QoE-oirented 360-degree Video Streaming

In recent years,Virtual Reality(VR)technology has become more and more popular.360-degree video,as the fundamental technology of VR,has drawn more attention for its immersive experience.During playback,the user is only shown a portion of the panoramic video,which is called viewport.So the transmission of the content out of the viewport wastes a lot of bandwidth resources.Due to its ability to reduce bandwidth consumption,tile-based 360-degree video streaming has gained popularity in recent years,in which panoramic view of the 360-degree video is first projected into a twodimensional(2D)perspective.Following that,each frame of the 2D video is spatially separated into several identical tiles which can be encoded into distinct bitrate levels.When transmitting the 360-degree video to the user,each tile is assigned a unique bitrate level according to the viewport.In the tile-based 360-degree video streaming,it is essential to predict future viewport and to allocate higher bitrates to tiles inside the predicted viewport to optimize the QoE of the users.However,it is challenging to transmit high-bitrate video and achieve the high Quality of Experience(QoE)under limit and fluctuating network conditions.The majority of existing work focuses on short-term viewport prediction,which is prone to rebuffering in dynamic network conditions.On the other hand,the recently developed on-policy Deep Reinforcement Learning(DRL)-based bitrate allocation approaches suffer from the Exploration-Exploitation dilemma,and it’s hard to find the optimal solution.To address these issues,this paper presents a tile-based adaptive 360-degree video streaming system,named LS360,which consists of long-term viewport prediction and adaptive bitrate allocation.First,a Long Short-Term Memory(LSTM)-based viewport prediction model is proposed for long-term viewport prediction.A precise long-term viewport prediction is advantageous for coping with bandwidth changes and avoiding rebuffering by pre-fetching additional tiles into the buffer.We conducted a study on VR movement dataset and found the similarity of viewing pattern among users.According to this,our viewport prediction makes use of the similarity feature from all users’ previous movement information and the target user’s fixation movement feature to improve prediction accuracy,especially for long-term.Next,we employ the off-policy Soft Actor-Critic(SAC)algorithm to make optimal tile bitrate allocation decisions.We formulate the user QoE and propose the QoE-aware bitrate allocation problem.The QoE optimization problem is transferred into the online bitrate allocation based on the predicted viewport.The SAC agent makes bitrate decisions by taking the predicted viewport,playback buffer,and bandwidth-related information into account.Finally,we conduct simulations on the real head movement dataset and real network to evaluate the performance of LS360.The experimental results demonstrate that LS360 outperforms state-of-the-art streaming algorithms in terms of long-term viewport prediction accuracy and QoE under different bandwidth conditions.