Variable Step Size Lms Adaptive Equalization Algorithm On Dsp,

When signal is transmitted, the limited channel bandwith and the multipath wireless channel will both introduce the ISI, thus the symbol error rate is increased and the performance of the communication system is worsened. Adaptive equalization is the most commonly used technology to control the ISI. The algorithms, which include LMS and RLS, are the key points of the adaptive equalization. The LMS algorithm is widely used because of its simplicity and good performance. The fixed step size LMS could not get both the good performance of the convergence speed and the steady state error. The longer step size results faster convergence speed and higher steady state error while the shorter step size results slower convergence speed and lower steady state error. The variable step size LMS algorithms(VSS-LMS) are introduced to solve the problem. But they have many drawbacks, such as the slower convergence speed and the sensitivity to noise.After reading a large amount of the papers about the VSS-LMS, the dissertation analyzes several LMS algorithms and sets a nonlinear function between estimation error and step size factor and presents the HTLMS algorithm based on modified hyperbolic tangent function. In the algorithm, the step size factor is adjusted by the absolute value of the product of the current and former errors. The algorithm also introduces the disturbance of the absolute estimation error to update the tapping vectors of the adaptive filter. The dissertation then compares HTLMS with several other LMS algorithms in different SNR conditions on MATLAB. The simulation shows that HTLMS algorithm has faster convergence speed, better performance of noise suppression and lower steady state error.The performance of the HTLMS algorithm is then verified on the TMS320C6713 DSK. The paper forms an equalizer based on the algorithm and apply it in a QAM digital communication system in C program. The result shows that the algorithm could restrain the ISI quite well.