s3158985
/
MAP_Gait_Dynamics
Template
1
0
Fork 0
Code Issues Pull Requests 1 Releases Wiki Activity

TEST remote repository

This commit is contained in:
Lars 2020-11-07 16:46:05 +01:00
parent 5491738fff
commit ba8df083e2
32 changed files with 6008 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

7
AutocorrRegSymmSteps.m Normal file
View File

@ -0,0 +1,7 @@
% autocorrelation for step symmetrie
function [c, lags] = AutocorrRegSymmSteps(x)
[c,lags] = xcov(x,200,'unbiased');
c = c/c(lags==0);
c = c(lags>=0);
lags = lags(lags>=0);

43
AutocorrStrides.m Normal file
View File

@ -0,0 +1,43 @@
function [ResultStruct] = AutocorrStrides(data,FS, StrideTimeRange,ResultStruct);
%% Stride-times measures
% Stride time and regularity from auto correlation (according to Moe-Nilssen and Helbostad, Estimation of gait cycle characteristics by trunk accelerometry. J Biomech, 2004. 37: 121-6.)
RangeStart = round(FS*StrideTimeRange(1));
RangeEnd = round(FS*StrideTimeRange(2));
[AutocorrResult,Lags]=xcov(data,RangeEnd,'unbiased');
AutocorrSum = sum(AutocorrResult(:,[1 5 9]),2); % This sum is independent of sensor re-orientation, as long as axes are kept orthogonal
AutocorrResult2= [AutocorrResult(:,[1 5 9]),AutocorrSum];
IXRange = (numel(Lags)-(RangeEnd-RangeStart)):numel(Lags);
% check that autocorrelations are positive for any direction,
AutocorrPlusTrans = AutocorrResult+AutocorrResult(:,[1 4 7 2 5 8 3 6 9]);
IXRangeNew = IXRange( ...
AutocorrPlusTrans(IXRange,1) > 0 ...
& prod(AutocorrPlusTrans(IXRange,[1 5]),2) > prod(AutocorrPlusTrans(IXRange,[2 4]),2) ...
& prod(AutocorrPlusTrans(IXRange,[1 5 9]),2) + prod(AutocorrPlusTrans(IXRange,[2 6 7]),2) + prod(AutocorrPlusTrans(IXRange,[3 4 8]),2) ...
> prod(AutocorrPlusTrans(IXRange,[1 6 8]),2) + prod(AutocorrPlusTrans(IXRange,[2 4 9]),2) + prod(AutocorrPlusTrans(IXRange,[3 5 7]),2) ...
);
if isempty(IXRangeNew)
StrideTimeSamples = Lags(IXRange(AutocorrSum(IXRange)==max(AutocorrSum(IXRange)))); % to be used in other estimations
StrideTimeSeconds = nan;
ResultStruct.StrideTimeSamples = StrideTimeSamples;
ResultStruct.StrideTimeSeconds = StrideTimeSeconds;
else
StrideTimeSamples = Lags(IXRangeNew(AutocorrSum(IXRangeNew)==max(AutocorrSum(IXRangeNew))));
StrideRegularity = AutocorrResult2(Lags==StrideTimeSamples,:)./AutocorrResult2(Lags==0,:); % Moe-Nilssen&Helbostatt,2004
RelativeStrideVariability = 1-StrideRegularity;
StrideTimeSeconds = StrideTimeSamples/FS;
ResultStruct.StrideRegularity_V = StrideRegularity(1);
ResultStruct.StrideRegularity_ML = StrideRegularity(2);
ResultStruct.StrideRegularity_AP = StrideRegularity(3);
ResultStruct.StrideRegularity_All = StrideRegularity(4);
ResultStruct.RelativeStrideVariability_V = RelativeStrideVariability(1);
ResultStruct.RelativeStrideVariability_ML = RelativeStrideVariability(2);
ResultStruct.RelativeStrideVariability_AP = RelativeStrideVariability(3);
ResultStruct.RelativeStrideVariability_All = RelativeStrideVariability(4);
ResultStruct.StrideTimeSamples = StrideTimeSamples;
ResultStruct.StrideTimeSeconds = StrideTimeSeconds;
end

118
CalcMaxLyapConvGait.m Normal file
View File

@ -0,0 +1,118 @@
function [L_Estimate,ExtraArgsOut] = CalcMaxLyapConvGait(ThisTimeSeries,FS,ExtraArgsIn)
if nargin > 2
if isfield(ExtraArgsIn,'J')
J=ExtraArgsIn.J;
end
if isfield(ExtraArgsIn,'m')
m=ExtraArgsIn.m;
end
if isfield(ExtraArgsIn,'FitWinLen')
FitWinLen=ExtraArgsIn.FitWinLen;
end
end
%% Initialize output args
L_Estimate=nan;ExtraArgsOut.Divergence=nan;ExtraArgsOut.J=nan;ExtraArgsOut.m=nan;ExtraArgsOut.FitWinLen=nan;
%% Some checks
% predefined J and m should not be NaN or Inf
if (exist('J','var') && ~isempty(J) && ~isfinite(J)) || (exist('m','var') && ~isempty(m) && ~isfinite(m))
warning('Predefined J and m cannot be NaN or Inf');
return;
end
% multidimensional time series need predefined J and m
if size(ThisTimeSeries,2) > 1 && (~exist('J','var') || ~exist('m','var') || isempty(J) || isempty(m))
warning('Multidimensional time series needs predefined J and m, can''t determine Lyapunov');
return;
end
%Check that there are no NaN or Inf values in the TimeSeries
if any(~isfinite(ThisTimeSeries(:)))
warning('Time series contains NaN or Inf, can''t determine Lyapunov');
return;
end
%Check that there is variation in the TimeSeries
if ~(nanstd(ThisTimeSeries) > 0)
warning('Time series is constant, can''t determine Lyapunov');
return;
end
%% Determine FitWinLen (=cycle time) of ThisTimeSeries
if ~exist('FitWinLen','var') || isempty(FitWinLen)
if size(ThisTimeSeries,2)>1
for dim=1:size(ThisTimeSeries,2),
[Pd(:,dim),F] = pwelch(detrend(ThisTimeSeries(:,dim)),[],[],[],FS);
end
P = sum(Pd,2);
else
[P,F] = pwelch(detrend(ThisTimeSeries),[],[],[],FS);
end
MeanF = sum(P.*F)./sum(P);
CycleTime = 1/MeanF;
FitWinLen = round(CycleTime*FS);
else
CycleTime = FitWinLen/FS;
end
ExtraArgsOut.FitWinLen=FitWinLen;
%% Determine J
if ~exist('J','var') || isempty(J)
% Calculate mutual information and take first local minimum Tau as J
bV = min(40,floor(sqrt(size(ThisTimeSeries,1))));
tauVmax = FitWinLen;
[mutMPro,cummutMPro,minmuttauVPro] = MutualInformationHisPro(ThisTimeSeries,(0:tauVmax),bV,1); % (xV,tauV,bV,flag)
if isnan(minmuttauVPro)
display(mutMPro);
warning('minmuttauVPro is NaN. Consider increasing tauVmax.');
return;
end
J=minmuttauVPro;
end
ExtraArgsOut.J=J;
%% Determine m
if ~exist('m','var') || isempty(m)
escape = 10;
max_m = 20;
max_fnnM = 0.02;
mV = 0;
fnnM = 1;
for mV = 2:max_m % for m=1, FalseNearestNeighbors is slow and lets matlab close if N>500000
fnnM = FalseNearestNeighborsSR(ThisTimeSeries,J,mV,escape,FS); % (xV,tauV,mV,escape,theiler)
if fnnM <= max_fnnM || isnan(fnnM)
break
end
end
if fnnM <= max_fnnM
m = mV;
else
warning('Too many false nearest neighbours');
return;
end
end
ExtraArgsOut.m=m;
%% Create state space based upon J and m
N_ss = size(ThisTimeSeries,1)-(m-1)*J;
StateSpace=nan(N_ss,m*size(ThisTimeSeries,2));
for dim=1:size(ThisTimeSeries,2),
for delay=1:m,
StateSpace(:,(dim-1)*m+delay)=ThisTimeSeries((1:N_ss)'+(delay-1)*J,dim);
end
end
%% Parameters for Lyapunov
WindowLen = floor(min(N_ss/5,10*FitWinLen));
if WindowLen < FitWinLen
warning('Not enough samples for Lyapunov estimation');
return;
end
WindowLenSec=WindowLen/FS;
%% Calculate divergence
Divergence=div_calc(StateSpace,WindowLenSec,FS,CycleTime,0);
ExtraArgsOut.Divergence=Divergence;
%% Calculate slope of first FitWinLen samples of divergence curve
p = polyfit((1:FitWinLen)/FS,Divergence(1:FitWinLen),1);
L_Estimate = p(1);

134
CalcMaxLyapWolfFixedEvolv.m Normal file
View File

@ -0,0 +1,134 @@
function [L_Estimate,ExtraArgsOut] = CalcMaxLyapWolfFixedEvolv(ThisTimeSeries,FS,ExtraArgsIn)
%% Description
% This function calculates the maximum Lyapunov exponent from a time
% series, based on the method described by Wolf et al. in
% Wolf, A., et al., Determining Lyapunov exponents from a time series.
% Physica D: 8 Nonlinear Phenomena, 1985. 16(3): p. 285-317.
%
% Input:
% ThisTimeSeries: a vector or matrix with the time series
% FS: sample frequency of the ThisTimeSeries
% ExtraArgsIn: a struct containing optional input arguments
% J (embedding delay)
% m (embedding dimension)
% Output:
% L_Estimate: The Lyapunov estimate
% ExtraArgsOut: a struct containing the additional output arguments
% J (embedding delay)
% m (embedding dimension)
%% Copyright
% COPYRIGHT (c) 2012 Sietse Rispens, VU University Amsterdam
%
% This program is free software: you can redistribute it and/or modify
% it under the terms of the GNU General Public License as published by
% the Free Software Foundation, either version 3 of the License, or
% (at your option) any later version.
%
% This program is distributed in the hope that it will be useful,
% but WITHOUT ANY WARRANTY; without even the implied warranty of
% MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
% GNU General Public License for more details.
%
% You should have received a copy of the GNU General Public License
% along with this program. If not, see <http://www.gnu.org/licenses/>.
%% Author
% Sietse Rispens
%% History
% April 2012, initial version of CalcMaxLyapWolf
% 23 October 2012, use fixed evolve time instead of adaptable
if nargin > 2
if isfield(ExtraArgsIn,'J')
J=ExtraArgsIn.J;
end
if isfield(ExtraArgsIn,'m')
m=ExtraArgsIn.m;
end
end
%% Initialize output args
L_Estimate=nan;ExtraArgsOut.J=nan;ExtraArgsOut.m=nan;
%% Some checks
% predefined J and m should not be NaN or Inf
if (exist('J','var') && ~isempty(J) && ~isfinite(J)) || (exist('m','var') && ~isempty(m) && ~isfinite(m))
warning('Predefined J and m cannot be NaN or Inf');
return;
end
% multidimensional time series need predefined J and m
if size(ThisTimeSeries,2) > 1 && (~exist('J','var') || ~exist('m','var') || isempty(J) || isempty(m))
warning('Multidimensional time series needs predefined J and m, can''t determine Lyapunov');
return;
end
%Check that there are no NaN or Inf values in the TimeSeries
if any(~isfinite(ThisTimeSeries(:)))
warning('Time series contains NaN or Inf, can''t determine Lyapunov');
return;
end
%Check that there is variation in the TimeSeries
if ~(nanstd(ThisTimeSeries) > 0)
warning('Time series is constant, can''t determine Lyapunov');
return;
end
%% Determine J
if ~exist('J','var') || isempty(J)
% Calculate mutual information and take first local minimum Tau as J
bV = min(40,floor(sqrt(size(ThisTimeSeries,1))));
tauVmax = 70;
[mutMPro,cummutMPro,minmuttauVPro] = MutualInformationHisPro(ThisTimeSeries,(0:tauVmax),bV,1); % (xV,tauV,bV,flag)
if isnan(minmuttauVPro)
display(mutMPro);
warning('minmuttauVPro is NaN. Consider increasing tauVmax.');
return;
end
J=minmuttauVPro;
end
ExtraArgsOut.J=J;
%% Determine m
if ~exist('m','var') || isempty(m)
escape = 10;
max_m = 20;
max_fnnM = 0.02;
mV = 0;
fnnM = 1;
for mV = 2:max_m % for m=1, FalseNearestNeighbors is slow and lets matlab close if N>500000
fnnM = FalseNearestNeighborsSR(ThisTimeSeries,J,mV,escape,FS); % (xV,tauV,mV,escape,theiler)
if fnnM <= max_fnnM || isnan(fnnM)
break
end
end
if fnnM <= max_fnnM
m = mV;
else
warning('Too many false nearest neighbours');
return;
end
end
ExtraArgsOut.m=m;
%% Create state space based upon J and m
N_ss = size(ThisTimeSeries,1)-(m-1)*J;
StateSpace=nan(N_ss,m*size(ThisTimeSeries,2));
for dim=1:size(ThisTimeSeries,2),
for delay=1:m,
StateSpace(:,(dim-1)*m+delay)=ThisTimeSeries((1:N_ss)'+(delay-1)*J,dim);
end
end
%% Parameters for Lyapunov estimation
CriticalLen=J*m;
max_dist = sqrt(sum(std(StateSpace).^2))/10;
max_dist_mult = 5;
min_dist = max_dist/2;
max_theta = 0.3;
evolv = J;
%% Calculate Lambda
[L_Estimate]=div_wolf_fixed_evolv(StateSpace, FS, min_dist, max_dist, max_dist_mult, max_theta, CriticalLen, evolv);

48
CalculateNonLinearParametersFunc.m Normal file
View File

@ -0,0 +1,48 @@
function [ResultStruct] = CalculateNonLinearParametersFunc(ResultStruct,dataAccCut,WindowLen,FS,Lyap_m,Lyap_FitWinLen,Sen_m,Sen_r)
%% Calculation non-linear parameters;
% cut into windows of size WindowLen
N_Windows = floor(size(dataAccCut,1)/WindowLen);
N_SkipBegin = ceil((size(dataAccCut,1)-N_Windows*WindowLen)/2);
LyapunovWolf = nan(N_Windows,3);
LyapunovRosen = nan(N_Windows,3);
SE= nan(N_Windows,3);
for WinNr = 1:N_Windows;
AccWin = dataAccCut(N_SkipBegin+(WinNr-1)*WindowLen+(1:WindowLen),:);
for j=1:3
[LyapunovWolf(WinNr,j),~] = CalcMaxLyapWolfFixedEvolv(AccWin(:,j),FS,struct('m',Lyap_m));
[LyapunovRosen(WinNr,j),outpo] = CalcMaxLyapConvGait(AccWin(:,j),FS,struct('m',Lyap_m,'FitWinLen',Lyap_FitWinLen));
[SE(WinNr,j)] = funcSampleEntropy(AccWin(:,j), Sen_m, Sen_r);
% no correction for FS; SE does increase with higher FS but effect is considered negligible as range is small (98-104HZ). Might consider updating r to account for larger ranges.
end
end
LyapunovWolf = nanmean(LyapunovWolf,1);
LyapunovRosen = nanmean(LyapunovRosen,1);
SampleEntropy = nanmean(SE,1);
ResultStruct.LyapunovWolf_V = LyapunovWolf(1);
ResultStruct.LyapunovWolf_ML = LyapunovWolf(2);
ResultStruct.LyapunovWolf_AP = LyapunovWolf(3);
ResultStruct.LyapunovRosen_V = LyapunovRosen(1);
ResultStruct.LyapunovRosen_ML = LyapunovRosen(2);
ResultStruct.LyapunovRosen_AP = LyapunovRosen(3);
ResultStruct.SampleEntropy_V = SampleEntropy(1);
ResultStruct.SampleEntropy_ML = SampleEntropy(2);
ResultStruct.SampleEntropy_AP = SampleEntropy(3);
if isfield(ResultStruct,'StrideFrequency')
LyapunovPerStrideWolf = LyapunovWolf/ResultStruct.StrideFrequency;
LyapunovPerStrideRosen = LyapunovRosen/ResultStruct.StrideFrequency;
end
ResultStruct.LyapunovPerStrideWolf_V = LyapunovPerStrideWolf(1);
ResultStruct.LyapunovPerStrideWolf_ML = LyapunovPerStrideWolf(2);
ResultStruct.LyapunovPerStrideWolf_AP = LyapunovPerStrideWolf(3);
ResultStruct.LyapunovPerStrideRosen_V = LyapunovPerStrideRosen(1);
ResultStruct.LyapunovPerStrideRosen_ML = LyapunovPerStrideRosen(2);
ResultStruct.LyapunovPerStrideRosen_AP = LyapunovPerStrideRosen(3);
end

48
CalculateStrideParametersFunc.m Normal file
View File

@ -0,0 +1,48 @@
function [ResultStruct] = CalculateStrideParametersFunc(dataAccCut_filt,FS,ApplyRemoveSteps,dataAcc,dataAcc_filt,StrideTimeRange)
%% Calculate stride parameters
ResultStruct = struct; % create empty struct
% Run function AutoCorrStrides, Outcomeparameters: StrideRegularity,RelativeStrideVariability,StrideTimeSamples,StrideTime
[ResultStruct] = AutocorrStrides(dataAccCut_filt,FS, StrideTimeRange,ResultStruct);
StrideTimeSamples = ResultStruct.StrideTimeSamples; % needed as input for other functions
% Calculate Step symmetry --> method 1
ij = 1;
dirSymm = [1,3]; % Gait Synmmetry is only informative in AP/V direction: See Tura A, Raggi M, Rocchi L, Cutti AG, Chiari L: Gait symmetry and regularity in transfemoral amputees assessed by trunk accelerations. J Neuroeng Rehabil 2010, 7:4.
for jk=1:length(dirSymm)
[C, lags] = AutocorrRegSymmSteps(dataAccCut_filt(:,dirSymm(jk)));
[Ad,p] = findpeaks(C,'MinPeakProminence',0.2, 'MinPeakHeight', 0.2);
if size(Ad,1) > 1
Ad1 = Ad(1);
Ad2 = Ad(2);
GaitSymm(:,ij) = abs((Ad1-Ad2)/mean([Ad1+Ad2]))*100;
else
GaitSymm(:,ij) = NaN;
end
ij = ij +1;
end
% Save outcome in struct;
ResultStruct.GaitSymm_V = GaitSymm(1);
ResultStruct.GaitSymm_AP = GaitSymm(2);
% Calculate Step symmetry --> method 2
[PksAndLocsCorrected] = StepcountFunc(dataAccCut_filt,StrideTimeSamples,FS);
LocsSteps = PksAndLocsCorrected(2:2:end,2);
if rem(size(LocsSteps,1),2) == 0; % is number of steps is even
LocsSteps2 = LocsSteps(1:2:end);
else
LocsSteps2 = LocsSteps(3:2:end);
end
LocsSteps1 = LocsSteps(2:2:end);
DiffLocs2 = diff(LocsSteps2);
DiffLocs1 = diff(LocsSteps1);
StepTime2 = DiffLocs2(1:end-1)/FS; % leave last one out because it is higher
StepTime1 = DiffLocs1(1:end-1)/FS;
SI = abs((2*(StepTime2-StepTime1))./(StepTime2+StepTime1))*100;
ResultStruct.GaitSymmIndex = nanmean(SI);
end

18
FilterandRealignFunc.m Normal file
View File

@ -0,0 +1,18 @@
function [dataAcc, dataAcc_filt] = FilterandRealignFunc(inputData,FS,ApplyRealignment)
%% Filter and Realign Accdata
% Apply Realignment & Filter data
if ApplyRealignment % apply relignment as described in Rispens S, Pijnappels M, van Schooten K, Beek PJ, Daffertshofer A, van Die?n JH (2014).
data = inputData(:, [3,2,4]); % reorder data to 1 = V; 2= ML, 3 = AP%
% Consistency of gait characteristics as determined from acceleration data collected at different trunk locations. Gait Posture 2014;40(1):187-92.
[RealignedAcc, ~] = RealignSensorSignalHRAmp(data, FS);
dataAcc = RealignedAcc;
[B,A] = butter(2,20/(FS/2),'low');
dataAcc_filt = filtfilt(B,A,dataAcc);
else % we asume tat data is already reorderd to 1 = V; 2= ML, 3 = AP in an earlier stage;
[B,A] = butter(2,20/(FS/2),'low');
dataAcc = inputData;
dataAcc_filt = filtfilt(B,A,dataAcc);
end

210
GaitOutcomesTrunkAccFuncIH.m Normal file
View File

@ -0,0 +1,210 @@
function [ResultStruct] = GaitOutcomesTrunkAccFuncIH(inputData,FS,LegLength,WindowLen,ApplyRealignment,ApplyRemoveSteps)
% DESCRIPTON: Trunk analysis of Iphone data without the need for step detection
% CL Nov 2019
% Adapted IH feb-april 2020
% koloms data of smartphone
% 1st column is time data;
% 2nd column is X, medio-lateral: + left, - right
% 3rd column is Y, vertical: + downwards, - upwards
% 4th column is Z, anterior- posterior : + forwards, - backwards
%% Input Trunk accelerations during locomotion in VT, ML, AP direction
% InputData: Acceleration signal with time and accelerations in VT,ML and
% AP direction.
% FS: sample frequency of the Accdata
% LegLength: length of the leg of the participant in m;
%% Output
% ResultStruct: structure coninting all outcome measured calculated
% Spectral parameters, spatiotemporal gait parameters, non-linear
% parameters
% fields and subfields: include the multiple measurements of a subject
%% Literature
% Richman & Moorman, 2000; [ sample entropy]
% Bisi & Stagni Gait & Posture 2016, 47 (6) 37-42
% Kavagnah et al., Eur J Appl Physiol 2005 94: 468?475; Human Movement Science 24(2005) 574?587 [ synchrony]
% Moe-Nilsen J Biomech 2004 37, 121-126 [ autorcorrelation step regularity and symmetry
% Kobsar et al. Gait & Posture 2014 39, 553?557 [ synchrony ]
% Rispen et al; Gait & Posture 2014, 40, 187 - 192 [realignment axes]
% Zijlstra & HofGait & Posture 2003 18,2, 1-10 [spatiotemporal gait variables]
% Lamoth et al, 2002 [index of harmonicity]
% Costa et al. 2003 Physica A 330 (2003) 53–60 [ multiscale entropy]
% Cignetti F, Decker LM, Stergiou N. Ann Biomed Eng. 2012
% May;40(5):1122-30. doi: 10.1007/s10439-011-0474-3. Epub 2011 Nov 25. [
% Wofl vs. Rosenstein Lyapunov]
%% Settings
Gr = 9.81; % Gravity acceleration, multiplication factor for accelerations
StrideFreqEstimate = 1.00; % Used to set search for stride frequency from 0.5*StrideFreqEstimate until 2*StrideFreqEstimate
StrideTimeRange = [0.2 4.0]; % Range to search for stride time (seconds)
IgnoreMinMaxStrides = 0.10; % Number or percentage of highest&lowest values ignored for improved variability estimation
N_Harm = 12; % Number of harmonics used for harmonic ratio, index of harmonicity and phase fluctuation
LowFrequentPowerThresholds = ...
[0.7 1.4]; % Threshold frequencies for estimation of low-frequent power percentages
Lyap_m = 7; % Embedding dimension (used in Lyapunov estimations)
Lyap_FitWinLen = round(60/100*FS); % Fitting window length (used in Lyapunov estimations Rosenstein's method)
Sen_m = 5; % Dimension, the length of the subseries to be matched (used in sample entropy estimation)
Sen_r = 0.3; % Tolerance, the maximum distance between two samples to qualify as match, relative to std of DataIn (used in sample entropy estimation)
NStartEnd = [100];
M = 5; % maximum template length
ResultStruct = struct();
%% Filter and Realign Accdata
% Apply Realignment & Filter data
if ApplyRealignment % apply relignment as described in Rispens S, Pijnappels M, van Schooten K, Beek PJ, Daffertshofer A, van Die?n JH (2014).
data = inputData(:, [3,2,4]); % reorder data to 1 = V; 2= ML, 3 = AP%
% Consistency of gait characteristics as determined from acceleration data collected at different trunk locations. Gait Posture 2014;40(1):187-92.
[RealignedAcc, ~] = RealignSensorSignalHRAmp(data, FS);
dataAcc = RealignedAcc;
[B,A] = butter(2,20/(FS/2),'low');
dataAcc_filt = filtfilt(B,A,dataAcc);
else % we asume tat data is already reorderd to 1 = V; 2= ML, 3 = AP in an earlier stage;
[B,A] = butter(2,20/(FS/2),'low');
dataAcc = inputData;
dataAcc_filt = filtfilt(B,A,dataAcc);
end
%% Step dectection
% Determines the number of steps in the signal so that the first 30 and last 30 steps in the signal can be removed
if ApplyRemoveSteps
% In order to run the step detection script we first need to run an autocorrelation function;
[ResultStruct] = AutocorrStrides(dataAcc_filt,FS, StrideTimeRange,ResultStruct);
% StrideTimeSamples is needed as an input for the stepcountFunc;
StrideTimeSamples = ResultStruct.StrideTimeSamples;
% Calculate the number of steps;
[PksAndLocsCorrected] = StepcountFunc(dataAcc_filt,StrideTimeSamples,FS);
% This function selects steps based on negative and positive values.
% However to determine the steps correctly we only need one of these;
LocsSteps = PksAndLocsCorrected(1:2:end,2);
%% Cut data & remove currents results
% Remove 20 steps in the beginning and end of data
dataAccCut = dataAcc(LocsSteps(31):LocsSteps(end-30),:);
dataAccCut_filt = dataAcc_filt(LocsSteps(31):LocsSteps(end-30),:);
% Clear currently saved results from Autocorrelation Analysis
clear ResultStruct;
clear PksAndLocsCorrected;
clear LocsSteps;
else;
dataAccCut = dataAcc;
dataAccCut_filt = dataAcc_filt;
end
%% Calculate stride parameters
ResultStruct = struct; % create empty struct
% Run function AutoCorrStrides, Outcomeparameters: StrideRegularity,RelativeStrideVariability,StrideTimeSamples,StrideTime
[ResultStruct] = AutocorrStrides(dataAccCut_filt,FS, StrideTimeRange,ResultStruct);
StrideTimeSamples = ResultStruct.StrideTimeSamples; % needed as input for other functions
% Calculate Step symmetry --> method 1
ij = 1;
dirSymm = [1,3]; % Gait Synmmetry is only informative in AP/V direction: See Tura A, Raggi M, Rocchi L, Cutti AG, Chiari L: Gait symmetry and regularity in transfemoral amputees assessed by trunk accelerations. J Neuroeng Rehabil 2010, 7:4.
for jk=1:length(dirSymm)
[C, lags] = AutocorrRegSymmSteps(dataAccCut_filt(:,dirSymm(jk)));
[Ad,p] = findpeaks(C,'MinPeakProminence',0.2, 'MinPeakHeight', 0.2);
if size(Ad,1) > 1
Ad1 = Ad(1);
Ad2 = Ad(2);
GaitSymm(:,ij) = abs((Ad1-Ad2)/mean([Ad1+Ad2]))*100;
else
GaitSymm(:,ij) = NaN;
end
ij = ij +1;
end
% Save outcome in struct;
ResultStruct.GaitSymm_V = GaitSymm(1);
ResultStruct.GaitSymm_AP = GaitSymm(2);
% Calculate Step symmetry --> method 2
[PksAndLocsCorrected] = StepcountFunc(dataAccCut_filt,StrideTimeSamples,FS);
LocsSteps = PksAndLocsCorrected(2:2:end,2);
if rem(size(LocsSteps,1),2) == 0; % is number of steps is even
LocsSteps2 = LocsSteps(1:2:end);
else
LocsSteps2 = LocsSteps(3:2:end);
end
LocsSteps1 = LocsSteps(2:2:end);
DiffLocs2 = diff(LocsSteps2);
DiffLocs1 = diff(LocsSteps1);
StepTime2 = DiffLocs2(1:end-1)/FS; % leave last one out because it is higher
StepTime1 = DiffLocs1(1:end-1)/FS;
SI = abs((2*(StepTime2-StepTime1))./(StepTime2+StepTime1))*100;
ResultStruct.GaitSymmIndex = nanmean(SI);
%% Calculate spatiotemporal stride parameters
% Measures from height variation by double integration of VT accelerations and high-pass filtering
[ResultStruct] = SpatioTemporalGaitParameters(dataAccCut_filt,StrideTimeSamples,ApplyRealignment,LegLength,FS,IgnoreMinMaxStrides,ResultStruct);
%% Measures derived from spectral analysis
AccVectorLen = sqrt(sum(dataAccCut_filt(:,1:3).^2,2));
[ResultStruct] = SpectralAnalysisGaitfunc(dataAccCut_filt,WindowLen,FS,N_Harm,LowFrequentPowerThresholds,AccVectorLen,ResultStruct);
%% Calculation non-linear parameters;
% cut into windows of size WindowLen
N_Windows = floor(size(dataAccCut,1)/WindowLen);
N_SkipBegin = ceil((size(dataAccCut,1)-N_Windows*WindowLen)/2);
LyapunovWolf = nan(N_Windows,3);
LyapunovRosen = nan(N_Windows,3);
SE= nan(N_Windows,3);
for WinNr = 1:N_Windows;
AccWin = dataAccCut(N_SkipBegin+(WinNr-1)*WindowLen+(1:WindowLen),:);
for j=1:3
[LyapunovWolf(WinNr,j),~] = CalcMaxLyapWolfFixedEvolv(AccWin(:,j),FS,struct('m',Lyap_m));
[LyapunovRosen(WinNr,j),outpo] = CalcMaxLyapConvGait(AccWin(:,j),FS,struct('m',Lyap_m,'FitWinLen',Lyap_FitWinLen));
[SE(WinNr,j)] = funcSampleEntropy(AccWin(:,j), Sen_m, Sen_r);
% no correction for FS; SE does increase with higher FS but effect is considered negligible as range is small (98-104HZ). Might consider updating r to account for larger ranges.
end
end
LyapunovWolf = nanmean(LyapunovWolf,1);
LyapunovRosen = nanmean(LyapunovRosen,1);
SampleEntropy = nanmean(SE,1);
ResultStruct.LyapunovWolf_V = LyapunovWolf(1);
ResultStruct.LyapunovWolf_ML = LyapunovWolf(2);
ResultStruct.LyapunovWolf_AP = LyapunovWolf(3);
ResultStruct.LyapunovRosen_V = LyapunovRosen(1);
ResultStruct.LyapunovRosen_ML = LyapunovRosen(2);
ResultStruct.LyapunovRosen_AP = LyapunovRosen(3);
ResultStruct.SampleEntropy_V = SampleEntropy(1);
ResultStruct.SampleEntropy_ML = SampleEntropy(2);
ResultStruct.SampleEntropy_AP = SampleEntropy(3);
if isfield(ResultStruct,'StrideFrequency')
LyapunovPerStrideWolf = LyapunovWolf/ResultStruct.StrideFrequency;
LyapunovPerStrideRosen = LyapunovRosen/ResultStruct.StrideFrequency;
end
ResultStruct.LyapunovPerStrideWolf_V = LyapunovPerStrideWolf(1);
ResultStruct.LyapunovPerStrideWolf_ML = LyapunovPerStrideWolf(2);
ResultStruct.LyapunovPerStrideWolf_AP = LyapunovPerStrideWolf(3);
ResultStruct.LyapunovPerStrideRosen_V = LyapunovPerStrideRosen(1);
ResultStruct.LyapunovPerStrideRosen_ML = LyapunovPerStrideRosen(2);
ResultStruct.LyapunovPerStrideRosen_AP = LyapunovPerStrideRosen(3);
end

74
GaitVariabilityAnalysisIH.m Normal file
View File

@ -0,0 +1,74 @@
% Gait Variability Analysis
% Script created for BAP students 2020
% Iris Hagoort
% April 2020
% Input: needs mat file which contains all raw accelerometer data
% Input: needs excel file containing the participant information including
% leg length.
%% Clear and close;
clear;
close all;
%% Load data
load('Phyphoxdata.mat'); % loads accelerometer data, is stored in struct with name AccData
load('ExcelInfo.mat');
Participants = fields(AccData);
%% Settings
FS = 100; % sample frequency
LegLengths = excel.data.GeneralInformation(:,5); % leglength info is in 5th column
LegLengthsM = LegLengths/100; % convert to m
%% Calculate parameters;
for i = 1: length(Participants);
tic;
LegLength = LegLengthsM(i);
WalkingConditions = fields(AccData.([char(Participants(i))]));
for j = 1: length(WalkingConditions);
if strcmp(char(WalkingConditions(j)),'Treadmill')
SubConditions = fieldnames(AccData.([char(Participants(i))]).([char(WalkingConditions(j))]));
for k = 1: length(SubConditions);
inputData = AccData.([char(Participants(i))]).([char(WalkingConditions(j))]).([char(SubConditions(k))]);
WindowLength = FS*10;
ApplyRealignment = true;
ApplyRemoveSteps = true;
[ResultStruct] = GaitOutcomesTrunkAccFuncIH(inputData,FS,LegLength,WindowLen,ApplyRealignment,ApplyRemoveSteps);
OutcomesAcc.([char(Participants(i))]).([char(WalkingConditions(j))]).([char(SubConditions(k))]) = ResultStruct;
end
elseif strcmp(char(WalkingConditions(j)),'Balance') || strcmp(char(WalkingConditions(j)),'TwoMWT')
disp('Files are not used for current analysis');
elseif strcmp(char(WalkingConditions(j)),'InsideStraight')
inputData = AccData.([char(Participants(i))]).([char(WalkingConditions(j))]);
ApplyRealignment = true;
ApplyRemoveSteps = false; % don't remove steps for the straight conditions
% function specific for the walking conditions containing a lot
% of turns
[ResultStruct] = GaitVariabilityAnalysisIH_WithoutTurns(inputData,FS,LegLength,ApplyRealignment,ApplyRemoveSteps);
OutcomesAcc.([char(Participants(i))]).([char(WalkingConditions(j))]) = ResultStruct;
else
inputData = AccData.([char(Participants(i))]).([char(WalkingConditions(j))]);
ApplyRealignment = true;
ApplyRemoveSteps = true;
WindowLen = FS*10;
[ResultStruct] = GaitOutcomesTrunkAccFuncIH(inputData,FS,LegLength,WindowLen,ApplyRealignment,ApplyRemoveSteps)
OutcomesAcc.([char(Participants(i))]).([char(WalkingConditions(j))]) = ResultStruct;
end
end
toc;
end
% Save struct as .mat file
save('GaitVarOutcomes30pril.mat', 'OutcomesAcc');

108
GaitVariabilityAnalysisIH_WithoutTurns.m Normal file
View File

@ -0,0 +1,108 @@
function [ResultStruct] = GaitVariabilityAnalysisIH_WithoutTurns(inputData,FS,LegLength,ApplyRealignment,ApplyRemoveSteps);
% SCRIPT FOR ANAlysis straight parts
% NOG GOEDE BESCHRIJVING TOEVOEGEN.
%% Realign data
data = inputData(:, [3,2,4]); % reorder data to 1 = V; 2= ML, 3 = AP
%Realign sensor data to VT-ML-AP frame
if ApplyRealignment % apply relignment as described in Rispens S, Pijnappels M, van Schooten K, Beek PJ, Daffertshofer A, van Die?n JH (2014).
% Consistency of gait characteristics as determined from acceleration data collected at different trunk locations. Gait Posture 2014;40(1):187-92.
[RealignedAcc, ~] = RealignSensorSignalHRAmp(data, FS);
dataAcc = RealignedAcc;
end
%% Filter data strongly & Determine location of steps
% Filter data
[B,A] = butter(2,3/(FS/2),'low'); % Filters data very strongly which is needed to determine turns correctly
dataStepDetection = filtfilt(B,A,dataAcc);
% Determine steps;
%%%%%%% HIER MISSCHIEN ALTERNATIEF VOOR VAN RISPENS %%%%%%%%%%%%%
% Explanation of method: https://nl.mathworks.com/help/supportpkg/beagleboneblue/ref/counting-steps-using-beagleboneblue-hardware-example.html
% From website: To convert the XYZ acceleration vectors at each point in time into scalar values,
% calculate the magnitude of each vector. This way, you can detect large changes in overall acceleration,
% such as steps taken while walking, regardless of device orientation.
magfilt = sqrt(sum((dataStepDetection(:,1).^2) + (dataStepDetection(:,2).^2) + (dataStepDetection(:,3).^2), 2));
magNoGfilt = magfilt - mean(magfilt);
minPeakHeight2 = std(magNoGfilt);
[pks, locs] = findpeaks(magNoGfilt, 'MINPEAKHEIGHT', minPeakHeight2); % for step detection
numStepsOption2_filt = numel(pks); % counts number of steps;
%%%%%%%%%%%%%%%%%%%%%%%% TOT HIER %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%% Determine locations of turns;
diffLocs = diff(locs); % calculates difference in step location
avg_diffLocs = mean(diffLocs); % average distance between steps
std_diffLocs = std(diffLocs); % standard deviation of distance between steps
figure;
findpeaks(diffLocs, 'MINPEAKHEIGHT', avg_diffLocs, 'MINPEAKDISTANCE',5); % these values have been chosen based on visual inspection of the signal
line([1 length(diffLocs)],[avg_diffLocs avg_diffLocs])
[pks_diffLocs, locs_diffLocs] = findpeaks(diffLocs, 'MINPEAKHEIGHT', avg_diffLocs,'MINPEAKDISTANCE',5);
locsTurns = [locs(locs_diffLocs), locs(locs_diffLocs+1)];
%% Visualizing turns
% Duplying signal + visualing
% to make second signal with the locations of the turns filled with NaN, so
% that both signals can be plotted above each other in a different colour
magNoGfilt_copy = magNoGfilt;
for k = 1: size(locsTurns,1);
magNoGfilt_copy(locsTurns(k,1):locsTurns(k,2)) = NaN;
end
% visualising signal;
figure;
subplot(2,1,1)
hold on;
plot(magNoGfilt,'b')
plot(magNoGfilt_copy, 'r');
title('Inside Straight: Filtered data with turns highlighted in blue')
hold off;
%% Calculation
% VRAAG LAURENS zie blauwe blaadje
startPos = 1;
for i = 1: size(locsTurns,1);
endPos = locsTurns(i,1)-1;
inputData = dataAcc(startPos:endPos,:);
WindowLen = size(inputData,1);
ApplyRealignment = false;
[ResultStruct] = GaitOutcomesTrunkAccFuncIH(inputData,FS,LegLength,WindowLen,ApplyRealignment,ApplyRemoveSteps); % Naam van deze moet nog aangepast.
if i ==1 % only the firs time
Parameters = fieldnames(ResultStruct);
NrParameters = length(Parameters);
end
for j = 1:NrParameters % only works if for every bin we get the same outcomes (which is the case in this script)
DataStraight.([char(Parameters(j))])(i) = ResultStruct.([char(Parameters(j))]);
end
startPos = locsTurns(i,2)+1;
end
clear ResultStruct;
% Calculate mean over the bins without turns to get 1 outcome value per parameter for inside
% straight;
for j = 1:NrParameters;
ResultStruct.([char(Parameters(j))]) = nanmean(DataStraight.([char(Parameters(j))]))
end
end

1709
GaitVariabilityAnalysisLD.html Normal file
View File

File diff suppressed because one or more lines are too long

BIN
GaitVariabilityAnalysisLD.mlx Normal file
View File

Binary file not shown.

1944
GaitVariabilityAnalysisLD.tex Normal file
View File

File diff suppressed because it is too large Load Diff

103
GaitVariabilityAnalysisLD_v1.m Normal file
View File

@ -0,0 +1,103 @@
%% Gait Variability Analysis CLBP
% Gait Variability Analysis
% Script created for MAP 2020-2021
% adapted from Claudine Lamoth and Iris Hagoort
% version1 October 2020
% Input: needs mat file which contains all raw accelerometer data
% Input: needs excel file containing the participant information including
% leg length.
%% Clear and close;
clear;
close all;
%% Load data;
% Select 1 trial.
% For loop to import all data will be used at a later stage
[FNaam,FilePad] = uigetfile('*.xls','Load phyphox data...');
filename =[FilePad FNaam];
PhyphoxData = xlsread(filename)
%load('Phyphoxdata.mat'); % loads accelerometer data, is stored in struct with name AccData
%load('ExcelInfo.mat');
%Participants = fields(AccData);
%% Settings;
%adapted from GaitOutcomesTrunkAccFuncIH
LegLength = 98; % LegLength info not available!
FS = 100; % Sample frequency
Gr = 9.81; % Gravity acceleration, multiplication factor for accelerations
StrideFreqEstimate = 1.00; % Used to set search for stride frequency from 0.5*StrideFreqEstimate until 2*StrideFreqEstimate
StrideTimeRange = [0.2 4.0]; % Range to search for stride time (seconds)
IgnoreMinMaxStrides = 0.10; % Number or percentage of highest&lowest values ignored for improved variability estimation
N_Harm = 12; % Number of harmonics used for harmonic ratio, index of harmonicity and phase fluctuation
LowFrequentPowerThresholds = ...
[0.7 1.4]; % Threshold frequencies for estimation of low-frequent power percentages
Lyap_m = 7; % Embedding dimension (used in Lyapunov estimations)
Lyap_FitWinLen = round(60/100*FS); % Fitting window length (used in Lyapunov estimations Rosenstein's method)
Sen_m = 5; % Dimension, the length of the subseries to be matched (used in sample entropy estimation)
Sen_r = 0.3; % Tolerance, the maximum distance between two samples to qualify as match, relative to std of DataIn (used in sample entropy estimation)
NStartEnd = [100];
M = 5; % Maximum template length
ResultStruct = struct(); % Empty struct
inputData = (PhyphoxData(:,[1 2 3 4])); % matrix with accelerometer data
ApplyRealignment = true;
ApplyRemoveSteps = true;
WindowLen = FS*10;
%% Filter and Realign Accdata
% dataAcc depends on ApplyRealignment = true/false
% dataAcc_filt (low pass Butterworth Filter + zerophase filtering
[dataAcc, dataAcc_filt] = FilterandRealignFunc(inputData,FS,ApplyRealignment);
%% Step dectection
% Determines the number of steps in the signal so that the first 30 and last 30 steps in the signal can be removed
% StrideTimeSamples is needed for calculation stride parameters!
[dataAccCut,dataAccCut_filt,StrideTimeSamples] = StepDetectionFunc(FS,ApplyRemoveSteps,dataAcc,dataAcc_filt,StrideTimeRange);
%% Calculate Stride Parameters
% Outcomeparameters: StrideRegularity,RelativeStrideVariability,StrideTimeSamples,StrideTime
[ResultStruct] = CalculateStrideParametersFunc(dataAccCut_filt,FS,ApplyRemoveSteps,dataAcc,dataAcc_filt,StrideTimeRange);
%% Calculate spatiotemporal stride parameters
% Measures from height variation by double integration of VT accelerations and high-pass filtering
% StepLengthMean; Distance; WalkingSpeedMean; StrideTimeVariability; StrideSpeedVariability;
% StrideLengthVariability; StrideTimeVariabilityOmitOutlier; StrideSpeedVariabilityOmitOutlier; StrideLengthVariabilityOmitOutlier;
[ResultStruct] = SpatioTemporalGaitParameters(dataAccCut_filt,StrideTimeSamples,ApplyRealignment,LegLength,FS,IgnoreMinMaxStrides,ResultStruct);
%% Measures derived from spectral analysis
% IndexHarmonicity_V/ML/AP/ALL ; HarmonicRatio_V/ML/AP ; HarmonicRatioP_V/ML/AP ; FrequencyVariability_V/ML/AP; Stride Frequency
AccVectorLen = sqrt(sum(dataAccCut_filt(:,1:3).^2,2));
[ResultStruct] = SpectralAnalysisGaitfunc(dataAccCut_filt,WindowLen,FS,N_Harm,LowFrequentPowerThresholds,AccVectorLen,ResultStruct);
%% calculate non-linear parameters
% Outcomeparameters: Sample Entropy, Lyapunov exponents
[ResultStruct] = CalculateNonLinearParametersFunc(ResultStruct,dataAccCut,WindowLen,FS,Lyap_m,Lyap_FitWinLen,Sen_m,Sen_r);
% Save struct as .mat file
% save('GaitVarOutcomesParticipantX.mat', 'OutcomesAcc');
%% AggregateFunction (seperate analysis per minute);
% see AggregateEpisodeValues.m
%
%

95
GaitVariabilityStructToExcelIH.m Normal file
View File

@ -0,0 +1,95 @@
%% Description file
%% Clear and close
clear;
close all;
%% Load data;
load('GaitVarOutcomes30april.mat')
Participants = fields(OutcomesAcc);
%% Settings;
counter = 1;
run = 1;
ParticipantNr = 0;
ConditionNames = {'TwoMWT','InsidePath','InsideStraight','Outside','Treadmill_Comfortable','Treadmill_Condition1','Treadmill_Condition2','Treadmill_Condition3','Treadmill_Condition4','Treadmill_Condition5','Treadmill_Condition6','Treadmill_Condition7','Treadmill_Condition8'};
%% Reorder data;
for i = 1: length(Participants);
% Participant Nr;
ParticipantNr = ParticipantNr + 1;
% Group Nr;
if contains(Participants(i),'Y');
Group = 1;
else
Group = 2;
end
WalkingConditions = fields(OutcomesAcc.([char(Participants(i))]));
for j = 1: length(WalkingConditions);
if strcmp(char(WalkingConditions(j)),'Treadmill')
SubConditions = fieldnames(OutcomesAcc.([char(Participants(i))]).([char(WalkingConditions(j))]));
for k = 1: length(SubConditions);
NameTreadmill = [char(WalkingConditions(j)),'_',char(SubConditions(k))];
ConditionNr = find(strcmp(ConditionNames, NameTreadmill));
Parameters = fieldnames(OutcomesAcc.([char(Participants(i))]).([char(WalkingConditions(j))]).([char(SubConditions(k))]));
for l = 1: length(Parameters);
Data(counter+1,1) = {([char(Participants(i))])};
Data(counter+1,2) = {ParticipantNr};
Data(counter+1,3) = {Group};
Data(counter+1,4) = {([char(NameTreadmill)])};
Data(counter+1,5) = {ConditionNr};
Data(counter+1,l+5) = {OutcomesAcc.([char(Participants(i))]).([char(WalkingConditions(j))]).([char(SubConditions(k))]).([char(Parameters(l))])(1)};
end
counter = counter+1;
end
else
ConditionNr = find(strcmp(ConditionNames, WalkingConditions(j)));
Parameters = fieldnames(OutcomesAcc.([char(Participants(i))]).([char(WalkingConditions(j))]));
for l = 1: length(Parameters);
Data(counter+1,1) = {([char(Participants(i))])};
Data(counter+1,2) = {ParticipantNr};
Data(counter+1,3) = {Group};
Data(counter+1,4) = {([char(WalkingConditions(j))])};
Data(counter+1,5) = {ConditionNr};
Data(counter+1,l+5) = {OutcomesAcc.([char(Participants(i))]).([char(WalkingConditions(j))]).([char(Parameters(l))])(1)};
end
counter = counter+1;
end
if run == 1;
Data(1,1) = {'ParticipantCode'};
Data(1,2) = {'ParticipantNr'};
Data(1,3) = {'Group'};
Data(1,4) = {'ConditionName'};
Data(1,5) = {'ConditionNr'};
Data(1,6:(size(Parameters,1)+5)) = Parameters'; % First row with variable names;
run = run+1;
end
end
end
if ispc;
xlswrite('GaitVariabilityOutcomes.xls', 'Data');
elseif ismac;
filename = 'GaitVariabilityOutcomes.xlsx';
writecell(Data,filename);
else
end

83
HarmonicityFrequency.m Normal file
View File

@ -0,0 +1,83 @@
function [ResultStruct] = HarmonicityFrequency(dataAccCut_filt,P,F, StrideFrequency,dF,LowFrequentPowerThresholds,N_Harm,FS,AccVectorLen,ResultStruct)
% Add sum of power spectra (as a rotation-invariant spectrum)
P = [P,sum(P,2)];
PS = sqrt(P);
% Calculate the measures for the power per separate dimension
for i=1:size(P,2);
% Relative cumulative power and frequencies that correspond to these cumulative powers
PCumRel = cumsum(P(:,i))/sum(P(:,i));
PSCumRel = cumsum(PS(:,i))/sum(PS(:,i));
FCumRel = F+0.5*dF;
% Derive relative cumulative power for threshold frequencies
Nfreqs = size(LowFrequentPowerThresholds,2);
LowFrequentPercentage(i,1:Nfreqs) = interp1(FCumRel,PCumRel,LowFrequentPowerThresholds)*100;
% Calculate relative power of first 10 harmonics, taking the power of each harmonic with a band of + and - 10% of the first
% harmonic around it
PHarm = zeros(N_Harm,1);
PSHarm = zeros(N_Harm,1);
for Harm = 1:N_Harm,
FHarmRange = (Harm+[-0.1 0.1])*StrideFrequency;
PHarm(Harm) = diff(interp1(FCumRel,PCumRel,FHarmRange));
PSHarm(Harm) = diff(interp1(FCumRel,PSCumRel,FHarmRange));
end
% Derive index of harmonicity
if i == 2 % for ML we expect odd instead of even harmonics
IndexHarmonicity(i) = PHarm(1)/sum(PHarm(1:2:12));
elseif i == 4
IndexHarmonicity(i) = sum(PHarm(1:2))/sum(PHarm(1:12));
else
IndexHarmonicity(i) = PHarm(2)/sum(PHarm(2:2:12));
end
% Calculate the phase speed fluctuations
StrideFreqFluctuation = nan(N_Harm,1);
StrSamples = round(FS/StrideFrequency);
for h=1:N_Harm,
CutOffs = [StrideFrequency*(h-(1/3)) , StrideFrequency*(h+(1/3))]/(FS/2);
if all(CutOffs<1) % for Stride frequencies above FS/20/2, the highest harmonics are not represented in the power spectrum
[b,a] = butter(2,CutOffs);
if i==4 % Take the vector length as a rotation-invariant signal
AccFilt = filtfilt(b,a,AccVectorLen);
else
AccFilt = filtfilt(b,a,dataAccCut_filt(:,i));
end
Phase = unwrap(angle(hilbert(AccFilt)));
SmoothPhase = sgolayfilt(Phase,1,2*(floor(FS/StrideFrequency/2))+1); % This is in fact identical to a boxcar filter with linear extrapolation at the edges
StrideFreqFluctuation(h) = std(Phase(1+StrSamples:end,1)-Phase(1:end-StrSamples,1));
end
end
FrequencyVariability(i) = nansum(StrideFreqFluctuation./(1:N_Harm)'.*PHarm)/nansum(PHarm);
if i<4,
% Derive harmonic ratio (two variants)
if i == 2 % for ML we expect odd instead of even harmonics
HarmonicRatio(i) = sum(PSHarm(1:2:end-1))/sum(PSHarm(2:2:end)); % relative to summed 3d spectrum
HarmonicRatioP(i) = sum(PHarm(1:2:end-1))/sum(PHarm(2:2:end)); % relative to own spectrum
else
HarmonicRatio(i) = sum(PSHarm(2:2:end))/sum(PSHarm(1:2:end-1));
HarmonicRatioP(i) = sum(PHarm(2:2:end))/sum(PHarm(1:2:end-1));
end
end
end
ResultStruct.IndexHarmonicity_V = IndexHarmonicity(1); % higher smoother more regular patter
ResultStruct.IndexHarmonicity_ML = IndexHarmonicity(2); % higher smoother more regular patter
ResultStruct.IndexHarmonicity_AP = IndexHarmonicity(3); % higher smoother more regular patter
ResultStruct.IndexHarmonicity_All = IndexHarmonicity(4);
ResultStruct.HarmonicRatio_V = HarmonicRatio(1);
ResultStruct.HarmonicRatio_ML = HarmonicRatio(2);
ResultStruct.HarmonicRatio_AP = HarmonicRatio(3);
ResultStruct.HarmonicRatioP_V = HarmonicRatioP(1);
ResultStruct.HarmonicRatioP_ML = HarmonicRatioP(2);
ResultStruct.HarmonicRatioP_AP = HarmonicRatioP(3);
ResultStruct.FrequencyVariability_V = FrequencyVariability(1);
ResultStruct.FrequencyVariability_ML = FrequencyVariability(2);
ResultStruct.FrequencyVariability_AP = FrequencyVariability(3);
ResultStruct.StrideFrequency = StrideFrequency;

146
MutualInformationHisPro.m Normal file
View File

@ -0,0 +1,146 @@
function [mutM,cummutM,minmuttauV] = MutualInformationHisPro(xV,tauV,bV,flag)
% [mutM,cummutM,minmuttauV] = MutualInformationHisPro(xV,tauV,bV,flag)
% MUTUALINFORMATIONHISPRO computes the mutual information on the time
% series 'xV' for given delays in 'tauV'. The estimation of mutual
% information is based on 'b' partitions of equal probability at each dimension.
% A number of different 'b' can be given in the input vector 'bV'.
% According to a given flag, it can also compute the cumulative mutual
% information for each given lag, as well as the time of the first minimum
% of the mutual information.
% INPUT
% - xV : a vector for the time series
% - tauV : a vector of the delays to be evaluated for
% - bV : a vector of the number of partitions of the histogram-based
% estimate.
% - flag : if 0-> compute only mutual information,
% : if 1-> compute the mutual information, the first minimum of
% mutual information and the cumulative mutual information.
% if 2-> compute (also) the cumulative mutual information
% if 3-> compute (also) the first minimum of mutual information
% OUTPUT
% - mutM : the vector of the mutual information values s for the given
% delays.
% - cummutM : the vector of the cumulative mutual information values for
% the given delays
% - minmuttauV : the time of the first minimum of the mutual information.
%========================================================================
% <MutualInformationHisPro.m>, v 1.0 2010/02/11 22:09:14 Kugiumtzis & Tsimpiris
% This is part of the MATS-Toolkit http://eeganalysis.web.auth.gr/
%========================================================================
% Copyright (C) 2010 by Dimitris Kugiumtzis and Alkiviadis Tsimpiris
% <dkugiu@gen.auth.gr>
%========================================================================
% Version: 1.0
% LICENSE:
% This program is free software; you can redistribute it and/or modify
% it under the terms of the GNU General Public License as published by
% the Free Software Foundation; either version 3 of the License, or
% any later version.
%
% This program is distributed in the hope that it will be useful,
% but WITHOUT ANY WARRANTY; without even the implied warranty of
% MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
% GNU General Public License for more details.
%
% You should have received a copy of the GNU General Public License
% along with this program. If not, see http://www.gnu.org/licenses/>.
%=========================================================================
% Reference : D. Kugiumtzis and A. Tsimpiris, "Measures of Analysis of Time Series (MATS):
% A Matlab Toolkit for Computation of Multiple Measures on Time Series Data Bases",
% Journal of Statistical Software, in press, 2010
% Link : http://eeganalysis.web.auth.gr/
%=========================================================================
nsam = 1;
n = length(xV);
if nargin==3
flag = 1;
elseif nargin==2
flag = 1;
bV = round(sqrt(n/5));
end
if isempty(bV)
bV = round(sqrt(n/5));
end
bV(bV==0)=round(sqrt(n/5));
tauV = sort(tauV);
ntau = length(tauV);
taumax = tauV(end);
nb = length(bV);
[oxV,ixV]=sort(xV);
[tmpV,ioxV]=sort(ixV);
switch flag
case 0
% Compute only the mutual information for the given lags
mutM = NaN*ones(ntau,nb);
for ib=1:nb
b = bV(ib);
if n<2*b
break;
end
mutM(:,ib)=mutinfHisPro(xV,tauV,b,ioxV,ixV);
end % for ib
cummutM=[];
minmuttauV=[];
case 1
% Compute the mutual information for all lags up to the
% largest given lag, then compute the lag of the first minimum of
% mutual information and the cumulative mutual information for the
% given lags.
mutM = NaN*ones(ntau,nb);
cummutM = NaN*ones(ntau,nb);
minmuttauV = NaN*ones(nb,1);
miM = NaN*ones(taumax+1,nb);
for ib=1:nb
b = bV(ib);
if n<2*b
break;
end
miM(:,ib)=mutinfHisPro(xV,[0:taumax]',b,ioxV,ixV);
mutM(:,ib) = miM(tauV+1,ib);
minmuttauV(ib) = findminMutInf(miM(:,ib),nsam);
% Compute the cumulative mutual information for the given delays
for i=1:ntau
cummutM(i,ib) = sum(miM(1:tauV(i)+1,ib));
end
end % for ib
case 2
% Compute the mutual information for all lags up to the largest
% given lag and then sum up to get the cumulative mutual information
% for the given lags.
cummutM = NaN*ones(ntau,nb);
miM = NaN*ones(taumax+1,nb);
for ib=1:nb
b = bV(ib);
if n<2*b
break;
end
miM(:,ib)=mutinfHisPro(xV,[0:taumax]',b,ioxV,ixV);
% Compute the cumulative mutual information for the given delays
for i=1:ntau
cummutM(i,ib) = sum(miM(1:tauV(i)+1,ib));
end
end % for ib
mutM = [];
minmuttauV=[];
case 3
% Compute the mutual information for all lags up to the largest
% given lag and then compute the lag of the first minimum of the
% mutual information.
minmuttauV = NaN*ones(nb,1);
miM = NaN*ones(taumax+1,nb);
for ib=1:nb
b = bV(ib);
if n<2*b
break;
end
miM(:,ib)=mutinfHisPro(xV,[0:taumax]',b,ioxV,ixV);
minmuttauV(ib) = findminMutInf(miM(:,ib),nsam);
end % for ib
mutM = [];
cummutM=[];
end

BIN
Phyphoxdata.mat Normal file
View File

Binary file not shown.

275
RealignSensorSignalHRAmp.m Normal file
View File

@ -0,0 +1,275 @@
function [RealignedAcc,RotationMatrixT,Flags] = RealignSensorSignalHRAmp(Acc, FS)
% History
% 2013/05/31 SR solve problem of ML-AP rotation of appr. 90 degrees ("Maximum number of function evaluations has been exceeded")
% 2013/08/15 SR use literature HR (i.e. amplitude not power, and harmonic frequencies not sin^4 windows)
% 2013/08/16 SR correct: use abs(X) instead of sqrt(X.*X)
% 2013/09/11 SR use only relevant F (containing odd or even harmonics) and speed up determination of harmonic frequencies
%% Define VT direction (RVT) as direction of mean acceleration
MeanAcc = mean(Acc);
RVT = MeanAcc'/norm(MeanAcc);
RMLGuess = cross([0;0;1],RVT); RMLGuess = RMLGuess/norm(RMLGuess);
RAPGuess = cross(RVT,RMLGuess); RAPGuess = RAPGuess/norm(RAPGuess);
%% Estimate stride frequency
StrideFrequency = StrideFrequencyFrom3dAcc(Acc, FS);
AccDetrend = detrend(Acc,'constant');
N = size(AccDetrend,1);
%% calculate complex DFT
F = FS*(0:(N-1))'/N;
dF = FS/N;
FHarmRange = [(1:20)'-0.1, (1:20)'+0.1]*StrideFrequency;
EvenRanges = FHarmRange(2:2:end,:);
OddRanges = FHarmRange(1:2:end,:);
EvenHarmonics = zeros(size(F));
for j=1:size(EvenRanges,1)
IXRange = [EvenRanges(j,1)/dF, EvenRanges(j,2)/dF]+1;
IXRangeRound = min(N,round(IXRange));
EvenHarmonics(IXRangeRound(1):IXRangeRound(2)) = EvenHarmonics(IXRangeRound(1):IXRangeRound(2)) + 1;
EvenHarmonics(IXRangeRound(1)) = EvenHarmonics(IXRangeRound(1)) - (IXRange(1)-IXRangeRound(1)+0.5);
EvenHarmonics(IXRangeRound(2)) = EvenHarmonics(IXRangeRound(2)) - (IXRangeRound(2)-IXRange(2)+0.5);
end
OddHarmonics = zeros(size(F));
for j=1:size(OddRanges,1)
IXRange = [OddRanges(j,1)/dF, OddRanges(j,2)/dF]+1;
IXRangeRound = min(N,round(IXRange));
OddHarmonics(IXRangeRound(1):IXRangeRound(2)) = OddHarmonics(IXRangeRound(1):IXRangeRound(2)) + 1;
OddHarmonics(IXRangeRound(1)) = OddHarmonics(IXRangeRound(1)) - (IXRange(1)-IXRangeRound(1)+0.5);
OddHarmonics(IXRangeRound(2)) = OddHarmonics(IXRangeRound(2)) - (IXRangeRound(2)-IXRange(2)+0.5);
end
%% select only frequencies that contain odd or even harmonics
RelevantF = OddHarmonics~=0 | EvenHarmonics~=0;
OddHarmonics = OddHarmonics(RelevantF);
EvenHarmonics = EvenHarmonics(RelevantF);
DFT = fft(AccDetrend,N,1);
DFT = DFT(RelevantF,:);
DFTMLG = DFT*RMLGuess;
DFTAPG = DFT*RAPGuess;
%% Initial estimation of ML-AP realignment (educated guess)
% We start of by finding the directions ML = MLguess + x*APguess for which
% the 'harmonic ratio' for the ML directions is maximal or minimal, and we
% do this by varying x in still the power variant of HR instead of amplitude:
HRML = @(x) real(((OddHarmonics'.*(DFTMLG'+x*DFTAPG'))*(DFTMLG+x*DFTAPG))...
/((EvenHarmonics'.*(DFTMLG'+x*DFTAPG'))*(DFTMLG+x*DFTAPG)));
% or by substititutions
a = real((OddHarmonics.*DFTAPG)'*DFTAPG);
b = real((OddHarmonics.*DFTMLG)'*DFTAPG + (OddHarmonics.*DFTAPG)'*DFTMLG);
c = real((OddHarmonics.*DFTMLG)'*DFTMLG);
d = real((EvenHarmonics.*DFTAPG)'*DFTAPG);
e = real((EvenHarmonics.*DFTMLG)'*DFTAPG + (EvenHarmonics.*DFTAPG)'*DFTMLG);
f = real((EvenHarmonics.*DFTMLG)'*DFTMLG);
% and by setting d/dx(HRML(x))==0 we get the quadratic formula
aq = a*e-b*d;
bq = 2*a*f-2*c*d;
cq = b*f-c*e;
% DT = bq^2-4*aq*cq = 4*(a*f-*c*d)*(a*f-c*d) - 4*(a*e-b*d)*(b*f-c*e)
% = 4aaff +4ccdd -8acdf -4abef -4bcde +4acee + 4bbdf
% = 4aaff +4ccdd -4abef -4bcde +4ac(ee-df) +4(bb-ac)df
x = (-bq +[1 -1]*sqrt(bq.^2-4*aq*cq))/(2*aq);
% with hopefully two solutions for x.
% from here on use amplitude HR:
HRML = @(x) real((OddHarmonics'*abs(DFTMLG+x*DFTAPG))...
/(EvenHarmonics'*abs(DFTMLG+x*DFTAPG)));
% now we take the x for which the HR is maximal, and the x rotated 90
% degrees fo which it is minimal, and take the mean angle between them
HR1 = HRML(x(1));
HR2 = HRML(x(2));
if HR1 > HR2
thetas = atan([x(1),-1/x(2)]);
else
thetas = atan([-1/x(1),x(2)]);
end
theta = mean(thetas);
xeg = tan(theta);
if abs(diff(thetas)) > pi/2 % thetas are more than 90 degrees apart, and the 'mean of the two axes' should be rotated 90 degrees
xeg = -1/xeg;
end
%% Numerical search for best harmonic ratios product
% Now we want to improve the solution, since the two above directions are
% not necessarily orthogonal. Thus we want to find the orthogonal
% directions ML = MLguess + x*APguess, and AP = APguess - x*MLguess,
% for which the product of 'harmonic ratios' for the ML and AP directions
% is maximal, and we do this by varying x, starting with the educated guess
% above using power instead of amplitude:
% HRML = ((OddHarmonics'.*(DFTMLG'+x*DFTAPG'))*(DFTMLG+x*DFTAPG))...
% /((EvenHarmonics'.*(DFTMLG'+x*DFTAPG'))*(DFTMLG+x*DFTAPG));
% HRAP = ((EvenHarmonics'.*(-x*DFTMLG'+DFTAPG'))*(-x*DFTMLG+DFTAPG))...
% /((OddHarmonics'.*(-x*DFTMLG'+DFTAPG'))*(-x*DFTMLG+DFTAPG));
% and
% HR_Prod = HRML*HRAP = ...
% Now use amplitude
Minus_HR_Prod = @(x) -real(...
(OddHarmonics'*abs(DFTMLG+x*DFTAPG))...
/(EvenHarmonics'*abs(DFTMLG+x*DFTAPG))...
*(EvenHarmonics'*abs(-x*DFTMLG+DFTAPG))...
/(OddHarmonics'*abs(-x*DFTMLG+DFTAPG)));
[xopt1,Minus_HR_Prod_opt1,exitflag1]=fminsearch(Minus_HR_Prod,xeg);
if exitflag1 == 0 && abs(xopt1) > 100 % optimum for theta beyond +- pi/2
[xopt1a,Minus_HR_Prod_opt1a,exitflag1a]=fminsearch(Minus_HR_Prod,-xopt1);
if ~(exitflag1a == 0 && abs(xopt1a) > 100)
xopt1org = xopt1;
xopt1 = xopt1a;
Minus_HR_Prod_opt1 = Minus_HR_Prod_opt1a;
end
end
options = optimset('LargeScale','off','GradObj','off','Display','notify');
[xopt2,Minus_HR_Prod_opt2,exitflag2]=fminunc(Minus_HR_Prod,xopt1,options);
if exitflag2 == 0 && abs(xopt2) > 100 % optimum for theta beyond +- pi/2
[xopt2a,Minus_HR_Prod_opt2a,exitflag2a]=fminsearch(Minus_HR_Prod,-xopt2);
if ~(exitflag2a == 0 && abs(xopt2a) > 100)
xopt2 = xopt2a;
Minus_HR_Prod_opt2 = Minus_HR_Prod_opt2a;
end
end
RML = (RMLGuess+xopt2*RAPGuess)/sqrt(1+xopt2^2);
RAP = cross(RVT,RML); RAP = RAP/norm(RAP);
RT = [RVT,RML,RAP];
if nargout > 1
RotationMatrixT = RT;
end
RealignedAcc = Acc*RT;
if nargout > 2
Flags = [exitflag1,exitflag2];
end
if nargin > 2 && plotje
Thetas = pi*(-0.49:0.01:0.49)';
Xs = tan(Thetas);
HR_Prod = nan(size(Xs));
HRMLs = nan(size(Xs));
for i=1:numel(Xs),
HR_Prod(i,1) = -Minus_HR_Prod(Xs(i));
HRMLs(i,1) = HRML(Xs(i));
end
figure();
semilogy(Thetas,[HR_Prod,HRMLs.^2]);
hold on;
semilogy(atan(x),[HR1,HR2].^2,'r.');
semilogy(atan([xopt1;xopt2]),-[Minus_HR_Prod_opt1;Minus_HR_Prod_opt2],'g.');
semilogy(atan(xeg),-Minus_HR_Prod(xeg),'kx');
end
%% attempt to solve it analytically:
% % We can rewrite HR_Prod as:
% % HR_Prod =
% % ( (x.^2)*((OddHarmonics.*DFTAPG)'*DFTAPG) ...
% % + x*((OddHarmonics.*DFTMLG)'*DFTAPG + (OddHarmonics.*DFTAPG)'*DFTMLG)...
% % + ((OddHarmonics.*DFTMLG)'*DFTMLG) )...
% % /
% % ( (x.^2)*((EvenHarmonics.*DFTAPG)'*DFTAPG) ...
% % + x*((EvenHarmonics.*DFTMLG)'*DFTAPG + (EvenHarmonics.*DFTAPG)'*DFTMLG)...
% % + ((EvenHarmonics.*DFTMLG)'*DFTMLG) )
% % *
% % ( (x.^2)*((EvenHarmonics.*DFTMLG)'*DFTMLG) ...
% % - x*((EvenHarmonics.*DFTMLG)'*DFTAPG + (EvenHarmonics.*DFTAPG)'*DFTMLG)...
% % + ((EvenHarmonics.*DFTAPG)'*DFTAPG) )...
% % /
% % ( (x.^2)*((OddHarmonics.*DFTMLG)'*DFTMLG) ...
% % - x*((OddHarmonics.*DFTMLG)'*DFTAPG + (OddHarmonics.*DFTAPG)'*DFTMLG)...
% % + ((OddHarmonics.*DFTAPG)'*DFTAPG) )
% %
% % or by making these substitutions:
% a2 = (OddHarmonics.*DFTAPG)'*DFTAPG;
% a1 = (OddHarmonics.*DFTMLG)'*DFTAPG + (OddHarmonics.*DFTAPG)'*DFTMLG;
% a0 = (OddHarmonics.*DFTMLG)'*DFTMLG;
% c2 = (EvenHarmonics.*DFTAPG)'*DFTAPG;
% c1 = (EvenHarmonics.*DFTMLG)'*DFTAPG + (EvenHarmonics.*DFTAPG)'*DFTMLG;
% c0 = (EvenHarmonics.*DFTMLG)'*DFTMLG;
% b2 = c0;
% b1 = c1;
% b0 = c2;
% d2 = a0;
% d1 = a1;
% d0 = a2;
% % we get:
% % HR_Prod = (x.^2*a2 + x*a1 + a0) * (x.^2*b2 + x*b1 + b0) / (x.^2*c2 + x*c1 + c0) * (x.^2*d2 + x*d1 + d0)
% % or:
% % HR_Prod = N/D
% % with
% % N = ( x.^4*(a2*b2)...
% % +x.^3*(a2*b1+a1*b2)...
% % +x.^2*(a2*b0+a1*b1+a0*b2)...
% % +x *(a1*b0+a0*b1)...
% % + a0*b0 )...
% % and with
% % D = ( x.^4*(c2*d2)...
% % +x.^3*(c2*d1+c1*d2)...
% % +x.^2*(c2*d0+c1*d1+c0*d2)...
% % +x *(c1*d0+c0*d1)...
% % + c0*d0 )
% % or D = ( x.^4*(a0*b0)...
% % +x.^3*(a1*b0+a0*b1)...
% % +x.^2*(a2*b0+a1*b1+a0*b2)...
% % +x *(a2*b1+a1*b2)...
% % + a2*b2 )
% % or by substition of
% n4 = a2*b2;
% n3 = a2*b1+a1*b2;
% n2 = a2*b0+a1*b1+a0*b2;
% n1 = a1*b0+a0*b1;
% n0 = a0*b0;
% % d4 = n0;
% % d3 = n1;
% % d2 = n2;
% % d1 = n3;
% % d0 = n4;
% % we get
% % N = sum(ni*x^i) and D=sum(di*x^i)
% %
% % Then HR_Prod reaches a maximum or minimum when d/dx (N/D) = 0
% % or (dN/dx)/D - N/(D^2)*(dD/dx) = 0
% % or (dN/dx)*D - (dD/dx)*N = 0
% % or
% % x^7*(4*n4*d4 - 4*n4*d4)
% % + x^6*(4*n4*d3 - 4*n3*d4 + 3*n3*d4 - 3*n4*d3)
% % + x^5*(4*n4*d2 - 4*n2*d4 + 3*n3*d3 - 3*n3*d3 + 2*n2*d4 - 2*n4*d2)
% % + x^4*(4*n4*d1 - 4*n1*d4 + 3*n3*d2 - 3*n2*d3 + 2*n2*d3 - 2*n3*d2 + 1*n1*d4 - 1*n4*d1)
% % + x^3*(4*n4*d0 - 4*n0*d4 + 3*n3*d1 - 3*n1*d3 + 2*n2*d2 - 2*n2*d2 + 1*n1*d3 - 1*n3*d1)
% % + x^2*( 3*n3*d0 - 3*n0*d3 + 2*n2*d1 - 2*n1*d2 + 1*n1*d2 - 1*n2*d1)
% % + x^1*( 2*n2*d0 - 2*n0*d2 + 1*n1*d1 - 1*n1*d1)
% % + x^0*( 1*n1*d0 - 1*n0*d1)
% % = 0
% % or
% % x^6*(1*n4*d3 - 1*n3*d4)
% % + x^5*(2*n4*d2 - 2*n2*d4)
% % + x^4*(3*n4*d1 - 3*n1*d4 + 1*n3*d2 - 1*n2*d3)
% % + x^3*(4*n4*d0 - 4*n0*d4 + 2*n3*d1 - 2*n1*d3)
% % + x^2*( 3*n3*d0 - 3*n0*d3 + 1*n2*d1 - 1*n1*d2)
% % + x^1*( 2*n2*d0 - 2*n0*d2)
% % + x^0*( 1*n1*d0 - 1*n0*d1)
% % = 0
% % or
% % (x^6+x^0) * (1*n4*n1 - 1*n3*n0)
% % + (x^5+x^1) * (2*n4*n2 - 2*n2*n0)
% % + (x^4+x^2) * (3*n4*n3 - 3*n1*n0 + 1*n3*n2 - 1*n2*n1)
% % + x^3 * (4*n4*n4 - 4*n0*n0 + 2*n3*n3 - 2*n1*n1)
% % = 0
% % or
% % (x^6+x^0) * (1*(a2*b2)*(a1*b0+a0*b1) - 1*(a2*b1+a1*b2)*(a0*b0))
% % + (x^5+x^1) * (2*(a2*b2)*(a2*b0+a1*b1+a0*b2) - 2*(a2*b0+a1*b1+a0*b2)*(a0*b0))
% % + (x^4+x^2) * (3*(a2*b2)*(a2*b1+a1*b2) - 3*(a1*b0+a0*b1)*(a0*b0) + 1*(a2*b1+a1*b2)*(a2*b0+a1*b1+a0*b2) - 1*(a2*b0+a1*b1+a0*b2)*(a1*b0+a0*b1))
% % + x^3 * (4*(a2*b2)*(a2*b2) - 4*(a0*b0)*(a0*b0) + 2*(a2*b1+a1*b2)*(a2*b1+a1*b2) - 2*(a1*b0+a0*b1)*(a1*b0+a0*b1))
% % = 0
% % or
% % (x^6+x^0) * (a2*b2*a1*b0 + a2*b2*b1*a0 - a2*b1*a0*b0 - b2*a1*a0*b0)
% % + (x^5+x^1) * (2*(a2*a2*b2*b0+a2*b2*b2*a0+a2*b2*a1*b1) - 2*(a2*a0*b0*b0+b2*a0*a0*b0+a1*b1*a0*b0))
% % + (x^4+x^2) * (3*a2*a2*b2*b1 +a2*a2*b1*b0 +3*a2*b2*b2*a1 +a2*b2*a1*b0 +a2*b2*b1*a0 +a2*a1*b1*b1 -a2*a1*b0*b0 -a2*a0*b1*b0 +b2*b2*a1*a0 +b2*a1*a1*b1 -b2*a1*a0*b0 -b2*b1*a0*a0 -a1*a1*b1*b0 -a1*b1*b1*a0 -3*a1*a0*b0*b0 -3*b1*a0*a0*b0)
% % + x^3 * (4*a2*a2*b2*b2 +2*a2*a2*b1*b1 +4*a2*b2*a1*b1 +2*b2*b2*a1*a1 -2*a1*a1*b0*b0 -4*a1*b1*a0*b0 -2*b1*b1*a0*a0 -4*a0*a0*b0*b0)
% % = 0
% % which can probably not be solved analytically?.

118
SpatioTemporalGaitParameters.m Normal file
View File

@ -0,0 +1,118 @@
function [ResultStruct] = SpatioTemporalGaitParameters(dataAccCut_filt,StrideTimeSamples,ApplyRealignment,LegLength,FS,IgnoreMinMaxStrides,ResultStruct);
Cutoff = 0.1;
MinDist = floor(0.7*0.5*StrideTimeSamples); % Use StrideTimeSamples estimated above
DatalFilt = dataAccCut_filt;
% From acceleration to velocity
Vel = cumsum(detrend(DatalFilt,'constant'))/FS;
[B,A] = butter(2,Cutoff/(FS/2),'high');
Pos = cumsum(filtfilt(B,A,Vel))/FS;
PosFilt = filtfilt(B,A,Pos);
PosFiltVT = PosFilt(:,1);
% Find minima and maxima in vertical position
% if ~ApplyRealignment % Signals were not realigned, so it has to be done here
% MeanAcc = mean(AccLoco);
% VT = MeanAcc'/norm(MeanAcc);
% PosFiltVT = PosFilt*VT;
% end
% Find minima and maxima in vertical position
[PosPks,PosLocs] = findpeaks(PosFiltVT(:,1),'minpeakdistance',MinDist);
[NegPks,NegLocs] = findpeaks(-PosFiltVT(:,1),'minpeakdistance',MinDist);
NegPks = -NegPks;
if isempty(PosPks) && isempty(NegPks)
PksAndLocs = zeros(0,3);
else
PksAndLocs = sortrows([PosPks,PosLocs,ones(size(PosPks)) ; NegPks,NegLocs,-ones(size(NegPks))], 2);
end
% Correct events for two consecutive maxima or two consecutive minima
Events = PksAndLocs(:,2);
NewEvents = PksAndLocs(:,2);
Signs = PksAndLocs(:,3);
FalseEventsIX = find(diff(Signs)==0);
PksAndLocsToAdd = zeros(0,3);
PksAndLocsToAddNr = 0;
for i=1:numel(FalseEventsIX),
FIX = FalseEventsIX(i);
if FIX <= 2
% remove the event
NewEvents(FIX) = nan;
elseif FIX >= numel(Events)-2
% remove the next event
NewEvents(FIX+1) = nan;
else
StrideTimesWhenAdding = [Events(FIX+1)-Events(FIX-2),Events(FIX+3)-Events(FIX)];
StrideTimesWhenRemoving = Events(FIX+3)-Events(FIX-2);
if max(abs(StrideTimesWhenAdding-StrideTimeSamples)) < abs(StrideTimesWhenRemoving-StrideTimeSamples)
% add an event
[M,IX] = min(Signs(FIX)*PosFiltVT((Events(FIX)+1):(Events(FIX+1)-1)));
PksAndLocsToAddNr = PksAndLocsToAddNr+1;
PksAndLocsToAdd(PksAndLocsToAddNr,:) = [M,Events(FIX)+IX,-Signs(FIX)];
else
% remove an event
if FIX >= 5 && FIX <= numel(Events)-5
ExpectedEvent = (Events(FIX-4)+Events(FIX+5))/2;
else
ExpectedEvent = (Events(FIX-2)+Events(FIX+3))/2;
end
if abs(Events(FIX)-ExpectedEvent) > abs(Events(FIX+1)-ExpectedEvent)
NewEvents(FIX) = nan;
else
NewEvents(FIX+1) = nan;
end
end
end
end
PksAndLocsCorrected = sortrows([PksAndLocs(~isnan(NewEvents),:);PksAndLocsToAdd],2);
% Find delta height and delta time
DH = abs(diff(PksAndLocsCorrected(:,1),1,1));
DT = diff(PksAndLocsCorrected(:,2),1,1);
% Correct outliers in delta h
MaxDH = min(median(DH)+3*mad(DH,1),LegLength/2);
DH(DH>MaxDH) = MaxDH;
% Estimate total length and speed
% (Use delta h and delta t to calculate walking speed: use formula from
% Z&H, but divide by 2 (skip factor 2)since we get the difference twice
% each step, and multiply by 1.25 which is the factor suggested by Z&H)
HalfStepLen = 1.25*sqrt(2*LegLength*DH-DH.^2);
Distance = sum(HalfStepLen);
WalkingSpeedMean = Distance/(sum(DT)/FS);
% Estimate variabilities between strides
StrideLengths = HalfStepLen(1:end-3) + HalfStepLen(2:end-2) + HalfStepLen(3:end-1) + HalfStepLen(4:end);
StrideTimes = PksAndLocsCorrected(5:end,2)-PksAndLocsCorrected(1:end-4,2);
StrideSpeeds = StrideLengths./(StrideTimes/FS);
WSS = nan(1,4);
STS = nan(1,4);
for i=1:4,
STS(i) = std(StrideTimes(i:4:end))/FS;
WSS(i) = std(StrideSpeeds(i:4:end));
end
StepLengthMean=mean(StrideLengths);
StrideTimeVariability = min(STS);
StrideSpeedVariability = min(WSS);
StrideLengthVariability = std(StrideLengths);
% Estimate Stride time variability and stride speed variability by removing highest and lowest part
if ~isinteger(IgnoreMinMaxStrides)
IgnoreMinMaxStrides = ceil(IgnoreMinMaxStrides*size(StrideTimes,1));
end
StrideTimesSorted = sort(StrideTimes);
StrideTimeVariabilityOmitOutlier = std(StrideTimesSorted(1+IgnoreMinMaxStrides:end-IgnoreMinMaxStrides));
StrideSpeedSorted = sort(StrideSpeeds);
StrideSpeedVariabilityOmitOutlier = std(StrideSpeedSorted(1+IgnoreMinMaxStrides:end-IgnoreMinMaxStrides));
StrideLengthsSorted = sort(StrideLengths);
StrideLengthVariabilityOmitOutlier = std(StrideLengthsSorted(1+IgnoreMinMaxStrides:end-IgnoreMinMaxStrides));
ResultStruct.StepLengthMean = StepLengthMean;
ResultStruct.Distance = Distance;
ResultStruct.WalkingSpeedMean = WalkingSpeedMean;
ResultStruct.StrideTimeVariability = StrideTimeVariability;
ResultStruct.StrideSpeedVariability = StrideSpeedVariability;
ResultStruct.StrideLengthVariability = StrideLengthVariability;
ResultStruct.StrideTimeVariabilityOmitOutlier = StrideTimeVariabilityOmitOutlier;
ResultStruct.StrideSpeedVariabilityOmitOutlier = StrideSpeedVariabilityOmitOutlier;
ResultStruct.StrideLengthVariabilityOmitOutlier = StrideLengthVariabilityOmitOutlier;

15
SpectralAnalysisGaitfunc.m Normal file
View File

@ -0,0 +1,15 @@
function [ResultStruct] = SpectralAnalysisGaitfunc(dataAccCut_filt,WindowLen,FS,N_Harm,LowFrequentPowerThresholds,AccVectorLen,ResultStruct)
P=zeros(0,size(dataAccCut_filt,2));
for i=1:size(dataAccCut_filt,2)
[P1,~] = pwelch(dataAccCut_filt(:,i),hamming(WindowLen),[],WindowLen,FS);
[P2,F] = pwelch(dataAccCut_filt(end:-1:1,i),hamming(WindowLen),[],WindowLen,FS);
P(1:numel(P1),i) = (P1+P2)/2;
end
dF = F(2)-F(1);
% Calculate stride frequency and peak widths
[StrideFrequency, ~, PeakWidth, MeanNormalizedPeakWidth] = StrideFrequencyRispen(P,F);
[ResultStruct] = HarmonicityFrequency(dataAccCut_filt, P,F, StrideFrequency,dF,LowFrequentPowerThresholds,N_Harm,FS,AccVectorLen,ResultStruct);
end

35
StepDetectionFunc.m Normal file
View File

@ -0,0 +1,35 @@
function [dataAccCut,dataAccCut_filt,StrideTimeSamples] = StepDetectionFunc(FS,ApplyRemoveSteps,dataAcc,dataAcc_filt,StrideTimeRange)
%% Step detecection
% Determines the number of steps in the signal so that the first 30 and last 30 steps in the signal can be removed
ResultStruct = struct();
if ApplyRemoveSteps
% In order to run the step detection script we first need to run an autocorrelation function;
[ResultStruct] = AutocorrStrides(dataAcc_filt,FS,StrideTimeRange,ResultStruct);
% StrideTimeSamples is needed as an input for the stepcountFunc;
StrideTimeSamples = ResultStruct.StrideTimeSamples;
% Calculate the number of steps;
[PksAndLocsCorrected] = StepcountFunc(dataAcc_filt,StrideTimeSamples,FS);
% This function selects steps based on negative and positive values.
% However to determine the steps correctly we only need one of these;
LocsSteps = PksAndLocsCorrected(1:2:end,2);
%% Cut data & remove currents results
% Remove 20 steps in the beginning and end of data
dataAccCut = dataAcc(LocsSteps(31):LocsSteps(end-30),:);
dataAccCut_filt = dataAcc_filt(LocsSteps(31):LocsSteps(end-30),:);
% Clear currently saved results from Autocorrelation Analysis
clear ResultStruct;
clear PksAndLocsCorrected;
clear LocsSteps;
else;
dataAccCut = dataAcc;
dataAccCut_filt = dataAcc_filt;
end

65
StepcountFunc.m Normal file
View File

@ -0,0 +1,65 @@
function [PksAndLocsCorrected] = StepcountFunc(dataAcc_filt,StrideTimeSamples,FS);
% Step count funciton extracted from other function called:
Cutoff = 0.1;
MinDist = floor(0.7*0.5*StrideTimeSamples); % Use StrideTimeSamples estimated above
DataLFilt = dataAcc_filt;
% From acceleration to
Vel = cumsum(detrend(DataLFilt,'constant'))/FS;
[B,A] = butter(2,Cutoff/(FS/2),'high');
Pos = cumsum(filtfilt(B,A,Vel))/FS;
PosFilt = filtfilt(B,A,Pos);
PosFiltVT = PosFilt(:,1);
% Find minima and maxima in vertical position
[PosPks,PosLocs] = findpeaks(PosFiltVT(:,1),'minpeakdistance',MinDist);
[NegPks,NegLocs] = findpeaks(-PosFiltVT(:,1),'minpeakdistance',MinDist);
NegPks = -NegPks;
if isempty(PosPks) && isempty(NegPks)
PksAndLocs = zeros(0,3);
else
PksAndLocs = sortrows([PosPks,PosLocs,ones(size(PosPks)) ; NegPks,NegLocs,-ones(size(NegPks))], 2);
end
% Correct events for two consecutive maxima or two consecutive minima
Events = PksAndLocs(:,2);
NewEvents = PksAndLocs(:,2);
Signs = PksAndLocs(:,3);
FalseEventsIX = find(diff(Signs)==0);
PksAndLocsToAdd = zeros(0,3);
PksAndLocsToAddNr = 0;
for i=1:numel(FalseEventsIX),
FIX = FalseEventsIX(i);
if FIX <= 2
% remove the event
NewEvents(FIX) = nan;
elseif FIX >= numel(Events)-2
% remove the next event
NewEvents(FIX+1) = nan;
else
StrideTimesWhenAdding = [Events(FIX+1)-Events(FIX-2),Events(FIX+3)-Events(FIX)];
StrideTimesWhenRemoving = Events(FIX+3)-Events(FIX-2);
if max(abs(StrideTimesWhenAdding-StrideTimeSamples)) < abs(StrideTimesWhenRemoving-StrideTimeSamples)
% add an event
[M,IX] = min(Signs(FIX)*PosFiltVT((Events(FIX)+1):(Events(FIX+1)-1)));
PksAndLocsToAddNr = PksAndLocsToAddNr+1;
PksAndLocsToAdd(PksAndLocsToAddNr,:) = [M,Events(FIX)+IX,-Signs(FIX)];
else
% remove an event
if FIX >= 5 && FIX <= numel(Events)-5
ExpectedEvent = (Events(FIX-4)+Events(FIX+5))/2;
else
ExpectedEvent = (Events(FIX-2)+Events(FIX+3))/2;
end
if abs(Events(FIX)-ExpectedEvent) > abs(Events(FIX+1)-ExpectedEvent)
NewEvents(FIX) = nan;
else
NewEvents(FIX+1) = nan;
end
end
end
end
PksAndLocsCorrected = sortrows([PksAndLocs(~isnan(NewEvents),:);PksAndLocsToAdd],2);
end

150
StrideFrequencyFrom3dAcc.m Normal file
View File

@ -0,0 +1,150 @@
function [StrideFrequency, QualityInd, PeakWidth, MeanNormalizedPeakWidth] = StrideFrequencyFrom3dAcc(AccXYZ, F)
%% Description
% Estimate stride frequency in 3d accelerometer data, using multi-taper and
% pwelch spectral densities
%
% Input:
% AccXYZ: a three-dimensional time series with trunk accelerations
% FS: the sample frequency of the time series
% StrideFreqGuess: a first guess of the stride frequency
%
% Output:
% StrideFrequency: the estimated stride frequency
% QualityInd: a number (0-1, 0=no confidence, 1=fully confident) indicating how much confidence we have in the
% estimated stride frequency
%% Copyright
% COPYRIGHT (c) 2012 Sietse Rispens, VU University Amsterdam
%
% This program is free software: you can redistribute it and/or modify
% it under the terms of the GNU General Public License as published by
% the Free Software Foundation, either version 3 of the License, or
% (at your option) any later version.
%
% This program is distributed in the hope that it will be useful,
% but WITHOUT ANY WARRANTY; without even the implied warranty of
% MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
% GNU General Public License for more details.
%
% You should have received a copy of the GNU General Public License
% along with this program. If not, see <http://www.gnu.org/licenses/>.
%% Author
% Sietse Rispens
%% History
% February 2013, version 1.1, adapted from StrideFrequencyFrom3dAcc
%% Check input
if size(AccXYZ,2) ~= 3
error('AccXYZ must be 3-d time series, i.e. contain 3 columns');
elseif size(AccXYZ,1) < 10*F
error('AccXYZ must be at least ten seconds long');
end
%% Get PSD
if numel(F) == 1, % Calculate the PSD from time series AccXYZ, F is the sample frequency
AccFilt = detrend(AccXYZ,'constant'); % Detrend data to get rid of DC component in most of the specific windows
LenPSD = 10*F;
for i=1:3,
[P1,Fwf] = pwelch(AccFilt(:,i),hamming(LenPSD),[],LenPSD,F);
[P2,Fwf] = pwelch(AccFilt(end:-1:1,i),hamming(LenPSD),[],LenPSD,F);
Pwf(:,i) = (P1+P2)/2;
end
elseif numel(F)==size(AccXYZ,1), % F are the frequencies of the power spectrum AccXYZ
Fwf = F;
Pwf = AccXYZ;
end
Pwf(:,4) = sum(Pwf,2);
%% Estimate stride frequency
% set parameters
HarmNr = [2 1 2];
CommonRange = [0.6 1.2];
% Get modal frequencies and the 'mean freq. of the peak'
for i=1:4,
MF1I = find([zeros(5,1);Pwf(6:end,i)]==max([zeros(5,1);Pwf(6:end,i)]),1);
MF1 = Fwf(MF1I,1);
IndAround = Fwf>=MF1*0.5 & Fwf<=MF1*1.5;
MeanAround = mean(Pwf(IndAround,i));
PeakBeginI = find(IndAround & Fwf<MF1 & Pwf(:,i) < mean([MeanAround,Pwf(MF1I,i)]),1,'last');
PeakEndI = find(IndAround & Fwf>MF1 & Pwf(:,i) < mean([MeanAround,Pwf(MF1I,i)]),1,'first');
if isempty(PeakBeginI), PeakBeginI = find(IndAround,1,'first'); end
if isempty(PeakEndI), PeakEndI = find(IndAround,1,'last'); end
ModalF(i) = sum(Fwf(PeakBeginI:PeakEndI,1).*Pwf(PeakBeginI:PeakEndI,i))/sum(Pwf(PeakBeginI:PeakEndI,i));
if i==4
HarmNr(4) = HarmNr(find(Pwf(MF1I,1:3)==max(Pwf(MF1I,1:3)),1));
end
end
% Get stride frequency and quality indicator from modal frequencies
StrFreqFirstGuesses = ModalF./HarmNr;
StdOverMean = std(StrFreqFirstGuesses)/mean(StrFreqFirstGuesses);
StrideFrequency1 = median(StrFreqFirstGuesses(1:3));
if StrideFrequency1 > CommonRange(2) && min(StrFreqFirstGuesses(1:3)) < CommonRange(2) && min(StrFreqFirstGuesses(1:3)) > CommonRange(1)
StrideFrequency1 = min(StrFreqFirstGuesses(1:3));
end
if StrideFrequency1 < CommonRange(1) && max(StrFreqFirstGuesses(1:3)) > CommonRange(1) && max(StrFreqFirstGuesses(1:3)) < CommonRange(2)
StrideFrequency1 = min(StrFreqFirstGuesses(1:3));
end
HarmGuess = ModalF/StrideFrequency1;
StdHarmGuessRoundErr = std(HarmGuess - round(HarmGuess));
if StdOverMean < 0.1
QI1 = 1;
StrideFrequency = mean(StrFreqFirstGuesses);
else
if StdHarmGuessRoundErr < 0.1 && all(round(HarmGuess) >= 1)
QI1 = 0.5;
StrideFrequency = mean(ModalF./round(HarmGuess));
else
QI1 = 0;
StrideFrequency = StrideFrequency1;
end
end
if nargout >= 2
QualityInd = QI1;
end
if nargout >= 3
N_Harm = 20;
PeakWidth = nan(1,3);
if nargout >= 4
MeanNormalizedPeakWidth = nan(1,3);
end
%% Get (mean) widths of harmonic peaks
for i=1:3,
WidthHarm = nan(1,N_Harm);
PowerHarm = nan(1,N_Harm);
for HarmonicNr = 1:N_Harm,
FreqRangeIndices = ...
Fwf >= StrideFrequency*(HarmonicNr-0.5) ...
& Fwf <= StrideFrequency*(HarmonicNr+0.5);
PeakPower = sum(Pwf(FreqRangeIndices,i));
PeakMean = sum(Pwf(FreqRangeIndices,i).*Fwf(FreqRangeIndices))/PeakPower;
PeakMeanSquare = sum(Pwf(FreqRangeIndices,i).*Fwf(FreqRangeIndices).^2)/PeakPower;
WidthHarm(HarmonicNr) = sqrt(PeakMeanSquare-PeakMean.^2);
PowerHarm(HarmonicNr) = PeakPower;
end
PeakWidth(i) = WidthHarm(HarmNr(i)); % Take the 1st or 2nd harmonic width as original measure
if nargout >= 4
MeanNormalizedPeakWidth(i) = sum(WidthHarm./(1:N_Harm).*PowerHarm)/sum(PowerHarm);
end
end
end
if nargout == 0
IXplotw = Fwf<10;
figure();
for i=1:3,
subplot(2,2,i);
plot(Fwf(IXplotw,1),Pwf(IXplotw,i));
end
subplot(2,2,4);
plot(Fwf(IXplotw,1),Pwf(IXplotw,1:3));
end

150
StrideFrequencyRispen.m Normal file
View File

@ -0,0 +1,150 @@
function [StrideFrequency, QualityInd, PeakWidth, MeanNormalizedPeakWidth] = StrideFrequencyRispen(AccXYZ, F)
%% Description
% Estimate stride frequency in 3d accelerometer data, using multi-taper and
% pwelch spectral densities
%
% Input:
% AccXYZ: a three-dimensional time series with trunk accelerations
% FS: the sample frequency of the time series
% StrideFreqGuess: a first guess of the stride frequency
%
% Output:
% StrideFrequency: the estimated stride frequency
% QualityInd: a number (0-1, 0=no confidence, 1=fully confident) indicating how much confidence we have in the
% estimated stride frequency
%% Copyright
% COPYRIGHT (c) 2012 Sietse Rispens, VU University Amsterdam
%
% This program is free software: you can redistribute it and/or modify
% it under the terms of the GNU General Public License as published by
% the Free Software Foundation, either version 3 of the License, or
% (at your option) any later version.
%
% This program is distributed in the hope that it will be useful,
% but WITHOUT ANY WARRANTY; without even the implied warranty of
% MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
% GNU General Public License for more details.
%
% You should have received a copy of the GNU General Public License
% along with this program. If not, see <http://www.gnu.org/licenses/>.
%% Author
% Sietse Rispens
%% History
% February 2013, version 1.1, adapted from StrideFrequencyFrom3dAcc
%% Check input
if size(AccXYZ,2) ~= 3
error('AccXYZ must be 3-d time series, i.e. contain 3 columns');
elseif size(AccXYZ,1) < 10*F
error('AccXYZ must be at least ten seconds long');
end
%% Get PSD
if numel(F) == 1, % Calculate the PSD from time series AccXYZ, F is the sample frequency
AccFilt = detrend(AccXYZ,'constant'); % Detrend data to get rid of DC component in most of the specific windows
LenPSD = 10*F;
for i=1:3,
[P1,Fwf] = pwelch(AccFilt(:,i),hamming(LenPSD),[],LenPSD,F);
[P2,Fwf] = pwelch(AccFilt(end:-1:1,i),hamming(LenPSD),[],LenPSD,F);
Pwf(:,i) = (P1+P2)/2;
end
elseif numel(F)==size(AccXYZ,1), % F are the frequencies of the power spectrum AccXYZ
Fwf = F;
Pwf = AccXYZ;
end
Pwf(:,4) = sum(Pwf,2);
%% Estimate stride frequency
% set parameters
HarmNr = [2 1 2];
CommonRange = [0.6 1.2];
% Get modal frequencies and the 'mean freq. of the peak'
for i=1:4,
MF1I = find([zeros(5,1);Pwf(6:end,i)]==max([zeros(5,1);Pwf(6:end,i)]),1);
MF1 = Fwf(MF1I,1);
IndAround = Fwf>=MF1*0.5 & Fwf<=MF1*1.5;
MeanAround = mean(Pwf(IndAround,i));
PeakBeginI = find(IndAround & Fwf<MF1 & Pwf(:,i) < mean([MeanAround,Pwf(MF1I,i)]),1,'last');
PeakEndI = find(IndAround & Fwf>MF1 & Pwf(:,i) < mean([MeanAround,Pwf(MF1I,i)]),1,'first');
if isempty(PeakBeginI), PeakBeginI = find(IndAround,1,'first'); end
if isempty(PeakEndI), PeakEndI = find(IndAround,1,'last'); end
ModalF(i) = sum(Fwf(PeakBeginI:PeakEndI,1).*Pwf(PeakBeginI:PeakEndI,i))/sum(Pwf(PeakBeginI:PeakEndI,i));
if i==4
HarmNr(4) = HarmNr(find(Pwf(MF1I,1:3)==max(Pwf(MF1I,1:3)),1));
end
end
% Get stride frequency and quality indicator from modal frequencies
StrFreqFirstGuesses = ModalF./HarmNr;
StdOverMean = std(StrFreqFirstGuesses)/mean(StrFreqFirstGuesses);
StrideFrequency1 = median(StrFreqFirstGuesses(1:3));
if StrideFrequency1 > CommonRange(2) && min(StrFreqFirstGuesses(1:3)) < CommonRange(2) && min(StrFreqFirstGuesses(1:3)) > CommonRange(1)
StrideFrequency1 = min(StrFreqFirstGuesses(1:3));
end
if StrideFrequency1 < CommonRange(1) && max(StrFreqFirstGuesses(1:3)) > CommonRange(1) && max(StrFreqFirstGuesses(1:3)) < CommonRange(2)
StrideFrequency1 = min(StrFreqFirstGuesses(1:3));
end
HarmGuess = ModalF/StrideFrequency1;
StdHarmGuessRoundErr = std(HarmGuess - round(HarmGuess));
if StdOverMean < 0.1
QI1 = 1;
StrideFrequency = mean(StrFreqFirstGuesses);
else
if StdHarmGuessRoundErr < 0.1 && all(round(HarmGuess) >= 1)
QI1 = 0.5;
StrideFrequency = mean(ModalF./round(HarmGuess));
else
QI1 = 0;
StrideFrequency = StrideFrequency1;
end
end
if nargout >= 2
QualityInd = QI1;
end
if nargout >= 3
N_Harm = 20;
PeakWidth = nan(1,3);
if nargout >= 4
MeanNormalizedPeakWidth = nan(1,3);
end
%% Get (mean) widths of harmonic peaks
for i=1:3,
WidthHarm = nan(1,N_Harm);
PowerHarm = nan(1,N_Harm);
for HarmonicNr = 1:N_Harm,
FreqRangeIndices = ...
Fwf >= StrideFrequency*(HarmonicNr-0.5) ...
& Fwf <= StrideFrequency*(HarmonicNr+0.5);
PeakPower = sum(Pwf(FreqRangeIndices,i));
PeakMean = sum(Pwf(FreqRangeIndices,i).*Fwf(FreqRangeIndices))/PeakPower;
PeakMeanSquare = sum(Pwf(FreqRangeIndices,i).*Fwf(FreqRangeIndices).^2)/PeakPower;
WidthHarm(HarmonicNr) = sqrt(PeakMeanSquare-PeakMean.^2);
PowerHarm(HarmonicNr) = PeakPower;
end
PeakWidth(i) = WidthHarm(HarmNr(i)); % Take the 1st or 2nd harmonic width as original measure
if nargout >= 4
MeanNormalizedPeakWidth(i) = sum(WidthHarm./(1:N_Harm).*PowerHarm)/sum(PowerHarm);
end
end
end
if nargout == 0
IXplotw = Fwf<10;
figure();
for i=1:3,
subplot(2,2,i);
plot(Fwf(IXplotw,1),Pwf(IXplotw,i));
end
subplot(2,2,4);
plot(Fwf(IXplotw,1),Pwf(IXplotw,1:3));
end

1
div_calc.m Normal file
View File

@ -0,0 +1 @@
function Divergence=div_calc(state,ws_sec,fs,period_sec,progress) % calculate divergence curve (needed for max lyapunov exponent). % Input: % state: state space % ws_sec: window size over which divergence will be calculated(in seconds) % fs: sample frequency % period_sec: indicates period of time-wise near samples to exclude in % nearest neighbour search % progress: show progress or not % % Output: % Divergence: the divergence curve % % History: % August 2011: v1, Sietse Rispens based on version KvS en SMB % 1 November 2012, Sietse Rispens: Use div_calc_shorttimeseries for short % time series if size(state,1) <= 10000 Divergence=div_calc_shorttimeseries(state,ws_sec,fs,period_sec,progress); return; end [N,D]=size(state); ws=round(ws_sec*fs); Period=round(period_sec*fs); if N<ws, error('ws shall not be larger than N'); end NCompleteWindow = N-ws+1; Divergence_sum=zeros(1,ws); Divergence_count=zeros(1,ws); SumStd = sqrt(sum(std(state,1).^2)); DistLimStart = SumStd*nthroot(1/NCompleteWindow,D); DistLim = DistLimStart; TreeRoot=kdtree_build(state(1:NCompleteWindow,:)); k0 = 2; % The initial number of nearest neighbors to look for kmax = min(2*Period + 2, NCompleteWindow); % The maximum number of nearest neighbors to look for for i = 1:NCompleteWindow StateI = state(i,:); if ~isnan(StateI) DistLim = DistLim*nthroot(1/5,D); DistLim = min(DistLim,DistLimStart); Index = []; k=k0; kmaxreached = 0; while isempty(Index) && ~kmaxreached [Idx, Dist] = kdtree_k_nearest_neighbors(TreeRoot,StateI,k); Dist(abs(i-Idx) < Period)=nan; Index = Idx(find(Dist==min(Dist),1,'first')); if k >= kmax kmaxreached = 1; elseif isempty(Index) k = min(kmax,k*2); end end if ~isempty(Index) if i+ws>N || Index+ws>N % do not use these data else if ~isempty(Index) DistCurve = log(sqrt(sum((state(i:i+ws-1,:)-state(Index:Index+ws-1,:)).^2,2))); NotNan = ~isnan(DistCurve); Divergence_sum(NotNan) = Divergence_sum(NotNan) + DistCurve(NotNan)'; Divergence_count(NotNan) = Divergence_count(NotNan) + 1; end end end end if progress > 0 if mod(i,round(progress))==0 if i>round(progress) fprintf('\b\b\b\b\b\b\b\b\b\b'); end fprintf('i=%8d', i); end end end kdtree_delete(TreeRoot); % Free the pointer to k-d-tree if progress > 0 fprintf('\n'); end Divergence= (Divergence_sum./Divergence_count);

1
div_calc_shorttimeseries.m Normal file
View File

@ -0,0 +1 @@
function divergence=div_calc_shorttimeseries(state,ws_sec,fs,period_sec,progress) % calculate local dynamic stability (max lyapunov exponent). % Input: % state: appropriate state space % ws: window size over which divergence should be calculated(in seconds) % fs: sample frequency % period: ..... (can be dominant period in the signal(in seconds)) % plotje: show a graph. % % Note that in this version ws should be larger than 4*period, as the long % term divergence is calculated from 4*period to ws*fs. % % Output: % divergence: the divergence curve % lds: the 2 estimates of the maximum lyapunov exponents (short term and long term) % % Earlier versions by KvS en SMB. Made the routine faster, 14/01/2011, Sietse Rispens % Use less memory to prevent memory error, 01/02/2011, Sietse Rispens % Take mean of the log of divergence instead of log of the mean of divergence, 26/05/2011, Sietse Rispens % Allow non-integer ws_sec and period_sec and change fitting-range, 01/08/2011, Sietse Rispens % Do not try to find neighbours that need to be followed beyond end of time series, 01/11/2012, Sietse Rispens % Exclude neighbours closer than Period in time (instead of period/2), 01/11/2012, Sietse Rispens [m,n]=size(state); ws=round(ws_sec*fs); period=round(period_sec*fs); mcompletewindow = m-ws+1; statecw = state(1:mcompletewindow,:); divergence_sum=zeros(1,ws); divergence_count=zeros(1,ws); diff_state = statecw*0; diff_state_sqr = diff_state; diff_total = zeros(size(diff_state,1),1); for i = 1:mcompletewindow if ~isnan(state(i,:)) start=round(max([1,i-period+1])); stop=round(min([mcompletewindow,i+period-1])); for j=1:n, diff_state(:,j) = statecw(:,j)-statecw(i,j); diff_state_sqr(:,j)=diff_state(:,j).^2; end diff_total(:,1)=sum(diff_state_sqr,2); diff_total(start:stop,1)=NaN; [mini,index]=min(diff_total); if i+ws>m || index+ws>m % do not use these data else divergence_sum = divergence_sum + log(sqrt(sum((state(i:i+ws-1,:)-state(index:index+ws-1,:)).^2,2)))'; divergence_count = divergence_count + 1; end end if progress > 0 if mod(i,round(progress))==0 if i>round(progress) fprintf('\b\b\b\b\b\b\b\b\b\b'); end fprintf('i=%8d', i); end end end if progress > 0 fprintf('\n'); end divergence= (divergence_sum./divergence_count);

1
div_wolf_fixed_evolv.m Normal file
View File

File diff suppressed because one or more lines are too long

55
findminMutInf.m Normal file
View File

@ -0,0 +1,55 @@
function minmuttau = findminMutInf(miV,nsam)
% minmuttau = findminMutInf(miV,nsam)
% findminMutInf finds the lag tau of the first local minimum of mutual
% information 'miV' (given from lag 0 and up to a maximum lag 'taumax')
% and using a sliding window of length 2*nsam+1.
%========================================================================
% <findminMutInf.m>, v 1.0 2010/02/11 22:09:14 Kugiumtzis & Tsimpiris
% This is part of the MATS-Toolkit http://eeganalysis.web.auth.gr/
%========================================================================
% Copyright (C) 2010 by Dimitris Kugiumtzis and Alkiviadis Tsimpiris
% <dkugiu@gen.auth.gr>
%========================================================================
% Version: 1.0
% LICENSE:
% This program is free software; you can redistribute it and/or modify
% it under the terms of the GNU General Public License as published by
% the Free Software Foundation; either version 3 of the License, or
% any later version.
%
% This program is distributed in the hope that it will be useful,
% but WITHOUT ANY WARRANTY; without even the implied warranty of
% MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
% GNU General Public License for more details.
%
% You should have received a copy of the GNU General Public License
% along with this program. If not, see http://www.gnu.org/licenses/>.
%=========================================================================
% Reference : D. Kugiumtzis and A. Tsimpiris, "Measures of Analysis of Time Series (MATS):
% A Matlab Toolkit for Computation of Multiple Measures on Time Series Data Bases",
% Journal of Statistical Software, in press, 2010
% Link : http://eeganalysis.web.auth.gr/
%=========================================================================
taumax = length(miV)-1;
minmuttau = NaN;
if taumax>2*nsam
i=nsam+1;
found = 0;
while i<taumax+1-nsam & found==0
winx = miV([i-nsam:i-1 i+1:i+nsam]);
check = find(miV(i) < winx);
if length(check) == length(winx)
found=1;
else
i=i+1;
end
end
if found
minmuttau = i-1;
end
end

79
funcSampleEntropy.m Normal file
View File

@ -0,0 +1,79 @@
function [SE] = funcSampleEntropy(DataIn, m, r)
%% Description
% Calculate the sample entropy as described in
% Richman JS, Moorman JR (2000)
% "Physiological time-series analysis using approximate entropy and sample entropy"
% American Journal of Physiology. Heart and Circulatory Physiology [2000, 278(6):H2039-49]
%
% The sample entropy is calculated as the natural logarithm of the
% probability that two samples of length m that are within a distance of r,
% remain within a distance of r when adding one additional sample. Note
% that distance is considered as the maximum of the distances for the
% individual dimensions 1 to m, and that the input data is normalised.
%
% Input:
% DataIn: a one-dimensional time series
% m: the dimension of the vectors to be used. The vectors consist of m
% consecutive samples
% r: the maximum distance between two samples to qualify as a
% mathch, relative to the std of DataIn
%
% Output:
% SE: the calculated sample entropy
%
%% Copyright
% COPYRIGHT (c) 2012 Sietse Rispens, VU University Amsterdam
%
% This program is free software: you can redistribute it and/or modify
% it under the terms of the GNU General Public License as published by
% the Free Software Foundation, either version 3 of the License, or
% (at your option) any later version.
%
% This program is distributed in the hope that it will be useful,
% but WITHOUT ANY WARRANTY; without even the implied warranty of
% MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
% GNU General Public License for more details.
%
% You should have received a copy of the GNU General Public License
% along with this program. If not, see <http://www.gnu.org/licenses/>.
%% Author
% Sietse Rispens
%% History
% 7 May 2012, version 1.0
%% Check input
if size(DataIn,1) ~= 1 && size(DataIn,2) ~= 1
error('DataIn must be a vector');
end
DataIn = DataIn(:)/std(DataIn(:));
N = size(DataIn,1);
if N-m <= 0
error('m must be smaller than the length of the time series DataIn');
end
%% Create the vectors Xm to be compared
Xm = zeros(N-m,m);
for i = 1:m,
Xm(:,i) = DataIn(i:end-1-m+i,1);
end
%% Count the numbers of matches for Xm and Xmplusone
CountXm = 0;
CountXmplusone = 0;
XmDist = nan(size(Xm));
for i = 1:N-m,
for j=1:m,
XmDist(:,j)=abs(Xm(:,j)-Xm(i,j));
end
IdXmi = find(max(XmDist,[],2)<=r);
CountXm = CountXm + length(IdXmi) - 1;
CountXmplusone = CountXmplusone + sum(abs(DataIn(IdXmi+m)-DataIn(i+m))<=r) - 1;
end
%% Return sample entropy
SE = -log(CountXmplusone/CountXm);

104
matlab.sty Normal file
View File

@ -0,0 +1,104 @@
\NeedsTeXFormat{LaTeX2e}
\ProvidesPackage{matlab}
\RequirePackage{verbatim}
\RequirePackage{fancyvrb}
\RequirePackage{alltt}
\RequirePackage{upquote}
\RequirePackage[framemethod=tikz]{mdframed}
\RequirePackage{hyperref}
\RequirePackage{color}
\newcommand{\maxwidth}[1]{\ifdim\linewidth>#1 #1\else\linewidth\fi}
\newcommand{\mlcell}[1]{{\color{output}\verbatim@font#1}}
\definecolor{output}{gray}{0.4}
% Unicode character conversions
\DeclareUnicodeCharacter{B0}{\ensuremath{^\circ}}
\DeclareUnicodeCharacter{21B5}{\ensuremath{\hookleftarrow}}
% Paragraph indentation
\setlength{\parindent}{0pt}
% Hyperlink style
\hypersetup{
colorlinks=true,
linkcolor=blue,
urlcolor=blue
}
% environment styles for MATLAB code and output
\mdfdefinestyle{matlabcode}{%
outerlinewidth=.5pt,
linecolor=gray!20!white,
roundcorner=2pt,
innertopmargin=.5\baselineskip,
innerbottommargin=.5\baselineskip,
innerleftmargin=1em,
backgroundcolor=gray!10!white
}
\newenvironment{matlabcode}{\verbatim}{\endverbatim}
\surroundwithmdframed[style=matlabcode]{matlabcode}
\newenvironment{matlaboutput}{%
\Verbatim[xleftmargin=1.25em, formatcom=\color{output}]%
}{\endVerbatim}
\newenvironment{matlabsymbolicoutput}{%
\list{}{\leftmargin=1.25em\relax}%
\item\relax%
\color{output}\verbatim@font%
}{\endlist}
\newenvironment{matlabtableoutput}[1]{%
{\color{output}%
\hspace*{1.25em}#1}%
}{}
% Table of Contents style
\newcounter{multititle}
\newcommand{\matlabmultipletitles}{\setcounter{multititle}{1}}
\newcounter{hastoc}
\newcommand{\matlabhastoc}{\setcounter{hastoc}{1}}
\newcommand{\matlabtitle}[1]{
\ifnum\value{multititle}>0
\ifnum\value{hastoc}>0
\addcontentsline{toc}{section}{#1}
\fi
\fi
\section*{#1}
}
\newcommand{\matlabheading}[1]{
\ifnum\value{hastoc}>0
\addcontentsline{toc}{subsection}{#1}
\fi
\subsection*{#1}
}
\newcommand{\matlabheadingtwo}[1]{
\ifnum\value{hastoc}>0
\addcontentsline{toc}{subsubsection}{#1}
\fi
\subsubsection*{#1}
}
\newcommand{\matlabheadingthree}[1]{
\ifnum\value{hastoc}>0
\addcontentsline{toc}{paragraph}{#1}
\fi
\paragraph*{#1}
}
\newcommand{\matlabtableofcontents}[1]{
\renewcommand{\contentsname}{#1}
\tableofcontents
}

71
mutinfHisPro.m Normal file
View File

@ -0,0 +1,71 @@
function mutV=mutinfHisPro(xV,tauV,b,ioxV,ixV)
% mutV=mutinfHisPro(xV,tauV,b,ioxV,ixV)
% mutinfHisPro computes the mutual information on the time series 'xV'
% for given delays in 'tauV'. The estimation of mutual information is
% based on 'b' partitions of equal probability at each dimension.
% The last two input parameters are the ordered time series and the
% corresponding indices that will be used in the equiprobable binning
% (they both have been computed before and therefore they are passed
% here rather than computing it again).
%========================================================================
% <mutinfHisPro.m>, v 1.0 2010/02/11 22:09:14 Kugiumtzis & Tsimpiris
% This is part of the MATS-Toolkit http://eeganalysis.web.auth.gr/
%========================================================================
% Copyright (C) 2010 by Dimitris Kugiumtzis and Alkiviadis Tsimpiris
% <dkugiu@gen.auth.gr>
%========================================================================
% Version: 1.0
% LICENSE:
% This program is free software; you can redistribute it and/or modify
% it under the terms of the GNU General Public License as published by
% the Free Software Foundation; either version 3 of the License, or
% any later version.
%
% This program is distributed in the hope that it will be useful,
% but WITHOUT ANY WARRANTY; without even the implied warranty of
% MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
% GNU General Public License for more details.
%
% You should have received a copy of the GNU General Public License
% along with this program. If not, see http://www.gnu.org/licenses/>.
%=========================================================================
% Reference : D. Kugiumtzis and A. Tsimpiris, "Measures of Analysis of Time Series (MATS):
% A Matlab Toolkit for Computation of Multiple Measures on Time Series Data Bases",
% Journal of Statistical Software, in press, 2010
% Link : http://eeganalysis.web.auth.gr/
%=========================================================================
n = length(xV);
ntau = length(tauV);
mutV = NaN*ones(ntau,1);
hM = NaN*ones(b,b);
cumhM = zeros(b,b+1);
cpxV = [1/b:1/b:1]';
for itau=1:ntau
tau = tauV(itau);
ntotal = n-tau;
rxV = [0;round(cpxV*ntotal)];
ix1V = ixV;
ix1V(ioxV(end-tau+1:end)) = [];
x2prV = prctile(xV(ix1V+tau),cpxV*100);
for i = 1:b
for j = 1:b
cumhM(i,j+1) = length(find(xV(ix1V(rxV(i)+1:rxV(i+1))+tau)<=x2prV(j)));
end
hM(i,:) = diff(cumhM(i,:));
end
% The use of formula H(x)=1, when log_b is used.
mutS = 2;
for j=1:b
for i=1:b
if hM(i,j) > 0
mutS=mutS+(hM(i,j)/ntotal)*log(hM(i,j)/ntotal)/log(b);
end
end
end
mutV(itau) = mutS;
end