I have a working python script for Fast Fourier Transform (fft) signal which plots the graph and fft correctly, I am fetching data from postgre so I ommited that code. And due to limit of paste i pasted shorter version of signal, but the signal is preatty much the similar on longer timeframe.
The x scale of fft is fine, but what i get is a large y scale which i do not understand.
I have an fft processed by machine directly that looks correct and i have another measurement done by hand held machine. The ffts look almost identical on x scale and relative sizes on y scale, but the actual calculated y scale differ a lot.
In script data_step is in micro second between two timestamps. The n and datastep cant be wrong since the x axis is fine. From the data i get:
Min of signal(value_points_sensor1): -0.057881687
Max of signal: 0.07810274
Max of fft y: 5522.93
Shouldnt the size of fft be similar to size of amplitude of signal? Also arent the unit of amplitude on fft and signal on y axis the same?
import matplotlib.pyplot as plt
from scipy.fft import rfft, rfftfreq, irfft
import numpy as np
figure, axis = plt.subplots(2, 1)
plt.subplots_adjust(hspace=1)
time_points_sensor1 = np.array(row[4])
value_points_sensor1 = np.array(row[3])
axis[0].plot(time_points_sensor1, value_points_sensor1)
n = len(row[2])
data_step = (abs((int(row[2][-2] - row[2][-1]))) / 1000000)
yf = rfft(value_points_sensor1)
xf = rfftfreq(n,data_step)
axis[1].plot(xf,abs(yf))
plt.show()
The value of rfft is proportional both to the magnitude of the data and the number of points. So you should expect the values in yf to be large.
Compare:
# One cycle of a cosine curve
x = rfft(np.cos(np.linspace(0, 2 * np.pi, 1000)))
max(abs(x))
500.2493021844711
# Two cycles of a cosine curve, again 1000 points per cycle
x = rfft(np.cos(np.linspace(0, 4 * np.pi, 2000)))
max(abs(x))
1000.2484171710441