function Serr=specerr(S,J,err,trialave,numsp)
% Function to compute lower and upper confidence intervals on the spectrum
% Usage: Serr=specerr(S,J,err,trialave,numsp)
% Outputs: Serr (Serr(1,...) - lower confidence level, Serr(2,...) upper confidence level)
%
% Inputs:
% S - spectrum
% J - tapered fourier transforms
% err - [errtype p] (errtype=1 - asymptotic estimates; errchk=2 - Jackknife estimates;
% p - p value for error estimates)
% trialave - 0: no averaging over trials/channels
% 1 : perform trial averaging
% numsp - number of spikes in each channel. specify only when finite
% size correction required (and of course, only for point
% process data)
%
% Outputs:
% Serr - error estimates. Only for err(1)>=1. If err=[1 p] or [2 p] Serr(...,1) and Serr(...,2)
% contain the lower and upper error bars with the specified method.
if nargin < 4; error('Need at least 4 input arguments'); end;
if err(1)==0; error('Need err=[1 p] or [2 p] for error bar calculation. Make sure you are not asking for the output of Serr'); end;
[nf,K,C]=size(J);
errchk=err(1);
p=err(2);
pp=1-p/2;
qq=1-pp;
if trialave
dim=K*C;
C=1;
dof=2*dim;
if nargin==5; dof = fix(1/(1/dof + 1/(2*sum(numsp)))); end
J=reshape(J,nf,dim);
else
dim=K;
dof=2*dim*ones(1,C);
for ch=1:C;
if nargin==5; dof(ch) = fix(1/(1/dof + 1/(2*numsp(ch)))); end
end;
end;
Serr=zeros(2,nf,C);
if errchk==1;
Qp=chi2inv(pp,dof);
Qq=chi2inv(qq,dof);
Serr(1,:,:)=dof(ones(nf,1),:).*S./Qp(ones(nf,1),:);
Serr(2,:,:)=dof(ones(nf,1),:).*S./Qq(ones(nf,1),:);
elseif errchk==2;
tcrit=tinv(pp,dim-1);
for k=1:dim;
indices=setdiff(1:dim,k);
Jjk=J(:,indices,:); % 1-drop projection
eJjk=squeeze(sum(Jjk.*conj(Jjk),2));
Sjk(k,:,:)=eJjk/(dim-1); % 1-drop spectrum
end;
sigma=sqrt(dim-1)*squeeze(std(log(Sjk),1,1)); if C==1; sigma=sigma'; end;
conf=repmat(tcrit,nf,C).*sigma;
conf=squeeze(conf);
Serr(1,:,:)=S.*exp(-conf); Serr(2,:,:)=S.*exp(conf);
end;
Serr=squeeze(Serr);