conditioning

Gravitational Wave Data Conditioning

Data conditioning is essential before matched filtering. Raw gravitational wave detector data contains low-frequency noise, instrumental artifacts, and needs proper sampling rates for computational efficiency.

Overview

The conditioning pipeline typically involves:

1. High-pass filtering (remove low-frequency noise below ~15 Hz)
2. Resampling (downsample to appropriate sampling rate)
3. Crop filter wraparound (remove edge artifacts from filtering)
4. PSD estimation (calculate power spectral density for matched filtering)

High-Pass Filtering

Remove low-frequency noise and instrumental artifacts:

from pycbc.filter import highpass

# High-pass filter at 15 Hz (typical for LIGO/Virgo data)
strain_filtered = highpass(strain, 15.0)

# Common cutoff frequencies:
# 15 Hz: Standard for ground-based detectors
# 20 Hz: Higher cutoff, more aggressive noise removal
# 10 Hz: Lower cutoff, preserves more low-frequency content

Why 15 Hz? Ground-based detectors like LIGO/Virgo have significant low-frequency noise. High-pass filtering removes this noise while preserving the gravitational wave signal (typically >20 Hz for binary mergers).

Resampling

Downsample the data to reduce computational cost:

from pycbc.filter import resample_to_delta_t

# Resample to 2048 Hz (common for matched filtering)
delta_t = 1.0 / 2048
strain_resampled = resample_to_delta_t(strain_filtered, delta_t)

# Or to 4096 Hz for higher resolution
delta_t = 1.0 / 4096
strain_resampled = resample_to_delta_t(strain_filtered, delta_t)

# Common sampling rates:
# 2048 Hz: Standard, computationally efficient
# 4096 Hz: Higher resolution, better for high-mass systems

Note: Resampling should happen AFTER high-pass filtering to avoid aliasing. The Nyquist frequency (half the sampling rate) must be above the signal frequency of interest.

Crop Filter Wraparound

Remove edge artifacts introduced by filtering:

# Crop 2 seconds from both ends to remove filter wraparound
conditioned = strain_resampled.crop(2, 2)

# The crop() method removes time from start and end:
# crop(start_seconds, end_seconds)
# Common values: 2-4 seconds on each end

# Verify the duration
print(f"Original duration: {strain_resampled.duration} s")
print(f"Cropped duration: {conditioned.duration} s")