UC San Diego

This dissertation is concerned with digital communications systems operating over channels that vary in time due to mobility of the transmitter or receiver. Velocity of the transmitter or receiver causes a Doppler shift or smearing in frequency which hinders parameter estimation and degrades data detection performance. To aid in our goal of improved estimation and detection performance, we discuss several characterization methods, both probabilistic and deterministic, that can accurately model the time-varying channel and that enable efficient parameter estimation and data detection. The first topic we address is that of carrier frequency offset estimation in the presence of a time-varying frequency-selective channel. The scheme is data-aided, meaning that a block of pilot data known to the receiver is first transmitted through the channel. Joint estimators of both the CFO and the time-varying frequency-selective channel are developed using both probabilistic and deterministic channel models. The probabilistic CFO estimator is shown to achieve the Cramer -Rao Lower Bound, and the deterministic estimator can perform well, but depends heavily on the choice of channel parameterization in the deterministic model. The second topic we focus on is concerned with blind data detection over time-varying channels. The problem we consider is recovery of a short block of time-domain data symbols transmitted over a channel with time-varying multiplicative distortion. The receiver does not know the data, the channel, or any statistics that could parameterize a probabilistic description of the channel. We design data detection algorithms based on a deterministic parameterization of the channel that can be implemented efficiently with the sphere decoding algorithm. Additionally we provide techniques for a priori selection of the channel parameterization that result in near-optimal performance over a wide range of channel conditions. In our final topic we consider blind detection over a frequency-selective time-varying channel. A short block of symbols is transmitted in the frequency domain, as either an OFDMA frequency allocation or as a subset of a larger OFDM symbol. The multiplicative distortion across the symbols is varying, due to frequency-selectivity. In addition, the transmitted symbols interfere with one another due to the channel time-variation, manifested as intercarrier interference. We introduce a novel method of correlatively coding the symbols across frequency before transmission that reduces the power of the ICI at the receiver. The scheme results in transmitted symbols that are constant in magnitude, which enables the application of efficient detection algorithms also based either probabilistic or deterministic parameterization of the channel. The resulting blind detector is efficient and can outperform an OFDM system which knows the channel perfectly