Cooperative Multi-AP Sensing and Dynamic-Echo Suppression for Static Environment Reconstruction in Cell-Free ISAC
TL;DR
This work combines covariance-informed state fitting with temporal echo estimation from the observations to establish reconstruction gains from the complete pipeline and surface refinement, with a distinct suppression benefit under strong dynamic interference.
Abstract
Distributed access points (APs) provide complementary bistatic views for cell-free integrated sensing and communication (ISAC), but moving-object echoes can obscure static surfaces in the same channel state information (CSI). We combine covariance-informed state fitting with temporal echo estimation from the observations. Their residual CSI means feed a shared multi-AP network with surface supervision and forward measurement consistency. On 24 held-out simulated geometries, the complete pipeline increased mean surface F-score from 0.446 to 0.480 and reduced squared Chamfer distance from 1.409 to 1.136 m2. This pipeline combines expanded training data, surface-oriented fine-tuning and phase averaging. With decoder and inference settings fixed, a controlled +10 dB moving-path test on independently ray-traced channels increased F-score from 0.256 to 0.391 through combined suppression (paired gain 0.1353; 95% interval [0.0960, 0.1729]). In the CSI-domain sensing-error test, combined input retained F-score of 0.4793 under an added velocity-error standard deviation of 8 m/s. A separate spatial-prior study increased F-score from 0.468 to 0.657 through surface refinement. On 24 independent ray-tracing test geometries, receiver adaptation yielded F-score of 0.412 versus 0.369 for a no-CSI occupancy prior (paired gain 0.0427; 95% interval [0.0108, 0.0772]). These results establish reconstruction gains from the complete pipeline and surface refinement, with a distinct suppression benefit under strong dynamic interference.