STeP-C: Real-Time Neural Compression for Distributed Fiber-Optic Sensing at Scale
A live fiber-optic sensing test cut file size by 98.85% while processing 535 MB/s—above the stream’s 324 MB/s requirement.
The authors report 31.04 dB full-band PSNR on a live 1 kHz, 81 km distributed acoustic sensing stream. They identify entropy coding as the main streaming bottleneck and accelerate it alongside GPU inference. The model also handled unseen channel ranges within the same acquisition system without retraining. Downstream detection retained 96.62% high-confidence class agreement and identical sets of detected event types.
ArXiv · AI/CL/LG's note
The authors report 31.04 dB full-band PSNR on a live 1 kHz, 81 km distributed acoustic sensing stream. They identify entropy coding as the main streaming bottleneck and accelerate it alongside GPU inference. The model also handled unseen channel ranges within the same acquisition system without retraining. Downstream detection retained 96.62% high-confidence class agreement and identical sets of detected event types.
ArXiv · AI/CL/LG's note
score 4