Cooperative Protocols In Dense Wireless Networks For Broadcast And Consensus

摘要

Network congestion is a communication bottleneck in large wireless networks which use packet-switched communications, impeding their performance in critical applications like network broadcast and in-network data processing. This forms the motivation for this thesis. We propose low complexity physical layer communication protocols that use cooperative transmission to ameliorate this problem. We show that cooperation at the physical layer can significantly improve performance with relatively small coordination overheads by exploiting the broadcast nature of the wireless channel. In the first part of the thesis, we address congestion in broadcast and propose decode and forward protocols that can accommodate multiple users broadcasting their content simultaneously. These protocols incorporate state of the art techniques like power control, successive interference cancellation, use of sideinformation for decoding, and interference alignment. By considering the two user linear network in depth we obtain the necessary conditions for the successful broadcast of the content of multiple users. It is shown that these physical layer cooperative protocols achieve lower broadcast latency than packet-switched flooding protocols and are suitable for broadcast in environments with fading. Additionally, we examine the performance of single user broadcast in the presence of decoding errors and show that the errors in its information flow need not be catastrophic. Supporting numerical results are included. In the second part of the thesis, we employ a similar approach and propose an efficient physical layer architecture for average consensus gossiping algorithms which perform in-network computation. This architecture relies on structured codes which combine channel and source coding. These codes result in consensus updates that are data driven, where transmissions are scheduled based on the states of the nodes rather than their index. Through this simple strategy we show that in spite of bandwidth and power limitations, nodes in an increasingly dense network can converge to the average with bounded delay and precision. Simulations show that this strategy outperforms packet-switched protocols even for moderately sized networks.