Adaptive Interference Alignment For Wireless Networks

Along with the rapid development of the global economy,there has been an ever increasing demand for high capacity wireless communication networks.While various technologies have been utilized to promote the spectrum efficiency of the networks,many of them have also caused severe interference problems.For example,reuse of the same set of subbands may lead to severe inter-cell interference in 4G/LTE netowrks when the co-networking mode is applied,remarkably reducing the achievable rate of cell edge users.Existing interference mitigation techniques,such as Fractional/Soft Frequency Reuse,are mainly inspired by the idea to depart the potential interferers from the intended receivers,which to some extent may help but they still somewhat decrease the overall throughput because some of the subbands cannot be reused.Based on the signal space principles,interference alignment(IA)has been regarded as a new breakthrough in information theory in recent years,which may provide an alternative solution to the interference problem.This dissertation presents the results of a study on the applications of the IA technique to various wireless communications scenarios such as cognitive radio networks,cellular networks and NOMA networks.It features the user-demands-oriented closed-form IA algorithms and low complexity power allocation methods.A summary of the contributions made is as follows:1.For cognitive networks,a universal IA algorithm is proposed to allow spectrum-sharing among a pair of primary users and multiple pairs of secondary users,in accordance with their Do F demands.The average achievable Do F of the scheme is then analysed and a lower bound derived.A comparative study shows that the proposed scheme outperforms all existing ones in both effective Do F or the served user ratio.2.For downlink cellular networks,we propose an adaptive IA algorithm(AIA)to cope with various data rate demands from mobile users.Taking advantage of the prior knowledge of a unitary matrix,the proposed AIA not only requires less CSI feedback than ZF-IA,but it is also robust to CSI error and reduces the calculation payload as well.Next,to maximize the sum rate,yet another geometry-based power allocation algorithm(GPA)of low complexity is derived which closely matches the optimal and well-known water-filling(WF)algorithm even at medium SNRs.Furthermore,the proposed GPA is designed in closed form and has better fairness than WF.3.In the scenario of downlink NOMA networks,we design a closed-form IA algorithm to simultaneously remove ICI,inter-cluster and intra-cluster interferences.Since the direction of the ICI is pre-set to be known the complexity of the algorithm is greatly reduced.Although the proposed scheme just serves as many users as ICA-CBF NOMA,it requires less CSI feedback and is superior in sum rate.Moreover,a sub-optimal 2-stage power allocation algorithm(SINPA)is proposed to maximize the average satisfaction index of the mobile users which has been proved as optimal for intra-cluster power allocation.Comparative studies with exhaustive search schemes have shown its effectiveness.4.Finally,a deep neural network(DNN)structured IA scheme is presented to accomplish interference alignment in cellular networks,in which a DNN network learns to design beamforming matrices from the proposed AIA scheme.With the DNN-IA scheme implemented onboard a GTX 1060 GPU driven by Tensorflow+Keras platform,simulation results have confirmed the effectiveness and potential of DNNs for application to future interference networks.It is worth mentioning that all the algorithms proposed in this study are in closed-form and are in general easy to implement and analyze,which has been verified by thorough simulation.