As a consequence of the 5G network densification and heterogeneity, there is a competitive relationship between the sufficient satisfaction of the cell users and the powerefficiency of 5G transmissions. This paper proposes a Deep Q-Learning (DQL) based power configuration algorithm by jointly optimizing the energy-efficiency (EE) and throughputadequacy (JET) of 5G cells. The algorithm exploits the user demands to effectively learn-and-improve the user fulfillment rate, while ensuring cost-efficient power adjustment. To evaluate the potency of the developed methodology, several validation setups were conducted comparing the outcomes of the JET-DQL with those derived from conventional power control schemes, namely a Water-filling (WF) algorithm, a weighted minimum mean squared error (WMMSE) method, a heuristic solution and three fixed power allocation policies. JET-DQL algorithm exhibits a remarkable trade-off between the allocated throughput (ensuring high user satisfaction rates and average behavior in total allocated throughput relative to baselines), while resulting into low-valued (almost minimum) power configurations. In particular, even for strict demand scenarios, JET-DQL outperforms the other baselines with respect to EE showing a gain of 2.9-4.5 relative to others, although it does not provide the optimal sum-rate utility and minimum power levels.

No abstract

The goal of the study presented in this paper is to evaluate the performance of a proposed Reinforcement Learning (RL) power allocation algorithm. The algorithm follows a demand-driven power adjustment approach aiming at maximizing the number of users inside a coverage area that experience the requested throughput to accommodate their needs. In this context, different Quality of Service (QoS) classes, corresponding to different throughput demands, have been taken into account in various simulation scenarios. Considering a realistic network configuration, the performance of the RL algorithm is tested under strict user demands. The results suggest that the proposed modeling of the RL parameters, namely the state space and the rewarding system, is promising when the network controller attempts to fulfill the user requests by regulating the power of base stations. Based on comparative simulations, even for strict demands requested by multiple users (2.5 -5 Mbps), the proposed scheme achieves a performance rate of about 96%.

No abstract

No abstract