Low Power Algorithm-Architecture Co-Design of Fast Independent Component Analysis (FICA) for Multi-Gas Sensor Applications

Abstract

For miniaturized multi-gas sensors, the detected multi-gas signals would be self-interfered by responses to multiple gases. In this paper, a fast Independent Component Analysis ( FICA) is proposed to restore the original source signals from the mixed signals received by different gas sensors. This FICA is designed and implemented by low power algorithm-architecture co-design considering the trade-offs among power, delay and accuracy of extracted signals for multigas sensor applications. To further reduce the power consumption, a data-length controller is designed to adjust the calculated data-length. Additionally, a stability check unit is utilized to terminate the ICA execution for reduction of the computation time and total energy. Compared with the conventional ICA design, the proposed low-power FICA realizes energy reduction by 75% for multi-gas sensor applications.