Dr. Amir Weiss Presents a Surprising Result in the Field of Compression and Localization

A new paper by Dr. Amir Weiss has been accepted for publication in the prestigious IEEE Signal Processing Letters journal. In the paper, On the Suboptimality of Rate–Distortion-Optimal Compression: Fundamental Accuracy Limits for Distributed Localization, Dr. Weiss presents surprising results on the interplay between data compression and localization via distributed sensing systems.

The paper examines the fundamental accuracy limits of distributed localization, in which a computing center receives information from multiple sensors after the information had been compressed due to communication constraints. In such scenarios, the amount of information sent to the computing center must be significantly reduced, but (potentially) in a different way for each task. Through comparison with the Cramér–Rao lower bound on the localization accuracy obtained based on information from classical optimal compression, Dr. Weiss shows that localization accuracy can be dramatically improved even by simple compression schemes that are aware of the fact that at the end of the processing chain, the information is specifically intended for localization.

The Advantage of Task-Oriented Compression

In his paper, Dr. Amir Weiss presents an important result for systems in which communication is integrated with spatial sensing capability (such as radar): compression that is optimal in the classical rate-distortion sense, generally considered the natural choice for recovering the compressed signal, can be significantly inferior when the goal islocalization. "Standard compression methods, which typically aim to minimize the mean squared error in recovering the signal itself, are not necessarily optimal for localization tasks, and in fact any other tasks apart from the recovery of the compressed signal itself," explains Dr. Weiss. "Moreover, it turns out that compression that is optimal in the rate-distortion sense can, in the process of compression, eliminate critical components of the received signal containing rich information about the location, thereby significantly degrading the resulting localization accuracy. It goes without saying how important the task of localization is, both in civilian and in military/security applications."

The compression scheme proposed in the paper is a simple alternative method for selecting specific frequency bands, which can achieve far better results at the same data rate imposed by communication constraints on the channels between the sensors and the computing center. One of the key insights from this work, as presented in the paper, is the need to develop new "localization-aware" compression methods for integrated sensing and communication (ISAC) systems.

Read the full paper on the IEEE website