Slope-assisted super-resolution distributed temperature sensing based on a Raman adaptive denoising network

Raman optical time-domain reflectometry (R-OTDR) based distributed temperature sensing (DTS) offers high temperature sensitivity for long-range monitoring but is fundamentally constrained by the spatial-resolution (SR)–temperature-resolution (TR) trade-off: narrowing the pulse width enhances SR while reducing the Raman photon budget, thereby degrading signal-to-noise ratio (SNR) and TR, especially for heating zones shorter than the system SR. To overcome this long-standing limitation, we propose a unified Raman adaptive denoising network (RADNet) with a slope-assisted high-fidelity Raman convolutional simulation model. The simulation model accurately reproduces the physically critical descending-edge slopes of real R-OTDR traces, providing training data with authentic edge characteristics. RADNet improves Raman SNR and temperature precision while retaining slope information required for temperature recovery in sub-spatial-resolution heated sections. Experiments on an 8.2 km multimode fiber demonstrate that, with a 40 ns pulse width corresponding to a system SR of approximately 6.6 m, the proposed method enables temperature recovery for a 20 cm heated section and improves the standard TR from 9.05 °C to 0.52 °C under the 10,000-averaging condition. In addition, an extreme temperature-discrimination test shows that RADNet processed traces can distinguish temperature intervals of 0.2 °C. By simultaneously improving temperature precision and preserving edge information for sub-resolution demodulation without hardware upgrades, this approach alleviates the impact of the SR–TR trade-off, offering a practical pathway toward high-resolution and high-precision Raman DTS for applications such as geothermal monitoring and power-system early warning.