Joint source / channel coding forwwan multiview video multicast with cooperative peer-to-peer repair

WWAN video multicast is challenging because of unavoidable packet losses, and inability to perform retransmission per packet for every user due to the NAK implosion problem. Previously proposed cooperative peer-to-peer repair (CPR) strategy, leveraging on the broadcast nature of wireless transmission and "uncorrelatedness" of receivers' channels, calls for peers with good WWAN channels to streaming server (rich peers) to locally relay packets lost to peers with bad WWAN channels (poor peers) over a secondary ad-hoc WLAN network. In the interactive multiview video streaming (IMVS) scenario, however, where users can each periodically select one out of many available views for decoding and display, a poor peer may not have a neighboring rich peer watching the same view for packet recovery via CPR. In this paper, we propose a new CPR strategy for peers to repair lost packets to neighbors watching different views. The key idea is for server to transmit depth maps in addition to texture maps, so that lost frames in a different view can be reconstructed using depth-image-based rendering (DIBR). Like Forward Error Correction (FEC) packets, encoded depth maps incur an overhead in redundant information transmission to counter network losses, and we allocate optimal amount of bits for FEC packets (for same-view direct path protection via WWAN source) and depth map encoding (for different-view indirect path via CPR peers) to minimize expected distortion. Experimental results show our proposed CPR scheme offers a 3.4dB PSNR improvement over a non-CPR scheme that relies on FEC only.