how to level up TURBO decoder performance

HI all.

I am working with MATLAB R2011b, and have a problem with turbo decode. It work well with small data block (only less than s 20 characters per input) when length of data increasing it will less accuracy of decoding.

I have try with both system object of Mat lab and external code from CML library but don't have any improvement.

My example is here http://www.mediafire.com/?bib485vw1816doq. Any one can help have any way to correct that.

Thanks.

Any one help me.

My example is

>> data = round( rand( 1, 1000 ) );
>> code = f_turbo_code(data);
>> de_code = f_turbo_deco(code);
>>  err_r = sum(xor(data,de_code))

err_r =

       178

and my code as following:

function out = f_turbo_code( in )

        % The generator polynomials we are using
        feedback_polynomial = [1 0 1 1];
        feedforward_polynomial = [1 1 0 0];

        %interleave matrix
        interleave_matrix = [1:1:length(in)];

        %Make polynomial
        genPoly = [feedback_polynomial; feedforward_polynomial];

        % upper out put for original data
        upper_output = ConvEncode( in, genPoly, 0);

        %lower input data is interleaved version of upper
        lower_output = ConvEncode( [in(interleave_matrix)], genPoly, 0);

        % put out put with first upper and second one with rate 1/3
        out = [upper_output(1:2:end) upper_output(2:2:end) lower_output(2:2:end)];
    end

function out = f_turbo_deco( in )

        % The generator polynomials we are using
        feedback_polynomial = [1 0 1 1];
        feedforward_polynomial = [1 1 0 0];

        %Make polynomial
        genPoly = [feedback_polynomial; feedforward_polynomial];

        turbo_iterations = 10;
        number_tails = 3;

        %interleave matrix
        interleave_matrix = [1:1:(length(in)/3-number_tails)];

        % split data from  input

        x1 = [in(1:(length(in)/3))];
        x2 = [in((length(in)/3+ 1):(length(in)/3*2))];
        x3 = [in((length(in)/3*2+ 1):(length(in)))];

        input_upper_c = zeros(1 , 2*length(x1));
        input_upper_c(1:2:end) = x1;
        input_upper_c(2:2:end) = x2;

        input_lower_c = zeros(1 , 2*length(x1));
        x1_data = [x1(1:(length(x1)-number_tails))];
        x1_tails = [x1((length(x1)-number_tails+1):length(x1))];
        input_lower_c(1:2:end) = [x1_data(interleave_matrix) x1_tails];
        input_lower_c(2:2:end) = x3;

        % estimate of original data
        data = [x1(1:(length(x1)-number_tails))];
        input_upper_u = zeros(1, length(x1)-number_tails);

        %Automatically stop when no more errors
        autostop = 1;
        % decoding        

        for turbo_iter=1:turbo_iterations
            % Pass through upper decoder
            [output_upper_u output_upper_c] = SisoDecode( input_upper_u, input_upper_c, genPoly, 0, 0 );

            % Extract Extrinsic information
            ext = output_upper_u - input_upper_u;

            % Interleave this information, which organizes it in the same manner
            % which the lower decoder sees bits
            input_lower_u = ext(interleave_matrix);    

            % Pass through lower decoder
            [output_lower_u output_lower_c] = SisoDecode( input_lower_u, input_lower_c, genPoly, 0, 0 );

            % Interleave LLR - not part of turbo algorithm, we interleave it so we
            % can measure the mutual information, since it needs to match the order
            % of bits in data. 
            interleaved_output_lower_u = output_lower_u(interleave_matrix);    

            % Interleave and extract Extrinsic information
            input_upper_u = output_lower_u - input_lower_u;         
            input_upper_u = input_upper_u(interleave_matrix);

            % Hard decision based on LLR: if < 0 bit is 0, if > 0 bit is 1
            detected_data = (sign(interleaved_output_lower_u) + 1) / 2;
            error_positions = xor( detected_data, data);

            % exit if all the errors are corrected
            temp_errors = sum(error_positions);

            if (temp_errors==0) && (autostop)
                break;
            end        
        end 
        out = detected_data;
    end

and ConvEncode and SiSoDecode is take from CML file.

And as you see the performance is very low, it take about 200 error per 1000 character.

So my question is any way to make a better one. I ready try with system object from R2011b but it don't have any improve

[Walter Roberson]

When .rar files are posted, you lose most of the regular volunteers. Some of them do not bother to download archives at all; some of them cannot handle .rar specifically.

If your code is a reasonable length it is best to post it here. (And if it isn't a reasonable length then the volunteers are likely to say "That's too much like work; isolate the problem more and get back to us.")

[Ryan G]

Try using the MATLAB Profiler to find out where it is taking the longest, then try to DIY from there. If you have an issue at that point you can come back with a more specific line that is the issue.

[Walter Roberson]

I don't think efficiency is the issue: it appears to me that the question is about ability to reconstruct the original data stream. That points to a problem in the encoder or a problem in the decoder, but I do not know enough about turbo coding methods to trace the problem in the time I have available.