In modulation schemes using orthogonal signaling, one can make the error probability arbitrary small by increasing the symbol alphabet size M at the cost of increased bandwidth. However, without the penalty of the increased bandwidth, we can improve the error rate performance of the modulation schemes when we employ L-orthogonal signaling(LOS), which we call the orthogonal signaling(OS)and the biorthogonal signaling(BS) when L=1 and 2, respectively.
Detection of LOS can be accomplished in either a coherent or a noncoherent manner. The ideal coherent detection requires accurate knowledge of the channel. On the other hand, noncoherent detection is typically used in situations where it is difficult to obtain reliable channel estimate. Although the noncoherent detection eliminates the need for channel estimate, it suffers from performance penalty compared to the ideal coherent detection. For example, at BER of $P_b = 10^{-3}$, binary OS(2FSK) requires about 1.2dB more $E_b/N_o$ when we use the noncoherent detection instead of the coherent one in AWGN channel.
To achieve better performance in noncoherent detection, Divsalar and Simon proposed the multi-symbol differential detection technique for MPSK based on maximum likelihood sequence estimation(MLSE) and derived a union bound on the bit error probability. They also examined the technique for full-response continuous phase modulation. Recently, with the increasing interest in application of LOS to mobile radio communication, there have been reported multi-symbol receiver structures for OS and performance analysis results of the noncoherent LOS over Rayleigh fading channels.
In this dissertation, we first propose an efficient noncoherent detection scheme for biorthogonal signals, which uses multi-symbol observation interval. Then we analyse the performance of the proposed detection method and verify the performance with the simulation results. Instead of adopting the symbol-by-symbol detection as in the previous noncoherent detection methods, the proposed method jointly estimates both the demodulated data and the channel only from the received signal. In the joint estimation process, we can reduce the computation complexity by using approximate maximum-likelihood sequence estimation. As a result, the proposed detection scheme achieves performance close to that of the ideal coherent detection with perfect channel estimates. When we select the appropriate observation symbol interval N in the given symbol alphabet size M according to the available bandwidth, we can implement the proposed detection scheme with the comparable computational complexity to that of the ideal coherent detection scheme. The analysis as well as simulation results show that the required average signal-to-noise ratio per bit $γ_b$ can be reduced by as much as 1.4dB and the capacity can be increased by as much as 38% when we use the proposed system in the CDMA cellular reverse link.
