Learning to Transmit Fresh Information in Energy Harvesting Networks Using Supervised Learning
Shiyang Leng, Aylin Yener · 2021 55th Asilomar Conference on Signals, Systems, and Computers · 2021
This paper provides a study of age of information (AoI) based scheduling in a wireless network over orthogonal channels and transmitters that are energy harvesting. The scheduling problem with associated interference and energy constraints is formulated as a mixed integer linear program which is known to be computationally hard. This paper considers utilizing supervised learning to obtain transmission policies that preserve freshness of information. Specifically, scheduling of status update transmissions of energy harvesting transmitters to their intended receivers is interpreted as a time-series classification problem, for which a bidirectional recurrent neural network is trained to solve. Experimental results for relatively small networks are provided for which the optimal policy can be determined and serves as a benchmark. It is observed that learning-based policies perform well in terms of average AoI with much faster runtime, suggesting that the approach may be worthwhile for larger networks.