Presuming have the Stat Toolbox, ninfit makes it pretty simple directly...
If you write
function dhat=model(coef,x)
% some convenient internal shorthand definitions...
a=coef(1);
b=coef(2);
X=x(:,1);
Y=x(:,2);
% predicted model
dhat=a*exp(b./x).*y;
then
coeff=nlinfit([X,Y],D1,@model,coeff0);
where coeff0 is an initial guess/estimate for the coefficients a and b
Alternatively, the above form can be linearized by log transformation into
ln(A) + B/X + ln(Y)
which can be solved by OLS for coefficients in transformed space.
