The accuracy of optical measurements at low light levels is limited by the
quantum noise of the source and by the random nature of the interaction wit
h the measured object. The source noise may be reduced by use of nonclassic
al photon-number squeezed light. We consider the use of two photon-correlat
ed beams (generated, for example, by spontaneous parametric downconversion)
to measure the optical transmittance of an object. The photons of each bea
m obey a random Poisson process but are synchronized in time. One beam is u
sed to probe the abject, and the other is used as a reference providing inf
ormation on the realization of the random arrival of photons at the object.
The additional information available by such measurement may be exploited
to improve the accuracy of the measurement. Various estimators, including t
he maximum-likelihood estimator, are considered, and their performance is e
valuated and compared with the measurement based on a single-beam conventio
nal (Poissonian) source and a maximally squeezed (fixed-photon-number) sour
ce. The performance advantage that is established depends on parameters suc
h as the intensity of the source, the transmittance of the object, the quan
tum efficiency of the detectors, the background noise, and the degree of co
rrelation of the photon numbers in the two beams. (C) 1999 Optical Society
of America [S0740-3232(99)00702-4].