Mercurial > hg > ltpda
view m-toolbox/classes/@ao/gapfilling.m @ 38:3aef676a1b20 database-connection-manager
Keep backtrace on error
author | Daniele Nicolodi <nicolodi@science.unitn.it> |
---|---|
date | Mon, 05 Dec 2011 16:20:06 +0100 |
parents | f0afece42f48 |
children |
line wrap: on
line source
% GAPFILLING fills possible gaps in data. %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% % % DESCRIPTION: GAPFILLING interpolated data between two data % segments. This function might be useful for possible % gaps or corrupted data. Two different types of % interpolating are available: linear and spline, the latter % results in a smoother curve between the two data segments. % % CALL: b = gapfilling(a1, a2, pl) % % INPUTS: a1 - data segment previous to the gap % a2 - data segment posterior to the gap % pl - parameter list % % OUTPUTS: b - data segment containing a1, a2 and the filled data % segment, i.e., b=[a1 datare_filled a2]. % % <a href="matlab:utils.helper.displayMethodInfo('ao', 'gapfilling')">Parameters Description</a> % % VERSION: $Id: gapfilling.m,v 1.21 2011/11/07 15:34:34 miquel Exp $ % %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% function varargout = gapfilling(varargin) %%% Check if this is a call for parameters if utils.helper.isinfocall(varargin{:}) varargout{1} = getInfo(varargin{3}); return end if nargout == 0 error('### cat cannot be used as a modifier. Please give an output variable.'); end % Collect input variable names in_names = cell(size(varargin)); for ii = 1:nargin,in_names{ii} = inputname(ii);end % Collect all AOs [as, ao_invars] = utils.helper.collect_objects(varargin(:), 'ao', in_names); pli = utils.helper.collect_objects(varargin(:), 'plist', in_names); pls = parse(pli, getDefaultPlist()); if length(as)~=2 error('only two analysis objects are needed!') end % go through each input AO for i=1:numel(as) % check this is a time-series object if ~isa(as(i).data, 'tsdata') error(' ### Gap filling requires tsdata (time-series) inputs.') end end %--- check input parameters method = find(pls, 'method'); % method definition: linear or 'spline' addnoise = find(pls, 'addnoise'); % decide whether add noise or not in the filled data %--------------------------- % rename AOs a1 = as(1); a2 = as(2); if a1.fs ~= a2.fs warning('Sampling frequencies of the two AOs are different. The sampling frequency of the first AO will be used to reconstruct the gap.') end % Different units error if a1.xunits ~= a2.xunits error('!!! Data has different X units !!!'); end if a1.yunits ~= a2.yunits error('!!! Data has different Y units !!!'); end a1_length = len(a1); a2_length = len(a2); start_a2 = a2.t0.utc_epoch_milli/1000 + a2.x(1); end_a1 = (a1.t0.utc_epoch_milli/1000 + a1.x(1) + a1.nsecs - 1); gaptime = (start_a2 - end_a1); gapn = gaptime*a1.fs-1; t = (0:1:gapn-1)'/a1.fs; %--- gapfilling process itself if strcmp(method,'linear') % linear interpolation method ---xfilled=(deltay/deltax)*t+y1(length(y1))--- if len(a1)>10 && len(a2)>10 dy = mean(a2.y(1:10))-mean(a1.y(a1_length-10:a1_length)); filling_data = (dy/gaptime)*t + mean(a1.y(a1_length-10:a1_length)); filling_time = (1:1:gapn)'/a1.fs + a1.x(a1_length); else error('!!! Not enough data in the data segments (min=11 for each one for the linear method) !!!'); end elseif strcmp(method,'spline') % spline method xfilled = a*T^3 + b*T^2 + c*T +d if len(a1)>1000 && len(a2)>1000 % derivatives of the input data are calculated da1 = diff(a1.y(1:100:a1_length))*(a1.fs/100); da1 = tsdata(da1, a1.fs/100); da1 = ao(da1); da2 = diff(a2.y(1:100:a2_length))*(a2.fs/100); da2 = tsdata(da2, a2.fs/100); da2 = ao(da2); % This filters the previous derivatives % filters parameters are obtained plfa1 = getlpFilter(a1.fs/100); plfa2 = getlpFilter(a2.fs/100); lpfa1 = miir(plfa1); lpfpla1 = plist(param('filter', lpfa1)); lpfa2 = miir(plfa2); lpfpla2 = plist(param('filter', lpfa2)); % derivatives are low-pass filtered da1filtered = filtfilt(da1, lpfpla1); da2filtered = filtfilt(da2, lpfpla2); % coefficients are calculated c = mean(da1filtered.y(len(da1filtered)... -10:len(da1filtered))); d = mean(a1.y(len(a1)-10:len(a1))); a=(2*d+(c+mean(da2filtered.y(1:10)))... *gaptime-2*mean(a2.y(1:10)))/(gaptime.^3); b=-(3*d+2*c*gaptime+mean(da2filtered.y(1:10))... *gaptime-3*mean(a2.y(1:10)))/(gaptime^2); % filling data is calculated with the coefficients a, b, c and d filling_data = a*t.^3+b*t.^2+c*t+d; filling_time = (1:1:gapn)'/a1.fs + a1.x(a1_length); else error('!!! Not enough data in data segments (min=1001 in spline method)'); end end % this add noise (if desired) to the filled gap if strcmp(addnoise,'yes'); % calculation of the standard deviation after eliminating the low-frequency component phpf = gethpFilter(a1.fs); ax = tsdata(a1.y, a1.fs); ax = ao(ax); hpf = miir(phpf); hpfpl = plist(param('filter', hpf)); xhpf = filter(ax, hpfpl); hfnoise = std(xhpf); % noise is added to the filling data filling_data = filling_data + randn(length(filling_data),1)*hfnoise.data.getY; end % join data filling_data = [a1.y; filling_data; a2.y]; filling_time = [a1.x; filling_time; a2.x]; % create new output tsdata ts = tsdata(filling_time, filling_data); ts.setYunits(a1.yunits); ts.setXunits(a1.xunits); % make output analysis object b = ao(ts); b.setT0(a1.t0) b.name = sprintf('gapfilling(%s,%s)', ao_invars{1}, ao_invars{2}); b.addHistory(getInfo('None'), pls, [ao_invars(1) ao_invars(2)], [as(1).hist as(2).hist]); % Set output if nargout == numel(b) % List of outputs for ii = 1:numel(b) varargout{ii} = b(ii); end else % Single output varargout{1} = b; end end function plf = getlpFilter(x) plf = plist(); plf = append(plf, param('gain', 1)); plf = append(plf, param('ripple', 0.5)); plf = append(plf, param('type', 'lowpass')); plf = append(plf, param('order', 2)); plf = append(plf, param('fs', x)); plf = append(plf, param('fc', 0.1/100)); end function phf = gethpFilter(x) phf = plist(); phf = append(phf, param('gain', 1)); phf = append(phf, param('ripple', 0.5)); phf = append(phf, param('type', 'highpass')); phf = append(phf, param('order', 2)); phf = append(phf, param('fs', x)); phf = append(phf, param('fc', 0.1/100)); end %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% % Local Functions % %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% % % FUNCTION: getInfo % % DESCRIPTION: Get Info Object % % HISTORY: 11-07-07 M Hewitson % Creation. % %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% function ii = getInfo(varargin) if nargin == 1 && strcmpi(varargin{1}, 'None') sets = {}; pl = []; else sets = {'Default'}; pl = getDefaultPlist(); end % Build info object ii = minfo(mfilename, 'ao', 'ltpda', utils.const.categories.sigproc, '$Id: gapfilling.m,v 1.21 2011/11/07 15:34:34 miquel Exp $', sets, pl); ii.setModifier(false); end %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% % % FUNCTION: getDefaultPlist % % DESCRIPTION: Get Default Plist % % HISTORY: 11-07-07 M Hewitson % Creation. % %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% function plout = getDefaultPlist() persistent pl; if exist('pl', 'var')==0 || isempty(pl) pl = buildplist(); end plout = pl; end function pl = buildplist() pl = plist(); % Method p = param({'method', 'The method used to interpolate data.'}, {1, {'linear', 'spline'}, paramValue.SINGLE}); pl.append(p); % Add noise p = param({'addnoise', ... ['Noise can be added to the interpolated data.<br>'... 'This noise is defined as random variable with<br>'... 'zero mean and variance equal to the high-frequency<br>'... 'noise of the first input.']}, paramValue.YES_NO); p.val.setValIndex(2); pl.append(p); end % PARAMETERES: 'method' - method used to interpolate data between a1 and a2. % Two options can be used: 'linear' and 'spline'. % Default values is 'linear'. % 'addnoise' - noise can be added to the interpolated data. % This noise is defined as random variable with % zero mean and variance equal to the high-frequency % noise if a1. 'Yes' adds noise. Default value % is 'no'.