Research On Factor Graph Based Phase Estimation

The reliability and efficiency are the pursuit of communications, and the Shannon limit is approached as the development of channel coding such as Turbo code which decodes by iteration. However the impact of phase noise on the system performance emerges since the codes mentioned above usually work at a very low signal-to-noise-ratio (SNR). Besides some other implementations like satellite communications require strong phase noise cancellation. Thus phase estimation comes under the spot light, with papers proposing iterative code-aided phase estimation recently. And factor graph based phase estimation is dug into for it describes th system structure and carries out message passing, as well as iterative calculation well.Based on the existing research, this paper focuses on factor graph based phase estimation by message passing, with analysis on algorithm performance, including:1) Introduction of communications system model, with focus on the constant and random-walk phase models. Simulation results give a clear comparison of the influence of b(?)h models on the original symbols.2) Introduction of factor graphs. Factor graph based marginals, message passing, and s(?)m-product algorithm (SPA) are given and appraised by comparison with mathematical w(?)ys. And message representation on factor graph is appended.3) Concentration is given to the phase estimation part, with constant phase FFG and random-walk FFG presented. Firstly, numerical integration (NI) on both models is introduced and a hierarchy quantification NI is proposed due to the contradiction of quantification level and calculation complexity of the original NI. The proposed NI reduces the calculation c(?)plexity with no claim on performance. Secondly, gradient method on constant and random-walk phase FFG's is presented and a simplified gradient method on random-walk p(?)e model is proposed to tackle the instability, calculation complexity, and reliance on a(?)cent estimates of the original method. The simplified method claims no performance with reduction on calculation. Thirdly, EM algorithm on both models is given and a simplified EM o(?) random-walk phase model is proposed to cope with the instability, calculation complexity, and eliance on adjacent estimates of the original. The simplified EM reduces calculation with a(?)st no claim on performance. Simulation proves the analysis and algorithm performances.