Dynamic Compensation of Intelligent Sensors

Commercial sensors have generally, due to their own characteristics, some undesirable influences on the measured quantity and its precision. In particular, the dynamic characteristics can be reflected on to the measured quantity and lead to false or delayed interpretation of the underlying physical process. The quality and, therefore, the cost of the sensor are often tied to the dynamic performance of these instruments. Intelligent sensors are able to adapt to changing environments, calibrate themselves, and predict the pattern of the future signal. This paper presents algorithms to improve the dynamic performance of sensors, to identify the dynamic characteristics of the sensor, and to predict the future pattern of the measured quantity. In particular, two inverse filters are proposed for the improvement of the sensors dynamic performance. One filter incorporates an optimal constant feedback gain that reduces the computational cost and increases the accuracy. A system identification method is used to identify the sensors dynamic properties and allows for adaptation of the inverse filters parameters. This identification algorithm computes the optimum input to the system, i.e., the sensor. The optimization is based on the inverse covariance matrix of the information matrix. A genetic algorithm is used to perform both optimizations for the computation of the optimal input and for the optimal constant feedback gain. In addition, a predictive filter formulation is given that is based on the identified system. Simulation results indicate that both inverse filters are capable of recovering the original or true signal. The second filter shows superiority in terms of convergence, lower computational cost, and lower error due to its optimized parameters. The predictive filter indicates good working accuracy for the signal prediction.