Abstract
Blind identification (BID) of a multiple-input multiple-output (MIMO) linear time-invariant system is a problem of estimating the unknown system with only a set of non-Gaussian vector output measurements of the system. Most of the existing BID algorithms are developed under the noise-free assumption and require stringent conditions on the unknown MIMO system as well as on the spectra of the driving inputs of the system. With noise effects taken into account, this thesis studies blind single-input multiple-output (SIMO) and MIMO system identification with noisy measurements. Based on the relationship between the unknown system and the optimum equalizer associated with the higher-order statistics based inverse filter criteria (HOS-IFC) proposed by Tugnait and Chi et al., a BID algorithm for an SIMO system driven by a temporally independent input signal is proposed together with the proof of the identifiability for an SIMO system using the cross-spectral matrix of the measurements and the optimum equalizer associated with the HOS-IFC. Then, a two-step BID algorithm for an MIMO FIR system driven by spatially independent and temporally colored or independent inputs is further proposed using the HOS-IFC and Qiu et al.'s greatest common divisor computation algorithm together with the proof of the system identifiability. The proposed two-step BID algorithm only requires coprimeness of the components in each column of the system transfer function and is thus more flexible than most of the existing BID algorithms. On the other hand, some applications of the BID algorithms for MIMO systems using the HOS-IFC are also presented in this thesis including simultaneous estimation of multiple time delays, blind beamforming, blind source separation in multipath and multiuser detection for asynchronous direct sequence/code division multiple access (DS/CDMA) systems using multiple antennas.