soft viterbi implementation

[shameem]

function encoder_out=cnv_encd(g,k,encoder_in)

hi,
Can anybody please upload some good material on implementation of soft viterbi decoder?
I need it very much. Olease help.

thanks

 

[ahmedseu]

These two functions are for conv. encoder and soft viterbi decoder
%---------------------------------------------------------------------------

function encoder_out=cnv_encd(G,k,encoder_in)
% cnv_encd(G,k,encoder_in)
% determines the output sequence of a binary convolutional encoder
% G is the generator matrix of the convolutional code
% with n0 rows and l*k columns. Its rows are g1,g2,...,gn.
% k is the number of bits entering the encoder at each clock cycle.
% encoder_in The binary input seq.

% check to see if extra zero padding is necessary
if rem(length(encoder_in),k) > 0
encoder_in=[encoder_in,zeros(size(1:k-rem(length(encoder_in),k)))];
end
n=length(encoder_in)/k;
% check the size of matrix G
if rem(size(G,2),k) > 0
error('Error, g is not of the right size.')
end
% determine l and n0
l=size(G,2)/k;
n0=size(G,1);
% add extra zeros
u=[zeros(size(1l-1)*k)),encoder_in,zeros(size(1l-1)*k))];
% generate uu, a matrix whose columns are the contents of
% conv. encoder at various clock cycles.
u1=u(l*k:-1:1);
for i=1:n+l-2
u1=[u1,u((i+l)*k:-1:i*k+1)];
end
uu=reshape(u1,l*k,n+l-1);
% determine the output
encoder_out=reshape(rem(G*uu,2),1,n0*(l+n-1));



%---------------------------------------------------------------------------------

function [decoder_out,survivor_state,cumulated_metric]=viterbi_soft(G,k,decoder_in)
% VITERBI The Viterbi decoder for convolutional codes
% [decoder_out,survivor_state,cumulated_metric]=viterbi(G,k,decoder_in)
% G is a n x Lk matrix each row of which
% determines the connections from the shift register to the
% n-th output of the code, k/n is the rate of the code.
% survivor_state is a matrix showing the optimal path through
% the trellis. The metric is given in a separate function metric(x,y)
% and can be specified to accomodate hard and soft decision.
% This algorithm minimizes the metric rather than maximizing
% the likelihood.

n=size(G,1);
% check the sizes
if rem(size(G,2),k) ~=0
error('Size of G and k do not agree')
end
if rem(size(decoder_in,2),n) ~=0
error('channel output not of the right size')
end
L=size(G,2)/k;
number_of_states=2^((L-1)*k);
% generate state transition matrix, output matrix, and input matrix
for j=0:number_of_states-1
for l=0:2^k-1
[next_state,memory_contents]=nxt_stat(j,l,L,k);
input(j+1,next_state+1)=l;
branch_output=rem(memory_contents*G',2);
nextstate(j+1,l+1)=next_state;
output(j+1,l+1)=bin2deci(branch_output);
end
end
state_metric=zeros(number_of_states,2);
depth_of_trellis=length(decoder_in)/n;
decoder_in_matrix=reshape(decoder_in,n,depth_of_trellis);
survivor_state=zeros(number_of_states,depth_of_trellis+1);
% start decoding of non-tail channel outputs
for i=1:depth_of_trellis-L+1
flag=zeros(1,number_of_states);
if i <= L
step=2^((L-i)*k);
else
step=1;
end
for j=0:step:number_of_states-1
for l=0:2^k-1
branch_metric=0;
%binary_output=(ones(1,n)-2*deci2bin(output(j+1,l+1),n));
binary_output=deci2bin(output(j+1,l+1),n);
for ll=1:n
branch_metric=branch_metric+abs(decoder_in_matrix(ll,i)-binary_output(ll));
end
if((state_metric(nextstate(j+1,l+1)+1,2) > state_metric(j+1,1)...
+branch_metric) | flag(nextstate(j+1,l+1)+1)==0)
state_metric(nextstate(j+1,l+1)+1,2) = state_metric(j+1,1)+branch_metric;
survivor_state(nextstate(j+1,l+1)+1,i+1)=j;
flag(nextstate(j+1,l+1)+1)=1;
end
end
end
state_metric=state_metric,2:-1:1);
end
% start decoding of the tail channel-outputs
for i=depth_of_trellis-L+2:depth_of_trellis
flag=zeros(1,number_of_states);
last_stop=number_of_states/(2^((i-depth_of_trellis+L-2)*k));
for j=0:last_stop-1
branch_metric=0;
%binary_output=(ones(1,n)-2*deci2bin(output(j+1,1),n));
binary_output=deci2bin(output(j+1,1),n);
for ll=1:n
branch_metric=branch_metric+abs(decoder_in_matrix(ll,i)-binary_output(ll));
end
if((state_metric(nextstate(j+1,1)+1,2) > state_metric(j+1,1)...
+branch_metric) | flag(nextstate(j+1,1)+1)==0)
state_metric(nextstate(j+1,1)+1,2) = state_metric(j+1,1)+branch_metric;
survivor_state(nextstate(j+1,1)+1,i+1)=j;
flag(nextstate(j+1,1)+1)=1;
end
end
state_metric=state_metric,2:-1:1);
end
% generate the decoder output from the optimal path
state_sequence=zeros(1,depth_of_trellis+1);
state_sequence(1,depth_of_trellis)=survivor_state(1,depth_of_trellis+1);
for i=1:depth_of_trellis
state_sequence(1,depth_of_trellis-i+1)=survivor_state((state_sequence(1,depth_of_trellis+2-i)...
+1),depth_of_trellis-i+2);
end
decodeder_output_matrix=zeros(k,depth_of_trellis-L+1);
for i=1:depth_of_trellis-L+1
dec_output_deci=input(state_sequence(1,i)+1,state_sequence(1,i+1)+1);
dec_output_bin=deci2bin(dec_output_deci,k);
decoder_out_matrix,i)=dec_output_bin(k:-1:1)';
end
decoder_out=reshape(decoder_out_matrix,1,k*(depth_of_trellis-L+1));
cumulated_metric=state_metric(1,1);

 

[shameem]

hi ahmedseu,
thank u very much for giving me the code of soft viterbi decoder.

Can u please give me a document, so that i can understand it easily.

thanks

 

[mamerm77]

dear All,
could any one provide me by a master documentation in implementation of th eviterbi decoder in VHDL , or arch

 

[fm_com_28]

Dear,
it is possible if we can contact via email if u want.I have written a code for a 4state K=2 with rate1/2 hard decision viterbi decoder for the example mentioned in the first paper of Viterbi.
I will now begin to try another one but for soft decision , so we can interchange our knowledge in this field.
Best Regards

 

[yasiranwar86]

plz i need viterbi decoder implementation in c language plz if any boby can provide any kind of help to implement it on dsp processor