Low-complexity ML channel estimation schemes for OFDM

Orthogonal frequency division multiplexing (OFDM) with cyclic prefix allows for low-complexity frequency domain equalization of frequency selective fading channels when the channel frequency response is available at the receiver. The channel frequency response can be obtained by using the frequency correlation properties or equivalently the finite delay spread of the channel at the cost of increased computational complexity. Significant portion of this increased complexity is due to multiple conversions from time domain to frequency domain and vice-versa. In this paper we derive the conditions and approximations under which the maximum-likelihood (ML) estimate of the time domain channel can be obtained directly from the received signal thus avoiding multiple time-frequency domain conversions. The frequency response can then be obtained by taking a single (partial) FFT. In doing so, we obtain a fundamental constraint on the number of pilots, N_u, used to estimate a channel with delay spread D. We show that N_u ≥ D for error-free estimation in the absence of noise. Simulations for 802.11a systems are then presented to compare the different estimation schemes in terms of PER.