Design of Neural Network Based Adaptive CIM Function Channel Equalizer Using Learning Algorithm for Rate of Convergence
2017
Hochschulschrift
Zugriff:
105
The purpose of this thesis is to design a channel equalizer to compensate signal distortion during transmission. The channel equalizer consists of functional link artificial neural network (FLANN) structure as well as correntropy induced metric (CIM) algorithm and learning algorithm for rate of convergence. This thesis introduces a new FLANN based equalizer that utilizes CIM. The performance depends on the weight updating formulae in the channel equalizer. Generally used MSE algorithm directly applies the output error to the weight updating calculation. However, CIM algorithm uses output error to revise the sign of weight updating values. And system uses gradually learning algorithm for rate of convergence. In the beginning, the learning rate parameter is set to a larger value so that the system has a larger weight value in early training. In this case, the system can quickly achieve faster convergence. According to estimated amount of error, system gradually corrected the learning rate parameter. The convergence value is liable to stable and achieves zero at the later stage of training, so the system does not produce violent errors because of excessive weight value. The purpose of this thesis is to design a channel equalizer obtains more accurate weight updating value and the faster convergence rate than the conventional MSE algorithm equalizer. For example, we see from simulation results that learning rate of convergence FLANNCIM achieves stabilization after 1000 iterations, learning rate of stability FLANNCIM requires 1600 iterations to attain similar results, but FLANNMSE requires 3000 iterations to attain similar results. It can also be seen that learning rate of convergence FLANNCIM exhibits lower steady state error. When bit error rate is at, The required signal-to-noise ratio (SNR) for FLANNCIM with variable learning rate is located at 9.27 dB, and that for FLANNCIM with fixed learning rate is located at 10 dB. However, FLANNMSE is located at 10.3 dB. In addition, learning rate of convergence FLANNCIM error value is lower than learning rate of stability FLANNCIM and FLANNMSE in various channels.
Titel: |
Design of Neural Network Based Adaptive CIM Function Channel Equalizer Using Learning Algorithm for Rate of Convergence
|
---|---|
Autor/in / Beteiligte Person: | JHANG, WE-LUN ; 張維倫 |
Link: | |
Veröffentlichung: | 2017 |
Medientyp: | Hochschulschrift |
Sonstiges: |
|