A System Identification Procedure Using Compressive Sensing

The conventional system identification, which is a branch of machine learning, takes advantages of the whole sampling data to identify the system. To identify a system with less sampling density, compressive sensing is applied on system identification, which randomly extracts the sampling data from the system response. Hence a novel identification procedure is proposed using compressive sensing techniques. Then a second order system is selected as the system to be identified using such identification procedure. The identification performances of estimated systems are investigated from the scenario randomly extracting 10% of total sampling data to the scenario using 90% of total sampling data. Each scenario consists of three noise cases with different levels of SNRs to test the robustness of the signal recovery algorithms of compressive sensing. The results show that the system identification using compressive sensing has are relatively high identification performance and is robust to noise when using 30% or more of total sampling data.