Hardware Platform For Active Acoustic Spectroscopy Sensors

Abstract

Real-time estimation of the frequency response function of fluids is one of the critical building blocks for cost-effective application of active acoustic spectroscopy technology in process control industry. The motivation was to study and develop real-time algorithm for non-parametric estimation of frequency response function on embedded platforms, according to the specific requirements and application scenario involved in active acoustic spectroscopy sensors. In addition to that, implementations of the algorithm on fixed-point and floating-point digital signal processors have been made in order to benchmark the performance and evaluate different platforms. The up-to-date ADSP-BF561 and ADSP-21489 processors, designed for signal processing tasks around audio frequency range, served as the implementation and evaluation platforms. The embedded programs for each have been developed based on the distinct processor architectures and memory hierarchies. It has been observed that the floating-point processor which runs at much slower core frequency, delivers better performance over the other. By the benchmark results, it could be identified that the less efficient memory substructure of the fixed-point processor constitutes the bottleneck of its total performance.