Overview
========

DeepExtractor is a deep learning framework for reconstructing transient noise artefacts
(*glitches*) and gravitational-wave signals in LIGO detector strain data.

The key idea
------------

LIGO strain data contains both astrophysical signals and instrumental glitches — short-duration
noise transients that can mimic or obscure real gravitational-wave events. DeepExtractor addresses
this by framing glitch reconstruction as a **supervised denoising problem**:

Given a 2-second stretch of whitened strain containing a glitch, the model learns to output
a clean background estimate. Subtracting that background from the input recovers the glitch
(or signal) waveform.

.. code-block:: text

    Input:  h(t)  = background noise  +  glitch/signal
    Output: n̂(t)  = predicted background
    Result: ĝ(t)  = h(t) − n̂(t)   ← reconstructed glitch or signal

Architecture
------------

The model is a **U-Net** operating on STFT spectrograms (magnitude + phase). The input strain
is transformed into a 2-channel time-frequency representation, processed by the U-Net encoder-
decoder, then converted back to the time domain via iSTFT. The default model,
``DeepExtractor_257``, uses a 4-level U-Net with feature maps ``[64, 128, 256, 512]`` and
produces 257×257 spectrograms.

Pretrained models
-----------------

Two pretrained variants of ``DeepExtractor_257`` are provided, both fine-tuned on LIGO O3 data:

.. list-table::
   :header-rows: 1
   :widths: 20 40 40

   * - Variant
     - Trained on
     - Best for
   * - ``bilby_noise``
     - Simulated LIGO/Virgo noise (bilby)
     - Simulated data, injection studies
   * - ``real_noise``
     - Real LIGO O3 strain
     - Real LIGO O3 detector data

Weights are downloaded automatically from Hugging Face Hub on first use.

Citation
--------

If you use DeepExtractor in your work, please cite:

.. code-block:: bibtex

   @article{s91m-c2jw,
     title     = {Time-domain reconstruction of signals and glitches in gravitational wave data with deep learning},
     author    = {Dooney, Tom and Narola, Harsh and Bromuri, Stefano and Curier, R. Lyana and Van Den Broeck, Chris and Caudill, Sarah and Tan, Daniel Stanley},
     journal   = {Phys. Rev. D},
     volume    = {112},
     issue     = {4},
     pages     = {044022},
     numpages  = {24},
     year      = {2025},
     month     = {Aug},
     publisher = {American Physical Society},
     doi       = {10.1103/s91m-c2jw},
     url       = {https://link.aps.org/doi/10.1103/s91m-c2jw}
   }