cWB-space: a model-free search for short gravitational-wave signals in LISA data
This paper introduces cWB-space, a pipeline designed to find short or transient gravitational-wave signals in data from the Laser Interferometer Space Antenna (LISA) without assuming a precise waveform in advance. The method looks for concentrated patches of power in time and frequency, then checks how those patches appear across LISA’s different detector outputs and reconstructs parts of the waveform. In a realistic simulated test called Sangria, the pipeline recovered all six injected massive-black-hole binary signals as the six highest-ranked candidates, and the reconstructed waveforms matched the injections over the time intervals studied.
At a high level the pipeline works in stages. First it forms the usual LISA data combinations by applying time-delay interferometry (TDI) — a way of combining delayed laser measurements to cancel much larger laser noise. It then makes time–frequency maps using the Wilson–Daubechies–Meyer (WDM) transform and whitens those maps with empirically measured backgrounds at each frequency. Bright, connected pixels in these maps become candidate events. The candidates are ranked by how strong they are, how their power is split between two main LISA channels, and how their frequency changes in time. A separate zeta-WDM stage uses a Sagnac-based observable (called the zeta channel) as an additional check for instrumental disturbances.
The authors tested cWB-space on several simulated scenarios. They injected merging massive-black-hole binaries (MBHBs), other prescribed bursts such as sine–Gaussian and white-noise bursts, and a variety of instrumental disturbances into link-level simulated data. For MBHB waveforms they used the phenomxpy tool with IMRPhenomTHM for aligned-spin cases and IMRPhenomTPHM for precessing examples. The simulated instrument used the LISA Instrument model and PyTDI to form observables. In controlled tests the pipeline associated candidates with the injections, compared detector responses, and recovered partial waveforms. In the Sangria challenge it ranked the six injected MBHBs at the top and produced waveform reconstructions that agreed well with the injected signals over the intervals it examined.