Research On Theories And Methods Of Traffic Queue-based Dynamic Radio Resource Optimization

To support the rapidly increasing requirements for mobile multimedia service and to satisfy the objectives of high spectral efficiency and high energy efficiency, the cloud radio access network(C-RAN) and heterogeneous cloud radio access network (H-CRAN) are proposed successively. Due to the random traffic arrivals and time-varying channel conditions in practical networks, the research on theories and methods of traffic queue-based dynamic radio resource optimization should take the delay performance into account. This dissertation works on the theories and methods of traffic queue-based dynamic radio resource optimization in C-RAN and H-CRAN. First, for the C-RAN, a thorough research on the radio resource allocation in multi-point transmission, radio resource optimization for precoding optimization and radio resource optimization with node selection is presented. Then, for the H-CRAN, a thorough research on the radio resource optimization with congestion control and corresponding performance bounds is presented. The main contents and contributions of this dissertation are summarized as follows:1. Traffic queue-based dynamic radio resource optimization in coordinated multi-point transmissionThe fronthhaul capacity in C-RAN is usually limited, and the channel state information is usually non-ideal. This dissertation first proposes a stream splitting-based hybrid coordinated multi-point(H-CoMP) transmission scheme, which can balance the cooperation gain the fronthaul consumption effectively. Considering the limited fronthaul capacity and non-ideal channel state information, the traffic queue-based dynamic power and rate allocation optimization problem for H-CoMP is formulated by using the Markov decision process. To reduce the complexity of solving the constrained Markov decision process problem,the online learning algorithm that estimates the post-decision value function and the stochastic gradient algorithm that obtains the power and rate allocation policy are proposed. The simulation results show that the proposed H-CoMP scheme can improve the cooperation gain with limited fronthaul consumption, the proposed resource allocation algorithm can achieve better delay performance, and the proposed online learning of value functions can finally converge.2. Traffic queue-based dynamic radio resource optimization for precoding optimizationRegarding the issue of low-delay and low-power cooperative transmission in C-RAN with limited fronthaul consumption, the traffic queue-based dynamic radio resource optimization for precoding optimization is studied. The linear constraints for cooperative precoding are first established by implementing a simple user-centric cooperative set selection scheme. The problem that minimizes the queuing delay and transmit power consumption is then formulated as a constrained discrete-time Markov decision process. By describing the traffic dynamics with differential equations, the equivalence between continuous-time and discrete-time Markov decision process is derived,based on which the closed-form approximated expression of value function is calculated. The transformed problem is finally solved with the semi-definite positive relaxation and alternated optimization algorithm.The simulation results show that the proposal achieves significantly better delay and power performance due to its adaptiveness to both queue state information and channel state information.3. Traffic queue-based dynamic radio resource optimization with node selectionIn C-RAN, not only the transmit power consumption but also the fronthaul power consumption and static circuit power consumption should be considered into the power minimization, therefore, the power savings can be achieved by switching off some remote radio heands(RRHs). Considering the impact of node selection on network power consumption and queuing delay, this dissertation investigates the traffic queue-based dynamic radio resource optimization with node selection in large-scale C-RAN. Since the formulated stochastic optimization problem is a complex non-convex problem, it is first transformed by exploiting the Lyapunov optimization, two algorithms that can both be implemented in a parallelized manner are then proposed based on the group sparse beamforming and the relaxed integer programming, respectively. The simulation results show that the two proposed algorithms can flexibly control the tradeoff between queuing delay and network power consumption, and they are highly scalable to different network sizes.4. Traffic queue-based dynamic radio resource optimization with congestion controlConsidering the arbitrary traffic arrival rates in H-CRAN that adopts orthogonal frequency division multiple access (OFDMA), the congestion control should be incorporated into the resource optimization to guarantee certain energy efficiency (EE) requirement and delay performance. In this dissertation, with the average network EE constraint and the queue stability constraints, the problem that maximizes the utility of average throughput is formulated as a stochastic optimization problem. By utilizing the Lyapunov optimization, the stochastic optimization problem is decomposed into three separate subproblems in each slot. Both the theoretical analysis and simulation results validate the tradeoff between throughput and queuing delay under certain EE requirement. The simulation results also show that the proposed algorithm outperforms the optimization algorithm that does not consider congestion control in terms of delay and energy savings.