[Question] A problem originated in engineering: H is a (N*SF+L-1)*(N*SF) complex matrix and is generated with a Matlab command H=filter(h,1,[eye(N*SF);zeros(L-1,N*SF)]) where h is a L*1 complex vector. C is a (N*SF)*(N*K) complex matrix and is generated with a Matlab command C=kron(eye(N),c) where c is a SF*K matrix whose columns are Walsh codes. I find out with a Matlab program that C'*H'*H*C always has a better condition than H'*H. Is the observation general and may it be proven rigorously? [Difficulty] I failed to express the eigenvalues of H'*H and C'*H'*H*C. [Thoughts] The phenomenon may be observed with the following code. cond_s=[]; cond_S=[]; k=0; while k<10000 c=[1,1;1,-1;1,1;1,-1]; h=randn(3,1); H=filter(h,1,[eye(8);zeros(2,8)]); s=H'*H; cond_s=[cond_s,cond(s)]; C=kron(eye(2),c); S=C'*s*C; cond_S=[cond_S,cond(S)]; k=k+1; end all(cond_s>cond_S) is always TRUE.
Walsh codes help improving condition of matrix?
Started by ●February 4, 2007