Performance Evaluation of Machine Learning Based Channel Equalization Techniques

Abstract

Wireless communication systems are evolving with the passage of time and resulted in the advanced communication systems such as mobile and Adhoc communication systems, sensor networks etc. Two fundamental parameters of their performance are higher data rates and better spectral efficiency. In order to achieve high data rate and robust communication over wireless the most important task to be performed at the receiver side is the channel equalization. The transmitted data symbols when pass through the wireless channel suffer various types of impairments, such as fading, Doppler shifts and Inter Symbol Interference (ISI), degrade the overall performance of the communication system. In order to mitigate the channel related impairments many channel equalization algorithms have been proposed in the communication systems domain. These algorithms are based on either the least squares methods or minimum mean square estimation methods. Frequency domain equalization methods are also used for this purpose. The channel equalization problem can also be solved as a classification problem using Machine Learning (ML) methods. Many researchers have addressed this problem however there is a need to compare the performance of the ML based equalizers by using the already established communication systems criterion. In this thesis the channel equalization has been performed using ML techniques and their Bit Error Rate (BER) performance has been compared. Radial Basis Functions (RBF), Multi Layer Perceptrons (MLP), Support Vector Machines (SVM), and Polynomial based NN’s have been used for channel equalization. The work is also extended to the multi-carrier systems where the channel equalization of OFDM system is carried out using Long Short Term Memory (LSTM) method. The simulation results show improved BER when compared with the LMS based traditional channel equalization method. Further the computational complexity of all the used ML algorithms has also been formulated theoretically and has been verified with the help of simulations.