Abstract:
The rapid advancement in making Machine Learning compact for Internet of Things devices
with constrained capabilities has created opportunities to deploy intelligent processing at the extreme periphery of networks, positioning computational power directly adjacent to sensing and control hardware. This emerging field, known as Tiny Machine Learning (TinyML), represents a research approach focused on implementing artificial intelligence and deep learning algorithms on highly power-efficient microcontroller systems. As parallel research efforts lead, the resource-constrained nature of TinyML have lead to research on new strategies ,notably distributing real time inference across the edge, yet these approaches are limited by their lack of long-range communication capabilities from commodity hardware , deployment on stationary environments and reliance on short distance protocols, which fail to address real-world dynamic environments and the constraints of TinyML based hardware. To address this issue, the author has proposed a novel long distance communication protocol derived from Wi-Fi, which uses a novel Wi-Fi frame injection technique utilising custom Wi-Fi frames with erasure encoding to mitigate the existing problems in legacy Wi-Fi protocol, which fails to perform long distance communication.
According to the test results, the proposed protocol achieve 86% (310m) more distance
compared to the traditional Wi-Fi (42.3m), while running same TinyML model with
distributed inferencing on both setups with identical parameters. This sugests that the
proposed protocol is ideal for long distance communication.
