-
Notifications
You must be signed in to change notification settings - Fork 5
/
Copy pathtestLAMP_CE.py
174 lines (158 loc) · 6.85 KB
/
testLAMP_CE.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
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
from LAMP_CE_Network import load_trainable_vars
import numpy as np
import tensorflow as tf
import myshrinkage
from scipy.io import savemat, loadmat
from os import path
import os
os.environ["CUDA_VISIBLE_DEVICES"] = "0"
os.environ['TF_CPP_MIN_LOG_LEVEL'] = '2' # BE QUIET!!!! 设置log日志级别，只显示warning和error
import math
pi = math.pi
shrink = 'bg' # 'soft'
type = 'ULA' # 'UPA'
MN = '40256'
M = 40
N = 256
G = 1024
K = 64
L = 8
SNRrange = [0, 5, 10, 15, 20]
# SNRrange = [15, 20]
def getNoise(y, snr):
ypower = tf.reduce_mean(tf.square(tf.abs(y)), axis=0, keepdims=True)
noise_var = tf.cast(ypower / ((10 ** (snr / 10)) * 2), tf.complex128)
noise = tf.complex(real=tf.random_normal(shape=tf.shape(y), dtype=tf.float64),
imag=tf.random_normal(shape=tf.shape(y), dtype=tf.float64))
n = tf.sqrt(noise_var) * noise
return n
D = loadmat(type + 'testdata' + str(N) + '_' + str(K) + '.mat')
ht = D['h']
ht = np.transpose(ht, [2, 0, 1])
U1 = D['U']
for snr in range(len(SNRrange)):
# get the trained network
trainedfilename = type + '_' + shrink + '_' + MN + '_train' + str(64) + 'carriers' + str(30) + 'dB.mat'
# trainedfilename = type + '_' + shrink + '_' + MN + str(SNRrange[snr]) + 'dB.mat'
saveresultsname = type + 'results' + '_' + shrink + '_' + MN + '_train64carriers_test' + str(K) + 'carriers.mat'
# saveresultsname = type + 'results' + '_' + shrink + '_' + MN + '_' + str(N_bits) + 'bits_quan_phase' + '.mat'
if not path.exists(saveresultsname):
# print('There is no file for saving the result before, and it has been generated!')
if shrink is 'bg':
LAMP_nmse = np.zeros(dtype=np.float64, shape=(len(SNRrange), 1))
D = dict(LAMP_nmse=LAMP_nmse)
savemat(saveresultsname, D)
else:
if shrink is 'gm':
GMLAMP_nmse = np.zeros(dtype=np.float64, shape=(5, 1))
D = dict(GMLAMP_nmse=GMLAMP_nmse)
savemat(saveresultsname, D)
T = 10
untied = False
eta, theta_init = myshrinkage.get_shrinkage_function(shrink)
var_all = []
OneOverM = tf.constant(float(1) / M, dtype=tf.float64)
Atheta = tf.random_uniform(shape=(M, N), minval=0, maxval=2*pi, dtype=tf.float64)
Atheta_ = tf.Variable(initial_value=Atheta, name='Atheta_' + str(30) + '_0')
var_all.append(Atheta_)
# q = tf.round(Atheta_ * N_Bits / (2 * pi))
# Atheta_quan = q * 2 * pi / N_Bits
Areal = tf.multiply(tf.cos(Atheta_), tf.sqrt(OneOverM))
Aimag = tf.multiply(tf.sin(Atheta_), tf.sqrt(OneOverM))
A_ = tf.complex(Areal, Aimag, name='A') # 恒模预编码矩阵
h_ = tf.placeholder(tf.complex128, (None, N, K))
h1 = tf.transpose(h_, [1, 0, 2])
y1_ = []
for k_1 in range(K):
y1 = tf.matmul(A_, h1[:, :, k_1])
noise = getNoise(y1, SNRrange[snr])
y1 = y1 + noise
y1_.append(y1)
ytemp_ = tf.stack(y1_, axis=2)
y_ = tf.transpose(ytemp_, [1, 0, 2])
v_ = y_
OneOverMK = tf.constant(float(1) / (M*K), dtype=tf.float64)
rvar_ = OneOverMK * tf.expand_dims(tf.square(tf.norm(tf.abs(v_), axis=[1, 2])), 1)
U_ = tf.placeholder(tf.complex128, (G, N))
A_H = tf.transpose(A_, conjugate=True)
B = tf.matmul(U_, A_H) # 初始化B
Breal_ = tf.Variable(tf.real(B), name='Breal_' + str(30) + '_1')
var_all.append(Breal_)
Bimag_ = tf.Variable(tf.imag(B), name='Bimag_' + str(30) + '_1')
var_all.append(Bimag_)
B_ = tf.complex(Breal_, Bimag_, name='B')
v1 = tf.transpose(v_, [1, 0, 2])
v1 = tf.reshape(v1, (M, -1))
Bvtemp = tf.matmul(B_, v1)
Bv = tf.reshape(Bvtemp, (G, tf.shape(h_)[0], K))
Bv_ = tf.transpose(Bv, [1, 0, 2])
theta_ = tf.Variable(theta_init, dtype=tf.float64, name='theta_' + str(30) + '_1')
var_all.append(theta_)
xhat_, dxdr_ = eta(Bv_, rvar_, K, theta_)
GOverM = tf.constant(float(G) / M, dtype=tf.complex128)
xhat_, dxdr_ = eta(Bv_, rvar_, K, theta_)
for t in range(2, T+1):
b_ = tf.expand_dims(GOverM * dxdr_, 1)
U_H = tf.transpose(U_, conjugate=True)
matrix = tf.matmul(A_, U_H)
x2 = tf.transpose(xhat_, [1, 0, 2])
x3 = tf.reshape(x2, (G, -1))
Axhat = tf.matmul(matrix, x3)
Axhat = tf.reshape(Axhat, (M, tf.shape(h_)[0], K))
Axhat_ = tf.transpose(Axhat, [1, 0, 2])
v_ = tf.reshape(v_, [tf.shape(h_)[0], M*K])
bv = tf.multiply(b_, v_)
bv_ = tf.reshape(bv, [tf.shape(h_)[0], M, K])
v_ = y_ - Axhat_ + bv_
rvar_ = OneOverMK * tf.expand_dims(tf.square(tf.norm(tf.abs(v_), axis=[1, 2])), 1)
if untied: # 表明每一层的B都会训练
Breal_ = tf.Variable(tf.real(B), name='Breal_' + str(snr) + '_' + str(t))
var_all.append(Breal_)
Bimag_ = tf.Variable(tf.imag(B), name='Bimag_' + str(snr) + '_' + str(t))
var_all.append(Bimag_)
B_ = tf.complex(Breal_, Bimag_, name='B')
Bv_ = tf.matmul(B_, v_)
rhat_ = xhat_ + Bv_
else:
v3 = tf.transpose(v_, [1, 0, 2])
v4 = tf.reshape(v3, (M, -1))
Bv = tf.matmul(B_, v4)
Bv = tf.reshape(Bv, (G, tf.shape(h_)[0], K))
Bv_ = tf.transpose(Bv, [1, 0, 2])
rhat_ = xhat_ + Bv_
xhat_, dxdr_ = eta(rhat_, rvar_, K, theta_)
U_H = tf.transpose(U_, conjugate=True)
xhat1 = tf.transpose(xhat_, [1, 0, 2])
xhat2 = tf.reshape(xhat1, (G, -1))
hhat = tf.matmul(U_H, xhat2)
hhat = tf.reshape(hhat, (N, tf.shape(h_)[0], K))
hhat_ = tf.transpose(hhat, [1, 0, 2])
nmse_ = tf.reduce_mean(
tf.square(tf.norm(tf.abs(hhat_ - h_), axis=[1, 2])) / tf.square(tf.norm(tf.abs(h_), axis=[1, 2])))
sess = tf.Session()
sess.run(tf.global_variables_initializer(), feed_dict={U_: U1})
load_trainable_vars(sess, trainedfilename)
nmse_SNR = []
xhat = sess.run(xhat_, feed_dict={h_: ht, U_: U1})
nmse = sess.run(nmse_, feed_dict={h_: ht, U_: U1})
nmse_dB = 10 * np.log10(nmse)
print(str(SNRrange[snr]) + 'dB:' + ' ' + 'NMSE = ' + str(nmse_dB) + 'dB')
nmse_SNR = np.append(nmse_SNR, nmse_dB)
# print(nmse_SNR)
results = loadmat(saveresultsname)
if shrink is 'gm':
GMLAMP_nmse = results['GMLAMP_nmse']
# GMLAMP_nmse = GMLAMP_nmse[0]
GMLAMP_nmse = np.append(GMLAMP_nmse, nmse_SNR)
# GMLAMP_nmse[ibegin:iend] = nmse_SNR
print(GMLAMP_nmse)
D = dict(GMLAMP_nmse=GMLAMP_nmse)
savemat(saveresultsname, D)
else:
if shrink is 'bg':
# print(results)
LAMP_nmse = results['LAMP_nmse']
LAMP_nmse[snr, :] = nmse_SNR
D = dict(LAMP_nmse=LAMP_nmse)
savemat(saveresultsname, D)
tf.reset_default_graph()