|Ph.D Student||Sanderovich Amichai|
|Subject||Information Theoretic Aspects of Decentralized Processing|
|Department||Department of Electrical Engineering||Supervisors||Professor Yossef Steinberg|
|? 18? Shlomo Shamai )Shitz(|
|Full Thesis text|
We study the achievable rates of a decentralized processing system, where a multi-antenna nomadic transmitter is received by multiple agents (the wireless MIMO channel). The transmission is to be decoded at a remote final destination, connected to the agents (hot spots) by finite capacity links (wired backhaul network). In the nomadic regime, we assume that the agents do not have any decoding ability and hence the channel observations are processed and forwarded to the final destination through lossless links with fixed capacities. We propose new achievable rates based on elementary compression and also on a Wyner-Ziv (CEO-like) processing, for both fast fading and block fading channels, as well as for general discrete channels. Further limiting the nomadic transmitter to a circular-symmetric complex Gaussian signaling, new upper bounds are derived for both fast fading and block fading, based on the vector version of the entropy power inequality. These bounds are then compared to the achievable rates in several scenarios, and demonstrated to be tight in certain cases.
The asymptotic setting with numbers of agents and transmitter's antennas taken to infinity is also analyzed. The superiority of the Wyner-Ziv compression over the elementary compression is shown, where only the former can achieve the full diversity-multiplexing tradeoff.
We proceed by applying the developed results to derive new achievable rates for the uplink channel of a cellular network with joint multi-cell processing, where finite capacity backhaul links connect the cell sites (base stations) to the central processing unit.
Symmetric cellular network models are used, rendering analytical treatment plausible. For these idealistic models, we present achievable rates for cell-sites that use compress-and-forward schemes combined with local decoding, for Gaussian channels. The rates are given in closed form for the classical Wyner model. For modest backhaul capacities, the demonstrated rates fall rather close to the rates corresponding to the ultimate ideal multi-cell processing.