MAPNET (Mathematical Modelling and optimization of programmable 5G networks) is an MSCA-IF funded project. The primary goal of MAPNET is to propose new modelling techniques for 5G-and-beyond networks. The emergence of a variety of futuristic applications such as augmented and virtual reality (AR/VR), autonomous driving, industry 4.0, and edge-robotics relying on wireless connectivity brings completely new requirements for telecommunication networks. To ensure the reliable, safe and secure operations of these application with heterogeneous requirements on data rates and latency, telecommunication networks are required to be highly adaptive, resource efficient (energy and bandwidth), ubiquitous and dependable. To this end, ultra-dense network deployment, use of millimeter wave (mmWave) spectrum and access network slicing have emerged as the main technology candidates to fulfill the stringent application requirements. However, performance optimization of ultra-dense mmWave networks using conventional mathematical techniques is very difficult due to dense-deployments, highly fluctuating channel conditions and directional transmissions. Furthermore, energy efficiency analysis is very important for mmWave networks as a large number of antennas together with other components (phase shifters, analog to digital converters, etc.) is likely to increase the energy consumption. Similarly, network slicing is a new concept that partition a single physical infrastructure into logically separated resource blocks, also called slices, such that the different requirements of different services can be satisfied. The emergence of non-permanent network elements such as drones bring new challenges to the access network slicing systems as the services employing these network elements demand very short-term slice requests but with a very stringent latency/data-rate requirements and limited energy budget. To solve aforementioned challenges of future networks, novel problem formulations and efficient solutions are desired. During this action (MSCA-IF MAPNET), application of both traditional mathematical optimization and machine learning based techniques is investigated for resource allocation in ultra-dense 5G-and-beyond networks to meet the requirements on data rates, latency and energy efficiency. Novel algorithms are devised to achieve better throughput, latency and energy efficiency.
Keywords: Millimeter wave networking, ultra-dense networks, multi-objective resource allocation, network slicing and RAN sharing.
Research objectives
The specific research objectives are listed as the following:
1) Mathematical modelling of ultra-dense networks
2) Energy-efficiency maximization of ultra-dense networks
3) Network slicing reformulations to satisfy user demands
1) Deep reinforcement learning based optimization of ultra-dense Millimeter wave networks: we developed a DRL based framework for power and beamwidth allocation in ultra-dense mmWave deployment. Instead of centralized DRL, we propose a distributed DRL framework in which each base station (BS)- user equipment(UE) link is , modeled as a DRL agent. Each agent makes its own decision about power and beamwidths to be used during a slot. The reward are calculated centrally and distributed to the all the agents at the end of each slot. The reward function relies on the network sum-rate and accounts for the contributions of each agent towards the network sum-rate by the data rate of each link and the amount of amount of interference generated by the link to all the other neighbor links.
2) Energy efficiency maximization of ultra-dense Millimeter wave networks: Energy efficiency is a very important metric for ultra-dense mmWave deployments catering for the high data-rate applications. To address this challenge, we define a novel energy consumption analysis model of mmWave UDNs. The model explicitly accounts for the impact of the transmitter and receiver beamwidths on the network power consumption. We use a bi-objective utility function and Tchebyshev method is used to derive the Pareto boundary between the two objectives. The competition among wireless mmWave transmitters is modelled using non-cooperative game theory. The uniqueness of Nash equilibrium (NE) is established and algorithm based on Deep Q-learning is developed to compute the NE.
3) Latency and energy Optimization of mobile edge/fog computing systems: We consider a fog/edge computing aided drone communication system where latency and energy consumption optimizations are highly important. We assume an infrastructure owner which allocates frequency and computing power to different tenants that operates their drones for emergency communications. It is assumed that the drones have limited compute power and heavy tasks are needed to be offloaded to fog/edge server. The goal is to optimally share the available spectrum and computing power from the energy efficiency and latency requirements perspectives. We formulate the joint minimization of both service latency and energy consumption as a bi-objective minimization problem
4) Admission control in radio access network slicing systems: We consider a network slicing scenario with high density of tenants that requires short-term slices. We assume that MNOs have a admission waiting queue for the tenants that can not be serviced immediately. Since it is obvious that the MNO may extend the services of its long-term associates first, wait-time in the admission queue can be more than the expected for the short-term requests. In this situation, tenants can choose alternate MNOs but the short-term tenants can face the similar situation in the waiting queue of the another MNOs. We propose strategies to avoid this frequent switching behavior using the tools of game theory resulting in a more stable tenant behavior and increased profit for the MNOs.
Ultra-dense network deployment, use of millimeter wave radio access technology and network slicing for resource allocation is becoming a reality with the ongoing 5G roll outs worldwide. The ER developed cutting edge techniques to address the challenges of resource allocation in ultra-dense mmWave access networks to support the futuristic 6G applications such as smart cities, wireless AR/VR and haptic teleportation where a very high data rate, ultra-low latency and energy efficiency are of the paramount importance. MAPNET’s innovations resulted in efficient resource allocation and network modelling techniques to ensure the stringent requirements on latency and data rates.The ER got opportunity to work on cutting edge topics in the area of wireless networks and attended many courses on hot-topic related to stochastic geometry, machine learning and artificial intelligence. During the overall stay at the host institute, ER acquired new scientific and transferable skills.
MAPNET has received funding from the European Union’s Horizon 2020 research and innovation programme under the Marie Sklodowska-Curie (Individual Fellowship) grant agreement number 796378(link).
