Synchronization and Channel Estimation in OFDM Systems

This thesis consists of a prologue and six parts that address two aspects of orthogonal frequency-division multiplexing (OFDM) communication systems: synchronization and channel estimation. For the synchronization aspect, focus is on estimators of time and frequency offsets that exploit the cyclic prefix. For the channel estimation aspect, a concept is presented and analyzed that exploits the frequency correlation of the channel and combines low complexity with high performance. Synchronization errors in OFDM can cause intersymbol and intercarrier interference. One way to suppress these interferences in OFDM systems is to track the carrier frequency of the received signal and the proper start of the OFDM symbols. The joint maximum likelihood estimator of symbol time and carrier frequency offsets for OFDM systems using a cyclic prefix is presented. The maximum likelihood estimator, derived for an additive white Gaussian noise channel, exploits the redundancy introduced by the prefix and is independent of how the subcarriers are modulated. Therefore, it does not require extra pilot information. Some properties of practical OFDM systems change the premises of this estimator. In particular, in systems with pulse shaping, channel dispersion, or slowly changing synchronization errors, estimator performance can gain from more suitably chosen signal models. Three extended signal models, the associated maximum likelihood estimators, and structures for their implementation are presented. Furthermore, tracking synchronization of a multiuser OFDM system, which often has been questioned, is feasible using a scheme incorporating the above estimator concept. Simulations show that the symbol error rate in the uplink of a multiuser OFDM system is not noticeably affected in a system with coherently demodulated subcarriers and only little in a differentially demodulated system. Channel estimation is usually needed in coherent OFDM receivers in order to compensate for a frequency-selective channel. For this purpose, a channel equalizer needs to be fed estimates of the subcarrier attenuations. This thesis presents low-rank approximations of the linear minimum mean-squared error channel estimator. These exploit the fact that the subcarrier attenuations are correlated. The approximation is based on a suitable transformation of the channel attenuations. Channel estimators based on the discrete Fourier transform are analyzed. The symbol error rate of systems employing these channel estimators may experience an irreducible error floor if the channel is not sample-spaced. This thesis also presents and analyzes lowrank channel estimators using a transform based on singular-value decomposition of the channel attenuations. These channel estimators are proposed to be designed generically, i.e., for particular fixed, assumed channel correlations and signal-to-noise ratio. The resulting channel estimators, with suitably chosen fixed design parameters, combine lowcomplexity and high performance.