Study On Blind Equalization Of Satellite Channel Based On Echo State Network

Nonlinear and group delay characteristics of satellite channel cause serious Inter-symbol interference(ISI)when the transmitted signals get to receiver side.Equalization on receiver side is one of the effective method for reducing ISI.Blind equalization works without training sequence.It makes signals self-recovery by signal characteristics,effectively saves the resource of satellite communication bandwidth which is increasingly tense.Echo State Network(ESN)is a new type of recurrent network,its unique reservoir structure and simple learning algorithm let it be widely applied in time series prediction,nonlinear system modeling and channel equalization.This paper mainly studies the application of echo state network in blind equalization of satellite channel.The paper illustrates Bussgang and neural network blind equalization algorithm principles based on brief introduction of satellite communication and blind equalization theory.The principle and characteristics of the classical ESN are studied.For the hardware implementation of the algorithm,several improved ESNs are analyzed and discussed.Considering the characteristics of ESN in time domain signal processing as well as blind equalization theory,the ESN blind equalization through predictor theory and Bussgang approach are separately carried out.The predictor theory approach only takes sinals on receiver side,makes use of nonlinear projection ability of network,utilizes ESN as a predictor filter,minimizes the prediction-error by the basic pseudoinverse training method,and achieves blind equalization result.It is easy to implement.The result shows the algorithm is suitable for both group delay and nonlinear satellite channel model.It's a zero-delay algorithm,while with phase rotation.The Bussgang approach is realized by support vector regression(SVR)for weights training,the statistics of Constant Modulus Algorithm(CMA)and Multi-Modulus Algorithm(MMA)are put into the loss function of SVR to construct cost function,and the optimal values of output weights are obtained by iteration,then the blind equalization result is achieved at last.The result shows that,for the group delay satellite channel model,compared with the classical Bussgang blind equalization algorithm,such as CMA and MMA,the output constellation of the ESN basedmulti-modulus algorithm equalizer is more clear and the bit error rate(BER)performance is greatly improved.For the nonlinear satellite channel model,compared with the predictor theory approach based on ESN,the output constellation of the equalizer is more clear,BER performance has been improved,phase rotation has been modified.