Research On The Blind Source Separation Techniques Under Complex Electromagnetic Environment

One challenging problem brought by the increasingly complex electromagnetic environment is that more and more signals generated by radiators including radars and communication devices are becoming overlapped in the time-frequency domain.One promising solution has been provided by the blind source separation techniques,which rely merely on few assumptions on sources or the mixing process and are capable of restituting original sources from observed mixtures.Hence,it should be effective in handling overlapped signals and enhancing the adaptability of existing systems against complex electromagnetic environment.In this paper,the blind source separation problem is studied under multi-channel and single-channel scenarios respectively.For the multi-channel scenario,the work is mainly about the performance analysis of existing typical algorithms in practice;as for the single-channel problem,single channel blind source separation algorithms based on the cyclic spectrum filter and the single-multi channel transformation are studied.For the analysis of the performance of classic multi-channel blind source separation methods,major attention has been paid to the analytic relationship between practical factors including the environment noise level,parameters of the receiving array,locations of sources and the restitution quality,which has been measured by the signal to jamming-plus-noise ratio at the separation output.Main results of the analysis have been consisted of the theoretical performance bound under certain noise level,impacts of different practical factors expressed in the form of instructive conclusions.Finally,main results are verified via simulation works.For traditional single channel blind source separation scheme based on the cyclic spectrum filter,a novel correction scheme has been put forward to improve its performance under the existence of cycle frequency error.First,a new cyclic feature function is raised on the basis of traditional cyclic spectrum density;then,the steepest descent correction algorithm is derived regarding the new feature function;the convergence property of the novel scheme is also discussed.Simulation works indicate that the novel method holds remarkable edges in accuracy,time consumption as well as robustness,comparing with existing schemes.And notable improvements in the filtering quality can be seen when combining the novel correction scheme with the traditional cyclic spectrum filter.For the single channel blind source separation based on the single-multi channel transformation,a novel way of constructing pseudo-MIMO(multi-input multi-output)mixtures has been proposed,exploiting the inherent temporal-spatial coupling property of digitally modulated sources.Based on the property,the idea of ?temporal linear array‘ is proposed,with which the construction of pseudo-MIMO mixtures is made possible.The problem of designing crucial parameters of the temporal linear array has also been given attention to.Simulation works have demonstrated the effectiveness of the novel scheme in separating time-frequency overlapped digitally modulated sources.