Capacity Optimization For Full-Duplex Massive Multi-User MIMO Systems Based On Random Matrix Methods

As two of the most popular candidate technologies for the next generation mobile communication systems, massive multi-user multiple-input multiple-output (MU-MIMO) and full-duplex transmission both have the potential to significantly improve wireless systems’spectral efficiency. The interesting characteristics of massive MIMO and full-duplex communication motivate us to analyze the feasibility and performance gains of the combination of both technologies.In cellular systems operating in the full-duplex mode, each base station (BS) will experience strong self-interference during its data transmission intervals. Besides, the BS-to-BS interference, which occurs between neighboring full-duplex BSs, would also severely impact the systems’uplink performance. In this work, deterministic approximations of ergodic uplink achievable rates of full-duplex massive MU-MIMO systems are derived by methods from the random matrix theory. Based on the approximations, we prove that both the self-interference and the BS-to-BS interference vanish asymptotically as the number of BS antennas increases to infinity, which suggests that the uplink achievable rate accomplished by using full-duplex transmission is asymptotically twice as high as that of half-duplex systems. However, when the number of BS antennas is finite, both interference would still exist and deteriorate the system’s uplink. In the worst case, the interference can be so strong that the performance of full-duplex may be even worse than that of half-duplex. In order for full-duplex systems to outperform half-duplex systems in the uplink, the number of BS antennas must be larger than a threshold, which is derived explicitly. The result is useful for choosing the optimum duplexing mode under certain circumstances. Moreover, the system’s uplink performance also depends on the detection technique utilized. Cases with the matched filter or the linear minimum mean square error (MMSE) detector are respectively analyzed. Additionally, in the system model, a series of practical factors that have notable influence on massive MIMO and full-duplex systems are considered, including the antenna correlation, the line-of-sight (LoS) components of self-interference channels and BS-to-BS channels, and the limited dynamic range of transceivers. These factors significantly complicate the analysis. Since it is difficult to compute the systems’ergodic uplink achievable rates under finite system dimensions, we use random matrix methods to perform large system analysis. All of the derived results are validated by numerical studies. Simulations show that the uplink achievable rates match very well with their deterministic approximations. Besides, it is shown that when the number of BS antennas is large enough, full-duplex might be a more competitive choice than half-duplex from the perspective of uplink performance.