Efficient hardware implementation of the l1 — Regularized least squares for IoT edge computing

As the use of compressed sensing (CS) in internet of things (IoT) wearable nodes increases, the need for high performance and low power CS reconstruction algorithms for the battery powered IoT Edge Devices also increases. This paper describes an efficient multicore hardware implementation of the ℓ ₁ Regularized Least Squares (LS) optimization problem based on the Alternating Direction Method of Multipliers (ADMM) algorithm. The use a decomposition technique, that exploits the special matrix structure to update its inverse, significantly reduced the processing time as well as the complexity of the algorithm. The average processing time on a parallel multicore Zynq System on Chip (SoC) device has improved by a factor of 2 compared to the C++ PC software implementation which makes it suitable for real time applications.