Single Channel Blind Separation Of Time-Frequency Overlapped Communication Signals

Blind separation of communication signals, which has wide applications in signal detection and spectrum management, is a hot research topic in the area of blind signal processing. With the rapid development and wide application of wireless communication technologe, spectrum resources are increasingly scarce and the electromagnetic environment is more complex. The receiver equipped with only one antenna receives multiple signals simultaneously, which happens often in non-corperative communication scenario. Consequently, some challenges are brought to signal processing and information acquisition. As the co-channel signals are time-frequency overlapped, the traditional time or frequency domain filter can’t separate signals effectively. Blind separation or extraction of these signals from the mixed signal is significant.This dissertation focuses on blind sepation algorithms of time-frequency overlapping signals from the viewpoint of pratical applications. First of all, modulation identification and source number estimation for single channel mixed communication signals are studied, which is the basis of blind signal separation. And then, blind separation algorithms is investigated for two different cases, mixed signals with same symbol rate and mixed signals with different symbol rate. The main works and contributions of the dissertation are listed as followes.1) A novel method for modulation identification and source number estimation is proposed for single-channel mixed communication signals. Firstly, cycle frequencies of second-order and fourth-order cyclic cumulants for digital modulation signals are investigated. Based on difference of cycle frequencies features between different modulation signals, the modulation identification algorithm is presented. The proposed approach doesn’t need the prior information such as power, carrier frequency, symbol rate, time recovery and so on. And then, carrier frequency and symbol rate estimation of component signals is studied. Carrier frequency and symbol rate estimation can be approximately as a frequency estimation problem of sinusoidal signals. An interpolation method is utilized to improve the estimation results obtained by spectral lines. The estimation precision is significantly improved.2) Two cyclic auto-correlation based time estimation methods are presented for single-channel mixed communication signals, which is based on the relationship of cyclic autocorrelation functions and signal’s modulation parameters. The two different methods are designed for scenarios of unknown and known shaping filter of both signals respectively. The methods treat both co-channel signals as desired signals. Firstly, cyclic autocorrelation function of the received mixed signal is analyzed, and then the linear equations on time delay parameters are established for the two scenarios. Secondly, the closed-form solution for time delay parameters of both co-channel signals is obtained by solving the equations, and two estimation algorithms are presented. The Cramer-Rao bound is also analyzed and performance of the two methods is demonstrated by simulation results.3) For the separation of mixed signals with same symbol rates, a single channel signal separation method is proposed based on lattice reduction and decision feedback. The idea behind the proposed algorithm is to form a multi-channel multi-user detection model through over-sampling and then adopts a decision feedback detection method based on lattice reduction. The performance of the proposed method is close to the performance of the jointly maxmum likelihood symbol detection method while the complexity is dramasticly decreased.4) For the separation of mixed signals with different symbol rates, a periodically time-varying based blind separation approach is proposed. As the symbol rates of each signal is different with each other, signal’s modulation parameters can be estimated by cyclic statistics. To solve the problem of signal separation, a periodic multi-in multi-out model is constructed by over-sampling. Based on the model, lattice reduction aided SIC detection is utilized for symbol detection. Finally, computer simulation results are provided to demonstrate the performance of the proposed methods.5) A prototype system of single channel blind separation of time-frequency overlapping communication signals is designed. The algorithm of this dissertation is verified by a great number of tests conducted in the prototype system. Test results showed the effectiveness of the proposed algorithm in this dissertation.