Design Tradeoffs for Cloud-Based Ambient Assisted Living Systems

Ambient assisted living (AAL) has received considerable attention due to its ability to provide services to the elderly by sensors and actuators. However, building such a system is challenging on two fronts. First, the tradeoff between accuracy and monetary cost should be understood. Accuracy of each sensor can be empirically estimated from its sample rate. Typically, higher rate indicates higher accuracy. As a result, higher rate requires more computation resources to process the sampled data, incurring more monetary cost. Second, user needs change frequently. Thus, we need a resource allocation scheme that is (a) able to strike a good balance between accuracy and monetary cost and (b) adaptive enough to meet the frequently changing needs. Unfortunately, several seemingly natural solutions fail on one or more fronts (e.g., simple one shot optimizations). As a result, the potential benefits promised by these prior efforts remain unrealized. To fill the gap, we address these challenges and present the design and analysis of a low-complexity online algorithm to minimize the long-term accuracy-monetary cost on a queue length based control. The design is driven by insights that queue-lengths can be viewed as Lagrangian dual variables and the queue-length evolutions play the role of subgradient updates. Therefore, the control decisions depend only on the instantaneous information and can adapt to the changing needs. Simulations demonstrate that the proposed algorithm can strike a good balance between accuracy and monetary costs. Moreover, the asymptotic optimality of the proposed algorithm has been shown by rigorous analysis and numerical results.