# Error about inner matrix dimensions must agree

Hi,

I have a problem using CVX.

I would really appreciate if somebody could help

This is reproducible example.

This keep saying that inner matrix dimensions must agree, even though they does agree.

The below shows reproducible example.

clear all

N_CP=15;
M=2;
N_RU=6;
N=3;
U=2;
F=3;
D=20;
f_c=3.6* 10^9;
lambda_rf=(3* 10^8)/f_c;
B_fso=1* 10^9;
B_rf=40* 10^6;
f_s=B_rf;
N0_dBm=-174;
NF_dB=5;
sigma2_rf_dBm=N0_dBm+10*log10(B_rf)+NF_dB;
sigma2_rf_dB=sigma2_rf_dBm-30;
sigma2_rf=10^(sigma2_rf_dB/10);
c_fso=[5.0804,3.7454];

F_rf=randn(M*N_RU,N_CP)+j*randn(M*N_RU,N_CP)/sqrt(2);

for m=1:M
F_rf_m(:,:,m)=F_rf((m-1)*N_RU+1:(m-1)*N_RU+N_RU,:,1);
end

H_ml=(randn(M*N_RU,N+F)+j*randn(M*N_RU,N+F))/sqrt(2);
for m=1:M
H_m(:,:,m)=H_ml((m-1)*N_RU+1:(m-1)*N_RU+N_RU,:,1);
end

initalP_m_dBm=23;
initalP_m_dB=initalP_m_dBm-30;
initalP_m=10^(initalP_m_dB/10);
p_rate=0.5;
for m=1:M e)
W_m(:,:,m)=H_m(:,:,m)*inv(H_m(:,:,m)'*H_m(:,:,m));
W_m(:,:,m)=(sqrt(p_rate * initalP_m)/norm(W_m(:,:,m),'fro'))*W_m(:,:,m);
Q_m(:,:,m)=(sqrt(((1-p_rate) * initalP_m)/N_RU)) * eye(N_RU);
Z_m(:,:,m)=inv((W_m(:,:,m)*W_m(:,:,m)')+(Q_m(:,:,m)*Q_m(:,:,m)));
W(N_RU*(m-1)+1:N_RU*(m-1)+N_RU,:)=W_m(:,:,m);
Q(N_RU*(m-1)+1:N_RU*(m-1)+N_RU,:)=Q_m(:,:,m);
Z(N_RU*(m-1)+1:N_RU*(m-1)+N_RU,:)=Z_m(:,:,m);
end

cvx_begin

variable W_m(N_RU,N+F,M)  complex;
variable Q_m_lower(N_RU-1,N_RU-1,M) lower_triangular  complex;
variable Q_m_diagonal(N_RU,N_RU,M) diagonal
variable V_m(N_CP,N_CP,M) hermitian

Q_m=cvx(zeros(N_RU,N_RU,M));
for m=1:M
for t=1:N_RU
Q_m(t,t,m)=Q_m_diagonal(t,t,m);
end
end

for m=1:M
for x=2:N_RU
for y=1:x-1
Q_m(x,y,m)= Q_m_lower(x-1,y,m);
end
end
end

powerofmthRU=cvx(zeros(1,M));
R=cvx(zeros(N_RU,N_RU,M));
W=cvx(zeros(N_RU*M,N+F));
Q=cvx(zeros(N_RU*M,N_RU));

for m=1:M
W(N_RU*(m-1)+1:N_RU*(m-1)+N_RU,:)=W_m(:,:,m);
powerofmthRU(m)=sum(sum_square_abs(W_m(:,:,m)))+sum(sum_square_abs(Q_m(:,:,m)));
for t=1:N_CP
powerofsiganltomthRU(m)=sum(V_m(t,t,m));
end
Q(N_RU*(m-1)+1:N_RU*(m-1)+N_RU,:)=Q_m(:,:,m);
R(:,:,m)=chol(Z_m(:,:,m));
end

minimize sum(powerofmthRU)

subject to

for m=1:M
-sum(sum_square_abs(R(:,:,m)*W_m(:,:,m)))-sum(sum_square_abs(R(:,:,m)*Q_m(:,:,m)))+log_det(Z_m(:,:,m))+N_RU...
>= -log_det(eye(N_RU)+((F_rf_m(:,:,m) * V_m(:,:,m)*F_rf_m(:,:,m)')/sigma2_rf))-2 * log_det(real(Q_m(:,:,m)));

for j=1:M
if(m~=j)
F_rf_m(:,:,j)*V_m(:,:,m)* F_rf_m(:,:,j)' == zeros(N_RU);
end
end
end
cvx_end

Before cvx begin, they are all constant and initial value.

The really strange thing is if I erase either one of the two constraint written above,

it works fine. Only when the two constraints are written, there is error.

Thank you

This appears to be a CVX bug. I recommend you submit a bug report per http://cvxr.com/cvx/doc/support.html#bug-reports .