Designing training sequences for carrier frequency estimation and adaptive predistortion with quadrature modulation/demodulation errors

Abstract

This thesis investigates the following three issues: designing training sequences (TSs) for frequency and channel estimation, an adaptive predistortion (PD) with quadrature modulation/demodulation (QM/QDM) errors, and a non-data-aided (NDA) I/Q mismatch estimation for low-intermediate (IF) receivers. First, a procedure for selecting a TS is developed for frequency estimation in frequency-selective channels. An expression for the unconditional Cramer-Rao bound (UCRB) is obtained by averaging the CRB for frequency estimation over the probability density function of Gaussian random channels, plus a necessary and sufficient condition for minimizing the UCRB is derived. Based on these results, a procedure for selecting TSs is developed. Through a computer search, binary TSs up to length 24 are found and tabulated. It is observed that periodic TSs tend to be selected when the TS length is twice the channel du-ration. Simulation results demonstrate that the proposed TSs can enhance the performance of the maximum-likelihood (ML) frequency estimate. Using the result mentioned above, a procedure for selecting a TS is developed for joint ML frequency and channel estimation. Conditions for minimizing the mean square error (MSE) of the joint estimate are derived, and a procedure developed for finding TSs. Computer simulation results demonstrate that the proposed TSs can enhance the performance of the joint ML estimate. Next, adaptive PD techniques are proposed that jointly compensate both the amplifier nonlinearity and QM/QDM errors, whereas for the QDM errors, followed by the extraction of the PD parameters. The proposed methods are derived based on the indirect learning architecture, which has been originally developed for PD. Two types of the PD schemes are considered: one is the memoryless polynomial and the other is memory polynomial. The proposed compensators update the parameter vector based on the recursive least squares (RLS) criterion. It is demonstrated through c...