TPF-I SWG Report - Exoplanet Exploration Program - NASA

C HAPTER 4
interferometer. How this information is processed determines such fundamental properties as the proper
geometry of the spacecraft and the size of the telescopes. It became clear early in the design stage that
understanding planetary signal extraction (PSE) would be important in the Pre-Phase A design, if realistic
designs were to be considered.
There is an extensive literature on signal processing from Earth-bound interferometers, both optical and
radio. When TPF-I was first proposed, the problem of detecting a signal was already realized as being
important. The work of the PSE Working Group was to determine whether these techniques were
adequate, and if not, to develop improved methods. In addition, a set of metrics was developed which
was independent of signal processing technique that could be used to evaluate different architectures
without the computationally expensive process of actually testing each proposed architecture with several
different PSE algorithms and many different target solar systems.
The basic performance parameter is SNR. In this effort, the signal is the planetary signal, which is the
output of the phase-chopped interferometer. Noise comes from two sources (the star itself and extraneous
light), and it comes in two ways (simple background and systematic errors). In the studies that we did,
we considered only astronomic sources of background (local zodiacal and exozodiacal light), and we did
not yet study systematic backgrounds and errors (imperfect maintenance of the null and thermal
backgrounds from other spacecraft). The difficulty of the problem is that the SNR is inherently low,
limited by finite observing times and the aperture of the telescopes, and that the most interesting signal
bands are at the high and low wavelengths, where the sensitivity of the interferometer is lowest. The
other fundamental problem is that any interferometer has limited coverage in the u-v plane, so the
information obtained from the planetary system is fundamentally incomplete. This is exacerbated by the
probability that there are multiple planets in the system, causing confusion in the signal. The combination
of low SNR and incomplete information makes identifying a planet and extracting a spectrum a difficult,
but not impossible, challenge. These challenges fundamentally drive the mission design.
The most basic source of noise is the stellar leakage from the finite suppression of the interferometric
null, which is determined by the architecture of the interferometer. For these preliminary studies, we
developed a simulation that included the effects of stellar leakage, exozodical dust, and local zodiacal
background, and that could simulate the light from arbitrary planetary systems. Systematic effects were
not included. This simulation was independently validated by comparison to several other calculations
and simulations.
Many methods have been proposed for extracting the signal from background by different members of the
TPF-I collaboration. In the literature, inconsistent assumptions have been made in testing various
algorithms, so we decided to have a blind comparison of various algorithms using the JPL standard
simulator and a small set of test cases which included solar systems which we thought would be both easy
and difficult.
A group of scientists from JPL, The University of Arizona, Ball Aerospace, and eventually several other
organizations formed a science working group and developed several new algorithms, in addition to a
baseline cross-correlation/CLEAN (CC/CLEAN) algorithm. Each of these was tested against simulated
planetary systems with a realistic SNR. It was found that the CC/CLEAN algorithm did reasonably well,
and improved algorithms were both proposed and implemented. While none of these have yet turned out
88