DNN for GW Hyperarameter Estimation

Using Depp Learning method to estimate GW parameter space by training DNN with a GW template bank.

Trying to extend the job done in Reference to hyperparameter space.

Data Preparation

1. Generate SMBHB GW signals as templates(~ 3000)
2. obtain real noise time series from LISA noise PSD(Power Spectral Density)
3. Whiten our data using LISA PSD before add multiple realizations of :
   1. White Gaussian Noise
   2. Real LISA Noise
4. Shift signal peak to left at some random time
5. Add multiple noise realizations to shifted(whitened(signal))

DNN Design

We use, at first, the same DNN used in Reference.

Usage

GW_Waveform_Generator

This script can quickly generate GW waveforms for LISA and TianQin using known noise PSD, this section briefly introduces the meaning of each parameter used in the script.

• N: number of templates intend to generate.

• Tc: the time when the signal peaks, can be taken within [0, Tobs].

• Tobs: time of the observation in the unit of year.

• Phic: the chirp phase, in the range of [0, $2\pi$].

• ThetaS: sky location, an angle in the range of [0, 2$\pi$].

• PhiS: sky location, an angle in the range of [0, 2$\pi$].

• Iota: an angle in the range of [0, 2$\pi$].

• Psi: an angle in the range of [0, 2$\pi$].

• Z: redshift, in the range of [0, 10].

• DL: distance from the source, calculated according to the specific cosmology model and redshift Z.

• M1sun: the mass of the primary blackhole, in the unit of solar mass, M1sun > M2sun. $10^6 M_\odot$ < M1sun < $10^9 M_\odot$.

• M2sun: the mass of the minor blackhole.

here are all the parameters you can play with, change the parameters and make your own waveforms.

LISA_Templates_Generator

Dependencies

AstroPy, SciPy, Pandas, Numpy, PyYaml,Matplolib

Config

Use config.yaml to set parameter configration

Config 1

The first objective of this project is to predict the merger time Tc, we set the data length to 3 month and the merger time is between 3 month and 6 month. We will try to predict the Tc with the 3 month long data before merging.

Config 2

The second objective is to estimate the key parameters of the GW which are the chirp mass ($\mathcal{M}_c$) and sky locations $\theta$ and $\phi$. For this case, we set the data length to 2 weeks and the merging happens in this window.

Runtime

After setting up paramters in config.yaml, just use python to run the script python LISA_Templates_Generator.py, it will create a folder named fig to save the generated figures.

Reference

@article{PhysRevD.97.044039,
  title = {Deep neural networks to enable real-time multimessenger astrophysics},
  author = {George, Daniel and Huerta, E. A.},
  journal = {Phys. Rev. D},
  volume = {97},
  issue = {4},
  pages = {044039},
  numpages = {23},
  year = {2018},
  month = {Feb},
  publisher = {American Physical Society},
  doi = {10.1103/PhysRevD.97.044039},
  url = {https://link.aps.org/doi/10.1103/PhysRevD.97.044039}
}