Blind synchronization technique for OFDM based on oversampling and the modified Bhattacharyya bounds

The evaluation of the Cramér-Rao bound (CRB) for problems of estimating deterministic parameters in the presence of random nuisance parameters has usually been hard, due to the total probability structure of the marginal likelihood function. Alternative bounds exist, but they are either not looser than the CRB, or apply only to restricted classes of estimators. In the first part of this dissertation, we present a family of tighter bounds, the modified Bhattacharyya bound (MBB) as extensions of an existing bound in the literature. Case studies on synchronization problems show that the MBB's can be much tighter than the Cramér-Rao bound (CRB). As an analogy, another family of bounds, the modified Bhattacharyya bound 2 (MBB2) is also derived for the class of uniformly unbiased estimators. Orthogonal frequency division multiplexing (OFDM) is being used or under consideration for a variety of new wireless applications. The communication channels for some of these applications are highly dispersive, requiring synchronization algorithms which are robust to channel dispersion. In the second part of this dissertation, we present a blind OFDM synchronization technique, the Minimum-Interference-Power (MIP) algorithms, which exploits the significant self-interference energy buildup in the presence of synchronization errors. The MIP algorithms generate timing and frequency offset estimates based on the FFT sequence of the oversampled received signal. For OFDM systems with virtual carrier (VC), the same principle can be applied based on the VC without oversampling the received signal. The VC-based MIP frequency estimator is identical to an existing blind frequency estimator. To mitigate the complexity issue, we propose a two-stage estimation structure as well as a variant of the MIP estimator. Performance study via simulation shows that the MIP estimators are very robust against channel dispersion, and produces excellent estimates. A three-step hybrid estimator which combines existing cyclic prefix (CP)-based technique and the MIP technique is presented and shown to have better performance than the both component techniques for SNR up to 32dB. The document concludes with some discussions on the use of MIP timing estimators for channel length determination.