Theoretical And Simulation Studies Of Detectors For Non-gaussian Noise

The noise sources in many communication systems such as wireless ad hoc sensor networks and ultra-wideband(UWB)communication are not Gaussian,as a result of human behavior including car ignition,mechanical switching and so on.There have been different and popular statistical distributions,instead of Gaussian distribution proposed to model such impulsive noise such as the symmetric a-stable(SoaS)noise and the Middleton's Class A noise model,which is proved better performance.However,the optimal receivers of signals based on these models are very complex,and hardly realistic for the practical engineering applications.Under this circumstance,it is necessary and valuable for the thesis to study a novel suboptimal detector which can reduce complexity and ensure high-quality performance.We discuss the problem of designing approximation of log-likelihood ratio(LLR)to simplify the optimal detector for two non-Gaussian noise models.Cauchy noise is a special case of S?S noise and since the optimal Cauchy detector is complex and consumes excessive power and memory,we propose a two-piece approximation of the Cauchy log-likelihood ratio nonlinearity.A 1/y curve segment is used as the approximation of the optimal Cauchy LLR curve in the large argument region.In the small argument region,the traditional linear approximation is used.The novel suboptimal detector based on this two-piece approximation is robust and performs closely to the optimal detector in a practical range of signal-to-noise ratio values,surpassing the performance of popular previously published suboptimal Cauchy detectors by 0.8 dB to 4.5 dB.When applied to S?S noise with different parameters,the novel suboptimal detector gives excellent performance.The optimal detector of Middleton's Class A noise needs n-order summation which is of complexity.In this thesis we propose a piecewise approximation of LLR to reduce the detection model to a linear model.This suboptimal detector outperforms Gaussian detector and the first improved detector and approaches the performance of the optimal detector closely while the impulse characteristic of noise is not particularly strong.