Abstract
The best precision that can be achieved to estimate the location of a stellar-like object is a topic of permanent interest in the astrometric community. We analyse bounds for the best position estimation of a stellar-like object on a CCD detector array in a Bayesian setting where the position is unknown, but where we have access to a prior distribution. In contrast to a parametric setting where we estimate a parameter from observations, the Bayesian approach estimates a random object from observations that are statistically dependent on the position. We characterize the Bayesian Cramer-Rao (CR) that bounds the minimum mean square error (MMSE) of the best estimator of the position of a point source on a linear CCD-like detector, as a function of the properties of detector, the source, and the background. We quantify and analyse the increase in astrometric performance from the use of a prior distribution of the object position, which is not available in the classical parametric setting. This gain is shown to be significant for various observational regimes, in particular in the case of faint objects or when the observations are taken under poor conditions. Furthermore, we present numerical evidence that the MMSE estimator of this problem tightly achieves the Bayesian CR bound. This is a remarkable result, demonstrating that all the performance gains presented in our analysis can be achieved with the MMSE estimator.
Introduction
In astrometry, determining the exact location of a celestial object on a detector is a fundamental task. Traditionally, this is treated as a parametric estimation problem where the star's position is a fixed but unknown parameter. The classical Cramér-Rao (CR) bound provides the theoretical limit for the variance of any unbiased estimator in this context. However, modern astrometry often leverages existing catalogs (e.g., Gaia, USNO-B1) that provide prior information about an object's position. This paper explores a Bayesian framework, treating the position as a random variable, to quantify the precision gains offered by such prior knowledge.
Methodology
The study utilizes the Bayesian Cramér-Rao (BCR) bound, derived from the Van Trees inequality. The authors model a 1-D linear CCD detector with a Gaussian Point Spread Function (PSF), accounting for both Poisson noise and additive background noise.
The total Bayes-Fisher Information ($I_{Bayes}$) is decomposed into two additive components:
- The average information provided by the new observations ($I_{data}$).
- The information inherent in the prior distribution ($I_{prior}$).
By comparing the BCR bound to the classical CR bound across different signal-to-noise ratios (SNR) and pixel resolutions, the research characterizes the theoretical "gain" provided by the prior.
Key Results
The primary finding is that the BCR bound is strictly lower than or equal to the classical CR bound. The advantage of using a prior is most pronounced when the new observations are of low quality—such as for faint objects or in high-background conditions.
As shown in the figure above, the ratio of the Bayesian to parametric performance bounds (represented as $F/F_{par}$) approaches unity as the prior uncertainty ($\sigma_{priori}$) increases. For small $\sigma_{priori}$, the Bayesian bound is significantly tighter, indicating a massive precision boost. Notably, the relative gain is much larger for small flux values ($F=200$ to $1000$ ADU) compared to large flux values ($F=5000$ to $50000$ ADU), where the data itself provides high information content and the X-axis scale for significant gain is much smaller.
Equivalent Brightness and MMSE Performance
A compelling interpretation of these results is the concept of "equivalent object brightness." A high-quality prior can make a faint star appear as precise as a much brighter one. For instance, a prior with 0.1 arcsec uncertainty can effectively double the perceived brightness of a faint star in terms of positional precision. This is particularly relevant for catalogs like USNO-B1, where incorporating prior data can increase the number of stars meeting a 0.1 arcsec precision threshold by 50% to 90% for faint magnitudes.
Crucially, numerical simulations demonstrate that the Minimum Mean Square Error (MMSE) estimator—the conditional mean—tightly achieves the BCR bound. Unlike the parametric case, where unbiased estimators often fall short of the CR bound, the Bayesian gains are fully realizable in practice.
Conclusion
The Bayesian CR bound serves as a powerful benchmark for the maximum achievable precision in astrometric measurements when prior information is available. By incorporating existing catalog data, astronomers can significantly improve the yield of high-precision positions, particularly for faint objects that would otherwise fall below precision thresholds. Future work will address potential biases in prior information and extend the model to joint estimation of flux and background parameters.
Summary updated