|
@@ -0,0 +1,79 @@
|
|
1
|
+import numpy as np
|
|
2
|
+from scipy.optimize import least_squares
|
|
3
|
+
|
|
4
|
+def harmonic ( sN, f0, fs, nK, t ):
|
|
5
|
+ """
|
|
6
|
+ Performs inverse calculation of harmonics contaminating a signal.
|
|
7
|
+ Args:
|
|
8
|
+ sN = signal containing noise
|
|
9
|
+ f0 = base frequency of the sinusoidal noise
|
|
10
|
+ fs = sampling frequency
|
|
11
|
+ nK = number of harmonics to calculate
|
|
12
|
+ t = time samples
|
|
13
|
+ """
|
|
14
|
+ print("building matrix ")
|
|
15
|
+ A = np.zeros( (len(t), 2*nK) )
|
|
16
|
+ for irow, tt in enumerate(t):
|
|
17
|
+ A[irow, 0::2] = np.cos( np.arange(nK)*2*np.pi*(f0/fs)*irow )
|
|
18
|
+ A[irow, 1::2] = np.sin( np.arange(nK)*2*np.pi*(f0/fs)*irow )
|
|
19
|
+ # brutal
|
|
20
|
+ #for k, ik in enumerate( np.arange(0, 2*nK, 2) ):
|
|
21
|
+ # A[irow, ik ] = np.cos( k*2*np.pi*(f0/fs)*irow )
|
|
22
|
+ # A[irow, ik+1] = np.sin( k*2*np.pi*(f0/fs)*irow )
|
|
23
|
+
|
|
24
|
+ v = np.linalg.lstsq(A, sN, rcond=None) #, rcond=1e-8)
|
|
25
|
+
|
|
26
|
+ alpha = v[0][0::2]
|
|
27
|
+ beta = v[0][1::2]
|
|
28
|
+
|
|
29
|
+ amp = np.sqrt( alpha**2 + beta**2 )
|
|
30
|
+ phase = np.arctan(- beta/alpha)
|
|
31
|
+
|
|
32
|
+ h = np.zeros(len(t))
|
|
33
|
+ for ik in range(nK):
|
|
34
|
+ h += np.sqrt(alpha[ik]**2 + beta[ik]**2) * np.cos( 2.*np.pi*ik * (f0/fs) * np.arange(0, len(t), 1 ) + phase[ik] )
|
|
35
|
+
|
|
36
|
+ #plt.matshow(A, aspect='auto')
|
|
37
|
+ #plt.colorbar()
|
|
38
|
+
|
|
39
|
+ #plt.figure()
|
|
40
|
+ #plt.plot(alpha)
|
|
41
|
+ #plt.plot(beta)
|
|
42
|
+ #plt.plot(amp)
|
|
43
|
+
|
|
44
|
+ #plt.figure()
|
|
45
|
+ #plt.plot(h)
|
|
46
|
+ #plt.title("modelled noise")
|
|
47
|
+
|
|
48
|
+ return h
|
|
49
|
+
|
|
50
|
+if __name__ == "__main__":
|
|
51
|
+ import matplotlib.pyplot as plt
|
|
52
|
+
|
|
53
|
+ f0 = 60 # Hz
|
|
54
|
+ delta = np.random.rand()
|
|
55
|
+ fs = 50000 #1e4
|
|
56
|
+ t = np.arange(0, 1, 1/fs)
|
|
57
|
+ phi = .234
|
|
58
|
+ A = 1.0
|
|
59
|
+ nK = 20
|
|
60
|
+ sN = A * np.sin( (delta+f0)*2*np.pi*t + phi ) + np.random.normal(0,.1,len(t))
|
|
61
|
+ sNc = A * np.sin( (delta+f0)*2*np.pi*t + phi )
|
|
62
|
+ h = harmonic(sN, f0, fs, nK, t)
|
|
63
|
+
|
|
64
|
+ plt.figure()
|
|
65
|
+ plt.plot(t, sN, label="sN")
|
|
66
|
+ plt.plot(t, sN-h, label="sN-h")
|
|
67
|
+ plt.plot(t, h, label='h')
|
|
68
|
+ plt.title("true noise")
|
|
69
|
+ plt.legend()
|
|
70
|
+
|
|
71
|
+ plt.figure()
|
|
72
|
+ plt.plot(t, sN-sNc, label='true noise')
|
|
73
|
+ plt.plot(t, sN-h, label='harmonic removal')
|
|
74
|
+ plt.legend()
|
|
75
|
+ plt.title("true noise")
|
|
76
|
+
|
|
77
|
+ plt.show()
|
|
78
|
+
|
|
79
|
+ print("hello")
|