Detection noise bias and variance in the power spectrum and bispectrum in optical interferometry

Context. Long-baseline optical interferometry uses the power spectrum and bispectrum constructs as fundamental observables. Noise arising in the detection of the fringe pattern gives rise to b oth variance and biases in the power spectrum and bispectrum. Previous work on correcting the biases and estimating the variances for these quantities typically includes restrictive assump tions about the sampling of the interferogram and/or about the relative importance of Poisson and Gaussian noise sources. Until now it has been diffi cult to accurately compensate for systematic biases in data which violates these assumptions. Aims. We seek a formalism to allow the construction of bias-free estimators of the bispectrum and power spectrum, and to estimate their variances, under less restrictive conditions which i nclude both unevenly-sampled data and measurements affected by a combination of noise sources with Poisson and Gaussian statistics. Methods. We used a method based on the moments of the noise distributions to derive formulae for the biases introduced to the power spectrum and bispectrum when the complex fringe amplitude is derived from an arbitrary linear combinations of a set of discrete interferogram measurements. We simulated interferograms with different combinations of photon noise and read noise and with different fringe encoding schemes to illustrate the effects of these biases. Results. We have derived formulae for bias-free estimators of the power spectrum and bispectrum which can be used with any linear estimator of the fringe complex amplitude. We have demonstrated the importance of bias-free estimators for the case of t he detection of faint companions (for example exoplanets) using closure phase nulling. We have derived formulae for the variance of the power spectrum and have shown how the variance of the bispectrum could be calculated. Conclusions. The bias on the power spectrum and bispectrum can be removed using the correct choice of estimator based on an understanding of the statistics of the noise.