Research On Adaptive Equalization Algorithm Based On Wavelet And Neural Network

In the field of wireless communication, due to the non-ideal properties of multipath propagation and time-varying characteristics, signal doppler spread and delay spread will occurred, and this will cause the inter-symbol interference in digital transmission. Usually we use equalization to compensate for the inter-symbol interference caused by limited bandwidth and dispersive channel.The traditional equilibrium method is to approximate the inverse of the channel by using a training sequence and certain criteria, so that to compensate the distortion caused by the channel. But this kind of adaptive equalization has the shortcomings as follows: the training sequence will occupy the transmission bandwidth, and in some occasions the signal source can not transmit the training sequence. So the blind equalization developed a lot, blind equalization does not need to know the original code sequence accurately, it's accomplished by approximating the statistic value of the original signal.This paper firstly analysis the multipath fading channel in theory, research the impact on wireless communication caused by multipath fading,and established the tapped-delay-line model to approximate the channel, and then simulated the model. That provides the theory basis and simulation environment for the research on equalization algorithm afterwords.Secondly, this paper research the Nyquist criterion that ideal channel should satisfy to ensure transmiting wihh no ISI,which is the ideal target of the equalization algorithm. Then analyses the criteria to design equalization algorithm and some classical blind equalization algorithm like Bussagang algorithm and CMA algorithm,but the CMA algorithm has the disadvantage as follows: the convergence speed is slow, the steady residual error is high, and it's insensitive to the phase of carrier wave.The CMA algorithm use minimum mean square error as the criterion for iteration, the convergence speed of the algorithm is related to the autocorrelation of the input signal. The wavelet has the property of time-frequency localization and multi-resolution analysis, thus it can reduce the autocorrelation of the signal and increase the algorithm's convergence speed.This paper research the blind CMA algorithm based on wavelet transform, and compare to the algorithm that combine the dynamic content and variable step length. The wavelet packet transform overcomes the disadvantage that the wavelet transform divide the signal only in scale space, so it can reduce the autocorrelation of the signal further. So, the paper research the blind CMA algorithm based on wavelet packet transform.Because of the ability to perform well in solving the nonlinear problem and large scale parallel computing, neural network has it's advantage in equalization algorithm, but it has the disadvantage as follows : it has slow convergence speed , and network structure is decided by experience, it can be overcome by combining the wavelet transform. This paper research the influence on convergence speed when introduce the super exponential iteration to the BP feed-forward neural network of the blind CMA algorithm, at last research and compare the influence on the capability of the algorithm if combine with the wavelet transform.