Reduced complexity detection for massive MIMO-OFDM wireless communication systems

Abstract

The aim of this thesis is to analyze the uplink massive multiple-input multipleoutput with orthogonal frequency-division multiplexing (MIMO-OFDM) communication systems and to design a receiver that has improved performance with reduced complexity. First, a novel receiver is proposed for coded massive MIMO-OFDM systems utilizing log-likelihood ratios (LLRs) derived from complex ratio distributions to model the approximate effective noise (AEN) probability density function (PDF) at the output of a zero-forcing equalizer (ZFE). These LLRs are subsequently used to improve the performance of the decoding of low-density parity-check (LDPC) codes and turbo codes. The Neumann large matrix approximation is employed to simplify the matrix inversion in deriving the PDF. To verify the PDF of the AEN, Monte-Carlo simulations are used to demonstrate the close-match fitting between the derived PDF and the experimentally obtained histogram of the noise in addition to the statistical tests and the independence verification. In addition, complexity analysis of the LLR obtained using the newly derived noise PDF is considered. The derived LLR can be time consuming when the number of receive antennas is very large in massive MIMO-OFDM systems. Thus, a reduced complexity approximation is introduced to this LLR using Newton’s interpolation with different orders and the results are compared to exact simulations. Further simulation results over time-flat frequency selective multipath fading channels demonstrated improved performance over equivalent systems using the Gaussian approximation for the PDF of the noise. By utilizing the PDF of the AEN, the PDF of the signal-to-noise ratio (SNR) is obtained. Then, the outage probability, the closed-form capacity and three approximate expressions for the channel capacity are derived based on that PDF. The system performance is further investigated by exploiting the PDF of the AEN to derive the bit error rate (BER) for the massive MIMO-OFDM system with different M-ary modulations. Then, the pairwise error probability (PEP) is derived to obtain the upper-bounds for the convolutionally coded and turbo coded massive MIMO-OFDM systems for different code generators and receive antennas. Furthermore, the effect of the fixed point data representation on the performance of the massive MIMO-OFDM systems is investigated using reduced detection implementations for MIMO detectors. The motivation for the fixed point analysis is the need for a reduced complexity detector to be implemented as an optimum massive MIMO detector with low precision. Different decomposition schemes are used to build the linear detector based on the IEEE 754 standard in addition to a user-defined precision for selected detectors. Simulations are used to demonstrate the behaviour of several matrix inversion schemes under reduced bit resolution. The numerical results demonstrate improved performance when using QR-factorization and pivoted LDLT decomposition schemes at reduced precision.