-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathbeamspace_selection.m
238 lines (223 loc) · 10.4 KB
/
beamspace_selection.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
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
function [beam_matrix] = beamspace_selection(h_SRS, scenario)
% Initialize system parameters
N_ports=scenario.N_ports;
Ndigital = scenario.Ndigital;
Nrx = scenario.Nrx;
N_target_UE = scenario.N_target_UE;
training = scenario.training;
training_mode = scenario.training_mode;
% Initialize antenna lattice structure
N_pol_total = 2;
if scenario.Nrx == 64
N_hor_total = 8;
N_vert_total = 4;
elseif scenario.Nrx == 128
N_hor_total = 8;
N_vert_total = 8;
elseif scenario.Nrx == 256
N_hor_total = 16;
N_vert_total = 8;
elseif scenario.Nrx == 1024
N_hor_total = 32;
N_vert_total = 16;
end
if ~scenario.hybrid
% Determine the antenna lattice parameters
if scenario.beamspace_mode == 2
Wfin = (dftmtx(Nrx));
D = abs(diag(Wfin'*Wfin));
Wfin = Wfin * diag(1.0./sqrt(D));
elseif scenario.beamspace_mode == -1
if numel(size(h_SRS)) > 2
h_SRS = reshape( permute( h_SRS ,[2 1 3]), Nrx, []);
end
[U,S,~] = svd(h_SRS);
assert(isequal(sort(diag(S), "descend"),diag(S)))
beam_matrix = U(:,1:N_ports);
return
end
Nspace = Nrx;
else
% h_SRS [Nue, Nrx, Nsc]
assert(scenario.Nrx == 1024);
Wanalog = zeros(scenario.Nrx, scenario.Ndigital);
% Create mapping between subarray and antennas
inds_1D = 1:1024;
inds_2D = reshape(inds_1D, [N_pol_total N_vert_total N_hor_total]);
if scenario.digital_pol % Here can be other structures (e.g. 16-2-2)
Nsa_vert = 4;
Nsa_hor = 8;
Nsa_pol = 2;
else
Nsa_vert = 8;
Nsa_hor = 8;
Nsa_pol = 1;
end
sa2rx_inds = zeros(Ndigital, Nrx/Ndigital);
for i = 1:scenario.Ndigital
[p,v,h] = ind2sub([Nsa_pol Nsa_vert Nsa_hor], i);
N_hor_perSA = N_hor_total / Nsa_hor;
N_vert_perSA = N_vert_total / Nsa_vert;
ind = inds_2D(p, (v-1)*N_vert_perSA+1:v*N_vert_perSA, (h-1)*N_hor_perSA+1:h*N_hor_perSA);
sa2rx_inds(i,:) = ind(:);
end
% Determine values at phaseshifters
phases = zeros(1024,1);
switch scenario.analog_mode
case 0 % Just looking forward
% same zeros
case 1 % Offline trained on 140 scenarios
phases = load("Beamforming/phases_analog.mat").phases;
case 2 % Online trained for each scenario
scen_index = (scenario.index-1)*16+scenario.seed;
% phases = gd_analog_call(h_SRS, scenario.Ndigital, load("Beamforming/Python_input.mat").ind2train);
% save("Beamforming/saved_phases/phases_"+num2str(scen_index)+"scen", 'phases');
phases = load("Beamforming/saved_phases/phases_"+num2str(scen_index)+"scen").phases;
case 3 % Random phases (no channel info?)
phases = (2*rand(1024,1)-1) * pi;
case 4 % Trained with genetic algorithm
phases = load("Beamforming/Trained_phases_GA2.mat").phases;
case 5 % Trained with uniformity term
phases = load("Beamforming/Trained_phases_uniform.mat").phases;
case 6 % Long trained with gradient descent and uniformity
phases = load("Beamforming/Analog_learningGD_UNIFORM_2.mat").phases;
case -2 % For subarrays SU whole SVD phases
[u,~,~]=svds(squeeze(h_SRS(1,:,:)),1);
phases = angle(u);
case -3 % For subarray set MU whole SVD phases
Nsa_perUE = Ndigital / N_target_UE;
for user_id = 0:N_target_UE-1
[u,~,~]=svd(squeeze(h_SRS(user_id+1,:,:)));
for i = int32(user_id*Nsa_perUE+1:(user_id+1)*Nsa_perUE)
phases(sa2rx_inds(i,:)) = angle(u(sa2rx_inds(i,:),1));
end
end
case -4 % For subarray set MU subarray SVD phases
Nsa_perUE = Ndigital / N_target_UE;
for user_id = 0:scenario.N_target_UE-1
size_subarray = Nrx / Ndigital;
arrays_ind = user_id*Nsa_perUE+1:(user_id+1)*Nsa_perUE;
antennas_inds = sa2rx_inds(arrays_ind,:);
channel_part = squeeze(h_SRS(user_id+1,antennas_inds,:));
[u,~,~]=svds(channel_part,1);
u = reshape(u, [size_subarray,Nsa_perUE]);
for i = 1:numel(arrays_ind)
phases(antennas_inds(i,:)) = angle(u(i,:));
end
end
case -5 % For subarrays MU whole SVD phases
H = reshape(permute(h_SRS, [2,1,3]), scenario.Nrx, []);
[u,~,~] = svds(H,1);
phases = angle(u);
case -6 % For subarrays MU cluster SVD phases
u = load("Clustering&Learning/pi_subarray_svd_RX1024.mat").groups_directions(scenario.index, scenario.seed,:);
phases = angle(squeeze(u));
case -7 % For subarrays MU cluster SVD phases (filtered, GA)
load('Clustering/consistent_inds', 'consistent_inds'); % channels with dominating LOS component
% mapping between all and consistent channels and cluster index
cl_i = load(scenario.clustering_path).cl_i(ismember(consistent_inds, scenario.precalculated_indices(scenario.index,scenario.seed,1)));
% choosing the group direction
u = load(scenario.clustering_path).cl_d(cl_i,:);
if scenario.clustering_path == "Dergachev\Clustering&Learning\HC_sub"
phases = -angle(squeeze(u));
else
phases = angle(squeeze(u));
end
case -8 % For subarrays MU whole steering phases
steering = reshape(get_steering_matrix(squeeze(h_SRS)), [512,1]);
steering_vector = zeros(1024,1);
steering_vector(1:2:1024) = steering;
steering_vector(2:2:1024) = steering;
phases = angle(steering_vector);
case -9 % For subarrays MU subarray steering phases
h_SRS2 = squeeze(h_SRS);
for i = 1:scenario.Ndigital
phases(sa2rx_inds(i,:) ) = angle( reshape( get_steering_matrix( h_SRS2(sa2rx_inds(i,:) )), [16,1]) );
end
case -10 % For subarrays SU subarray SVD values
h_SRS2 = squeeze(h_SRS);
for i = 1:scenario.Ndigital
Wanalog(sa2rx_inds(i,:), i) = svds(h_SRS2(sa2rx_inds(i,:)), 1);
end
case -12 % For whole SU whole SVD values
h_SRS2 = squeeze(h_SRS);
[u, ~, ~] = svds(h_SRS2,1);
Wanalog(:,1) = u;
end
% Put computed phases to analog beamforming matrix
if ~ismember(scenario.analog_mode, [-10, -12])
for i = 1:scenario.Ndigital
ind = sa2rx_inds(i,:);
if scenario.subarray_phase
norm_multiplier = 1;
else
norm_multiplier = exp(1i*phases(ind(1)));
end
Wanalog(ind(:), i) = exp(1i*phases(ind(:))) / norm_multiplier;
end
end
% To have the right scale in detector
if ismember(scenario.analog_mode, [1 -10:-1])
norm_coefficient = sqrt(16);
elseif scenario.analog_mode == -12
norm_coefficient = 1/8;
end
Wanalog = Wanalog / norm_coefficient;
% Digital part beamforming
switch scenario.beamspace_mode
case 2 % Simple DFT beams
Wdigital = (dftmtx(scenario.Ndigital));
D = abs(diag(Wdigital'*Wdigital));
Wdigital = Wdigital * diag(1.0./sqrt(D));
case 4 % Offline trained FFT (with offline trained phases)
Wdigital = load("WeightsHybrid_mode3.mat").T_N;
case 12
if training; [~, T_N] = gd_digital(h_SRS, Wanalog, training_mode, 32); end
scen_index = (scenario.index-1)*16+scenario.seed;
if scenario.analog_mode == 4
if training; save("Beamforming/saved_digital/offline_phases_digital_"+num2str(scen_index)+"scen", 'T_N'); end
Wdigital = load("Beamforming/saved_digital/offline_phases_digital_"+num2str(scen_index)+"scen").T_N';
elseif scenario.analog_mode == 2
if training; save("Beamforming/saved_digital/online_phases_digital_"+num2str(scen_index)+"scen", 'T_N'); end
Wdigital = load("Beamforming/saved_digital/online_phases_digital_"+num2str(scen_index)+"scen").T_N';
elseif scenario.analog_mode == 3
if training; save("Beamforming/saved_digital/random_phases_digital_"+num2str(scen_index)+"scen", 'T_N');end
Wdigital = load("Beamforming/saved_digital/random_phases_digital_"+num2str(scen_index)+"scen").T_N';
elseif scenario.analog_mode == 5
if training; save("Beamforming/saved_digital/uniform_phases_digital_"+num2str(scen_index)+"scen_mode"+num2str(scenario.training_mode), 'T_N'); end
Wdigital = load("Beamforming/saved_digital/uniform_phases_digital_"+num2str(scen_index)+"scen_mode"+num2str(scenario.training_mode)).T_N';
elseif scenario.analog_mode == 6
if training; save("Beamforming/saved_digital/uniformGD_phases_digital_"+num2str(scen_index)+"scen_mode"+num2str(scenario.training_mode), 'T_N');end
Wdigital = load("Beamforming/saved_digital/uniformGD_phases_digital_"+num2str(scen_index)+"scen_mode"+num2str(scenario.training_mode)).T_N';
elseif scenario.analog_mode == -7
if training; save("Beamforming/saved_digital/filtered_GAcl_phases_digital_"+num2str(scen_index)+"scen_mode"+num2str(scenario.training_mode), 'T_N');end
Wdigital = load("Beamforming/saved_digital/filtered_GAcl_phases_digital_"+num2str(scen_index)+"scen_mode"+num2str(scenario.training_mode)).T_N';
end
case -1
if numel(size(h_SRS)) > 2
h_SRS = reshape( permute( h_SRS ,[2 1 3]), scenario.Nrx, []);
end
h_SRS = Wanalog' * h_SRS;
[U,S,~] = svd(h_SRS);
assert(isequal(sort(diag(S), "descend"),diag(S)))
beam_matrix = Wanalog * U(:,1:N_ports);
return
case 0
Wdigital = eye(scenario.Ndigital);
end
Wfin = Wanalog*Wdigital;
Nspace = Ndigital;
end
% Choosing most powerful beams among digital ones
beam_power_norm = zeros(scenario.N_user, Nspace);
for i = 1 : scenario.N_user
if i <= N_target_UE
beam_power = sum( abs( squeeze(h_SRS(i,:,:)).' * conj(Wfin)).^2 , 1);
else
beam_power = 1;
end
beam_power_norm(i,:) = beam_power / sum(beam_power);
end
beam_power_MU = sum (beam_power_norm,1);
[~, beam_indx] = sort(beam_power_MU, 'descend');
beam_matrix=Wfin(:,beam_indx(1:N_ports));