Optimal pilot pattern for sparse channel estimation in TFT-OFDM systems

Recently, a new orthogonal frequency division multiplexing (OFDM) scheme called time-frequency training OFDM (TFT-OFDM) has been proposed. Compared to the conventional cyclic prefix OFDM (CP-OFDM) or time-domain synchronous OFDM (TDS-OFDM) which only relies on either frequency-domain pilots or time-domain pseudo noise (PN) sequence for channel estimation, the proposed TFT-OFDM uses PN sequence to obtain the channel sparsity level, and then the accurate channel impulse response (CIR) can be estimated by frequency-domain pilots. According to the compressive sensing (CS) theory, the pilot pattern has deterministic influence on the channel estimation perfromace. However, the pilot pattern has not been studied in the papers that proposed TFT-OFDM. In this paper, we introduce an optimal pilot pattern in TFT-OFDM. The method is based on minimizing the column correlation of matrices composed of rows of the discrete Fourier transform (DFT) maxtrix. The indices of the chosen rows are determined by the pilot pattern. The optimal pilot pattern can be generated from the optimal cyclic difference set (CDS). When the optimal CDS does not exist, we introduce an iterative method based on greedy algorithm to obtain a suboptimal pilot pattern.