nai

Project Progress

The Terrestrial Planet Finder (TPF) mission will use low-resolution, low-sensitivity data to detect and characterize extrasolar terrestrial planets. This task focuses on mathematical techniques to maximize the return from TPF data.

Image Reconstruction and Planet Signal Extraction for TPF-I nulling interferometer
Velusamy (with Ken Marsh) has developed an algorithm based on Richardson and Marsh (1983) to reconstruct the image from the SINE/COS chopped output of dual nulling interferometer. This algorithm treats the ensemble of possible images as a Gaussian random process, subject to a positivity constraint, and contains no limitations on the presence of negative lobes in the null response pattern with the chop. The Bayesian procedure maximizes the probability density of the image conditioned on the data. This algorithm will have application in analyzing the spectral features detectable by TPF-I

Using Artificial Neural Networks to Discriminate TPF Spectra

To determine the scientific impact of trade-offs in instrument characteristics (e.g., wavelength coverage, spectral resolution, and signal-to-noise (S/N)) for spectral characterization of terrestrial planets, this task focuses on using Artificial Neural Networks (ANN) to determine instrumentation limits for reliable classification of spectra. As the test set, we have used UV-FIR planetary spectra generated by the Virtual Planetary Laboratory (VPL) and convolved these with slit functions and a simplistic noise model to produce spectra with a range of spectral resolutions and S/N levels. Thousands of ANNs have been run to discriminate these spectra in four wavelength regions, at multiple resolutions (millimeter imaging radiometer (MIR): R = 5 to 100, vis: R = 20-200), and for S/N from 3-100. Multiple versions of the same ANNs have also been run to test for consistency of results. With our simplistic noise model, the ANNs clearly discriminate spectra at very modest resolutions and S/N. Ongoing work involves expanding the test set to other planetary types and working with the TPF design teams to build more realistic noise models.