Massive connectivity is one of the killer use cases for fifth generation (5G) wireless systems over the millimeter wave (mmWave) band. Due to the sparse nature of mmWave channel, several users may choose the same beam as the strongest one, leading to severe intra-beam interference (intra-BI). Although intra-BI can be mitigated through digital precoding, the required number of RF chains for fulfilling massive connectivity becomes large. By applying non-orthogonal multiple access (NOMA) and properly allocating transmit power to each beam and user, the sum rate can be maximized using a smaller number of RF chains. However, the power allocated to each user depends on that allocated to other users within the same beam. Since the joint beam selection and power allocation problem contains a non-convex objective function with a large set of coupled and mixed integer variables, direct solutions may not exist particularly when both the number of users and antennas are large. In this work, we decompose the aforementioned problem into two subproblems, where the beam selection sub-problem is first solved under an equal power allocation. The outcome of beam selection is then used as the input to the power allocation sub-problem. For both subproblems, we develop numerous efficient algorithms to find the active beam set and the user power allocation, respectively. The notions of embedded methods from machine learning and that of intelligent searching from metaheuristics are adopted in our work as the key ingredient for algorithm constructions. Numerical results demonstrate that in the region of large user population, the proposed beam selection and power allocation algorithms can effectively improve the sum rate using a less number of RF chains in comparison with some existing solutions.
All Science Journal Classification (ASJC) codes
- Computer Science(all)
- Materials Science(all)