LOW-COMPLEXITY CODING FOR THE CEO PROBLEM WITH MANY ENCODERS Gerhard Maierbacher and Jo˜ ao Barros Department of Computer Science University of Porto, Portugal URL: http://www.dcc.fc.up.pt/~barros Motivated by a sensor networking application, we consider the CEO prob- lem, in which noisy observations of a common source process are separately encoded and sent to a joint receiver. The goal of the decoder is to produce the best possiblereconstruction of the source subject to some distortion crite- rion. Focusingon the quadratic Gaussian case, for which the rate-distortion function has recently been determined, we present practical coding schemes that are feasible for a very large number of distributed encoders, allowing a desirable trade-off between rate-distortion performance and algorithmic complexity. 1 INTRODUCTION 1.1 The quadratic Gaussian CEO Problem Consider the following multiterminal source coding problem proposed in [7]. A CEO (Chief Executive Officer) is interested in some data U 0 , which is not di- rectely accessible. Therefore the CEO employs a team of M agents, who observe independently corrupted versions of U 0 . Because the estimation of U 0 is only one among many pressing matters to which the CEO must attend, the combined rate at which the agents may communicate information about their observation to the CEO is limited, say to R Σ bits per second. The agents are not allowed to convene. The CEO’s task is to estimate U 0 with the smallest possible average distortion D. The described setup offers a reasonable model for sensor networks, in which physical data is collected and encoded independently by a large number of geo- graphically separated sensor nodes, possibly observing different characteristics of