DEcoding Recursive systematic convolutional code

[emma2]

Hi,,
I have problem figuring out how to do decode recursive systematic convolutional code .. I thought of Maximum log liklehood but I'm not sure about it :roll:. Anyone have an idea that might help.
Thanks in advance

 

[kalyanasv]

There are 2 ways to go about it. i) The MLSE -viterbi ii) MLSE with SOVA which uses probability iii) Maximum aposteriori - BCJR. Now do want to compute probabilities or do you want to just decode. If it just decode then the Viterbi is the simplest..you want performance..etc you can go for the Max-Log MAP..but then the SOVA performs equally well.

 

[emma2]

Wow ,I didn't expect that someone will give me the time and response this fast,Thanks Kalyanasv.
next Ireally want to check the performance so I will go withMaxlog-MAP, but raising MA-BCJR brought to my mind a question ..I thought BCJR is only for turbo code because you need to cosider the three streams of data (SYS P1 P2), but what I have (in my case) (Sys and P1) how does that work? you know what i haven't done any coding class so expect to see stubid questions . and if you think the answer for this questin is long just tell me and I will hit the google.
Thanks again really appreciate it.

 

[kalyanasv]

Hi,
BCJR is only an algorithm which uses aposterior probability i.e. MAP. So essentially the Max-Log MAP is a more implementable version of the MAP algorithm.
Your using a recursive Convolution Code which is why I suggested it. I suppose you already know that the RCS codes can be written as a Form of Generator matrix G. To get output sequence v=u*G where the * denotes convolution. Turbo Codes are one of things you can decode with BCJR. I am assuming you have rate 1/3 coder.
The working methodology for the Max-Log MAP, starts with MAP then comes the Log-MAP then the Max-Log MAP. To start with you can look at Here.

One note about coding: Do have look at Galois Fields GF(). Galois found this stuff at the age of "18". Probably the most fundamental theory you need.

I suppose your doing coding for some use...if not for class.. though it can be fun....

 

[emma2]

Hi,
I'm using Costello's book as reference,actually I started with GFs ,then moved to block codes, convolutional codes. And in my system I'm doing something close to distributed turbo code in multiple access relay channel where at the mobile station will do systematic RSC and so at the relay channel first need to eliminate the channel noise by decoding the received signal before interleaving and encoding RSC again. the big picture is turbo code.. this is part from what I'm implementing in my master thesis (for now forget about fun ). I will check the lecture you send, appreciate it.

 

[kalyanasv]

Hi,
Costello book is as good as a reference as you can get. Besides which, you should also definitely look at this website.Here
It provides a lot of free code. Actually it is tiring on the brain but then the fun part is when it works. The there is whole "IEEE" schbang. I would suggest looking at encoding some data. Then look at MAP then shoot for implementing the Max-Log MAP. So that way you have a metric to measure.
Thats some neat stuff for a masters project.
Good luck.