I just started to dive into signal processing due to a free-time project. Therefore, this is my first question here, for which I searched many previous threads, but couldn't find what I was looking for as most of the questions are about vibration or sound signals.
The underlying process of my application: "Based on a pressure sensor signal over time, I am trying to classify whether the activity was carried out correctly. Based on this idea I am planning to use logistic regression, taking several features as input to the model. Each snippet to classify is of the length of one period. Before and after the activity the sensor signal remains in an idle stage.
In terms of feature extraction, I already implemented the extraction of time-domain features. Nevertheless, I am currently trying to derive features in the frequency domain. The length of the raw sequence is 45 +-5 seconds sampled with 10Hz. Plotting the frequency spectrum of the unpadded signal looks like this, wherein the green are correctly executed activities and the red ones are faulty executed activities. The frequency axis is in 1/min and the peak is as expected around 1.5 per min.
Due to difficulties to identify the differences, I windowed and zero-padded the data to a sequence of length 2048, which is 4.5 * the original length. Then the FFT looks like this (zoomed in to the interesting frequency band):
- Is it useful at all to build the FFT of one period of the signal?
- Is zero-padding a useful tool as it does not increase the frequency resolution, but instead interpolates, which does not provide me additional information about the signal?
- Which useful features can be extracted from the spectrum based on the result? The basics of deriving the frequency of the highest peak and the maximum peak value are already implemented. One more idea was to analyze the PSD calculated through the Welch algorithm, but unfortunately, I cannot get a smooth signal as even a larger window on the zero-padded data does not work for me.
Thank you for all your help in advance.