Suboptimal Resource Allocation for Multi-User MIMO-OFDMA Systems

Future wireless communication systems are expected to reliably provide data services with rate requirements ranging from a few kbit/s up to some Mbits/s and, due to the high costs of frequency spectrum, these systems also need to be extremely efficient in terms of the spectrum usage. In particular, the application of transmission schemes based on Orthogonal Frequency Division Multiple Access (OFDMA) and on Multiple Input Multiple Output (MIMO) is considered as a promising solution to meet these requirements.

On the one hand, MIMO-OFDMA systems are flexible and spectrally efficient. On the other hand, the considerably large number of subcarriers and the inclusion of the space dimension make the Resource Allocation (RA) in such systems very complex. In fact, the optimum RA that maximizes the sum rate of the system is often too complex for practical application and, consequently, suboptimal rather efficient and low-complexity RA strategies are required in order to allocate the frequency, time, and space resources of the system to the Mobile Stations (MSs).

This project deals with suboptimal RA strategies in the downlink of MIMO-OFDMA systems aiming at the maximization of the sum rate. In order to solve the problem of maximizing the sum rate with affordable complexity, a new formulation for the problem is proposed which divides it into four subproblems, namely the Space Division Multiple Access (SDMA) grouping problem, the precoding problem, the power allocation problem, and the resource assignment problem. For each subproblem, several existing or newly proposed algorithms are applied, which are also oriented towards the maximization of the sum rate of the system. Through the combination of these algorithms, new suboptimal rather efficient RA strategies are obtained.

For the SDMA grouping problem, four new SDMA algorithms are proposed: one algorithm based on convex optimization and three greedy algorithms based on simple heuristics. The proposed algorithms build the SDMA groups based only on the spatial correlation and channel gain of the MSs and depend neither on precoding nor on power allocation. The proposed algorithms are shown to perform as good as some existing SDMA algorithms in terms of the achieved average sum rate and to have considerably lower computational complexity than the existing ones.

For the precoding problem, two existing algorithms are adopted. For the power allocation problem, a new iterative Soft Dropping Algorithm (SDA) is proposed, which is subsequently combined with Generalized Eigen-Precoding (GEP) into a new iterative joint precoding and power allocation algorithm. Moreover, the proof for the convergence of the joint precoding and power allocation algorithm is provided. In particular, the SDA and, consequently, the joint precoding and power allocation algorithm can pursue either the maximization of the sum rate or the provision of Quality of Service (QoS) to the MS by means of a simple parameter setting. Also as part of the precoding and power allocation algorithm, a new Sequential Removal Algorithm (SRA) is proposed, which might remove MSs from SDMA groups as to enhance the sum rate.

For the resource assignment problem, algorithms performing either separated or joint SDMA grouping and resource assignment are proposed and compared. For the maximization of the sum rate of the system, it is shown that a sequential assignment of resources to SDMA groups performs as well as assignment algorithms performing joint SDMA grouping and resource assignment while being considerably simpler.

Moreover, different criteria to prioritize MSs or SDMA groups are considered by the assignment algorithms, and it is shown that by adopting a suitable priority criterion the throughput fairness among the MSs can be considerably improved at the expense of only small reductions of the average sum rate of the system.

The new suboptimal RA strategies that result from the combination of the proposed algorithms are shown to obtain a considerable fraction of the maximum achievable sum rate of the system with computational costs considerably lower than that of an optimum RA. Indeed, the proposed RA strategies achieve over 90% of the average sum rate obtained by an RA strategy performing an Exhaustive Search (ES) for the SDMA group that maximizes the sum rate.