CANDor: Continuous Adaptive Neighbor Discovery

Many applications require continuous awareness of the set of surrounding devices. A user entering a smart space benefits from the ability to quickly discover which devices they can control nearby. This ability to know "who is around" is commonly provided by continuous neighbor disco-very protocols, which entail a schedule on which devices alternate between beaconing to advertise their presence and listening to detect the presence of neighbors. In existing protocols, these schedules are carefully configured to achieve application-level objectives—a target discovery latency, discovery probability, and energy consumption. These configurations are determined statically based on assumptions about network conditions (e.g., the number of expected neighboring devices, or nodes). However, the conditions that a given node typically experiences are dynamic. To handle these dynamics, existing protocols tend to be configured for the worst case, which results in schedules that waste energy. In this paper, we identify a signal for adaptation of continuous neighbor discovery and show how existing protocols can incorporate this signal to adapt their behavior to achieve consistent discovery probability and latency. The signal we adapt to is changing node density, and our novel insight is that we can extract this signal directly from the performance of neighbor discovery itself. We show that our approach effectively senses and adapts even when using initially suboptimal schedules in dynamic mobile environments, allowing neighbor discovery protocols to maintain their performance guarantees without added sensing overhead.