Amplitude-Independent Robust Snapshot 6-D Radio SLAM via a Uniffed Angle-Delay Formulation

Technical Summary: Amplitude-Independent Robust Snapshot 6-D Radio SLAM via a Unified Angle-Delay Formulation

Problem Statement
The paper addresses the challenge of bistatic snapshot radio Simultaneous Localization and Mapping (SLAM) in cellular mmWave settings. In this scenario, a User Equipment (UE) with an unknown 6-D pose (3-D position, 3-D orientation) and clock bias must be localized, and environmental scattering points reconstructed, using a single multipath channel snapshot. The problem is complicated by mixed Line-of-Sight (LoS) and Non-Line-of-Sight (NLoS) propagation. Specifically, the system must distinguish between valid geometric inliers (LoS paths and single-bounce NLoS-1 paths) and outliers (multi-bounce NLoS-n paths) without prior knowledge of path identity.

Existing robust snapshot methods are largely validated in 2-D/planar settings and often rely on path-amplitude or path-loss information to classify LoS paths. This reliance makes them sensitive to calibration errors and propagation-model mismatches. Furthermore, confusing LoS and NLoS-1 paths can bias state estimates because LoS paths lack associated scattering points.

Methodology
The authors propose an amplitude-independent robust method built on a unified angle-delay formulation that treats LoS and NLoS-1 paths as geometrically consistent inliers without requiring amplitude-based preclassification. The approach consists of three main stages:

1. Unified Angle-Delay Formulation (Coarse Stage):

   • The method derives a geometric constraint directly linking the UE state (position, clock bias, orientation) to angle-delay observations, eliminating the need for path-wise latent variables (e.g., scattering point coordinates) in the initial solver.
   • For a given path $i$, the formulation uses auxiliary vector constructions to express the relationship between the Base Station (BS) and UE. The constraint is defined as $M_i t = d_i(v_i - u_i)$, where $t$ is the BS-to-UE displacement, $d_i$ is the propagation distance (corrected for clock bias), and $u_i, v_i$ are the global direction vectors derived from Angle of Departure (AoD) and Angle of Arrival (AoA).
   • This formulation applies to both LoS (as a degenerate case) and NLoS-1 paths, allowing them to be processed uniformly.

2. Robust Initialization and Outlier Rejection:

   • To handle unknown path identity and NLoS-n outliers, the coarse solver is embedded in a consensus-based minimal-set search.
   • The algorithm generates hypotheses from minimal subsets of measurements ($N_{min}$). For each hypothesis, it performs a feasibility check to reject geometrically inconsistent candidates (e.g., checking collinearity for LoS or same-side consistency for NLoS-1).
   • 3-D Pose Initialization: For general 3-D scenarios, the method employs a twist-swing two-stage traversal to initialize the UE orientation ($SO(3)$). This exploits the prior that UEs are typically upright, separating the dominant yaw (twist) from bounded out-of-plane tilts (swing). Position and clock bias are updated in closed form for each orientation candidate.
   • The best hypothesis is selected based on a truncated cost function (MSAC-style), yielding a coarse inlier set and an initial UE state estimate.

3. Refinement and Model Selection:

   • A Jacobian-row-equilibrated Iteratively Reweighted Least Squares (IRLS) refinement stage follows. This avoids reliance on calibrated path-wise covariance matrices.
   • LoS Detection via Model Selection: Since LoS and NLoS-1 paths have different parameterizations (LoS has no scattering point), the method treats LoS detection as a model selection problem. It evaluates $M+1$ candidate models (one all-NLoS-1 model and $M$ mixed models where one path is LoS) and selects the best fit using the quasi-Akaike Information Criterion (QAIC).
   • This stage jointly refines the UE state, detects the LoS path, and estimates scattering points.

Key Contributions

• Unified Angle-Delay Formulation: A novel coarse-stage formulation written directly in the UE state and angle-delay observations that handles LoS and NLoS-1 inliers without path-wise latent variables or amplitude-based preclassification.
• Amplitude-Independent Robust Initialization: A pipeline embedding the coarse solver in a consensus-based minimal-set search with geometric feasibility checks, supported by a local-rank analysis that determines minimal sample size implications ($N_{min} \ge 4$ for 3-D).
• General 3-D/6-D Extension: An extension to full 6-D pose estimation using a twist-swing traversal for orientation initialization and local refinement on $SO(3)$, treating the planar case as a special instance.
• Robust Refinement with QAIC: A refinement stage combining Jacobian-row-equilibrated IRLS with QAIC-based model comparison to jointly refine the state and detect the LoS path without calibrated path-loss information.

Simulation Results
The method was evaluated via Monte Carlo simulations in both planar and general 3-D scenarios, comparing it against amplitude-dependent baselines (MH-MLE, RSR) and a traditional NLoS-only method.

• Accuracy: The proposed method achieved competitive or superior P90 errors in UE position, orientation, clock bias, and scattering point estimation compared to baselines, particularly in low-to-moderate noise regimes.
• Robustness to Model Mismatch: Under simulated path-loss model uncertainty (calibration mismatch), the proposed method maintained stable accuracy, whereas amplitude-dependent baselines (MH-MLE, RSR) showed significant performance degradation.
• LoS Detection: The method achieved higher LoS detection accuracy than MH-MLE in mixed propagation scenarios with path-loss uncertainty, except when geometric measurements were severely degraded.
• Computation: The method demonstrated stable computation times (~0.52s per trial) across different propagation conditions (LoS, mixed, NLoS), whereas baseline runtimes increased significantly as LoS information became less available.
• 3-D Performance: In 3-D simulations, the two-stage design (Coarse Formulation + Least Squares refinement) proved effective, with the refinement stage significantly reducing errors in orientation and scattering point estimation compared to the coarse estimate alone.

Significance and Claims
The paper claims that the proposed amplitude-independent approach provides a robust solution for snapshot radio SLAM where calibration quality and propagation models may be imperfect. By deferring LoS identification to a model-selection stage after robust geometric initialization, the method avoids the biases introduced by early amplitude-based classification. The authors assert that the unified angle-delay formulation is sufficient to provide reliable initialization and inlier selection, making the system competitive with calibrated amplitude-dependent methods while being more robust to model mismatch. The work also provides theoretical insights into the local-rank properties and minimal sample requirements for this specific formulation under unknown path identity.