Premium
On achieving throughput optimality with energy prediction–based power allocation in 5G networks
Author(s) -
Gardazi Syed Fasih Ali,
Ahmad Rizwan,
Qureshi Hassaan Khaliq,
Ahmed Waqas
Publication year - 2018
Publication title -
transactions on emerging telecommunications technologies
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.366
H-Index - 47
ISSN - 2161-3915
DOI - 10.1002/ett.3438
Subject(s) - throughput , computer science , benchmark (surveying) , efficient energy use , energy consumption , mathematical optimization , wireless , telecommunications , engineering , mathematics , geodesy , electrical engineering , geography
The expected presence of huge number of low data rate devices in 5G/beyond 5G networks can compromise its overall capacity and cause an increase in carbon footprint. With continuous improvement in radio access technologies, these low data rate devices can be offloaded to self sustainable small cells operating in spectrum extensions (licensed/unlicensed). To ensure self‐sustainability and end‐to‐end connectivity of these low data rate devices naturally calls for energy harvesting (EH)–based cooperative communication solution. Therefore, in this paper, we propose an energy prediction–based power allocation (EPPA) policy to improve the utilization of intermittent EH arrivals to maximize throughput. Moreover, EH predictions are obtained through the training of an M ‐order Markov model, which are further used to derive estimated throughput optimal (EPPA‐NO) and suboptimal (EPPA‐I) power allocation/data rate selection policies. An EPPA‐K scheme with exact knowledge of future arrivals are used as an upper bound to benchmark the performance of EPPA‐NO and EPPA‐I in terms of throughput, delay, energy consumption per bit, and respective complexity. It is observed that the performance of these schemes is highly dependent on the prediction accuracy. To this end, we further propose a learning‐based weighted linear regression scheme. For an energy‐constrained scenario, using weighted linear regression further improves energy efficiency over EPPA‐NO and EPPA‐I and reaches within 1% of EPPA‐K for a buffer size of 700 bits.