Techniques for blind system identification and application to the equalization of PAM/QAM modulated signals

Haralampos Pozidis

doi:10.17918/00008465

We consider the problem of blind system identification, that is, estimating an unknown system excited by an unknown input, based solely on the system output. This problem appears in applications such as speech recognition, image restoration and data transmission. In this study we focus on digital data transmission. In modern digital communication systems which transmit millions of symbols per second from dense constellations, smearing of adjacent symbols is inevitable due to the dispersive nature of the propagation medium. This distortion effect is termed Intersymbol Interference (ISI). ISI (channel) identification and equalization are necessary to compensate for distortion. Two classes of techniques for blind channel estimation exist: Higher Order Statistics (HOS) methods applied to the received signal, and Second Order Statistics (SOS) methods applied to oversampled versions of the received signal (temporal diversity), or to multichannel versions of it obtained by several antennas (spatial diversity). Each class has different characteristics and limitations: HOS-based methods require long data and usually exhibit high complexity, while SOS-based ones are sensitive to noise and require knowledge of channel length. In this dissertation we propose novel methods in both classes, aimed at addressing the limitations of each class. We first propose a SOS-based channel estimation method that utilizes the information embedded in the Fourier-phase of the cross-spectrum of functions of two observations. Approximate analytical expressions of the statistics of the estimation error are provided. Both analysis and simulation results indicate that the proposed approach possesses the desirable features of data efficiency, robustness to noise, and insensitivity to channel length mismatch. We next propose a HOS-based method, which achieves channel identification based on HOS of the received signal. The key feature is the requirement of a pair of HOS slices as opposed to the entire HOS. This allows for significant computational savings. We establish channel identifiability conditions, and show that appropriate selection of slices leads to reduction of estimation variance. We also propose a criterion for selection of "good' slices. Both our SOS and HOS methods are tested on channels extracted from actual microwave measurements, and demonstrate robustness in highly dispersive and noisy environments.

Techniques for blind system identification and application to the equalization of PAM/QAM modulated signals

Files and links (1)

Abstract

Metrics

Details

Techniques for blind system identification and application to the equalization of PAM/QAM modulated signals

Files and links (1)

Abstract

Metrics

Details

Drexel University Social media