We present a new data-driven waveform model for spin-aligned, eccentric binary black hole waveforms, constructed using a Fourier analysis network (FAN) architecture. Compared to conventional multilayer perceptron (MLP) architecture, the FAN architecture achieves $2.55\times$ higher accuracy under the same parameter budget and reduces training time by 7%. Its trigonometric activation functions embed Fourier features into the network, mitigating MLPs’ spectral bias and enabling more accurate modeling of oscillatory waveform structures. Our model supports mass ratios $q<8$, eccentricities $e<0.5$, aligned spins $|{\chi }_{1,2}|<0.99$, and relativistic anamaly $\zeta \in [0,2\pi ]$, covering $\{(\ell ,m)=(2,2),(2,1),(3,3),(3,2),(4,4),(4,3)\}$ spin-weighted spherical harmonic modes. For (2,2)-mode, the median mismatch of our model is 1e-3 under advanced Laser Interferometer Gravitational Wave Observatory sensitivity, and the generation speed is $\sim 5\mathrm{ms}$, which is $50\times$ faster than seobnrv5ehm. In the low-mismatch region ($e<0.42$, $q<6.5$, $\zeta \in [0,2\pi ]$, and $|{\chi }_{1,2}|<0.8$), the maximum mismatch for the (2,2) mode is below 1.3%. Compared to existing eccentric surrogate or data-driven waveform models, it achieves broader parameter coverage while maintaining a comparable level of accuracy (mismatch $\sim {10}^{-3}$). These results demonstrate that the proposed FAN-based model offers a reliable and efficient framework for eccentric waveform modeling, with the potential to accelerate gravitational wave data analysis.

Close

PrevNext

View figure in article

• Gravitational wave sources
• Gravitational waves
• Binary stars