Hybrid Interference Mitigation for Cognitive Small Cells in 5G Cellular Networks

Abstract

Small cells (SCs) based ultra-dense heterogeneous networks (HetNets) are one of the promising solutions for increased coverage and capacity in 5G cellular networks (CN). In co-channel deployment mode, the performance of the multi-tiered SC HetNets is limited by the co-tier and cross-tier interferences (CoI and CrI). Although the efficient resource allocation techniques can handle these interferences effectively, their complexity linearly increases with the density of the HetNets resulting from the dynamic and unplanned deployment of SCs. Multiple schemes have been proposed for efficient radio resource management (RRM) to handle the CoI and CrI in HetNets but Quality of Service (QoS) provision to macrocell user equipment UE௜௠ and small cell user equipment UE௖,௞ ௦ simultaneously is still an open research problem. Recently, intelligent schemes for RRM using artificial intelligence (AI) and machine learning (ML) have proved their effectiveness in wireless communication systems due to their ability to adapt to dynamic conditions and self-optimization. Therefore, SC HetNets can be implemented through an algorithm that is self-organizing and adaptive to the dynamic conditions of SCs in multi-tiered HetNets. In this dissertation, a hybrid solution based on Q-Learning (QL) and adaptive power allocation is proposed and evaluated for ultra-dense SC HetNets for simultaneous CrI and CoI mitigation and QoS provision to both UE௜௠ and UE௖,௞ ௦ . QL-based SCs adapt their transmit power to mitigate interferences and meet the minimum required capacity of UE௜௠ (𝐶௠ ௜ ) and UE௖,௞ ௦ (𝐶௖,௞ ௦ ) simultaneously, to provide QoS. Distributed adaptive power allocation scheme based on QL has been explored using independent learning (IL) and cooperative learning (CL) paradigms. Although the proposed QL scheme in the IL paradigm (IQL) successfully mitigated the CrI and CoI simultaneously and ensured the QoS to all user equipment (UEs) in the SC HetNets but in the IL paradigm SCs consider the neighbouring SCs as part of the environment in QL. To further improve the self-optimization of SCs in HetNet to improve the capacities of UE௜௠ and UE௖,௞ ௦ , we proposed CL based QL (CQL) algorithm for adaptive power allocation to SCs. The proposed distributive algorithm in the CL paradigm interacts with neighbouring SC to exchange useful information for self-optimization. xi The proposed QL schemes, IQL and CQL, successfully mitigated the CrI and CoI simultaneously and provided the minimum required 𝐶௠ ௜ and 𝐶௖,௞ ௦ to ensure minimum QoS. The proposed schemes have also shown a significant improvement in the 𝐶௠ ௜ and 𝐶௖,௞ ௦ in high interference scenarios as compared to the prior works. A detailed analysis of IQL and CQL shows that CQL has performed significantly better as compared to the IQL in providing higher 𝐶௠ ௜ , 𝐶௖,௞ ௦ and sum capacities of UE௖,௞ ௦ , 𝐶௦௨௠ ௦ . The CQL also reduced the sum power of SCs 𝑃௦௨௠ in ultra-dense SC HetNets by effectively adapting to the density and dynamic conditions as compared to IQL. Although the CQL performance is better than the IQL due to the exchange of useful information in CL results in slightly increased computational time. The increase in computational time is proportional to the density of SCs in HetNets. To reduce the computational time for CQL in ultra-dense SC HetNets, we proposed an optimal clustering technique which effectively reduced the computational time for CQL by dividing a large number of SC into smaller clusters of SC operating in parallel. In future research, we will explore the user association and load balancing in ultra-dense HetNets through QL-based solution in CL paradigm to fully utilize the benefits of CQL. Small cells (SCs) based ultra-dense heterogeneous networks (HetNets) are one of the promising solutions for increased coverage and capacity in 5G cellular networks (CN). In co-channel deployment mode, the performance of the multi-tiered SC HetNets is limited by the co-tier and cross-tier interferences (CoI and CrI). Although the efficient resource allocation techniques can handle these interferences effectively, their complexity linearly increases with the density of the HetNets resulting from the dynamic and unplanned deployment of SCs. Multiple schemes have been proposed for efficient radio resource management (RRM) to handle the CoI and CrI in HetNets but Quality of Service (QoS) provision to macrocell user equipment UE௜௠ and small cell user equipment UE௖,௞ ௦ simultaneously is still an open research problem. Recently, intelligent schemes for RRM using artificial intelligence (AI) and machine learning (ML) have proved their effectiveness in wireless communication systems due to their ability to adapt to dynamic conditions and self-optimization. Therefore, SC HetNets can be implemented through an algorithm that is self-organizing and adaptive to the dynamic conditions of SCs in multi-tiered HetNets. In this dissertation, a hybrid solution based on Q-Learning (QL) and adaptive power allocation is proposed and evaluated for ultra-dense SC HetNets for simultaneous CrI and CoI mitigation and QoS provision to both UE௜௠ and UE௖,௞ ௦ . QL-based SCs adapt their transmit power to mitigate interferences and meet the minimum required capacity of UE௜௠ (𝐶௠ ௜ ) and UE௖,௞ ௦ (𝐶௖,௞ ௦ ) simultaneously, to provide QoS. Distributed adaptive power allocation scheme based on QL has been explored using independent learning (IL) and cooperative learning (CL) paradigms. Although the proposed QL scheme in the IL paradigm (IQL) successfully mitigated the CrI and CoI simultaneously and ensured the QoS to all user equipment (UEs) in the SC HetNets but in the IL paradigm SCs consider the neighbouring SCs as part of the environment in QL. To further improve the self-optimization of SCs in HetNet to improve the capacities of UE௜௠ and UE௖,௞ ௦ , we proposed CL based QL (CQL) algorithm for adaptive power allocation to SCs. The proposed distributive algorithm in the CL paradigm interacts with neighbouring SC to exchange useful information for self-optimization. xi The proposed QL schemes, IQL and CQL, successfully mitigated the CrI and CoI simultaneously and provided the minimum required 𝐶௠ ௜ and 𝐶௖,௞ ௦ to ensure minimum QoS. The proposed schemes have also shown a significant improvement in the 𝐶௠ ௜ and 𝐶௖,௞ ௦ in high interference scenarios as compared to the prior works. A detailed analysis of IQL and CQL shows that CQL has performed significantly better as compared to the IQL in providing higher 𝐶௠ ௜ , 𝐶௖,௞ ௦ and sum capacities of UE௖,௞ ௦ , 𝐶௦௨௠ ௦ . The CQL also reduced the sum power of SCs 𝑃௦௨௠ in ultra-dense SC HetNets by effectively adapting to the density and dynamic conditions as compared to IQL. Although the CQL performance is better than the IQL due to the exchange of useful information in CL results in slightly increased computational time. The increase in computational time is proportional to the density of SCs in HetNets. To reduce the computational time for CQL in ultra-dense SC HetNets, we proposed an optimal clustering technique which effectively reduced the computational time for CQL by dividing a large number of SC into smaller clusters of SC operating in parallel. In future research, we will explore the user association and load balancing in ultra-dense HetNets through QL-based solution in CL paradigm to fully utilize the benefits of CQL.