Making mincompare more global and updating RLS to Matlab2017

subfunctions/benchmark-tools/bibs/FECGSYN_QRSmincompare.m

@@ -73,8 +73,8 @@
 fqrs = cell(1,size(data,1));
 for j = 1:size(data,1) 
  fqrstmp = cell(1,ceil(size(data,2)/(60*fs)));
- for l = 1:5
- fqrstmp{l} = jqrs_mod(data(j,60*fs*(l-1)+1:60*fs*(l)),REFRAC,TH,fs) + 60*fs*(l-1); 
+ for l = 1:length(fqrstmp)
+ fqrstmp{l} = jqrs_mod(data(j,60*fs*(l-1)+1:min(length(data),60*fs*(l))),REFRAC,TH,fs) + 60*fs*(l-1); 
  end
  fqrs{j} = cell2mat(fqrstmp);
 end

subfunctions/extraction-methods/libs/FECGESN/FECGESN_lmsrls_canceller.m

@@ -1,10 +1,10 @@
 function residual = FECGESN_lmsrls_canceller(ref_ecg,tar_ecg,fs,metStruct,debug)
 % Adaptive Noise cancelling filter for FECG source separation.
 %
-% This function uses the least mean square (LMS) or the recursive least square 
-% (RLS) algorithm in order to remove the MECG contribution on the abdominal 
-% mixture. The LMS can be run 'online' (i.e the weights are left evolving) 
-% or 'offline' (the weights are optimised on a training set and then constant 
+% This function uses the least mean square (LMS) or the recursive least square
+% (RLS) algorithm in order to remove the MECG contribution on the abdominal
+% mixture. The LMS can be run 'online' (i.e the weights are left evolving)
+% or 'offline' (the weights are optimised on a training set and then constant
 % for the test set). The RLS is run online.
 %
 % inputs
@@ -16,14 +16,14 @@
 % metStruct.N
 % metStruct.learningMode ('online'/'offline')
 % metStruct.method ('LMS'/'RLS')
-% 
+%
 % tar_ecg
 % residual: FECG residual
 %
 % NOTE: time used for initialising the weights = 30sec (see TIME_INIT constant)
 %
 %
-% References 
+% References
 % 1. B. Widrow, J. Glover, J. McCool, J. Kaunitz, C. Williams, R. Hearn,
 % J. Zeidler, E. Dong, and R. Goodlin, Adaptive noise cancelling:
 % Principles and applications, Proceedings of the IEEE, vol. 63, no. 12,
@@ -59,66 +59,61 @@
 TIME_INIT = 30;
 
 % == core function
-try
- % == normalize the data
- % assumes that the first 5sec are representative (no huge noise) for normalizaiton purposes. 
- [input_norm,~] = FECGESN_normalise_ecg(ref_ecg,1,5*fs);
- [output_norm,~] = FECGESN_normalise_ecg(tar_ecg,1,5*fs);
-
- if strcmp(metStruct.method,'LMS')
+% == normalize the data
+% assumes that the first 5sec are representative (no huge noise) for normalizaiton purposes.
+[input_norm,~] = FECGESN_normalise_ecg(ref_ecg,1,5*fs);
+[output_norm,~] = FECGESN_normalise_ecg(tar_ecg,1,5*fs);
+
+if strcmp(metStruct.method,'LMS')
  % == least mean square
- if strcmp(metStruct.learningMode,'online')
- [y,residual,~] = FECGESN_lms(input_norm,output_norm,metStruct.mu,metStruct.Nunits);
- else
- % training readout on first 30sec and test on ALL signal using
- % constant weights
-
- % split into train and test 
- train_fraction = TIME_INIT/(length(input_norm)/fs);
- [train_in,~] = split_train_test(input_norm,train_fraction);
- [train_out,~] = split_train_test(output_norm,train_fraction);
-
- % find lms param using training sequence
- [y,~,W] = FECGESN_lms(train_in,train_out,metStruct.mu,metStruct.Nunits);
- w = W(:,end);
-
- % now apply trained lms to all the signal
- residual = output_norm-filter(w,1,input_norm);
-
- % [~,w] = build_design_matrix_and_whopf(train_in,train_out,metStruct.Nunits);
- % chestf = conv(input_norm,w,'valid'); % idem as filtering
- % plot(output_norm(1:end)); hold on, plot(chestf(metStruct.Nunits+1:end),'r');
- % residual = output_norm-chestf(metStruct.Nunits+1:end); 
- end
- elseif strcmp(metStruct.method,'RLS')
- % == recursive least square
- p0 = 2*eye(metStruct.Nunits);
- ha = adaptfilt.rls(metStruct.Nunits,metStruct.mu,p0);
- [y,residual] = filter(ha,input_norm,output_norm); 
+ if strcmp(metStruct.learningMode,'online')
+ [y,residual,~] = FECGESN_lms(input_norm,output_norm,metStruct.mu,metStruct.Nunits);
+ else
+ % training readout on first 30sec and test on ALL signal using
+ % constant weights
+
+ % split into train and test
+ train_fraction = TIME_INIT/(length(input_norm)/fs);
+ [train_in,~] = split_train_test(input_norm,train_fraction);
+ [train_out,~] = split_train_test(output_norm,train_fraction);
+
+ % find lms param using training sequence
+ [y,~,W] = FECGESN_lms(train_in,train_out,metStruct.mu,metStruct.Nunits);
+ w = W(:,end);
+
+ % now apply trained lms to all the signal
+ residual = output_norm-filter(w,1,input_norm);
+
+ % [~,w] = build_design_matrix_and_whopf(train_in,train_out,metStruct.Nunits);
+ % chestf = conv(input_norm,w,'valid'); % idem as filtering
+ % plot(output_norm(1:end)); hold on, plot(chestf(metStruct.Nunits+1:end),'r');
+ % residual = output_norm-chestf(metStruct.Nunits+1:end);
  end
- 
-catch ME
- rethrow(ME);
- for enb=1:length(ME.stack); disp(ME.stack(enb)); end;
- residual = [];
+elseif strcmp(metStruct.method,'RLS')
+ % == recursive least square
+ p0 = 2*eye(metStruct.Nunits);
+ RLSFilt = dsp.RLSFilter(metStruct.Nunits,'ForgettingFactor',metStruct.mu,'InitialInverseCovariance',p0);
+ [y,residual] = RLSFilt(input_norm,output_norm);
 end
 
-% == debug 
+
+
+% == debug
 if debug
  LINE_WIDTH = 2;
  FONT_SIZE = 15;
  nb_of_points = length(ref_ecg);
  tm = 1/fs:1/fs:nb_of_points/fs;
  figure('name','lmsrls canceller');
- ax(1) = subplot(311); plot(tm,input_norm,'LineWidth',LINE_WIDTH); 
-  hold on, plot(tm,output_norm,'r','LineWidth',LINE_WIDTH);
-  legend('normalised reference ecg (chest ECG)','normalised target ecg (abdominal ECG)');
+ ax(1) = subplot(311); plot(tm,input_norm,'LineWidth',LINE_WIDTH);
+ hold on, plot(tm,output_norm,'r','LineWidth',LINE_WIDTH);
+ legend('normalised reference ecg (chest ECG)','normalised target ecg (abdominal ECG)');
  ax(2) = subplot(312); plot(tm,output_norm); hold on, plot(tm,y,'r','LineWidth',LINE_WIDTH);
-  legend('normalised target ecg (abdominal ECG)','predicted target ecg');
-  set(findall(gcf,'type','text'),'fontSize',FONT_SIZE);
+ legend('normalised target ecg (abdominal ECG)','predicted target ecg');
+ set(findall(gcf,'type','text'),'fontSize',FONT_SIZE);
  ax(3) = subplot(313); plot(tm,residual,'r','LineWidth',LINE_WIDTH);
-  legend('residual - FECG');
-  set(findall(gcf,'type','text'),'fontSize',FONT_SIZE);
+ legend('residual - FECG');
+ set(findall(gcf,'type','text'),'fontSize',FONT_SIZE);
  linkaxes(ax);
 end