Research On Krylov-subspace Based Low-compleixty Detection Algorithm For OFDM Systems

Orthogonal Frequency Division Multiplexing(OFDM) has many advantages, which has been widely focus on in the field of communication and becomes one of the key technologies of the next generation mobile communication systems. In the actual, however, the varying characteristics of the communication channel is very obvious. When the channel is fast fading, the Doppler frequency shift leads to destroyed inter-carrier orthogonality, resulting in a serious subcarrier see interference(ICI).This paper introduces the basic principles of OFDM, and analyzes the double selection fading channel model as well as ICI generation mechanism. For the ICI mitigation issues, several classical detection equalization algorithms are introduced. For OFDM systems with basis expansion model(BEM), this paper proves the conclusion that the banded frequency domain channel is approximately equivalent to the time domain channel with complex exponential basis expansion model, from the conversion between frequency and time domain of channel matrix. OFDM system transmits in units of blocks, while the BEM is block-based channel model. Utilizing narrowband characteristics of varying channel, BEM channel modeling convert the time varying channel to the weighted linear sum of small amount of quadrature component, which is a new understanding of channel. In such cases, OFDM channel model has many similarities with BEM channel model. In response to these similarities, this paper describes a particular application of BEM e.g., pre-matrix design for krylov subspace iteration detection algorithm. We will talk about preconditioning matrix design issues next.Krylov subspace algorithm can approximately solve the least squares problem via system matrix iterative. The distribution of the eigenvalues of system matrix has influence on the number of iterations with the algorithm. At the same performance, the eigenvalues of system matrix are more concentrated, leading to the fewer iterations and the lower complexity. Usually, precondition matrix is used to achieve the effect of more concentrated eigenvalues. In this paper, we firstly introduced an realization of krylov subspace algorithms such as LSQR and conjugate LSQR. Secondly, we introduce the design in precondition matrix in krylov algorithms. We design regularized preconditioner that clusters only the largest eigenvalues of the relevant system matrix, effectively avoiding the gain elimination because of the mixture between signal and noise subspaces. The last but not the least, we make the regularized precondition matrix combined with LSQR. The innovative algorithm can achieve a good balance between computational complexity and performance.