-
Notifications
You must be signed in to change notification settings - Fork 4
Why PSD?
Why use PSD spectrogram to replace the FFT spectrogram? I guess this is a problem that many users will have.
In this Wiki, I will show you why I think PSD should be used as the GOLDEN standard for spectrum analysis.
In control theory, we only care about the amplitude and phase of the signal. There's nothing wrong with it since the control theory only cares about those two components.
While it's convenient and straightforward, the FFT amplitude is almost never used in the statistics of signals. This is because it doesn't fit well with the physics of how signals operate. In most cases, the important parameter is not the amplitude, but the power. For example, when random noise signals combine in an electronic circuit, the resultant noise is equal to the combined power of the individual signals, not their combined amplitude.
On the other hand, due to the statistically uniform distribution of the phase of random noise, when more time points are collected, the amplitude of random noise will be lower and lower.
Alas, that's another bad option.
The power spectrum can show us the power of the signal at various frequency components, but it is susceptible to frequency resolution.
Frequency precision = Sampling frequency / Number of time points
So for an example, if we intercept 512 points of data and 1024 points of data from the 3.2Khz sampling rate data of BMI270 sensor, the frequency precision is (3205/512)hz and (3205/1024)hz
Here's a picture shows how the frequency resolution changes the result of autopower:
And in real world signal (3215hz on BMI270 maximum FIFO mode, throttle stays on 99%):
PSD normalizes the result under different frequency resolutions, eliminates the influence of resolution.
This makes the spectrum data comparable between different different sampling rate, even between different gyroscopes, e.g. [email protected] VS MPU6000@8k