Implementing ambient backscatter communication system

I have developed the Matlab code for Max likelihood detector from the research paper. My problem is that the code is running but my BER curve is not correct i.e, my BER values sometimes decrese or increase. The title of the paper is "Semi-coherent Detection and Performance Analysis for Ambient Backscatter System".Thanks in advance. Also attaching the image of algorithm.Optimal detector

`clc;
clear all;
close all;

% Initialization


Nw = 1; % noise variance
M = 50 ; % symbol period of the tag
Mt = 4; % training symbols 
al = 0.5 ; % alpha is tag coefficient
L = 1000;
N_mc = 10^6;
N = 50; % comsecutive samples of s(n)
RCD = 0.5;% Relative channel difference.
SNRdB = 0:5:20; 

% source 
sn = sqrt(1/2)*(randn(1,L)+1i*randn(1,L));
Ps = 1; % RF source signal power


% channels
hst = sqrt(1/2)*(randn(1,L)+1i*randn(1,L));
hsr = sqrt(1/2)*(randn(1,L)+1i*randn(1,L));

vn = 10; % var of htr
htr = sqrt(vn/2)*(randn(1,L)+1i*randn(1,L));

wn =  sqrt(1/2)*(randn(1,L)+1i*randn(1,L));

ho = hsr;
ho_abs = sqrt(sum(abs(ho).^2));
h1 = (hsr)+ (al*(hst.*htr));
h1_abs = sqrt(sum(abs(h1).^2));


% Thresholds
so2 = ((ho_abs)^2*Ps)+Nw;% sigmao square

s12 = ((h1_abs)^2*Ps)+Nw; % sigma1 square

ThCG = ((N*so2*s12)/(s12-so2))*log(s12/so2); % threshold for complex gaussian optimal detector.

% Reader
I = eye(N);
vn_yo = so2*I;
y_Ho = sqrt(vn_yo/2).*(randn(N,N)+1i*randn(N,N)); % matrix

vn_y1 = s12*I;
y_H1 = sqrt(vn_y1/2).*(randn(N,N)+1i*randn(N,N)); % matrix

y_bar_H1 = diag(y_H1);
y_bar_Ho = diag(y_Ho);
z = norm(y_bar_H1);
Z = z^2;


Nid = 1000; % number of incorrect bit decisions

for i = 1:length(SNRdB)
errcnt = 0;
bitcnt = 0;
while errcnt < Nid
    d = (rand > 1/2);
        if d == 1
        y_bar_H1 = diag(y_H1);
        end

         if d == 0
            y_bar_Ho = diag(y_Ho);
         end
if so2 > s12
    if Z > ThCG
        dhat = 0; 
    else
        dhat = 1;
    end % for Z>ThCG
else
    if Z < ThCG
        dhat = 0;
    else
        dhat = 1;
    end
end

if dhat ~= d
     errcnt = errcnt+1;
end
     bitcnt = bitcnt+1;  
end
BER(i) = errcnt/bitcnt;  
end

figure
grid on;
semilogy(SNRdB, BER);
`