Distributed Data Compression in Sensor Clusters: A Maximum Independent Flow Approach

TL;DR

This paper presents a polynomial-time, distributed algorithm for optimal data compression in sensor networks using a maximum independent flow approach, effectively managing limited communication capacities.

Contribution

It introduces a novel application of the maximum independent flow algorithm to sensor data compression, including a faster integral version and insights into the problem's submodular structure.

Findings

The MIF algorithm finds optimal compression flows in polynomial time.

A distributed implementation of the MIF algorithm is feasible.

An integral MIF algorithm offers faster performance for integral capacities.

Abstract

Let a cluster (network) of sensors be connected by the communication links, each link having a capacity upper bound. Each sensor observes a discrete random variable in private and one sensor serves as a cluster header or sink. Here, we formulate the problem of how to let the sensors encode their observations such that the direction of compressed data is a feasible flow towards the sink. We demonstrate that this problem can be solved by an existing maximum independent flow (MIF) algorithm in polynomial time. Further, we reveal that this algorithm in fact determines an optimal solution by recursively pushing the remaining randomness in the sources via unsaturated communication links towards the sink. We then show that the MIF algorithm can be implemented in a distributed manner. For those networks with integral communication capacities, we propose an integral MIF algorithm which completes much faster than MIF. Finally, we point out that the nature of the data compression problem in a sensor cluster is to seek the maximum independent information flow in the intersection of two submodular polyhedra, which can be further utilized to improve the MIF algorithm in the future.