Graph based iterative detection and channel estimation for sparse frequency selective channels

Kaygusuz, Ahmet Baran
The wireless channel is not only susceptible to noise and interference but also other channel impediments like multipath spread, Doppler spread, Doppler shift, and these impediments may change over time. Iterative receiver structures are commonly studied and utilized as they can be very effective in mitigating detrimental effects of the wireless channel. Iterative detection that performs decoding and estimation jointly can be realized in various ways. However, some methods depend on the memory of the channel and hence become computationally complex especially for sparse frequency selective channels. In this thesis, a factor graph based iterative detection is investigated with channel estimation for sparse channels. The sum-product algorithm applied on a factor graph attains optimum or near-optimum performance while its complexity increases linearly with the number of nonzero channel taps in such channels. We study the performance of graph based detection with different channel estimation schedules. Our results indicate that the method studied here enables operation at higher Doppler spreads and possibly with shorter training sequences.