"If x is an M × N signal matrix representing N channels in its columns, then xcorr(x) returns a (2*M – 1) × N^2 matrix with the autocorrelations and mutual cross-correlations of the channels of x"
The only time that could give you a scalar output would be if M = 1 and N = 1 -- that is, if you had a scalar input. But unless your Pm is empty, [Pm,g] is going to be a vector, and so the minimum output size you could get would be a 3 x 1 column vector (and might be larger, if Pm is larger.). You cannot store a 3 x 1 column vector into the single location r(j)
