A new eigenstructure-based parameter estimation of multichannel moving average processes

A closed-form solution to the parameter identification of multichannel moving average processes is presented. An eigenstructure-based approach is proposed to solve the identification equations involving output cumulants of any order by further exploiting the eigenstructure of the output cumulant matrices. The identification equations are derived without using the Kroneker products, hence greatly reducing the computational complexity. This approach is also computationally simpler than that of L. Tong et al. (1991), and iterations are avoided. In addition, the proposed approach allows one to combine the statistics of different order to achieve better performance. Computer simulation has shown that a much smaller sample size, on the order of 5000, is needed to achieve even better performance than our previous approach