A linear least squares smoothing approach is proposed for the joint order detection and blind channel estimation. By exploitin the isomorphic relation between the observation and input spaces, a new geometrical approach to the blind estimation of multichannel moving average processes is developed. The proposed joint order detection and channel estimation method has the finite sample convergence property in the absence of noise. Minimizing the least squares smoothing error by jointly choosing the channel coefficients and the channel order, the proposed algorithm offers improved performance and robustness over existing methods.
展开▼
