Abstract
Blind identification (BID) of multiple input multiple output systems is purposed to estimate an unknow system by using only the measurements. In general, the approaches of BID can be partitioned into two categories─ second-order statistics based algorithms, and higher-order statistics based algorithms.The advantages of second-order statistics based algorithms are shorter data lenght requirement, existence of closed form solution while the disadvantages are sensitivity to noise, requirement of prior imformation about channel lengths, more restrictions on systems. The advantages of higher-order statistics based algorithms are insensitivity to noise, less restrictions on systems, while the disadvantages are longer data length requirement, high computational load.Gorokhov and Loubaton have proposed a second-order statistics based subspace method for blind identification of an unknown K-input P-output (P either larger than or equal to K and K larger than 1) FIR system with the K input signals being mutually independent and temporally independent identically distributed (i.i.d.). However, only a partial system estimate can be obtained using their method. That is, the system estimate will be the unknown system multiplied an unknown ambiguity matrix R(z). In the case of equal channels, R(z) will be a constant nonsingular matrix and in the case of different channel lengths, R(z) will be a nonsingular upper triangular matrix. This thesis is concerned with resolving the ambiguity matrix R(z) embedded in the system estimate obtained by Gorokhov and Louba-ton’s subspace method. An approach is proposed in this thesis by utilizing Chi and Chen’s fast kurtosis maximization algorithm to estimate the inverse of the ambiguity matrix; meanwhile, the unknown system can thus be perfectly identified.