Study On Svr-based Blind Source Separation And Equalization Methods For MIMO Systems

Blind source separation and equalization of multiple input and multiple output(MI-MO)systems are designed to reduce the intersymbol interference(ISI)and interchannel interference(ICI)without training sequence.At present,there are two main methods:single-stage and multi-stage method.With the multiple input single output(MISO)e-qualizers cascaded in turn,multi-stage methods use equalization term to realize blind channel equalization and usually rely on channel estimation to complete continuous in-terference cancellation to separate the sources blindly,however,they tend to require the information of channel order and exist the accumulation of inter-stage processing error.The MISO equalizers in single-stage methods are parallel to each other,which accomplish blind source separation and equalization of multiple signals simultaneously and overcome the shortcomings of multi-stage methods.This paper studies single-stage methods and the main contents are as follows:To overcome the high residual interference in traditional online algorithms,first,some blind equalization algorithms in single input single output(SISO)system with better performance are extended to MIMO systems,and they are added to the cross-correlation of equalizers' output signals to obtain a lower system interference.Then,according to the different characteristics of these new algorithms,two online dual-mode schemes are constructed to deal with higher order quadrature amplitude modulation(QAM)signals.Simulation results show that the proposed algorithm presents lower sys-tem interference compared with the traditional algorithms;the dual-mode schemes can further reduce system residual interference and possess the ability to deal with 64QAM signals.Although the above improved methods reduce system residual interference,they require lots of samples to converge.Support vector regression(SVR)is a small data block learning tool based on struc-tural risk minimization.Therefore,the error function of constant modulus algorithm C-MA(p,2)and radius directed algorithm(RDA)and the cross-correlation among equal-izers' output vectors are introduced into the SVR framework,respectively,proposing a family of batch processing algorithms SVR-CC-CMA(p,2)and dual-mode schemes SVR-CC-CM(p,2)-RD.For these new methods,iterative re-weighted least square(IR-WLS)are applied to optimize the cost function with low complexity.Simulation results show that SVR-CC-CMA(p,2)obtain lower system interference with very small data blocks compared to the traditional online algorithm;SVR-CC-CM(p,2)-RD can fur-ther greatly reduce residual interference with increasing little computation and own the capability to process 64QAM signals.The above SVR-based blind signal recovery methods exist the problem of phase rotation and can not restore more high order QAM signals.Therefore,under the SVR framework,this paper also combines the family of multimodulus algorithm MMA(p,2)and the direct decision algorithm(DDA),resulting in another family of batch pro-cessing algorithms SVR-CC-MMA(p,2)and dual-mode schemes SVR-CC-MM(p,2)-DD.Simulation results show that SVR-CC-MMA(p,2)correct the phase rotation of SVR-CC-CMA(p,2)and present lower system interference;SVR-CC-MM(p,2)-DD can significantly reduce system residual interference with only increasing a smaller computation cost,in which SVR-CC-MM(4,2)-DD possesses the ability to recovery 256/1024QAM signals.