The detection of Extreme Mass Ratio Inspirals (EMRIs) is challenging due to their complex waveforms, long duration, and low signal-to-noise ratio (SNR). This makes them harder to identify compared to compact binary coalescences. While matched filtering techniques are computationally demanding, existing deep learning methods primarily focus on time-domain data and are limited by data duration and SNR. Additionally, most previous work ignores time-delay interferometry (TDI) and uses approximations that limit their ability to handle laser frequency noise.
To address these limitations, we propose DECODE, an end-to-end model that focuses on detecting EMRI signals in the frequency domain. DECODE utilizes a dilated causal convolutional neural network trained on synthetic data considering TDI-1.5 detector response. This allows DECODE to efficiently process a year’s worth of multichannel TDI data with an SNR of approximately 50.
We evaluate DECODE on 1-year data with accumulated SNR ranging from 50 to 120 and achieve a true positive rate of 96.3% at a false positive rate of 1%. Importantly, DECODE maintains an inference time of less than 0.01 seconds. To provide interpretability and generalization, we visualize three showcased EMRI signals. Our results demonstrate that DECODE has strong potential for future space-based gravitational wave data analysis.