MIT Department of Electrical Engineering & Computer Science

Asymptotically Optimal Waterfilling in Multiple Antenna Multiple Access Channels

Pramod Viswanath
University of California at Berkeley

Tuesday, March 14, 2000
2:15 PM (refreshments 2:00)
Grier Room, Room 34-401B
EECS Special Seminar

Abstract

Multiple antennas are an important resource in increasing the capacity of wireless communication systems. We focus on the uplink of a single cell system with the base station equipped with an antenna array. Our performance measure is sum capacity or spectral efficiency of the uplink - this is time-averaged sum of rates at which users can jointly reliably communicate. We are interested in the scenario when the base station tracks the channels from the users to the antenna array. Our main focus is on the key design question: How much increase in sum capacity does feedback of the channel states to the users buy? Our main result is the conclusion that there is hardly any gain in sum capacity by the users having side information of the channel states when there are sufficient number of antennas. This result holds when the slow fading component is varying very slowly in the time scale of symbol intervals and the result does not depend on the fast fading distribution. This conclusion is rather surprising from the point of view of the single antenna result: Knopp and Humblet showed that in a single antenna uplink system, the loss in not feeding back channel state to the users is significant. When the slow fading component is also considered in the antenna array scenario, we show the optimality (in the regime of sufficiently large number of antennas) of a simple waterfilling policy which for every user depends only on the slow fading component from that user to the antenna array. Application of the techniques used to derive this result to power control in DS-CDMA (Direct Sequence Code Division Multiple Access) are mentioned. A brief overview of the speaker's contribution to resource allocation problems in DS- CDMA is also given.

This is a joint work with Professors Venkat Anantharam and David Tse at the University of California, Berkeley.