-
Notifications
You must be signed in to change notification settings - Fork 17
/
LiftingFmat.py
98 lines (85 loc) · 3.25 KB
/
LiftingFmat.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
#Folding matrix F implemented with 3 lifting steps, including rounding
#for the IntMDCT
#Gerald Schuller, Sep. 2018
import numpy as np
def LiftingFmat(fb):
#produces the 3 lifting matrices F0, L0, L1,
#whose product results in the F folding matrix for the IntMDCT.
#Usage: F0,L0,L1=LiftingFmat(fb)
#Argument fb: The 1.5N coefficients for the normal F matrix of the MDCT
N=int(len(fb)/1.5)
alpha=np.arcsin(fb[:N//2])
#print("alpha=", alpha)
#Lifting1:
F0=np.zeros((N,N))
#third quadrant anti-diagonal:
F0[int(N/2):N,0:int(N/2)] =(np.eye(int(N/2)))
#2nd quadrant anti-diagonal:
F0[0:int(N/2),int(N/2):N] =(np.eye(int(N/2)))
#4th quadrant:
F0[int(N/2):N,int(N/2):N]=np.flipud(np.diag((np.cos(alpha)-1)/np.sin(alpha),k=0))
#Lifting2:
L0=np.zeros((N,N))
#2nd quadrant anti-diagonal:
L0[0:int(N/2),int(N/2):N] =np.fliplr(np.eye(int(N/2)))
#3rd quadrant anti-diagonal:
L0[int(N/2):N,0:int(N/2)] =np.fliplr(np.eye(int(N/2)))
#4th quadrant:
L0[int(N/2):N,int(N/2):N]=(np.diag(np.sin(alpha),k=0))
#Lifting3:
L1=np.zeros((N,N))
#third quadrant anti-diagonal:
L1[int(N/2):N,0:int(N/2)] =np.fliplr(np.eye(int(N/2)))
#2nd quadrant anti-diagonal:
L1[0:int(N/2),int(N/2):N] =np.fliplr(np.eye(int(N/2)))
#4th quadrant:
L1[int(N/2):N,int(N/2):N]=np.diag((np.cos(alpha)-1)/np.sin(alpha),k=0)
return F0, L0, L1
#Testing:
if __name__ == '__main__':
#Rotation
#np.matrix([[cos(alpha), -sin(alpha)],[sin(alpha), cos(alpha)]])
#Fmatrix: Rotation matrix with rows flipped:
#np.matrix([[sin(alpha), cos(alpha)],[cos(alpha), -sin(alpha)]])
#Lifting implementation: rows of first matrix F0 flipped, L1 can also be flipped
#when L0 has its columns flipped. In this way we get all 1's on the anti-diagonal,
#like for the zero-delay matrices.
#For N=2:
"""
alpha=0.1
F0=np.matrix([[0,1],[1,(np.cos(alpha)-1)/np.sin(alpha)]])
L0=np.matrix([[0,1],[1,np.sin(alpha)]])
L1=np.matrix([[0,1],[1,(np.cos(alpha)-1)/np.sin(alpha)]])
Fmatrix=F0*L0*L1
print("N=2: flipped rotation matrix from lifting=\n", Fmatrix)
print("As comparison, flipped rotation matrix:\n", np.matrix([[np.sin(alpha), np.cos(alpha)],[np.cos(alpha), -np.sin(alpha)]]))
"""
#With sympy:
import sympy
alpha=sympy.symbols('alpha')
F0=sympy.Matrix([[0,1],[1,(sympy.cos(alpha)-1)/sympy.sin(alpha)]])
L0=sympy.Matrix([[0,1],[1,sympy.sin(alpha)]])
L1=F0
print("Using sympy: flipped lifting steps:")
#print(sympy.latex(F0)) #for LaTeX
sympy.pprint(F0)
sympy.pprint(L0)
sympy.pprint(L1)
print("Their product is a flipped rotation matrix:")
sympy.pprint(sympy.simplify(F0*L0*L1))
N=4 #number of subbands
fb=np.sin(np.pi/(2*N)*(np.arange(int(1.5*N))+0.5))
F0,L0,L1=LiftingFmat(fb)
print("Lifting matrices for N=4:")
print("F0=\n", F0)
print("L0=\n", L0)
print("L1=\n", L1)
Fmat=np.dot(F0,L0)
Fmat=np.dot(Fmat,L1)
print("N=4: Fmat from lifting=\n", Fmat)
#comparison to the usual symmetric F matrix:
from MDCTfb import symFmatrix
Fa=symFmatrix(fb)
print("for comparison: usual Fa= \n", Fa[:,:,0])
#The inverse is the transpose.
#Hence lifting steps then are in reverse order and each transposed.