Long Short-Term Memory Network-based Power Control for Green Wireless Communications

• Main Authors: Chia-Yu Wan, 萬家妤
• Other Authors: Meng-Lin Ku
• Format: Others
• Language: en_US
• Published: 2019
• Online Access: http://ndltd.ncl.edu.tw/handle/y4zw2x

碩士 === 國立中央大學 === 通訊工程學系 === 107 === With the increasing prevalence of Internet of things, the energy consumption problem in becomes a challenging issue due to the limited battery storage on wireless communication devices. Thanks to the emergence of energy harvesting techniques, this problem can be efficiently solved by scavenging and storing energy in rechargeable batteries from ambient environments like solar. A conventional approach to schedule the energy usage over the future of a period of time, which is also called control power in energy harvesting communications, is to apply the convex optimization technique. This scheme relies on the perfect knowledge of the future information of energy harvesting conditions and channel gains control, which makes it difficult to be implemented in real applications. In this thesis, deep learning design frameworks, which only require the past knowledge of energy harvesting and channel conditions, are proposed to predict the future power control values in single-user and multi-user energy harvesting communications. For the single-user case, we utilize the historical solar data to calculate the optimal solutions via convex optimization, which is then used in the training of a long short-term memory network (LSTM) for predicting power control values. As an extension, an LSTM-based power control scheme is investigated in the multi-user scenario by using the reference power control solutions obtained from a weighted sum minimum mean-square-error (WMMSE) algorithm. The effectiveness of the proposed LSTM-based power control schemes is finally evaluated by computer simulations, and the results show that the proposed schemes can achieve sufficient good system throughput as compared with the baseline solutions.