We consider a system consisting of a source node which sends updates to a monitoring station using multiple communication channels. These updates are measurements of a time-varying quantity that the source is tracking. The monitoring station uses these updates to make control decisions and therefore needs fresh and accurate updates for optimal performance. One natural way to deal with the case where the updates can be unreliable/noisy is to use multiple recent updates from the source to obtain a robust estimate of the current value of quantity of interest. A recently proposed metric well-suited for this setting is called Loss-of-Credibility (LoC). LoC is a measure of the credibility of the estimate the monitoring station can obtain using the updates it has received.In this work, we focus on the setting each attempted update is successful with a probability which is an unknown function of the channel used for communication. The goal is to design a scheduling algorithm to determine which channel to use to send updates in order to minimize LoC. We model the problem as a multi-arm bandit (MAB) with communication channels as arms and a suitably defined LoC regret metric. We show that UCB and Thompson Sampling are order-optimal for our LoC setting. In addition, we propose variants of these policies which outperform the original policies in simulations.

We study the task of scheduling in a multi-source system where sources report their time-varying information to a central monitoring station via multiple orthogonal channels. The Age-of-Information of a source is defined as the amount of time elapsed since the latest update from that source was received at the monitoring station. At each time instant, the system pays a cost that is a function of the current Ages-of-Information of the sources. Our goal is to design scheduling policies to minimize this cost. We draw a novel parallel between our scheduling problem and the minimum mean cost cycle problem in weighted graphs and use this insight to design optimal scheduling policies for a very general class of cost functions. In addition, we compare the performance of our policy with naive greedy policies. We show that while greedy policies can be optimal if the cost function is symmetric with respect to the sources, our policy strictly outperforms greedy policies when the cost function is asymmetric across sources.

We consider a system consisting of multiple sensors, a central monitoring station, and multiple orthogonal frequency channels. The sensors measure heterogeneous time-varying signals and report their measurements to the central monitoring station which uses them to make control decisions. Due to limited communication capacity, not all sensors can send updates to the monitoring station at all times. The cost the system pays is a function of the weighted sum of the ages-of-information of the various sensors over time. The goal is to design scheduling policies which minimize the time-average of this cost. We propose a policy called SQRT-Weight which fetches updates from sensors at a frequency proportional to the square-root of the corresponding weights and show that this policy is asymptotically 8-optimal. In addition, we compare the performance of the SQRT-Weight policy with other natural scheduling policies via simulations and through analysis for certain special cases.

We consider a system consisting of multiple sensors, a central monitoring station, and multiple orthogonal frequency channels. The sensors measure heterogeneous time-varying signals and report their measurements to the central monitoring station which uses them to make control decisions. Due to limited communication capacity, not all sensors can send updates to the monitoring station at all times. The cost of the system pays at any time is a weighted sum of the ages-of-information of the various sensors at that time. The goal is to design scheduling policies which minimize the time-average of this cost. We propose a policy called SQRT-Weight which schedules updates from sensors at a frequency proportional to the square-root of the corresponding weights and show that this policy is asymptotically 8--optimal. In addition, we compare the performance of the SQRT-Weight policy with other natural scheduling policies via simulations.