Channel Pattern Recognition: A New Method for FDD Downlink Channel Estimation

Estimation of downlink channel state information (CSI) of frequency division duplex (FDD) systems without feedback, based on the estimated uplink CSI, is addressed. Instead of using the parameterized algorithms, this paper proposes a new method, channel pattern recognition, through which the physical character of the channel is described by probability model rather than variables. In this way, although the uplink and downlink CSI in FDD systems do not have the reciprocity, the equivalence of the physical transmission environment of the uplink and downlink channel can be used better. Firstly, a coarse estimate of the direction-of-arrival (DOA) of the channel path is obtained by the multiple signal classification (MUSIC) algorithm. The error of this coarse estimate is considered in the next step of channel pattern recognition. In order to deal with the uncertainty of the factors of the wireless channel, especially the hidden variables, a general Gaussian mixture model (GMM) is used as a basic channel pattern, of which the key parameters are left as unknown variables. After the optimization problem is constructed, a message passing algorithm is derived using variational inference theory, through which the parameter information of the transmission path can be obtained from the calculated channel pattern. In this way, through calculations of the key parameters of the GMM, the pattern of each factor of channel fading is determined. Finally, the downlink CSI is obtained using the recognized wireless channel pattern.