An Investigation of Computational Techniques for Rapid Gravitational Wave Denoising

State-of-the-art LIGO signal analysis emphasizes precise parameter extraction via computationally expensive matched-filtering, which increases latency in gravitational-wave (GW) source localization. As a result, rapid sky-angle determination is rare, hampering follow-up of possible electromagnetic (EM) counterparts to binary black hole (BBH) mergers. Our project explored computational techniques for rapid denoising of LIGO signals, with an eye toward integration into our rapid localization algorithm - NorthStar. We assessed conventional tools (short-time Fourier and wavelet transforms) and found that time–frequency resolution trade-offs and loss of phase information made fast parameter extraction unreliable for transient BBH signals. We then investigated the utility of Convolutional Neural Networks (CNNs) for denoising LIGO strain data. Using PyCBC and GWpy, we trained a simple CNN on simulated BBH chirps and showed it can accurately recover a denoised waveform when tested on the real event GW150914. Future work will integrate these methods into a synchronous localization pipeline to enable real-time EM searches.