-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathmain_loop.m
192 lines (160 loc) · 5.75 KB
/
main_loop.m
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
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
addpath(genpath(pwd));
close all
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% PARAMETRES
Mx = 2; %My auto (ratio) -- nombre de pixels
J_b = 6; % On suppose le réseau d'antennes carré de côté
% max 11 GPU atm
c= 3e8;
f = 1e9;
lambda = c/f;
dist = 10*lambda/2.1; % J_b antennes espacées de dist
N = 7; % itérations Kalman
W = 1; % nombre de réalisations
SNR = 10;
iMEM = 2; % nombre itérations MEM
RATIO = 10e14; % ratio erreur Kalman
Nreal = 1000;
% A_ev matrice d'évolution définie plus loin (need D ligne 35)
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%% choix / gen image
img = imread('bh_square.jpg');
adapted = adaptImg(img, Mx);
vadapted = vectorize(adapted);
clear img
%%% Calcul paramètres
My = size(adapted, 2);
z = antennes(J_b, dist);
J = size(z, 1);
D = numel(adapted);
I = matI(Mx, My);
dz = dz_c(J,z);
A_ev = eye(D);
A = matA(z,I,D,f,c);
% erreurs = [0.001 0.003 0.005 0.01 0.03 0.05 0.1 0.3];
% erreurs = [0.001 0.003 0.005 0.008 0.01 0.03];
% erreurs = [0.001 0.3 0.05];
erreurs = [0.001];
npos = numel(erreurs);
% calcul invariants kalman
H = matF(J,D,z,lambda,I);
[nY,nR,nQ,tX] = dataGen_im(H,J,N,D,vadapted,A_ev,Nreal,SNR,RATIO);
% X_0 = true_image ;
% X_mvdr = MEM(MVDR(A,reshape(nY(:,1),J,J),Mx,My),iMEM);
X_mvdr = MVDR(A,reshape(nY(:,1),J,J),Mx,My);
X_mvdr = normarr(abs(X_mvdr));
% X_0 = vadapted;
X_0 = X_mvdr;
% tX(:,1) = vadapted;
P_0 = X_0*X_0' - mean(X_0,'all');
% Nouveau calcul X_0 et P_0 exactes
% NX_0 = repmat(vadapted,1,Nreal);
% vX_0 = abs(0.001*randn(D,Nreal) + NX_0);
% errX_0 = vX_0 - NX_0;
%
% X_0 = vX_0(:,randi(Nreal));
% P_0 = (errX_0 * errX_0')/Nreal;
clear NX_0 vX_0 errX_0
% def listes vides
wK = zeros(D,J^2,N,W);
whos wK
wK_err = zeros(D,J^2,N,W);
wX = zeros(D,N,W);
wX_err = zeros(D,N,W);
e_x = zeros(W,1);
e_K = zeros(W,1);
e_x_e = zeros(W,1);
e_th = zeros(W,1);
e_xx = zeros(W,1);
errX = zeros(N,1);
errK = zeros(N,1);
errX_e = zeros(N,1);
errth = zeros(N,1);
errXX = zeros(N,1);
optot = npos*W;
tic
for err = 1:npos
poids = erreurs(err);
for realisation = 1:W
elapsedTime = toc;
opDone = W*(err-1) + realisation;
opRemain = optot - opDone + 1;
tMoyOp = elapsedTime/opDone;
ETA = tMoyOp*opRemain;
fprintf('Erreur : %u/%u -- Réalisation : %u/%u -- Time : %.4f -- ETA : %.4f\n',err,npos,realisation,W,elapsedTime,ETA);
% 1. Gen données aléatoires
z_err = z;
% z_err = z+ randn(size(z))*poids;
% I_err = I;
I_err(4,:) = I_err(4,:) + randn(size(I_err(4,:)))*poids;
% I_err = I + randn(size(I))*poids;
H_err = matF(J,D,z_err,lambda,I_err);
% Calcul Hcorr
% 1. poser pb
% y = H_err*X_mvdr;
y = nY(:,1);
Iv = reshape(I_err,[],1);
prb = @(x)prb_f(x,y,Iv,dz,J,D,lambda,X_mvdr);
% 2. init
% dbiasPrio = zeros(1,2*D)*err/100;
% xparfait = I_err-I;
% dbiasPrioInit = reshape(xparfait,1,2*D) + dbiasPrio;
% x0 = [dbiasPrioInit reshape(X_mvdr,1,[])];
% lb = [-0.1*ones(1,2*D) zeros(1,D)];
% ub = [0.1*ones(1,2*D) ones(1,D)];
x0 = zeros(1,2*D);
lb = [-0.1*ones(1,2*D)];
ub = [0.1*ones(1,2*D)];
% 3. lsqnonlin
options = optimoptions(@lsqnonlin,'Display','Iter','FiniteDifferenceType','central');
% options = optimoptions(@lsqnonlin,'MaxIterations',1000,'StepTolerance',1e-18,'Display','Iter','FunctionTolerance',1e-20,'MaxFunctionEvaluations',1e6,'FiniteDifferenceType','central');
% ,'Algorithm','levenberg-marquardt'
x = lsqnonlin(prb,x0,lb,ub,options);
% 4. nouveau X_0, calcul Hcorr
% X_0 = x(2*D+1:3*D);
Icorr = - reshape(x(1:2*D),D,2) + I_err;
Hcorr = matF(J,D,z_err,lambda,Icorr);
amel(I_err,Icorr,I)
% C = H ./ H_err;
% 2. Tourner Kalman_XPU
[X,K,Xp] = Kalman_CPU_V6(A_ev,H,X_0,P_0,nY,nR,nQ,D,N);
[X_e,K_e,Xp_e] = Kalman_CPU_V6(A_ev,H_err,X_0,P_0,nY,nR,nQ,D,N);
% 3. Calcul erreurs
for k = 1:N
errX(k) = fro(abs(X(:,k)/max(abs(X(:,k)))) - tX(:,k));
errX_e(k) = fro(abs(X_e(:,k)/max(abs(X_e(:,k)))) - tX(:,k));
errXX(k) = fro(abs(X(:,k)/max(abs(X(:,k)))) - abs(X_e(:,k)/max(abs(X_e(:,k)))));
errK(k) = fro(K_e(:,:,k)-K(:,:,k));
if k>=2
errth(k) = fro((K_e(:,:,k) - K(:,:,k))*nY(:,k) + (eye(D) - K_e(:,:,k)*H_err)*Xp_e(:,k) - (eye(D) - K(:,:,k)*H)*Xp(:,k));
end
end % end calculs erreurs
% 4. append in vect errs
wK(:,:,:,realisation) = K;
wK_err(:,:,:,realisation) = K_e;
wX(:,:,realisation) = X;
wX_err(:,:,realisation) = X_e;
e_x(realisation) = mean(errX(5:end));
e_K(realisation) = mean(errK(5:end));
e_x_e(realisation) = mean(errX_e(5:end));
e_th(realisation) = mean(errth(5:end));
e_xx(realisation) = mean(errXX(5:end));
end % end real
de_x = std(errX);
de_K = std(errK);
de_x_e = std(e_x_e);
de_xx = std(e_xx);
% Sauvegarde données
nom = strcat(strrep(sprintf('%.3f',poids),'.','_'),'.mat');
% save(nom,"wK","wK_err","wX","wX_err","e_x","e_K","e_x_e","de_x","de_K","de_x_e","e_th","de_xx","e_xx"); % ajouter Xp Xp_e
end % end err
function s = prb_f(x,y,Iv,dz,J,D,lambda,xMVDR)
% prb = @(x)([real(y); imag(y)] - [cos(-2*(pi/lambda)*(dz(:,1)*(Iv(1:D) - x(1:D)) + dz(:,2)*(Iv(D+1:D*2) - x(D+1:2*D))));
% sin(-2*(pi/lambda)*(dz(:,1)*(Iv(1:D) - x(1:D)) + dz(:,2)*(Iv(D+1:D*2) - x(D+1:2*D))))] * x(2*D+1:3*D));
Hmodif = zeros(J^2,D);
for q=1:D
Hmodif(: ,q) = exp(-1i*2*pi/lambda*(dz(:,1)*(Iv(q) - x(q))+dz(:,2)*(Iv(D+q) - x(D+q))));
end % .* aussi non ?
ymodif = Hmodif*xMVDR;
s = [real(y);imag(y)] - [real(ymodif);imag(ymodif)];
end