In an ideal world, radio signals travel in straight lines at the speed of light. To find a device's location, you would simply measure the Time of Arrival (TOA) from three anchors and find the intersection—a process known as trilateration.
But real-world environments, from dense urban canyons to cluttered factory floors, are far from ideal. Signals bounce off metallic surfaces, penetrate walls, and are blocked by human bodies. These phenomena create Non-Line-of-Sight (NLOS) conditions and multipath interference, where the signal that reaches a receiver is either delayed, attenuated, or arriving from a completely misleading direction. Da Zhang and Feng Xia note that these physical challenges are the primary reason GPS fails indoors, necessitating a new generation of environment-aware localization techniques.
The Geometry of a Broken Signal
When a direct path is blocked, the "first arrival" signal detected by a receiver is often a reflection. Because this reflected path is longer than the true distance, the system "hallucinates" that the target is further away than it actually is.
To combat this, researchers like Nikos Deligiannis, Spiros Louvros, and Stavros Kotsopoulos have proposed "Enhanced" hybrid models. Instead of treating every signal as equally valid, they use Geographical Information Systems (GIS) to weight signals. If a signal's path is known to be blocked by a building, its weight is reduced or it is excluded entirely, reducing urban location errors by up to 40% Optimizing location positioning using hybrid TOA-AOA techniques in mobile cellular networks.
From Geometry to Statistical Robustness
If we cannot guarantee a clear line-of-sight, we must treat the measurement process as a statistical problem rather than a purely geometric one.
Luca Carlino, Michael Muma, and Abdelhak M. Zoubir developed the RD-ML algorithm, which models signal propagation as a two-component Gaussian mixture. This allows the system to adaptively handle both Line-of-Sight (LoS) and NLOS conditions without manual calibration Robust Distributed Cooperative RSS-based Localization for Directed Graphs in Mixed LoS/NLoS Environments.
Similarly, Guangyang Zeng, Biqiang Mu, and Zhiguo Shi 0001 address the non-convexity of range-based optimization. They propose a two-step framework: a computationally efficient initial estimate followed by a single Gauss-Newton iteration. This approach is proven to reach the Cramer-Rao Lower Bound—the theoretical limit of precision—even in noisy, obstructed environments Global and Asymptotically Efficient Localization from Range Measurements.
Learning the Radio Environment
Recent shifts in the field move away from explicit modeling toward data-driven "Channel Charting." Instead of trying to calculate a distance from a signal, these systems learn the unique "fingerprint" of a location.
- Signal as Image: Chen Tian and colleagues treat WiFi Channel State Information (CSI) as video data, using 3D Convolutional Networks (C3D) to extract spatial and temporal features simultaneously Simultaneous Implementation Features Extraction and Recognition Using C3D Network for WiFi-based Human Activity Recognition.
- Physics-Informed Learning: Tao Li introduced Physics-Informed Reinforcement Learning (PIRL). By embedding the physical laws of millimeter-wave propagation into the reward function of a neural network, robotic agents can navigate complex, unseen layouts without needing to be retrained for every new room Generalizable Wireless Navigation through Physics-Informed Reinforcement Learning in Wireless Digital Twin.
- Unsupervised Mapping: Maximilian Stahlke and Bjoern Eskofier use Siamese neural networks to learn "geodesic" distances. This allows the system to map an indoor environment using 5G or UWB signals without requiring massive amounts of labeled ground-truth data Indoor Localization with Robust Global Channel Charting: A Time-Distance-Based Approach.
Cooperative and Hybrid Architectures
When a single sensor fails, fusion provides a safety net. In industrial settings, where metal machinery creates a "nightmare" for radio signals, researchers are tightly coupling different sensing modalities.
Haifeng Zhang and Zhitian Li demonstrated a Leader-Follower architecture for AGVs that fuses Visual-Inertial Odometry (VIO) with Ultra-Wideband (UWB). When the UWB signal is spiked by an obstruction, the VIO maintains the trajectory, and when the camera is blinded by poor lighting, the UWB provides the global anchor LEADER-FOLLOWER COOPERATIVE LOCALIZATION BASED ON VIO / UWB LOOSE COUPLING FOR AGV GROUP.
For large-scale networks, the Peregrine system developed by Bryan Teague, Florian Meyer, and Moe Z. Win uses scalable Bayesian inference. It intelligently decides when and with whom to communicate, allowing a network of low-cost UWB devices to maintain sub-meter accuracy across multi-floor buildings Peregrine: Network Localization and Navigation with Scalable Inference and Efficient Operation.
The Role of Intelligent Surfaces
A radical new direction involves changing the environment itself. Qingqing Wu and Wen Chen explore Intelligent Reflecting Surfaces (IRS). By placing semi-passive sensors on walls that can "steer" radio reflections, researchers can effectively "see around corners," turning the very obstacles that cause NLOS errors into infrastructure that improves localization accuracy Intelligent Reflecting Surface Assisted Localization: Performance Analysis and Algorithm Design.