Imagine trying to find your exact location inside a complex building by measuring your distance to a few fixed radio anchors. In an ideal world with a clear Line-of-Sight (LOS), your distance measurements are highly accurate, corrupted only by minor, zero-mean Gaussian thermal noise. Geometrically, each measurement represents a perfect circle around an anchor; with three or more anchors, these circles intersect at a single, unambiguous point.
In the real world, however, walls, machinery, and structural pillars block these direct paths. When a radio signal encounters an obstacle, it must either penetrate the material—which slows it down—or bounce off surrounding surfaces to find an alternate path. This is the Non-Line-of-Sight (NLOS) problem. Because the signal takes a longer time to arrive, the positioning system calculates a distance that is significantly larger than the true geometric distance.
This introduces a severe, positive, non-Gaussian bias. If you feed these corrupted distances into traditional triangulation algorithms, the intersecting circles warp, dragging your estimated position meters away from reality. The core challenge in modern indoor positioning is mitigating these unknown, bounded NLOS errors without having prior knowledge of the environment's layout or the statistical profile of the noise.
The Convex Relaxation Breakthrough
Traditional localization frameworks often fail in obstructed environments because they treat distance measurements as strict equalities. When NLOS conditions inflate these distances, the equations become mathematically inconsistent—the circles no longer intersect at a single point, creating a non-convex optimization landscape riddled with local minima.
To solve this, researchers shifted their perspective from strict geometry to convex relaxation. The key insight is simple yet powerful: because NLOS propagation always delays a signal, the measured distance $\hat{d}_i$ is guaranteed to be an upper bound of the true distance $d_i$. Therefore, instead of trying to force the estimated position $x$ to lie exactly on the perimeter of a circle ($||x - a_i|| = \hat{d}_i$), the problem can be reframed as an inequality:
$$||x - a_i|| \le \hat{d}_i$$
Geometrically, this transforms the search space from the thin, non-convex shells of spheres into solid, convex balls. The intersection of multiple convex balls forms a well-behaved, convex feasible region.
Using this insight, Hongyang Chen et al. developed a robust localization framework in Non-Line-of-Sight Node Localization Based on Semi-Definite Programming in Wireless Sensor Networks. By applying Semi-Definite Programming (SDP), the authors relaxed the non-convex distance constraints into a convex model. This allows the system to utilize all available distance data—including heavily corrupted NLOS measurements—without needing to identify or discard them beforehand. The optimization naturally finds the best fit within the overlapping convex bounds.
Expanding on this convex paradigm, Chao Ding proposed a highly scalable approach in Convex Euclidean distance embedding for collaborative position localization with NLOS mitigation. Instead of relying on standard SDP solvers, which can become computationally prohibitive as the number of sensors grows, the author formulated the problem using Euclidean Distance Matrices (EDM) combined with an alternating direction method of multipliers (ADMM). This mathematical formulation allows the system to scale efficiently to thousands of sensors while remaining completely agnostic to prior error statistics or explicit link classifications.
Intelligent Filtering and Combinatorial Isolation
While convex relaxation absorbs NLOS errors mathematically, another prominent strategy focuses on dynamically isolating or filtering out the corrupted signals before they can skew the position estimate.
One intuitive way to handle unknown errors is to optimize for the worst-case scenario. In Robust Localization Using Range Measurements With Unknown and Bounded Errors, Xiufang Shi et al. introduced a "MinMax-SDP" algorithm. Operating under the assumption that measurement errors are unknown but bounded, their framework minimizes the maximum possible estimation error. This bounded-error approach provides a mathematical guarantee of robustness even when the underlying noise distribution deviates entirely from standard Gaussian models.
Alternatively, if a system receives signals from many anchors, it can treat the problem combinatorially. The authors of An Exhaustive Method of TOA-Based Positioning in Mixed LOS/NLOS Environments, Chengwen He et al., proposed an exhaustive search method that systematically evaluates subsets of base stations. By analyzing the consistency of the resulting position estimates, the algorithm isolates subsets that contain only clean, LOS paths, effectively bypassing the need to estimate the magnitudes of the NLOS biases altogether.
[Anchor 1 (LOS)] ----> \
[Anchor 2 (NLOS)] ---> \ ---> [Hierarchical Voting / Subset Selection] ---> Clean Position Estimate
[Anchor 3 (LOS)] ----> /
[Anchor 4 (LOS)] ----> /
Other researchers leverage statistical filtering and voting mechanisms to achieve resilience:
- Hierarchical Voting: In A Hierarchical Voting Based Mixed Filter Localization Method for Wireless Sensor Network in Mixed LOS/NLOS Environments, Jinquan Hang et al. combined hierarchical voting with a hybrid Unscented Kalman Filter and Particle Filter. The voting layer acts as a coarse filter to reject severe NLOS outliers, while the subsequent filters refine the position based on the remaining measurements.
- Adaptive Covariance Adjustment: In A Double Extended Kalman Filter Algorithm for Weakening Non-Line-of-Sight Errors in Complex Indoor Environments Based on Ultra-Wideband Technology, Sheng Xu et al. utilized a cascaded Double Extended Kalman Filter (DEKF) featuring a residual classification module. When a measurement exhibits a high residual (indicating a likely NLOS path), the filter dynamically inflates that measurement's covariance, reducing its influence on the final state estimate.
- Metadata and Environment Modeling: Rather than relying solely on range data, Hoon Choi et al. demonstrated in Location Information Reliability-Based Precision Locating System Using NLOS Condition Estimation that analyzing satellite metadata, such as Signal-to-Noise Ratio (SNR) and elevation angles, can yield a 97% success rate in detecting NLOS conditions. Similarly, the authors of Location of Moving Targets in Substation Non-Line-of-Sight Environment (Yübo Wang et al.) and A Robust Wireless Sensor Network Localization Algorithm in Mixed LOS/NLOS Scenario (Bing Li et al.) demonstrated that combining weighted least squares with Gaussian Mixture Models (GMM) allows systems to maintain high precision by mapping the complex, multimodal nature of indoor propagation errors.
Bypassing Infrastructure Constraints
The ultimate evolution of this research direction moves away from rigid, pre-mapped anchor networks entirely, addressing scenarios where neither the environment nor the infrastructure coordinates are known in advance.
In PEOPLEx: PEdestrian Opportunistic Positioning LEveraging IMU, UWB, BLE and WiFi, Pierre-Yves Lajoie et al. proposed an opportunistic framework that solves a simultaneous localization and mapping (SLAM) problem for radio signals. By fusing smartphone inertial sensors (IMUs) with opportunistic radio pings via non-linear factor graph optimization, the system simultaneously estimates the user's trajectory and the unknown coordinates of the surrounding signal anchors. This eliminates the deployment bottleneck of surveying anchor locations beforehand, allowing high-precision tracking using minimal, unmapped infrastructure.
[Unknown Anchor Location] <--- (UWB Ranging) ---> [User Trajectory (IMU Dead Reckoning)]
|
(Factor Graph Optimization)
|
v
Simultaneous Estimate of Both States
Taking a physical-layer approach to the problem, Weiguo Wang et al. re-engineered the nature of the signal itself in BIFROST: Reinventing WiFi Signals Based on Dispersion Effect for Accurate Indoor Localization. Instead of treating the environment as an adversary that corrupts signals, the authors utilized leaky wave antennas (LWAs) to introduce a frequency-dependent dispersion effect to WiFi transmissions. This physical modification causes different frequencies to emerge from the antenna at different angles. When these signals bounce through an NLOS environment, the unique frequency signature received by the client allows the system to map the path back to its origin. By exploiting this dispersion, the framework achieves stable performance in obstructed environments, effectively turning the physical complexity of NLOS propagation into a localization advantage.
Key Foundational Work
Non-line-of-sight Node Localization based on Semi-Definite Programming in Wireless Sensor Networks (2012) — contributed 1 concept elements to this direction.
Location Information Reliability-Based Precision Locating System Using NLOS Condition Estimation (2013) — contributed 1 concept elements to this direction.