Beamforming algorithms for information relaying in dense wireless networks

Thesis Abstract

In this project we develop a class of bandwidth-efficient algorithms for information relaying in large-scale wireless ad-hoc networks. The settings we consider involve a single source communicating its data to a destination via the aid of low-power low-cost relay nodes.

In its simplest two-hop relaying form, data directly broadcasted to the relays from the source are directly relayed to the destination through a shared fading channel. We assume that the relays employ decode-and-forward or amplify-and-forward preprocessing prior to forwarding their data to the destination via beamforming. The beamforming weights are formed at the destination and fedback to the relays via broadcasting. They are constructed using knowledge of the relay-destination channel coefficients and an m-bit description of each source-relay channel state information (CSI).

For both relay data preprocessing models, we present methods for optimizing the m-bit quantizer employed at each relay for encoding its source-relay channel quality level, and for choosing the beamforming weights at the destination, so as optimize the destination uncoded bit error rates (Pr(e)). We also study the effect of the relative source-relay relay-destination distances on the Pr(e) for both relay preprocessing models. We use our findings to develop locally-optimized adaptive data-preprocessing algorithms at the relays.

We also develop extensions involving multi-hop networks with hierarchical cluster-based relaying. At each hop of these relay networks, each of the receiving relays obtains a beamformed version of the data of a distinct subset of the transmitting relays. As our simulations and analysis reveal, making available at each cluster head (CH) an optimized one-bit description of the effective source-relay CSIs associated with the transmitting relays in its cluster is sufficient. Specifically, not making fully available to the CHs, the source-relay CSIs, results in only a minimal loss in the Pr(e).