Channel Tracking Based On Particle Filter And Radial Basis Function Neural Network In MIMO-OFDM Systems

Mobile communication networks put even more emphasis on bandwidth efficiency and transmission performance. Along with this rise of rate comes the even undesired wireless transmission environment. The increasing vehicle speed causes fast fading in which the signal amplitude and phase fluctuate violently during transmission. The major scattering components in the typical urban area diffuse the signal in the time domain, causing clustered inter-symbol interference. So to overcome these unfavorable conditions and realize high speed, high efficiency transmission will be the main task of current mobile wireless communication.MIMO and OFDM will be the key technologies in the next generation of wireless communication systems. The combination of these two technologies makes good use of space diversity, frequency diversity and time diversity to improve the system capacity and the ability of anti-interference. Meanwhile, the effectiveness of MIMO-OFDM systems relies on the availability of accurate multi-channel estimation at the receivers. Without channel estimation, single OFDM system can only work under non-coherent conditions. Within a MIMO-OFDM system any module other than channel estimation can be replaced by other modules. In this paper, the focus are channel estimation methods. The MATLAB simulations have been carried on the comprehensive analysis and comparison to these methods validity and feasibility.There are many channel estimation methods proposed for MIMO-OFDM systems which can be classified into two kinds: traditional training signal methods and blind channel estimation. But in practical wireless environment, the channel is changed. In traditional training signal methods, the pilots take the excessively resources. Meanwhile the blind channel estimation can save the channel bandwidth and improve spectral efficiency, but has a much higher computational complexity. So the above method is not suitable for the time-varying channels.In this paper, we use the particle filter to estimate channel parameters information by tracking channels, this method can be applied to any state space model which can be generalized by traditional Kalman filtering methods. Compared with the traditional methods, we have the conclusion that the proposed method has more effective tracking performance.For the particle filter method, the initial value is one of the important factors that affecting the final tracking performance. In this paper we use the radial basis function neural network to obtain the initial value. Though the computational complexity is much higher than the traditional methods, the algorithm needs no more training signals or pilots after obtaining the initial estimation values. Simulations confirm the effective performance of the proposed method.