D-SCAN : Toward collaborative multi-radio coexistence in mobile devices via deep learning

As the demand for efficient and reliable wireless connectivity continues to increase, mobile Internet-of-Things devices equipped with multiple heterogeneous radios including Wi-Fi and Bluetooth have become prevalent. However, collocated Wi-Fi and Bluetooth operate in the same 2.4 GHz industrial, scientific, and medical band and interfere with each other internally and externally. Although current devices avoid internal cross-technology interference by enabling either Wi-Fi or Bluetooth to transmit packets at any specific time slot in a time-division multiplexing manner, the dissonance with external interference mitigation schemes can result in severe performance degradation. In this paper, we present D-SCAN, a novel collaborative coexistence mechanism. D-SCAN infers nearby Wi-Fi information efficiently using a collocated Bluetooth radio, thereby offsetting the overhead of key Wi-Fi functions and preventing collisions between Wi-Fi and Bluetooth. To this end, D-SCAN adopts a data-driven approach that captures the unique temporal and spectral features of Wi-Fi signals from Bluetooth spectrum measure-ments by leveraging deep neural networks. A D-SCAN prototype in real-world experiments reduces the latency and energy consumption of legacy Wi-Fi scanning by 23% and 45%, respectively. It also promotes the agile interference avoidance of Bluetooth that coexists with Wi-Fi on a single device. Thus, D-SCAN demonstrates efficiency in resource management and effectiveness in mitigating cross-technology interferences.