We consider the problem of estimating the bearing of a remote object given
measurements on a particular type of non-scanning radar, namely a focal-pla
ne array. Such a system focuses incoming radiation through a lens onto an a
rray of detectors. The problem is to estimate the angular position of the r
adiation source given measurements on the array of detectors and knowledge
of the properties of the lens. The training data are essentially noiseless,
and an estimator is derived for noisy test conditions. An approach based o
n kernel basis functions is developed. The estimate of the basis function w
eights is achieved through a regularization or roughness penalty approach.
Choosing the regularization parameter to be proportional to the inverse of
the input signal-to-noise ratio leads to a minimum prediction error. Experi
mental results for a 12-element detector array support the theoretical pred
ictions.