Research On Fairness-Orients Pilot Allocation Schemes In Massive MIMO System

Huge array gain of massive MIMO will be able to effectively improve the signal-to-noise ratio (SNR) for each user in multiuser system, making channel capacity and spectrum efficiency of the communication system improved greatly. Therefore, massive MIMO is regarded as one of the key technologies of the fifth generation for mobile communication (5G). Considering time division duplex (TDD) communication scheme, the number of orthogonal pilots used for channel training is limited, thus with the rapid increase of users, pilot reuse in adjacent cells is inevitable, leading to serious inter-cell interference, namely, pilot contamination. Studies show that pilot contamination constitutes the main bottleneck of massive MIMO system performance. Considering that the minimum signal-to-interference-and-noise ratio (SINR) of users normally cannot meet the basic demand of communication due to pilot contamination, this thesis proposes fairness-oriented pilot allocation schemes in massive MIMO system, based on max-min fairness criterion. By maximizing the minimum SINR of users, it realizes the fairness among users, thereby alleviating pilot contamination.Firstly, this thesis studies the relationship between the SINR of users and pilot allocation. By introducing the pilot selection matrix, the expressions of uplink and downlink SINR are derived, respectively. Studies show that the uplink SINR depends on the correlation between pilot selection matrices, but the downlink SINR does not depend on the specific pilot allocation. Then, based on the uplink SINR expression and max-min fairness criterion, fairness-oriented pilot allocation problems for users in one target cell and for those in all cells are formulated, respectively. Finally, the pilot allocation problems are converted into the optimization problem of pilot selection matrices.Secondly, considering the problem of fairness among users in one target cell, this thesis proposes three pilot allocation schemes, including centralized pilot allocation scheme (CPA), large scale fading based pilot allocation scheme (DPA) and location based pilot allocation scheme (LPA). Among them, by equivalent transformation, CPA transforms the original problem into a Mixed Integer Linear Programming (MILP) problem, which can be solved by the optimization software; DPA and LPA first give the judgment method of user interference strength based on large scale fading and location information of users, respectively, then based on the criterion that the user suffering the severest interference had the highest priority, pilot allocation is done in a greedy way. Simulation results show that CPA can provide the optimal solution of the pilot allocation problem for fairness of users in one target cell. DPA gives the approximate optimal solution, and reduces the computational complexity and the system complexity. Meanwhile, given location information of users, LPA improves the minimum SINR performance, and when the standard deviation of shadow fading is small, its performance can be comparable to that of DPA.Finally, considering the problem of fairness among users in all cells, this thesis proposes a global pilot allocation scheme (GPA). First of all, the global user fairness pilot allocation problem is decomposed into several sub-problems, each of which obtains the optimal pilot allocation for one cell, and can be transformed into an MILP problem equivalently; then, the overall pilot allocation is realized by solving the sub-problems in a sequent and iterative way. Studies show that the optimization objective of the pilot allocation problem for fairness of users in all cells is a monotone bounded function about the number of iteration. Therefore, the scheme can converge to the optimal solution in a limited time. Simulation results show that the minimum SINR performance can be comparable to that of exhaustive search method (EPA), thus GPA achieves the approximate optimal solution. Meanwhile, because of the optimization methods including problem division, equivalence transformation of the sub-problems and the iteration scheme, and the characteristics of convergence with fast speed, GPA reduce the complexity of the pilot allocation problem for fairness of users in all cells greatly.