A Sigma Point-based Low Complexity Algorithm for Multipath-based SLAM in MIMO Systems

Anna Masiero, Alexander Venus, and Erik Leitinger · 2025

Read A Sigma Point-based Low Complexity Algorithm for Multipath-based SLAM in MIMO Systems by Anna Masiero, Alexander Venus, and Erik Leitinger (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2503-15286.txt.

Introduction

Anna Masiero, Alexander Venus, and Erik Leitinger present a 2025 paper on a cheaper algorithm for locating radios by the echoes of their signals. In multipath mapping, reflections off flat walls act like mirror copies of the base station, and tracking them maps both the device and the room. Particle methods do this flexibly but need huge particle counts in high dimensions.

Their alternative uses sigma points, a small set of carefully placed samples that carry the uncertainty through the model. The swap cuts computation sharply while keeping accuracy respectable. The method serves indoor positioning for future wireless networks, where echoes are plentiful and processors on devices are weak.

Catalog record

CREATORAnna Masiero, Alexander Venus, and Erik Leitinger
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE63 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2503-15286

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

A Sigma Point-based Low Complexity Algorithm for Multipath-based SLAM in MIMO Systems Thanks: This project was funded by the Christian Doppler Research Association. Anna Masiero, Alexander Venus, and Erik Leitinger Abstract mpslam ( mpslam ) is a promising approach in wireless networks to jointly obtain position information of transmitters/receivers and information of the propagation environment. mpslam models specular reflections at flat surfaces as va , which are mirror images of base stations. Particle-based methods offer high flexibility and can approximate posterior probability density functions of the mobile agent state and the map feature states, (i.e., va states) with complex shapes. However, they often require a large number of particles to counteract degeneracy in high-dimensional parameter spaces, leading to high computational complexity. Conversely using an insufficient number of particles leads to reduced estimation accuracy. In this paper, we introduce a low-complexity mpslam algorithm using a sp ( sp )-based implementation of the spa ( spa ). We model the messages of continuous states of the agent and the va as Gaussian distributions and approximate nonlinearities via sp -transformations. This approach substantially reduces the computational complexity without decreasing accuracy. Since probabilistic data association yields Gaussian mixtures for the agent and VA states, we use moment matching to combine each mixture into a single Gaussian. Numerical results using synthetic and real data demonstrate that our method achieves significantly reduced computational runtimes compared to particle-based schemes, while exhibiting comparable (or even superior) localization and mapping performance. I Introduction Emerging sensing and signal processing techniques that exploit multipath propagation promise advanced capabilities in autonomous navigation, asset localization, and situational awareness for future communication networks. mpslam effectively tracks mobile transmitters or receivers while mapping the environment in wireless systems by modelling specular reflections of RF signals as va , which are mirror images of bs or static transceivers called pa (see Fig. 1 ) [ 1 , 2 , 3 , 4 , 5 ] . mpslam falls under the umbrella of feature-based SLAM approaches, which focus on detecting and mapping distinct environmental features [ 6 , 7 ] . mpslam facilitates a fg ( fg )-based representation of the joint posterior density and uses the spa to solve the mpslam problem in a Bayesian manner. It allows to solve the pda ( pda ) problem inherent to mpslam with high scalability and was shown to offer a superior trade-off between robustness and runtime [ 3 , 4 , 8 , 9 ] . mpslam has been successfully applied to a variety of different scenarios, including cooperative localization [ 10 ] , the use of adaptive map feature models [ 11 ] , and environments that involve reflections from rough surfaces [ 12 ] . Most mpslam methods use particle-based implementations [ 13 ] to represent the joint posterior distribution [ 4 , 8 , 9 , 3 ] . Particle-based methods offer high flexibility and can provide an asymptotically optimal approximation of posterior pdf with complex shapes. This property is particularly useful for highly nonlinear and reduced information scenarios, such as toa ( toa )-only mpslam , where the inherent physics of the problem can induce strongly non-Gaussian pdf [ 4 ] . While the factor graph-based approach to mpslam allows for significant reduction of the problem complexity, it typically still requires a high number of particles to counteract particle degeneracy in high-dimensional parameter spaces, leading to high runtimes; conversely using too few particles leads to reduced estimation accuracy. In mimo ( mimo ) systems, array measurements enable jointly estimating toa , aoa ( aoa ) and aod ( aod ). The additional information contained in aoa and aod estimates can yield unambiguous measurement transformations, allowing the resulting joint posterior pdf to be approximated accurately by Gaussian densities. A popular method for approximating pdf that arise from nonlinear transformations is the unscented or sp transform [ 14 , 15 , 16 ] , which has been shown to offer superior approximation performance compared to first-order Taylor linearization employed by kf ( kf )-type methods. Fig. 1 : Exemplary indoor environment including the mobile agent at position \bm{p}_{n} , a pa at position \bm{p}^{(1)}_{\text{A}} and two corresponding va at position \bm{p}_{1,\text{va}}^{(1)} and \bm{p}_{2,\text{va}}^{(1)} . The visualization includes the array geometry used by the agent and pa , along with the geometric relationships between the objects. In this paper, we propose an sp -based implementation of the spa algorithm for mpslam . By approximating all pdf using sp , we efficiently evaluate the integrals required by the algorithm. We describe in detail the steps involved in this approximation, emphasizing the handling of nonlinearities in both the state transition and the measurement models, and discuss the use of moment matching to approximate Gaussian mixtures arising from data association. The main contributions of this paper are as follows. • We propose a novel sp -based implementation of the spa for mpslam leveraging Gaussian approximations of all pdf by means of sp . This approach allows the integrals required by the algorithm to be evaluated very efficiently. • We provide a detailed derivation of all approximated messages of the spa . • We validate our method using simulated data, demonstrating significantly lower runtimes compared to a particle-based implementation, as well as improved accuracy in certain cases. • Using real radio signals, we demonstrate the applicability of the proposed method to real-world scenarios. Notations and Definitions: column vectors and matrices are denoted by boldface lowercase and uppercase letters. Random variables are displayed in san serif, upright font, e.g., \mathsfbr{x} and \bm{\mathsfbr{x}} and their realizations in serif, italic font, e.g. x . f({x}) and p({x}) denote, respectively, the pdf or pmf ( pmf ) of a continuous or discrete random variable \mathsfbr{x} (these are short notations for f{x}({x}) or p{x}({x}) ). (\cdot)^{\mathrm{T}} , denotes the matrix transpose. \lVert\cdot\rVert is the Euclidean norm. |\cdot| represents the cardinality of a set. \bdiag\{\bm{A},\bm{B}\} denotes a block-diagonal matrix with \bm{A} and \bm{B} on the diagonal and zero matrices in the off-diagonal blocks. \bm{I}_{[\cdot]} is an identity matrix of dimension given in the subscript. Furthermore, {1}_{\mathbb{A}}(\bm{x}) denotes the indicator function that is {1}_{\mathbb{A}}(\bm{x})=1 if \bm{x}\in\mathbb{A} and 0 otherwise, for \mathbb{A} being an arbitrary set and \mathbb{R}^{\text{+}} is the set of positive real numbers. The Gaussian pdf is f_{\text{N}}(x;\hat{x},\sigma)=1/(\sqrt{2\pi}\sigma){e}^{(-(x-\hat{x})^{2}/(2\sigma^{2}))} with mean \mu , standard deviation \sigma and the uniform pdf f_{\mathrm{U}}(x;a,b)=1/(b-a){1}_{[a,b]}(x) . A selected list of acronyms is given in Table I . TABLE I : Selection of acronyms CEDA channel estimation and detection algorithm FG factor graph LHF likelihood function MPC multipath component PA physical anchor PVA potential virtual anchor RV random variable SP sigma point SPA sum-product algorithm VA virtual anchor II System Setup and Geometrical Relations At each time step n , we consider a mobile agent located at position \bm{p}_{n}\triangleq[{p}_{\text{x}\hskip 0.5ptn}\;{p}_{\text{y}\hskip 0.5ptn}]^{\text{T}} , equipped with \sqrt{N_{\text{ant}}} times \sqrt{N_{\text{ant}}} uniform planar array (UPA) with N_{\text{ant}} antenna elements spaced distance d_{\text{ant}} apart and oriented at angle \kappa_{n} . Similarly, J bs acting as pa and placed at fixed positions \bm{p}_{\text{A}}^{(j)}\triangleq[{p}_{\text{x\hskip 0.5ptA}}^{(j)}\;{p}_{\text{y\hskip 0.5ptA}}^{(j)}]^{\text{T}} are also equipped with an N_{\text{ant}}^{(j)} -element UPA, with spacing d_{\text{ant}}^{(j)} . A radio signal \bm{r}_{n}^{(j)} transmitted by the mobile agent at carrier frequency f_{\text{c}} and with signal-bandwidth B arrives at the receiver via

Read the complete text