New Hybrid Precoding for mmWave MIMO Systems: LADR and DALR Architectures

Hybrid precoding for fully-connected architectures (FA) delivers superior performance in millimeter-wave (mmWave) multiple-input multiple-output (MIMO) systems but comes at the cost of significantly higher complexity compared to sub-connected architectures (SA). This paper introduces two new sub-connected hybrid precoding architectures: localized-antennas distributed-RF (LADR) and distributed-antennas localized-RF (DALR), designed to balance the trade-off between performance and complexity. Both architectures divide transmitter antennas and RF chains into two groups, which are either distributed or localized. In LADR, localized antenna groups are connected to distributed RF chain groups, providing high beamforming precision, making it well-suited for dense urban deployments where performance demands are stringent. In contrast, DALR connects distributed antenna groups to localized RF chain groups, offering beamforming with lower precision compared to LADR, making it better suited for large-scale networks, such as rural or wide-area applications, where broader coverage and scalability are prioritized over high precision. The hybrid precoding is optimized and solved iteratively by decomposing the problem into two independent subproblems, referred to as the odd and even subproblems. Simulation results demonstrate that the proposed architectures achieve performance close to FA, while reducing the number of phase shifters by 50% and lowering computational complexity to $\mathcal {O}(N_{t})$ , compared to the $\mathcal {O}(N_{t}^{2})$ complexity of traditional FA designs, where $N_{t}$ is the number of transmitter antennas. Furthermore, the proposed architectures outperform traditional SA by approximately 3 dB with only a slight increase in complexity. The results also indicate that LADR offers slightly better performance than DALR when the number of data streams is high due to its superior beamforming capability, while both architectures perform similarly when the number of data streams is low.