My LDPC BER curve is bad

Hello,

For my Masters Thesis, I am trying to simulate Quasi Cyclic Low-density Parity-Check Codes (QC - LDPC) codes on Matlab. Here are the details regarding it.

Parity Matrix - Quasi Cyclic LDPC (93 x 155)

Encoder - Approximate Lower Triangular Encoding (rearrange matrix to ALT form)

Decoders - Sum-Product and Min Sum Decoders.

Message Size - 62 bits

Modulation - BPSK

Channel - AWGN

I believe that I have properly coded. And I have been unable to track where the mistake is happening. But for some unidentifiable reason, by error rates are not what they're expected to be. I have attached the BER curve for reference. Any sort of help would be really appreciated. I have been plucking out all my hair for the past 20+ days failing to create a proper simulation. Thanks in advance.

clear
clc

[bb, Ha] = qcldpc(3, 5, 2, 5, 31);                          %Creates PCM. 

H = alt(Ha);                                                %Rearrange PCM to ALT form for Encoding


[N1, N2] = size(H);
k = N2-N1;

pcs =1;
while (pcs ~= 0)
    y = round(rand(1, N2-N1));  
    encode = enc(H,y);
    msg = [y encode]; 
    cs = mod(msg * H', 2);                                  %to detect a proper codeword
    pcs = sum(cs);
end
                                          
[~, bits] = size(msg);

dB = 1:5;                                                   % range of SNR values in dB

R = k/N2;                                                   % code rate 

iter = 10;                                                  %Iterations
N0 = (10.^(-dB/10));                                        %Variance values

for i = 1 : length(dB)
        ber1(i) = 0;  ber2(i) = 0;
        

        for j = 1:bits
            mod = bpsk(msg);                                %bpsk modulation
            bp = mod';
            tx = bp + sqrt(N0(i)/2) * randn(size(bp));      %Transmitted with AWGN
            dp(i,:) = dbpsk(tx);                            %Demodulated
            Li = -dp(i,:);
            vhat1(i,:) = SPA(tx, Ha, N0(i), iter);          %Decoding Using Sum Product Algorithm
            vhat2(i,:) = MSA(Li, Ha, iter);                 %Decoding Using Min Sum Algorithm
            [n1, r1] = biterr(vhat1(i,:), msg);   
            [n2, r2] = biterr(vhat2(i,:), msg);
            ber1(i) = ber1(i) + r1; 
            ber2(i) = ber2(i) + r2;
        end
        
        BER1(i) = ber1(i) / (bits * length(dB));            %BER Calculation
        BER2(i) = ber2(i) / (bits * length(dB));
   
end

figure('Name','BER vs SNR','NumberTitle','off');
semilogy(dB, BER1, 'o-');
hold;
semilogy(dB, BER2, 'o--');
legend('SPA','MSA')
grid on;
hold off;

[Rahul Raj]

Can you tell what method you are using for creating a quasi-cyclc LDPC code matrix or parity check matrix?