Research On Channel Estimation Of New Generation MIMO System

Since the 1990s, the multi-input multi-output (MIMO) technology of wire-less communication system has been widely concerned and studied. Recently,It has become one of the key technologies in the fourth generation (4G) mo-bile communication system. In the fifth generation (5G) system, MIMO will trend to be multi-dimensional and large-scale. By using a much larger number of antennas in base station than it in all user terminals, large-scale MIMO sys-tem can effectively reduce multi-user interference and noise thus significantly improving the spectral efficiency and energy efficiency of the communication system.Before studying the future MIMO technology, the channel characteristic parameters should be recognized for designing corresponding transceiver. In addition, when wireless communication system implementing, channel estima-tion is used to obtain accurate channel state information, which is the basis for system resource allocation, transmitter pre-coding and receiver signal detec-tion. In this paper, the above channel characteristic parameter estimation and the channel estimation technique are called uniformly as channel estimation techniques and studied on the background of future MIMO technology. Our research work and its innovation are summarized as follows:1. When studying 3D MIMO system, the unknown channel characteristic parameters of Geometry-based Stochastic Models are usually estimated in the field measurement data through the Space- Alternating Generalized expectation-maximization (SAGE) algorithm. However, in the traditional SAGE algorithm,Channel characteristic parameters such as channel's multi-path departure angle and arrival angle were serially divided into four estimator spaces and each of these spaces have only one parameter which is estimated by lattice search algo-rithm. This increases the risk of estimating error propagation and also reduces the estimated resolution and accuracy. In this paper, the traditional SAGE algo-rithm is improved. More concretely, channel's multi-path departure angle and arrival angle are merged into one multi-dimensional vector estimator space, and the whole vector is estimated by Hopfield neural network. The simulation re-sults show that the proposed SAGE algorithm improved by Hopfield neural network can converge to the true value of the angle parameters faster and more accurate.2. In the large-scale MIMO. system, the accuracy of the received signal autocorrelation matrix directly affects the performance of the blind channel es-timation algorithm. This paper deduces the theoretical relationship between the number of received signal used to estimate the correlation matrix and the performance of the blind channel estimation algorithm in Markov time-varying channel. The numerical simulation results prove the correctness of this theo-retical relation formula through which the optimal signal points for blind es-timation can be obtained. In addition, according to pilot pollution, a channel interpolation algorithm based on neural network is proposed in this paper. The simulation results show that the algorithm can effectively reduce the influence of the estimation error at the pilot position.3. In pilot aided channel estimation, using of compressive sensing (CS)technology can allow the requirement for pilots to decrease effectively, which is particularly critical in large-scale MIMO system. However, the traditional CS-based channel estimation algorithm uses a well-defined dictionary matrix which inevitably leads to a lack of accuracy because of the continuity of the channel's multi-path departure angle. This inaccurate dictionary matrix would further lead to a deterioration of channel estimation. In this paper, particle swarm op-timization (PSO) algorithm is used to obtain a more exact dictionary matrix by which the orthogonal matching pursuit (OMP) algorithm in CS can get a more accurate channel estimation. The simulation results show the effectiveness of the proposed PSO-OMP algorithm.4. In the CS-based channel estimation, the measurement matrix' s re-stricted isometric property (RIP) which is depending on the pilot values and locations greatly affects the accuracy of channel estimation, Based on this con-clusion, this paper proposes two pilot design schemes for CS-based channel es-timation. In the first scheme, the multi-antenna is grouped, and the pilot values are optimized for any one group based on the alternating projection algorithm.The second scheme is a low complexity one. In the special case that the sub-carriers number is equal to the product of the number of tip in channel filter model and the number of transmit antennas, the optimal pilot values can be ob-tained directly by formula and then only pilot locations need to be optimized.The simulation results show that compared with the traditional orthogonal pilot pattern, the two pilot design schemes can give us a better performance of CS channel estimation.