This thesis addresses the problem of blind channel identification and equalizatlon of digital communication signals using the second-order cyclostationarity present in the received data. This problem can be rephrased as that of determining the impulse response/frequency response of a communication channel (i.e. channel identification) using only distorted and noisy estimates of the received data. There is no direct knowledge of any training or pilot sequence (thus the method is "blind"), although there is implied statistical information about the transmitted data (second order cyclostationarity). Once the channel identification is performed, a filter (or equalizer) can be applied to the received signal. Blind channel identification and estimation algorithms are described for both moving average (MA) and autoregressive (AR) system models (including a new method that is not strictly blind which uses prior knowledge of the transmitter shaping pulse). Also, Cramer-Rao lower bounds (CRB) on the performance are developed for both the blind case and the case using prior knowledge of the transmit pulse (a discussion of the drawback of this bound for this particular problem is also given). It is shown that the CRB can be orders of magnitude lower if prior knowledge of the shaping pulse is exploited (as it could be in commercial applications). Identifiability issues for algorithms which use second-order cyclostationarity are briefly discussed as well as implementation details such as synchronization and model order estimation (which are addressed via ad-hoc algorithms). The overall performance of these algorithms is characterized using a set of Monte Carlo simulations, where it is shown that these blind methods can in some cases perform as well as non-blind methods using training sequences. The algorithms are also compared where applicable to the previously derived performance bounds.
展开▼
