Abnormality Detection and Monitoring in Multi-Sensor Molecular Communication

Abstract

In this paper, we investigate the problem of detecting and monitoring changes (abnormality) in molecular communication (MC), using the quickest change detection (QCD) schemes. The objective is to watch an environment using a sensor network and make decisions on the time and location of changes based on the received signals from sensors in the fusion center (FC). Such assumptions call for considering spatial and temporal correlations among sensors’ transmitting signals. We use the framework of Partially Observable Markov Decision Processes (POMDPs) based on non-homogeneous Markov models. The metric in detection (stopping-time) scenario is to minimize the delay of announcing an abnormality occurrence from the real change time, in a two-hypothesis setting, subject to constraints on false alarm and missed identification probabilities. Since the optimum detector is very complicated, by utilizing a myopic policy, sub-optimum detectors with less complexity and acceptable performances are proposed. For monitoring scenario, the goal is to determine how the changes are spread in time, in a multi-hypothesis setting. Such goal asks for deciding on multiple changes in time, which is different from the stopping-time scenario. In the latter scenario, we decide only about whether one change has occurred or not, then we stop decision making. In the monitoring scenario, one optimum and two sub-optimum metrics are defined and their corresponding detectors are derived. In this scenario, the detectors make decisions only in the time slots that a reliable decision can be made. The defined metrics minimize the waiting-time , the time that the detector does not make decision about any of the hypotheses due to lack of assurance, subject to constraints on false alarm and missed identification probabilities. In both scenarios, we mathematically model the system and evaluate the performance of the proposed detectors. For performance evaluation, we consider modeling a previously reported system of tumor growth in a tissue. The results confirm out-performance of the proposed detectors in comparison with existing ones.