Adaptive Filter for Acceleration Sensor Signal

Abstract

When carrying out any measurements, there is always a problem of separating the useful signal from noise or interference. The interference is usually caused by an external influence on the sensor, for example, an acceleration sensor. It can be considered that such an effect is vibrations created by the propeller group of a multi-rotor aircraft, which significantly distort the results of measuring acceleration components by the sensor. The Kalman filter, which is based on an autoregressive model, has received the most widespread use in processing the signal from the acceleration sensor. The Kalman filter allows you to get good results, but requires a lot of computing power compared to other methods. At the same time, a priori information about interference can significantly reduce the amount of calculations while maintaining the quality of useful signal extraction. An acoustic signal can act as a source of such information, which is obviously correlated with vibrations generated by the propeller group. Moreover, it can be assumed that the source of the acoustic signal is vibrations. Note that the acceleration sensor is used to measure constant or relatively slowly changing acceleration values due to the evolution of an unmanned vehicle. Since a microphone is used to measure the sound correlated with interference, the received signal will be limited in frequency from the bottom at about 10 Hz due to the design features of the microphones, that is, the signal from the microphone will not contain data about the acceleration of the device, but only about the interference. These facts can be used to build an adaptive filter. A simulation of such an adaptive filter was carried out, and a signal representing a sequence of triangular pulses with a zero mean having a wide spectrum was used as useful data. At the same time, the average power of the useful signal was approximately 0.13. Gaussian noise was also chosen as an interference with a zero mean and a variance of 0.5. Thus, the signal-to-noise ratio was approximately -2.8 Db. According to the structural diagram of the device, before adding interference to the signal, it was passed through a low-pass filter of the first order. The adaptive filter had only eight pulse response weights. The developed adaptive filter makes it possible, based on a priori data on external interference, to significantly improve the accuracy of measuring the useful signal. The use of the proposed algorithm significantly reduces the amount of calculations. The paper shows that it is sufficient to use eight filter weight coefficients so that the relative error in determining the measured value does not exceed 10%.