Estratégias de estimação para canais variantes no tempo em sistemas OFDM

Abstract

This dissertation deals with the estimation of time-selective fading channels in OFDM (Orthogonal Frequency Division Multiplexing) coding systems. We considered that the channel obeys the WSS-US (Wide Sense Stationary–Uncorrelated Scattering) model, with an integral number of multi-paths. In presence of time-selective fading, we showed that the sub-carriers are the DFT (Discrete Fourier Transform) of the average of Nc channel samples in time-domain, where Nc is the number of employed sub-carriers. Under the assumption that the channel obeys the TDL (Tapped Delay Line) model, and the transmitted symbols at each sub-carrier have the same power, we also showed that the channel power is divided between the sub-carrier and ICI (Inter-Carrier Interference) power. For a grid placement of the pilot sub-carriers, we derived a MMSE (Minimum Mean Square Error) estimator that is able to exploit both time and frequency correlations. Supposing the sub-carrier correlations can be separated as the product between the time and frequency correlations, the filtering structure of the IIR (Infinite Impulse Response) estimator can be separated as the estimation at the pilot positions and the interpolation over time and frequency domain. Just taking into account that the maximum Doppler frequency and the channel and sub-carrier power are known at the receiver, we proposed a robust estimator that does not require the estimation of channel correlations and noise variance. We considered an adaptive estimator derived from the filtering structure obtained in the MMSE estimator. The filtering section that exploits the frequency-domain channel correlations is estimated by the LORAF (Low-Rank Adaptive Filter ) and OPAST(Orthogonal Projection Approximation Subspace Tracking) algorithms. The filters that exploits the time-domain channel correlations are estimated adaptively by an algorithm based on QR decomposition, which supports a fast version, whose computational complexity has the order of the filter length. All the analyzed and proposed estimation techniques are compared by computer simulation.