[讨论]关于自适应判决反馈均衡器的程序问题（DFE），请高人指点

本人写的DFE均衡的LMS算法曲线，不知道问题出在哪里？
希望各位高手能指点一二

DFE的判决输出数学上到底如何表示呢？
本文中，s(k)=sign(z(k))的作法正确吗？
s为dfe输出
另，附件为原理框图

% DFE LMS EQ

clear;
itn = 1e3; % # of data samples

sigman2 = .01; %noise variance (=1/SNR)
sigman=sqrt(sigman2);
ch = [-.1 .3 -.7 2 2 -.7 .3 -.1]; % channel coefficients

Nf = 12;
Nb = 4;
N = Nf+Nb;

mu=0.01; % step size

runs = 100; % number of independent triels to
mse=zeros(itn,1);

tic

for j=1:runs

x=randint(itn,1)*2-1; % random 1,-1
r=filter(ch,1,x)+sigman*randn(itn,1);

f=zeros(Nf,1);
b=zeros(Nb,1);
weights=zeros(N,1);

for k=N:itn %DD-LMS

R=r(k:-1:k-Nf+1);
S=s(k:-1:k-Nb+1);

z(k) =R'*f-S'*b;
s(k)=sign(z(k));

end

e(k)=s(k)-z(k);
f=f-mu*e*R;
b=b+mu*e*S;
weights=[f',b']';

mse(k)=mse(k)+e(k).^2;
end;
end;

toc

mse=mse/runs;
semilogy(mse)
xlabel('No. of iterations')
ylabel('MSE')