Decentralized Processing:
The Impact of a Common Finite Capacity Feedback Link

Shlomo Shamai
Electrical Engineering Dept.
Technion, Israel

We discuss the scenario where a remote nomadic user communicates to a destination via a set of intermediate agents. The agents are ignorant of the codebook used due to the nomadic regime and are connected to the destination via reliable links of finite capacity. We focus here on independent Gaussian channels to all agents, where the transmitter and all agents are equipped with a single antenna.

First we review the results associated with a single transmit antenna, invoking decentralized quantization, which yields the ultimate achievable rate, in the nomadic regime with Gaussian signalling.

We confine attention to the basic scheme with two agents and consider the impact of a finite capacity feedback link from the final destination to the agents, allowing for a single round of conferencing. Network coding is optimal here in the sense of facilitating the full exploitation of the conferencing phase on the feedback link (from the destination to the agents). The impact of this conferencing protocol on the ultimate performance is quantified, and implications of layered coding in this scenario are also considered.



The talk is based on joint work with Amichai Sanderovich, and Yossef Steinberg.