Near-optimum training sequences for OFDM systems

We propose a genetic algorithm based approach to obtain low PAPR near-optimum training sequences for channel estimation in an OFDM system. The use of a genetic algorithm is proposed since the search space rises exponentially as the number of sub-channels employed increases. In addition constant envelope time domain training sequences, with a PAPR of 0dB are also investigated with the aim of taking advantage of higher peak transmit powers and so obtaining an improved channel estimate. It is found that a better BER performance is achieved using an optimised constant envelope time domain training sequence compared to that achieved when using an optimised conventional low PAPR training sequence. The near-optimum training sequence for the former approach is found to be one with a relatively high power on each frequency domain sub-channel.