Real-Time Digital Filtering for IoT Data in Programmable Network Switches
Nathaniel Nauman, Ruochong Wu, Saurabh Bagchi · 2022
This study seeks to optimally approximate fixed-point multiplication on a programmable switch to perform digital IIR filtering. For a stream of real-time IoT data, offloading digital filtering computations allows edge devices to allocate more CPU cycles for processing analytics in parallel. Multiplication is approximated through an adaptive bit width precision lookup table, and the formulas for the maximum possible percentage error and the memory consumption are shown. The users can access and alter the 32-bit filter coefficients to reflect the specifications of the desired IIR filter. This offers extensive flexibility for users to alter the filter characteristics through software.