Liquid State Based Transfer Learning for 360° Image Transmission in Wireless VR Networks

Abstract

In this paper, the problem of 360° image transmission is studied for a wireless network of virtual reality (VR) users that communicate with cellular base stations (BSs). The VR users will send their uplink tracking information to the BS and receive the VR images in the downlink. To satisfy VR users' delay target, the BSs can change the image transmission format for each image requested by users so as to reduce the downlink traffic load. Meanwhile, the VR users can directly rotate the already received VR image and use the rotated VR images at a later time to further reduce the downlink traffic load. This 360° image transmission and image rotation problem is then formulated as an optimization problem whose goal is to maximize the users' successful transmission probability which is defined as the probability that the delay of tracking information and image transmission for each VR user satisfies the VR delay requirement. A liquid state machine (LSM) based transfer learning algorithm is proposed to solve this optimization problem. The proposed LSM-based transfer learning algorithm enables each BS to transfer the already learned successful transmission to the new successful transmission that must be learned so as to increase the convergence speed. Simulation results show that the proposed algorithm achieves 14.9% gain in terms of successful transmission probability compared to Q-learning.