Skip to content

GitLab

Commit e2158243 authored by Jan-Gerrit Richter's avatar Jan-Gerrit Richter
Browse files

changed interp 

- cleanup
- bugfixes
- changed code to codebase from my diplomarbeit
- added shift to ear
parent 9653e316
Hide whitespace changes
Inline Side-by-side
applications/Binaural-HRTF/HRTF_class/@itaHRTF/interp.m
View file @ e2158243
......@@ -19,6 +19,11 @@ function cThis = interp(this,varargin)
% set to 1 if no range extrapolation is required
% order ... order of spherical harmonics matrix (default: 50)
% epsilon ... regularization coefficient (default: 1e-8)
% shiftToEar to a shift to approximate ear position to
% improve sh transformation (see [2])
% shiftAxis shift along this axis ('x','y' (default),'z')
% shiftOffset shift ears (L - R) by these values
% (default: [-0.0725 0.0725])
%
% OUTPUT:
% itaHRTF object
......@@ -26,19 +31,24 @@ function cThis = interp(this,varargin)
% .timeData: interpolated / range-extrapolated HRIRs for defined field points
% .dirCoord: itaCoordinates object
%
% Required: SphericalHarmonics functions of ITA Toolbox
%
%
% [1] Pollow, Martin et al., "Calculation of Head-Related Transfer Functions
% for Arbitrary Field Points Using Spherical Harmonics Decomposition",
% Acta Acustica united with Acustica, Volume 98, Number 1, January/February 2012,
% pp. 72-82(11)
%
% [2] Richter, Jan-Gerrit et al. "Spherical harmonics based hrtf datasets:
% Implementation and evaluation for real-time auralization",
% Acta Acustica united with Acustica, Volume 100, Number 4, July/August 2014,
% pp. 667-675(9)
%
%
%
% Author: Florian Pausch <fpa@akustik.rwth-aachen.de>
% Version: 2016-02-05
% TODO: check why this is still not working (coordinate assignment???)
sArgs = struct('order',50,'eps',1e-5);
sArgs = struct('order',50,'eps',1e-5,'shiftToEar',false,'shiftAxis','y','shiftOffset',[-0.0725 0.0725]);
sArgs = ita_parse_arguments(sArgs,varargin,2);
if ~isa(varargin{1},'itaCoordinates'),error('itaHRTF:interp', ' An itaCoordinate object is needed!')
end
......@@ -47,7 +57,7 @@ field_in = varargin{1};
% only take unique direction coordinates (round to 0.01deg resolution)
tempfield = unique(round([field_in.phi_deg*100 field_in.theta_deg*100]),'rows'); % may cause problems with older Matlab versions (<=R2013)!
tempfield = tempfield./100;
temp_r = this.dirCoord.r(1);
temp_r = field_in.r(1);
field = itaCoordinates(size(tempfield,1));
field.r = repmat(temp_r,size(tempfield,1),1);
field.phi_deg = tempfield(:,1);
......@@ -67,46 +77,34 @@ if ~isequal(this.dirCoord.r(1),field.r(1))
kr0 = k*this.dirCoord.r(1); % measurement radius
kr1 = k*field.r(1); % extrapolation radius
hankel_r0 = ita_sph_besselh(1:Nmax,2,kr0);
hankel_r1 = ita_sph_besselh(1:Nmax,2,kr1);
hankel_r0 = ita_sph_besselh(ita_sph_linear2degreeorder(1:Nmax),2,kr0);
hankel_r1 = ita_sph_besselh(ita_sph_linear2degreeorder(1:Nmax),2,kr1);
hankel_div = hankel_r1 ./ hankel_r0;
hankel_rep = hankel_div(:,1);
end
dweights = 1 + (0:Nmax).*((0:Nmax)+1); % calculate regularization weights
nSH = (Nmax+1).^2;
I = sparse(eye(nSH));
n = ita_sph_linear2degreeorder(1:round(nSH)).';
dweights_rep = zeros(sum(2*(0:Nmax)'+1),1);
dweights_rep(1)=dweights(1);
counter = 2;
for n=1:Nmax
nTimes = 2*n+1;
dweights_rep(counter:counter+nTimes-1)=dweights(n+1)*ones(nTimes,1);
if ~isequal(this.dirCoord.r(1),field.r(1))
hankel_rep=[hankel_rep, repmat(hankel_div(:,n),1,2*n+1)];
%% move data from earcenter
copyData = this;
if sArgs.shiftToEar
ear_d = sArgs.shiftOffset;
for ear=1:2
movedData = moveFullDataSet(this.ch(ear:2:this.nChannels),sArgs,ear_d(ear));
copyData.freqData(:,ear:2:copyData.nChannels) = movedData;
end
counter = counter + nTimes;
end
%% move data from earcenter
ear_d = [-0.07 0.07];
%% Weights
[~,w]= this.dirCoord.spherical_voronoi; % calculate weighting coefficients (Voronoi surfaces <-> measurement points)
% sG_full = ita_sph_sampling_equiangular(73,144,'theta_type','[]','phi_type','[)');
% % sG_full = ita_sph_sampling_gaussian(71);
% sG = sG_full;
% isOnArc = sG.theta < 2.75;
% sG.cart = sG.cart(isOnArc,:);
% sG.weights = sG.weights(isOnArc);
% % rescale the weights for the sum to be 4*pi again
% w = sG.weights .* 4.*pi ./ sum(sG.weights);
W = sparse(diag(w)); % diagonal matrix containing weights
D = sparse(diag(dweights_rep)); % decomposition order-dependent Tikhonov regularization
Y = ita_sph_base(this.dirCoord,Nmax,'real'); % calculate real-valued SHs using the measurement grid
D = I .* diag(1 + n.*(n+1)); % decomposition order-dependent Tikhonov regularization
Y = ita_sph_base(this.dirCoord,Nmax,'orthonormal',true); % calculate real-valued SHs using the measurement grid
%% Calculate HRTF data for field points
if Nmax > 25
......@@ -117,73 +115,58 @@ end
hrtf_arbi = zeros(this.nBins,2*field.nPoints); % columns: LRLRLR...
for ear=1:2
freqData_temp = this.freqData(:,ear:2:end);
% sh transformation
freqData_temp = copyData.freqData(:,ear:2:end);
a0 = (Y.'*W*Y + epsilon*D) \ Y.'*W * freqData_temp.';
% newCoords = this.dirCoord;
% newCoords.y = newCoords.y + ear_d(ear);
%
% data.channelCoordinates = newCoords;
% data.freqData = freqData_temp;
% data.freqVector = this.freqVector;
% data.c_meas = 344;
% data = process_result_delay_correction(data,this.dirCoord.r(1));
% % calculate weighted SH coefficients using a decomposition order-dependent Tikhonov regularization
%
% freqData_temp = data.freqData;
% Y = ita_sph_base(data.channelCoordinates,Nmax,'real');
% %% test the sh transformation results by plotting both spatial and sh data with surf
% s = itaSamplingSph(field);
% s.nmax = Nmax;
% figure
% surf(s,a0(:,10))
% figure
% surf(s,freqData_temp(10,:))
a0 = (Y.'*W*Y + epsilon*D) \ Y.'*W * freqData_temp.';
%a0 = (Y.'*W*Y + epsilon*D) \ Y.' *
%this.freqData(:,ear:2:end).'; % fpa version
%a0 = pinv(Y)* this.freqData(:,ear:2:end).'; % jck version
% range extrapolation
if ~isequal(this.dirCoord.r(1),field.r(1))
% calculate range-extrapolated HRTFs
a1 = a0 .* hankel_rep.';
Yest = ita_sph_base(field,N,'real'); % use real-valued SH's
%%% test here to see extrapolation results in spatial domain
% surf(s,a1(:,10))
% reconstruction to spatial data
Yest = ita_sph_base(field,N,'orthonormal',true); % use real-valued SH's
hrtf_arbi(:,ear:2:end) = (Yest*a1).'; % interpolated + range-extrapolated HRTFs
else
Yest = ita_sph_base(field,Nmax,'real'); % use real-valued SH's
% reconstruction to spatial data
Yest = ita_sph_base(field,Nmax,'orthonormal',true); % use real-valued SH's
hrtf_arbi(:,ear:2:end) = (Yest*a0).'; % interpolated HRTFs
end
end
%% move back to head center
% todo
% for ear=1:2
%
% newCoords = field;
% newCoords.y = newCoords.y - ear_d(ear);
%
% freqData = hrtf_arbi(:,ear:2:end);
%
%
% data.channelCoordinates = newCoords;
% data.freqData = freqData;
% data.freqVector = this.freqVector;
% data.c_meas = 344;
% data = process_result_delay_correction(data,this.dirCoord.r(1));
%
% hrtf_arbi(:,ear:2:end) = data.freqData;
%
% end
% set new direction coordinates
sph = zeros(field.nPoints*2 ,3);
sph(1:2:end,:) = field.sph;
sph(2:2:end,:) = field.sph;
% write new HRTF data set
cAudio = itaAudio(hrtf_arbi, 44100, 'freq');
cAudio.channelCoordinates.sph= sph;
cThis = this;
cThis.freqData = hrtf_arbi;
cThis.channelCoordinates.sph= sph;
cThis = itaHRTF(cAudio);
cThis.freqData = hrtf_arbi;
%% move back to head center
if sArgs.shiftToEar
ear_d_back = -ear_d;
% movedData = zeros(size(hrtf_arbi));
for ear=1:2
movedData = moveFullDataSet(cThis.ch(ear:2:cThis.nChannels),sArgs,ear_d_back(ear));
cThis.freqData(:,ear:2:cThis.nChannels) = movedData;
end
end
if ~isequal(cThis.dirCoord.r(1),field.r(1))%???
cThis.dirCoord.r = field.r;
......@@ -195,15 +178,44 @@ end
end
function result = process_result_delay_correction(result, target_d)
% shift every measurement point to target_d by applying a
% phase shift to the channel: (simplified!)
freq_L = result.freqVector;
add_phase_L = (result.channelCoordinates.r - target_d)* (freq_L./result.c_meas)' .* 2.*pi;
function [ data ] = moveFullDataSet(data,options,offsetShift)
fullCoords = data.channelCoordinates;
freqVector = data.freqVector;
shiftedData = zeros(size(data.freqData));
axis = options.shiftAxis;
for index = 1:length(freqVector)
shiftedData(index,:) = moveHRTF(fullCoords,data.freqData(index,:),freqVector(index),axis,offsetShift);
end
data = shiftedData;
end
function [data] = moveHRTF(s, data, frequency, axis, offset)
% the offset is given in m
result.freqData = result.freqData .* exp(1i.*add_phase_L');
origAxis = s.r;
if (size(data,2) > size(data,1))
data = data.';
end
offset = real(offset); % ??
switch axis
case 'x'
s.x = s.x + offset;
case 'y'
s.y = s.y + offset;
case 'z'
s.z = s.z + offset;
end
result.channelCoordinates.r = target_d;
newAxis = s.r;
k = 2*pi*frequency/340;
% the phase is moved by the difference of the axis points
data = data .* exp(1i*k*(newAxis - origAxis));
% amplitude manipulation did not yield better results
% data = data .* newAxis ./ origAxis;
end
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment