Atomix: A Framework for Deploying Signal Processing Applications on Wireless Infrastructure

Multi-processor DSPs have become the platform of choice for wireless infrastructure. This trend presents an opportunity to enable faster and wider scale deployment of signal processing applications at scale. However, achieving the hardware-like performance required by signal processing applications requires interacting with bare metal features on the DSP. This makes it challenging to build modular applications.

We present Atomix, a modular software framework for building applications on wireless infrastructure. We demonstrate that it is feasible to build modular DSP software by building the application entirely out of fixedtiming computations that we call atoms. We show that applications built in Atomix achieve hardware-like performance by building an 802.11a receiver that operates at high bandwidth and low latency. We also demonstrate that the modular structure of software built with Atomix makes it easy for programmers to deploy new signal processing applications. We demonstrate this by tailoring the 802.11a receiver to long-distance environments and adding RF localization to it.