pith. sign in

arxiv: 2604.19723 · v3 · pith:D3QNHAMAnew · submitted 2026-04-21 · 📡 eess.SP

Soft-Coherent Direct Multipath SLAM

Benjamin J. B. Deutschmann , Klaus Witrisal , Erik Leitinger This is my paper

Pith reviewed 2026-05-10 01:24 UTC · model grok-4.3

classification 📡 eess.SP
keywords coherent multipath SLAMD-MIMOXL-MIMOdirect localizationphase coherenceBayesian inferencesurface feature vectorraw RF signals
0
0 comments X

The pith

A shared complex mean in the likelihood function enables coherent fusion for direct multipath SLAM in distributed MIMO systems.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper develops a Bayesian method that performs simultaneous localization and mapping directly from raw radio signals by building and using a map of the propagation environment. It addresses degraded performance in indoor and urban settings with heavy multipath and blocked line-of-sight by maintaining phase coherence across distributed base stations or subarrays. The approach uses a likelihood function that shares a complex mean for coherent processing while the variance term handles noncoherent power, integrated with a surface feature vector model that supports near-field effects and visibility. This joint inference aims for high-accuracy localization and scalable processing via GPU parallelism. Simulations show it outperforms noncoherent baselines and approaches the posterior Cramér-Rao lower bound.

Core claim

The central claim is that a phase-preserving nonzero-mean Type-II likelihood function, with a complex mean shared across base stations or subarrays, enables coherent data fusion in direct multipath SLAM for D-MIMO and XL-MIMO systems; when combined with a surface feature vector-based model for environment mapping, it supports robust joint inference of user position and propagation map directly from raw RF signals, with a GPU-parallel implementation for scalability.

What carries the argument

The phase-preserving nonzero-mean Type-II likelihood function that shares a complex mean across distributed base stations or subarrays to enable coherent fusion while variance captures noncoherent signal power, integrated with a surface feature vector (SFV)-based model for map feature fusion across the infrastructure.

If this is right

  • The method achieves performance gains over existing noncoherent multipath SLAM approaches in simulations.
  • Localization accuracy approaches the corresponding posterior CRLB under coherent conditions.
  • High-resolution sensing and localization become feasible using coherent distributed arrays in challenging propagation environments.
  • The GPU-parallel implementation supports scalable processing across large antenna arrays, potentially enabling real-time operation.
  • The surface feature vector model incorporates near-field propagation and visibility effects into the joint map and localization inference.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • If practical phase synchronization across distributed arrays can be achieved in hardware, the approach could extend robust localization to real-world GPS-denied indoor and urban deployments.
  • The coherent fusion technique may connect to array processing methods in radar and communications for combined sensing, mapping, and communication tasks.
  • Hardware experiments with synchronized distributed subarrays would provide a direct test of whether the simulated gains over noncoherent methods hold in physical channels.

Load-bearing premise

Phase coherence is maintained across base stations or subarrays, enabling the shared complex mean in the likelihood function for coherent fusion.

What would settle it

A set of simulations or measurements in which phase coherence across base stations is lost or deliberately broken, checking whether the reported performance gains over noncoherent methods and the approach to the posterior CRLB disappear.

Figures

Figures reproduced from arXiv: 2604.19723 by Benjamin J. B. Deutschmann, Erik Leitinger, Klaus Witrisal.

Figure 1
Figure 1. Figure 1: Factor graph representing the joint posterior PDF from (23). Light blue boxes indicate [PITH_FULL_IMAGE:figures/full_fig_p006_1.png] view at source ↗
Figure 3
Figure 3. Figure 3: Position RMSE ∥pn − pˆn∥ vs. PEB evaluated on synthetic data (left). Cumulative frequency of the position RMSE (right). antennas equally spaced at λ/2. This leads to an observation length Nz = 160. The environment in [PITH_FULL_IMAGE:figures/full_fig_p011_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Mapping error ∥p sfv k,n − pb sfv s,n∥ for k ∈ {1 . . . 4} (left to right) vs. MEB evaluated on synthetic data of experiment 1. Estimates pb sfv s,n and ground truth p sfv k,n were associated using the Hungarian method [53], missed detections and false alarms not evaluated. 0 1 2 3 4 5 Num. paths a) j = 1 b) j = 2 50 100 150 200 0 1 2 3 4 c) j = 3 Step n Num. paths 50 100 150 d) j = 4 Step n true num. vis.… view at source ↗
Figure 5
Figure 5. Figure 5: True number of visible paths in experiment 2 vs. the sum of [PITH_FULL_IMAGE:figures/full_fig_p012_5.png] view at source ↗
read the original abstract

Challenging indoor and urban environments with severe multipath propagation and obstructed line-of-sight degrade classical radio positioning. Multipath-based simultaneous localization and mapping (MP-SLAM) addresses this by building and exploiting propagation maps for robust localization. Emerging distributed multiple-input multiple-output (D-MIMO)/extremely large-scale MIMO (XL-MIMO) infrastructures provide large spatial apertures and high-resolution sensing, especially when phase coherence is maintained across base stations, subarrays, or distributed arrays. We propose a scalable Bayesian direct MP-SLAM method for coherent data fusion in D-MIMO/XL-MIMO systems that jointly infers the environment while performing robust, high-accuracy localization directly from raw radio signals. While commonly used zero-mean Type-II likelihood functions inherently lead to noncoherent processing across distributed arrays and thus to aperture loss, the proposed phase-preserving nonzero-mean Type-II likelihood shares a complex mean across distributed arrays. This enables coherent fusion and preserves the distributed aperture gain, while the variance captures noncoherent signal power. The method is combined with a surface model that enables map-feature fusion across the distributed infrastructure and supports near-field propagation and visibility effects. Bayesian inference is performed using belief propagation by means of the sum-product algorithm on a factor graph with particle-based messages. Parallelizing over particles and arrays, the GPU-accelerated implementation achieves millisecond-level runtimes even in large or distributed infrastructures. Simulation results show that the proposed method achieves performance gains over existing noncoherent methods and approaches the corresponding posterior CRLB, highlighting the potential of coherent processing for high-resolution sensing and localization.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The manuscript proposes a scalable Bayesian direct multipath SLAM (MP-SLAM) method for coherent data fusion in D-MIMO/XL-MIMO systems. It introduces a phase-preserving nonzero-mean Type-II likelihood function sharing a complex mean across base stations or subarrays to enable coherent fusion, combined with a surface feature vector (SFV) model that supports map feature fusion, near-field propagation, and visibility effects. A GPU-parallel particle implementation is described for scalability across distributed infrastructure. Simulation results are presented showing performance gains over noncoherent baselines and proximity to the corresponding posterior CRLB (PCRLB).

Significance. If the simulation outcomes are robust, the work demonstrates the value of phase coherence in distributed arrays for high-resolution RF sensing and localization in multipath environments. The coherent likelihood construction and SFV-based distributed mapping provide a principled way to fuse raw signals across infrastructure while handling near-field effects, with the parallel implementation addressing practical scalability. This could inform design of emerging XL-MIMO systems for indoor/urban applications.

major comments (2)
  1. [Simulation results] Simulation results section: The central performance claims (gains over noncoherent methods and approach to PCRLB) rest on simulations, but the manuscript lacks reported details on the number of Monte Carlo trials, error bars or variance estimates, and any data exclusion rules. These are load-bearing for assessing whether the results reliably support the coherent-fusion advantage.
  2. [Likelihood construction] Likelihood construction (around the nonzero-mean Type-II model): The shared complex mean enables coherent fusion only under maintained phase coherence across BSs/subarrays; the paper should include a sensitivity analysis or bound on phase synchronization errors, as this assumption directly underpins the reported performance gap versus noncoherent baselines.
minor comments (2)
  1. [Abstract] The abstract and introduction could explicitly state the array sizes, number of BSs, and carrier frequency used in the simulations to allow readers to contextualize the reported gains.
  2. [SFV model] Notation for the SFV model and visibility effects could be clarified with a small diagram or explicit mapping to the likelihood parameters.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the positive assessment and constructive comments, which will help improve the manuscript. We address each major comment below.

read point-by-point responses

  1. Referee: Simulation results section: The central performance claims (gains over noncoherent methods and approach to PCRLB) rest on simulations, but the manuscript lacks reported details on the number of Monte Carlo trials, error bars or variance estimates, and any data exclusion rules. These are load-bearing for assessing whether the results reliably support the coherent-fusion advantage.

    Authors: We agree that these details are essential for evaluating the reliability of the simulation results. In the revised manuscript we will report the exact number of Monte Carlo trials, include error bars (or standard deviation estimates) on all performance curves, and explicitly state any data exclusion criteria applied during post-processing. These additions will be placed in the simulation setup subsection and figure captions. revision: yes

  2. Referee: Likelihood construction (around the nonzero-mean Type-II model): The shared complex mean enables coherent fusion only under maintained phase coherence across BSs/subarrays; the paper should include a sensitivity analysis or bound on phase synchronization errors, as this assumption directly underpins the reported performance gap versus noncoherent baselines.

    Authors: The model is derived under the assumption of phase coherence, which is required for the shared complex mean to provide the reported gains. We acknowledge that practical synchronization errors could reduce this advantage. In the revision we will add a short analytical bound on tolerable phase mismatch (derived from the likelihood function) together with a brief discussion of its implications; a full Monte-Carlo sensitivity study is left for future work as it would exceed the scope of a minor revision. revision: partial

Circularity Check

0 steps flagged

No significant circularity detected in derivation chain

full rationale

The paper's core derivation uses a standard Bayesian framework with a nonzero-mean Type-II likelihood for coherent fusion across distributed arrays and an SFV model for map features and near-field effects. These are combined with GPU-parallel particle filtering and benchmarked against the independent posterior CRLB. No steps reduce predictions to fitted parameters by construction, no self-definitional loops appear, and no load-bearing self-citations or imported uniqueness theorems are quoted that collapse the central claims. The approach is self-contained relative to external theoretical bounds and simulation comparisons.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

Limited information from abstract only; method rests on domain assumptions about coherence and feature modeling with no free parameters or invented entities explicitly listed.

axioms (2)
  • domain assumption Phase coherence maintained across BSs or subarrays
    Required for shared complex mean in likelihood to enable coherent fusion
  • domain assumption SFV-based model accurately represents map features, near-field propagation, and visibility effects
    Used to support feature fusion across distributed infrastructure

pith-pipeline@v0.9.0 · 5580 in / 1279 out tokens · 44588 ms · 2026-05-10T01:24:43.543361+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.