Error reduction of channel estimation based on singular value decomposition

In this paper, a channel estimation algorithm for localized resource assignment is studied. Here, we exploit the singular values of coefficient matrix between the received demodulation reference signal (DMRS) and the channel impulse response (CIR), and discover that the singular values will attenuate rapidly after a particular index. Utilizing this characteristic, the proposed algorithm modifies the accuracy of channel estimation by filtering out the noise at the singular value domain. Firstly, it transforms the received DMRS into singular value domain by multiplying the left-singular matrix of coefficient matrix, so as to concentrate the energy of channel information on enumerable principle singular values. Then it eliminates the noise relying on the extremely small singular values to decrease the estimation error. The mean square error of channel estimation is simulated to evaluate the proposed CE algorithm. Results show that channel information can be estimated accurately, and the proposed algorithm outperforms other typical CE algorithms, such as LMMSE, DFT and DCT.