We have a project using accelerometers and gyroscopes and we are transforming the data into position and orientation. Unfortunately, we have a lot of drift in our position transformation. We are wondering which filtering method is the best to reduce the noise in the raw data (Butterworth, Kalman or something else). Once this decision is made, we then are looking for tips on how to apply the appropriate settings for the chosen filter method. Our sampling rate is every milisecond. We have calibrated the devices to minimise sum of rotation when stationary and get total Accel as close to Gravity as possible.
Noise Reduction techniques: Other than the Kalman or Butterworth you have some other choices too. See e.g.,
1) PCA, LDA --- https://stats.stackexchange.com/questions/35185/dimensionality-reduction-svd-or-pca-on-a-large-sparse-matrix?rq=1
2) Classifiying top 10 ------ https://stats.stackexchange.com/questions/453068/classifying-10-top-features-using-svm
3) Autoencoders ----https://towardsdatascience.com/autoencoders-vs-pca-when-to-use-which-73de063f5d7
4) Wavelet transforms etc
Dear Jamie, I believe, even though Butterworth passband frequency responce is as flat as possible, your best choice is Chebyshev filter design, with its smaller absolute error and sharp transition from passband to stopband. To filter data, use digital dopplegangers of good old-fashioned analog filters. But since you want to extract data, which is essentially unknown, perhaps, Kalman filter is the best choice.
The Butterworth filter is unstable, I would therefore suggest the Kalman filter is preferred. I attach a good related item that may interest you.
- Badges
- Science topic
- Similar topics
- Solid State Physics
- Devices