基于神经网络DNN实现信号均衡matlab完整源码.zip

clear close all gpuDevice(1) t = datetime('now'); %% Network parameters ori_rate = 30e6; rec_rate = 60e6; rate_times = rec_rate/ori_rate; equal_order = 30; % add_zero = rate_times*equal_order/2; headwindow = equal_order-(fix(equal_order/2)+1); pilot_length = 2047; zero_length = 3000; data_length = 10000; split_num = 10; % Cut a signal into split_num shares inputSize = equal_order; outputSize = 1; % y=h*x+n; y:(outputSize,m) h:(outputSize,inputSize) x:(inputSize,m) % miniBatchSize = 40; % bias_begin = 0.05; % bias_step = 0.04; % bias_end = 1.05; % bias_scope = bias_begin : bias_step : bias_end; bias_scope = 0.45; bias_begin = bias_scope(1); bias_end = bias_scope(end); bias_step = 1; bias_loop_num = (bias_end-bias_begin)/bias_step+1; bias_loop_num = round(bias_loop_num); % amp_scope = [0.1613 0.32106 0.48082 0.64058 0.8003 1]; % amp_scope = [0.005 0.007 0.015 0.024 0.034 0.045 0.08 0.18 0.25 0.3 0.48082 0.64058 0.8003 1]; amp_scope = 0.18; amp_loop_num = numel(amp_scope) ; load_begin = 61; load_end = 80; data_num = load_end-load_begin+1; train_percent = 0.05; save_folder = '6.18'; data_folder = '6.18'; net_folder = '7.11'; save_path = "data_save/light_data_"+save_folder; data_path_ini = "/home/oem/Users/ruoxu/equalization-using-DNN/data_save/light_data_"+data_folder; ver = 2; %% Loop parameter settings total_ver = 2; for net_ver = 1:total_ver net_type = [ver,1,net_ver]; net_path = save_path + "/result1/"+net_folder+"/mix_bias_amp/Threenonlinear"+net_type(1)+"/net/looptime"+net_type(2)+"/net"+net_type(3); test_num_amp = 0; band_power = zeros(1,round(amp_loop_num)); for amp_loop = 1:amp_loop_num %% Load data xTrain = []; yTrain = []; amp = amp_scope(amp_loop); save_amp = zeros(1,round(bias_loop_num)); save_bias = zeros(1,round(bias_loop_num)); data_path = data_path_ini + "/data/"+ori_rate/1e6+"M/amp"+amp; test_num_amp = test_num_amp +1; test_num_bias = 0; for bias = bias_begin : bias_step : bias_end test_num_bias = test_num_bias + 1; load_path = data_path+"/bias"+bias+"/mat"; fprintf(" load amp = %f ,bias = %f \n",amp,bias); load_data totalNum = data_num*split_num; trainNum = totalNum*train_percent; xTrain_tmp = data_received_load(1:trainNum); yTrain_tmp = data_ori_load(1:trainNum); xTest_bias_tmp = data_received_load(trainNum+1:end); yTest_bias_tmp = data_ori_load(trainNum+1:end); amp_nor = 32000*amp; % save_amp(test_num_bias) = 10*log10(amp_nor^2); save_amp(test_num_bias) = amp; save_bias(test_num_bias) = bias; yTrain_tmp = cellfun(@(cell1)(cell1*amp_nor),yTrain_tmp,'UniformOutput',false); yTest_bias_tmp = cellfun(@(cell1)(cell1*amp_nor),yTest_bias_tmp,'UniformOutput',false); xTest_bias_name = ['xTest_',num2str(test_num_bias)]; yTest_bias_name = ['yTest_',num2str(test_num_bias)]; eval([xTest_bias_name,'=xTest_bias_tmp;']); eval([yTest_bias_name,'=yTest_bias_tmp;']); xTrain = [xTrain xTrain_tmp]; yTrain = [yTrain yTrain_tmp]; clear x y end %% Normalize data % load_path = "data_save/light_data_3.10/data/10M/rand_bias0.3/"; load_norm_path = "/home/oem/Users/ruoxu/channel-estimation-using-DNN/data_save/norm_factor/"; norm_mat = load(load_norm_path+"/save_norm.mat"); norm_names = fieldnames(norm_mat); norm_factor = gather(eval(strcat('norm_mat.',norm_names{1}))); yTrain = cellfun(@(cell1)(cell1*norm_factor),yTrain,'UniformOutput',false); band_power(amp_loop) = bandpower(yTrain{2}); for i = 1:bias_loop_num yTest_nor = eval(['yTest_',num2str(i)]); yTest_nor = cellfun(@(cell1)(cell1*norm_factor),yTest_nor,'UniformOutput',false); eval([['yTest_',num2str(i)],'= yTest_nor;']); end %% Reshape data for i = 1:numel(xTrain) x_rate = []; for j = 1:rate_times x_rate_loop = xTrain{i}(j:rate_times:end); x_rate_loop = toeplitz(x_rate_loop(equal_order:-1:1),x_rate_loop(equal_order:end)); x_rate = [x_rate x_rate_loop]; end xTrain{i} = x_rate; % xTrain_loop{i} = [xTrain_loop{i}; single( bias_loop_data(floor((i-1)/trainNum)+1) )*ones(1,size(xTrain_loop{i},2) )]; yTrain{i} = reshape(yTrain{i},outputSize,[]); % yTrain_loop{i} = yTrain_loop{i}(:,1:size(xTrain_loop{i},2)); yTrain{i} = yTrain{i}(:,1:size(xTrain{i},2)/rate_times); yTrain_loop_tmp = yTrain{i}; yTrain{i} = []; for k = 1:rate_times yTrain{i} = [yTrain{i} yTrain_loop_tmp]; end end for i = 1:bias_loop_num xtop_tem = eval(['xTest_',num2str(i)]); ytop_tem = eval(['yTest_',num2str(i)]); for j = 1:numel(xtop_tem) x_rate = []; for k = 1:rate_times x_rate_loop = xtop_tem{j}(k:rate_times:end); x_rate_loop = toeplitz(x_rate_loop(equal_order:-1:1),x_rate_loop(equal_order:end)); x_rate = [x_rate x_rate_loop]; end xtop_tem{j} = x_rate; % xtop_tem{j} = [xtop_tem{j}; single( bias_loop_data(i) )*ones(1,size(xtop_tem{j},2) )]; ytop_tem{j} = reshape(ytop_tem{j},outputSize,[]); % ytop_tem{j} = ytop_tem{j}(:,1:size(xtop_tem{j},2)); ytop_tem{j} = ytop_tem{j}(:,1:size(xtop_tem{j},2)/rate_times); ytop_tem_2 = ytop_tem{j}; ytop_tem{j} = []; for r = 1:rate_times ytop_tem{j} = [ytop_tem{j} ytop_tem_2]; end end eval([['xTest_',num2str(i)],'= xtop_tem;']); eval([['yTest_',num2str(i)],'= ytop_tem;']); end %% Test performance with trained networks looptime = 2; load(net_path+"/net.mat"); fprintf(" load net = %s \n",net_path); for i = 1:bias_loop_num for j = 1:rate_times eval([['nmse_rate',num2str(j),'_bias',num2str(i),'_mat'],'= zeros(1,looptime);']); end end for i = 1:looptime for j = 1:bias_loop_num %% Custom network x_fortest = cell(1,rate_times); y_fortest = cell(1,rate_times); x_fortest_tmp = eval(['xTest_',num2str(j)]); y_fortest_tmp = eval(['yTest_',num2str(j)]); colNum_perRate = size(x_fortest_tmp{1},2)/rate_times; for shuffle_loop = 1:numel(x_fortest_tmp) for rate_loop = 1:rate_times x_fortest{rate_loop} = [ x_fortest{rate_loop} x_fortest_tmp{shuffle_loop}( :,(rate_loop-1)*colNum_perRate+1:rate_loop*colNum_perRate ) ]; y_fortest{rate_loop} = [ y_fortest{rate_loop} y_fortest_tmp{shuffle_loop}( :,(rate_loop-1)*colNum_perRate+1:rate_loop*colNum_perRate ) ]; end end for shuffle_loop = 1:numel(x_fortest) colNum_tmp = size(x_fortest{shuffle_loop},2); idx = randperm(colNum_tmp); x_fortest{shuffle_loop} = x_fortest{shuffle_loop}(:,idx); y_fortest{shuffle_loop} = y_fortest{shuffle_loop}(:,idx); end x_fortest = cell2mat(x_fortest); y_fortest = cell2mat(y_fortest); x_fortest(:,:,1) = x_fortest; y_fortest(:,:,1) = y_fortest; x_fortest = dlarray(single(x_fortest),'CBT');