Channel estimation and equalization for OFDM system in sparse time-variant channel

Channel impulse response (CIR) is supposed to be second-order polynomial (SOP) variation within two neighboring blocks with each consisting of several OFDM symbols for NDS<0.1, where NDS denotes normalized Doppler spread. Subsequently, channel estimator ML+SOPI is proposed, where ML denotes maximum likelihood used to estimate the CIR of pilot sequence (PS) whereas SOP interpolation (SOPI) is to compute CIR of each symbol by these estimated CIRs of three closest PSs. There are three implementations for SOPI: scheme A employs a symbol-level SOPI to compute CIR of each symbol where CIR is viewed as constant during single symbol; scheme B gets the CIR of symbol midpoint by point-level SOPI, which represents the CIR of entire symbol; scheme C uses the same point-level interpolation as scheme B, but CIR of each symbol is approximated as linear varying not constant, and a zero-forcing equalizer is used. To further improve performance, threshold matrix (TM) is introduced into these schemes. From simulation, we find: a) TM brings about 2dB SNR gain for three schemes; b) when NDS approaches 0.1, schemes B and C are equivalent and 1.5dB better than scheme A at 10
⁻⁴
BER; c) for NDS<0.01, they has the identical performance. Scheme B with TM is preferred for sparse time-variant (STV) channel due to its low complexity.