Study On Channel Equalization Method Of MIMO-OFDM System Based On Extreme Learning Machine

MIMO-OFDM system combines the advantages of MIMO technology and OFDM technology, this system can use finite frequency spectrum to provide more transmission rate and system capacity, which is one of the crucial technologies of the fourth generation mobile communication. MIMO technology uses multipath transmit and receive antennas for communication in the same frequency band, this scheme improves the channel capacity and reliability, However, it increases the complexity of equalizer on receiver side inevitably. OFDM technology has a strong ability to resist multi-path fading and frequency selective fading, reduce the inter symbol interference (ISI) by inserting cyclic prefix(CP), but the sub-intercarrier interference by time-variant characteristic of channel gives rise to the bit error rate platform effect. Therefore, the equalization methods of the MIMO-OFDM system is a issue that worthy to studying.This paper introduces the principle of the MIMO-OFDM system, provides the MIMO-OFDM model, and illustrates the traditional equalization methods:detailed analyse the mean square error algorithm, least squares algorithm and Maximum likelihood algorithm. Based on the Matlab simulation platform, aim at2multiply by2MIMO-OFDM system, compares the performance of the three conventional equalization algorithms under time-variant frequency fading channel model COST207TU. Considering the nonlinear property of MIMO-OFDM system and good processing ability of neural network systems, this paper focus on the equalization method of MIMO-OFDM system based on neural network.In this article, channel equalization is seen as a classification problem. Input signals using the same modulation mode, the modulation mode can be divided into different classes in accordance with its modulation constellation. Received signals via neural network equalizer recover the corresponding class and further recover the transmitted signals. This paper present the equalization principle frame chart of MIMO-OFDM system based on neural network, under the above simulation environment, compare the equalization performance of feedforward BP network and RBF network with the traditional equalization algorithms, simulation results show that traditional feedwork neural network based equalizer can obtain relatively low bit error rate. For the disadvantage of traditional feedwork neural network, such as parameter uncertainly, slow convergence, and local minima, etc. we propose using extreme learning machine to solve the channel equalization problem of MIMO-OFDM system in this paper.Extreme learning machine is a single hidden layer feed forward network developed in recent years, compared with the traditional feedforward network, it has more simple structure, fast computing speed, better global optimal characteristics, and has been widely used in the regression and classification problems. This paper established the channel equalization algorithm of MIMO-OFDM system based on extreme learning machine, and compare the BER performance and computing time of the new algorithm with the conventional feedforward neural network equalizer, Simulation results demonstrate that the proposed extreme learning machine based equalizer has lower BER than conventional feedforward neural network equalizer, and the equalization time is shorter.