the reply. I haven't access to code right now but I shifted the 13 bits table towards MSB then kept the same pattern. for example the lowest table entry of "1,A,B,C,D,xxx...] was moved to bit index 14 for '1' and bit index 13:10 for ABCD
Kaz,
Thus effectively discarding the 3 lowest bits of the input data?
Did you convert the 2's complement into sign + (unsigned) magnitude, before discarding the 3 lowest bits?
Regards,
Josy
I did first convert negative values to unsigned(though that is not stated in the doc) then all negative values were opposite in trend. So I removed the conversion and now for the range +32767 to -32768 I get +127 to -80 !! but correct in trend.
convert to unsigned as in:
def twos2sm( a ):''' accept a 16 bit signed input vector and
return the sign and magnitude '''
if a[15]:
# negative
return 1, 2**15 - a[15:0]
else:
return 0, a[15:0]
A-law is an algorithm of the log compression of the audio signals.
It attracts for its simplicity. But it works bad for other data compression.
You need other kind of compression like huffman method, lzw, DCT+huffman etc.
I follow the LTE standards for radio to ethernet packets. I Don't have choices here.
Anyway It turned out that documented table is ok for positive range. For negative numbers it needed conversion to unsigned then inverting those bits in the table as well. Thanks all.
