Open debate: Protocol for signal analysis

Question

This is an open question, just for discussion. During my years studying and working on signal processing, I have identified several factors that are relevant for understanding any type of signal. For example, PSD for analyzing frequency spectrum and distribution of fundamental frequencies, if any; envelope of the signal for "slow" signals (below lets say 1KHz), skewness and kurtosis analysis of probability distribution of the signal, and so on.

However, I have always been curious about what do other people focus on analyzing a new signal and what are the important elements that you first try to get out before starting to plan a signal processing technique and filter design.

Do you have any particular protocol or procedure that you follow when approaching a "new signal" analysis?

Answer 1

I think this is "too broad" for a generic answer but it's a fun one so here goes.

I'm an audio guy, so the first thing I often do is to listen to it. This way I can run a whole lot of rough analysis in just a few seconds. Next steps depends a lot on why I'm working with the signal in the first place and what I've heard during the first round.

Another big one is to check "signal integrity". Does the signal has measurement or other artefacts, contaminations, what's the signal to noise ratio, is it calibrated, is the meta data consistent, is time/phase/frequency alignment correct etc. It's depressing how often the signal isn't really "clean". You can safe yourself A LOT of time by not working on bad or questionable signals in the first place.

Answer 2

Personally, when working with audio signals I use:

• File metadata and parameters (if there's a calibration associated with it, bit-depth, sampling frequency, etc.)
• Metrics, such as RMS/peak dBFS level - those give me an overview of what I am dealing with.
• Waveform to check for any clipping, low-frequency content, etc.
• Spectrogram to check for discontinuities, lossy compression, overall frequency content, aliasing. In conjunction with waveform, I do tend to extract rough SNR and levels of the signal or spot if there are sections which are pure 0's.
• Spectrum to have an idea of overall frequency content and spot harmonic distortion for pure sinusoidal signals, typically I don't use PSD.
• Persistence spectrum, calculated from the background noise, which gives me an idea of self-noise, interference or even anti-aliasing filter applied to the signal.

For signals which are measured impulse responses, I go through the regular frequency and phase response, group delay. For room IR's I would also look at decay, waterfall and spectrogram/scalogram.

Answer 3

With audio, showing the waveform and listening goes without a saying for me. Spectrogram of some kind tends to be the first thing I do to dig deeper into what is going on.