%TODO: Find false detected peaks. This could be a not detected or an additional detected peak. Simple algorithm: % 1) Choose smallest diff(peaks) value smallest_peak (e.g. 48, 97, 49, [45], ..) % 2) Divide all diff_peaks values with smallest_peak to get factor_peaks (e.g. 1.1, 1.8, 1.2, 1, ..) and to whole numbers (e.g. 1, [2], 1, 1, ..) % 3) Divide diff_peaks / factor_peaks and split (e.g. 48, [48.5, 48.5], 49, 45, ..) diff_peaks_new % 4) Start at first peak and cumsum the correlation values at the lag positions given by diff_peaks_new to sum_peaks_new and sum_peaks for diff_peaks % 5) If sum_diff_new > sum_diff finish, else remove smallest_peak from diff(peaks) and goto 1) function idx = findFalseDetectedPeaks(idx_orig, lag_orig, corr_orig) idx = idx_orig; while 1 %1 diff_peaks = diff(lag_orig(idx_orig)); [smallest_peak, smallest_peak_idx] = min(diff_peaks); %2 factor_peaks = round(diff_peaks / smallest_peak); %3 %factor_matrix = [1:length(factor_peaks); factor_peaks]; %we need this since octave doesn't have a repelem only repelemS. %diff_peaks_new = repelems(diff_peaks ./ factor_peaks, factor_matrix); diff_peaks_new = repelem(diff_peaks ./ factor_peaks, factor_peaks); %4 idx_new = round([idx_orig(1), cumsum(diff_peaks_new) + idx_orig(1)]'); sum_peaks_new = sum(corr_orig(idx_new)); sum_peaks = sum(corr_orig(idx_orig)); %5 if(sum_peaks_new < sum_peaks) % TODO: idx + 1 is not always the correct one. if diff is very % small, the false peak could be left or right from the real peak. % Solution: Remove peak with lowest correlation value in % this area if(corr_orig(idx_orig(smallest_peak_idx)) > corr_orig(idx_orig(smallest_peak_idx + 1))) idx_orig(smallest_peak_idx + 1) = []; else idx_orig(smallest_peak_idx) = []; end else idx = idx_new; end if (sum_peaks_new >= sum_peaks || length(idx_orig) <= 1) break; end %TODO: Warning: could this be an infinite loop? end end