Service Migrations in TSCH Network using Wireless Channel Estimation and Prediction

Abstract

Industrial Wireless Sensors and Actuators Networks (IWSANs) are gateway to the Industrial 4.0, which promises to realize smart factory leading to the Industrial Internet of Things (IIoT). It employs Cyber-Physical Systems (CPSs) to enhance operational efficiency and flexibility while reducing cost. IWSANs are delay-sensitive and always require low latency and reliable connection from sensor to actuator to successfully perform a physical action. Reliability and low-latency complement each other to prevent expected failures in wireless medium. In this way, detecting and predicting failure before it actually occurs is key to actually avoid it well in time. Detection and predictions are imperative in locating faults and failures. The causes of failures in a sensor or actuator can include hardware malfunction, poor battery life, interference, accident, and short term wireless connectivity problems. Although, industrial environment mostly undertakes redundant resource to circumvent such issues, yet poor coordination among multiple resources and inaccurately predicting failures may result in losses. In such a scenario, migration of services come to be a rescue, where an intermediary can migrate service from one device, which cannot complete a task due to resource exhaustion, to a more resource-rich device. Thus, in this paper, we focus on wireless connectivity failures caused by interference in the 2.4GHz frequency band. We do it by designing an Multi Channel Sniffing Setup (MCSS) testbed, that acts as a spectrum observer and is deployed in different locations in industrial WSAN. Alongside, we use the concept of Cognitive Radio (CR) to predict interference and noise level in the spectrum by proposing an Intelligent Low-power Wireless Spectrum Prediction (ILPWSP) based on Deep Q Network (DQN). The MCSS testbed and the ILPWSP coordinate in assessing wireless connectivity risks, predict failures in sensor and actuator nodes and then make efficient decisions on the migration of services from one device to another device. Our results show the feasibility of spectrum prediction with an acceptable ratio for reliable IWSN.