In this work we propose the concept of successive refinement of information for multiple users. We give achievable rate-distortion regions for the Gaussian source. The performance of the proposed scheme is shown to be superior to conventional approaches based on multiplexed solutions of optimal point-to-point successively refinable transmission strategies. We also provide a more universal approach to multiuser successive refinement based on the Wyner-Ziv method of coding with side-information, where the source reconstruction based on the base layer is treated as side-information during the refinement phase.
With the exponential growth in communication bandwidth and computing speed, more users are getting networked than ever before for seamless transmission of information between each other. Furthermore, heterogeneity of connectivity in the networking world is ever-increasing. The bandwidth of a communication link might vary from a few tens of kilobits per second in hand-held devices to a few megabits per second in high speed data links in ADSL lines. In such scenarios, the management of information storage and transmission has to incorporate interaction of information signals of multiple users. This provides the motivation for the study undertaken in this paper.
Consider an encoder wishing to transmit an information source X to two users as shown in Figure 1. The two receivers are connected to the encoder through errorless channels of unequal capacities. We wish to have a successively refinable transmission of information to these two receivers. We assume the broadcasting of information such that only a part of the information transmitted to the user with higher capacity is received by the user with lower capacity. Thus this is a generalization of the successive refinement of information for a single user. Since the capacities of the channels are unequal, the refinement of the users must also be unequal and proportional to the capacities of the individual links. Note that this work deals with source coding only, and the individual links are assumed to be error-free.
Figure 1: Multiuser successive refinement: the encoder sends an information sequence at rate R bits/sample. This information is intended for both the decoders. Each decoder has access to a quantized version of X.
1Professor, University of Michigan