Design And Performance Analysis Of Downlink Precoding In Massive MIMO Systems

Theoretical research has shown that Massive MIMO systems are able to significantly increase the capac-ity or reduce the power consumption by deploying a large number of antennas at the BS. In practical imple-mentation, however, the imperfect hardware may bring about two performance-limiting issues, i.e., hardware mismatch and IQ imbalance. Hardware mismatch caused by different hardware configuration between the uplink and the downlink affects the channel characteristics, while the IQ imbalance at the transmitter induces distortion to the transmitted signals. In this dissertation, the design and performance analysis of downlink precoding are carried out to alleviate the impact of both hardware-related issues. The main contributions are listed below.Firstly, the SINR performance is analyzed for MF and RZF in the downlink of Massive MIMO systems. 1) By utilizing stochastic mathematics, the PDF of the SINR at each user is derived for MF; 2) Asymptotic analysis are carried out and it is proved that when the number of transmit antennas N and the number of users U grow to infinity with a fixed ratio ?= U/N, the asymptotic SINR of each user is given by 1/?; When the number of users is large enough, the asymptotic SINR is given by NP0/[P0(U-1)+ln(NU)]. When N grows to infinity, the asymptotic SINR is determined by the ratio of the average transmit power P0 and ln N/N; 3) The asymptotic SINR for RZF is derived in closed form.Secondly, the performance of downlink linear precoding schemes is analyzed for Massive MIMO sys-tems with hardware mismatch:1) With consideration of any power allocation schemes, the achievable rates of MF and RZF are derived under channels with Rayleigh fading and path loss; 2) The upper bounds of the achievable rates are given for MF and RZF, which are only related to the statistics of the hardware mismatch parameters and the ratio of the number of transmit antennas and the number of users; 3) The performance of different hardware mismatch calibration schemes are compared, and it is proved that Pre-Cal schemes outperform Post-Cal schemes.Thirdly, widely linear precoding schemes are proposed in order to handle the IQ imbalance at the trans-mitter in Massive MIMO systems with single-antenna users:1) The equivalent real-valued single model is derived with consideration of transmitter IQ imbalance, and WL-ZF-R precoding scheme is given based on this model, which does not require knowledge of IQ parameters and achieves the same multiplexing gain as that of ZF with perfect IQ branches; 2) With a complex-valued signal model, the MMSE based WL-MMSE-C precoding scheme is proposed. Assuming perfect knowledge of IQ parameters, WL-MMSE-C with IQ im-balance performs the same as MMSE precoding with no IQ imbalance; 3) Two low complexity IQ parameter estimation schemes are proposed, i.e., DLS-RNP and ALS-RNP, based on a method to reduce the number of estimated parameters; 4) A low complexity algorithm is proposed to simplify the matrix inversion operation involved in ZF and MMSE precoding schemes using Krylov subspace based method.Fourthly, widely linear precoding schemes are proposed in order to handle the IQ imbalance at the trans-mitter in Massive MIMO systems with multiple-antenna users:1) The expression of BD type precoding schemes are given based on a real-valued signal model. By using Taylor expansion, a low complexity method is proposed to simplify WL-BD precoding scheme; 2) The expression for sum date rate and the multiplexing gain are given for WL-BD. It is proved that when the transmitter has perfect IQ branches, WL-BD performs the same as BD. When the transmitter has IQ imbalance, however, WL-BD achieves the same multiplexing gain as that of BD with perfect IQ branches.