Performance Enhancing of Heterogeneous Network through Optimisation and Machine Learning Techniques

In the last two decades, by the benefit of advanced wireless technology, growing data service cause the explosive traffic demand, and it brings many new challenges to the network operators. In order to match the growing traffic demand, operators shall deploy new base stations to increase the total cellular network capacity. Meanwhile, a new type of low-power base stations are frequently deployed within the network, providing extra access points to subscribers. However, even the new base station can be operated in low power, the total network energy consumption is still increased proportional to the total number of base station, and considerable network energy consumption will become one of the main issues to the network operators. The way of reducing network energy consumption become crucial, especially in 5G when multiple antennas are deployed within one site. However, the base station cannot be always operated in low power because it will damage the network performance, and power can be only reduced in light-traffic period. Therefore, the way of balancing traffic demand and energy consumption will be come the main investigation direction in this thesis, and how to link the operated power of base station to the current traffic demand is investigated. In this thesis, algorithms and optimisations are utilised to reduce the network energy consumption and improve the network performance. To reduce the energy consumption in light-traffic period, base stations switch-off strategy is proposed in the first chapter. However, the network performance should be carefully estimated before the switch-off strategy is applied. The NP-hard energy efficiency optimisation problem is summarised, and it proposes the method that some of the base stations can be grouped together due to the limited interference from other Pico cells, reducing the complexity of the optimisation problem. Meanwhile, simulated annealing is proposed to obtain the optimal base stations combination to achieve optimal energy efficiency. By the optimisation algorithm, it can obtain the optimal PCs combination without scarifying the overall network throughput. The simulation results show that not only the energy consumption can be reduced but also the significant energy efficiency improvement can achieve by the switched-off strategy. The average energy efficiency improvement over thirty simulation is 17.06%. The second chapter will tackle the issue of how to raise the power of base stations after they are switched off. These base stations shall back to regular power level to prepare the incoming traffic. However, not all base stations shall be back to normal power due to the uneven traffic distribution. By analysing the information within the collected subscriber data, such as moving speed, direction, downlink and time, Naive Bayesian classifier will be utilised to obtain the user movement pattern and predict the future traffic distribution, and the system can know which base station will become the user's destination. The load adaptive power control is utilised to inform the corresponding base stations to increased the transmission power, base stations can prepare for the incoming traffic, avoiding the performance degradation. The simulation results show that the machine learning can accurately predict the destination of the subscriber, achieving average 90.8% accuracy among thirty simulation. The network energy can be saved without damage the network performance after the load adaptive function is applied, the average energy efficiency improvement among three scenarios is 4.3%, the improvement is significant. The significant improvement prove that the proposed machine learning and load adaptive power modification method can help the network reduce the energy consumption. In the last chapter, it will utilise cell range expansion to tackle the resources issue in cooperative base station in joint transmission, improving downlink performance and tackle the cell-edge problem. Due to the uneven traffic distribution, it will cause the insufficient resources problem in cooperative base station in joint transmission, and the system throughput will be influenced if cooperative base station executes joint transmission in high load. Therefore, the cell range expansion is utilised to solve the problem of unbalanced traffic between base station tier, and flow water algorithm is utilised to tackle the resources distribution issue during the traffic offloading. The simulation shows the NP-hard problem can be sufficiently solved by the flow water algorithm, and the downlink throughput gain can be obtained, it can obtain 26% gain in the M-P scenario, and the gain in P-M scenario is 24%. The result prove that the proposed method can provide significant gain to the subscriber without losing any total network throughput.