Multi-channel autoregressive modeling through orthogonal projection

A multi-channel autoregressive (MAR) modeling algorithm is introduced. The new method treats MAR modeling as a vector orthogonal projection, where the optimal MAR coefficient matrices lead to prediction error vectors whose linear dependency upon available measurement vectors is minimized. The standard Gram-Schmidt orthogonal transform was extended to multi-channel time series and utilized to calculate the MAR coefficient matrices. Simulation results show that multi-channel power spectra thus derived from the orthogonal projection method exhibit good frequency resolution, without line splitting and frequency bias, and the coherence was also accurately estimated