Traditional wireless positioning systems rely on a strict geometric requirement: a direct, unobstructed line-of-sight (LOS) between the target and the sensors. When an obstacle blocks this path, standard localization algorithms fail. Historically, signals that bounce off walls, ceilings, or surrounding objects have been treated as destructive interference—multipath noise to be filtered out and discarded, as noted by Wen et al. and Turpin et al..
A powerful paradigm shift turns this problem on its head. Instead of treating multipath reflections as noise, modern non-line-of-sight (NLOS) localization frameworks treat them as virtual anchors. By understanding the geometry of the environment, a single physical sensor can exploit these reflections to "see" around corners, mapping hidden spaces and tracking occluded targets with remarkable precision.
The Geometry of a Reflection
To understand why multipath exploitation works, imagine looking at a lightbulb reflected in a flat mirror. To your eyes, the light appears to originate from a "virtual bulb" located at an equal distance behind the mirror's surface. If you know where the mirror is, measuring the angle and arrival time of the reflected light allows you to calculate exactly where the real bulb is, even if a solid wall blocks your direct view of it.
Real Target (Hidden)
o
\
\
============X============ Reflecting Wall
\
\
o Virtual Anchor
In wireless and acoustic sensing, physical surfaces act exactly like these mirrors. As Kadambi et al. demonstrate in their work on Neural RF SLAM, modeling signal reflections as virtual transmitters located behind walls provides enough geometric constraints to enable precise localization using only a single physical antenna.
Instead of requiring multiple physical base stations to triangulate a position, the environment itself provides the necessary infrastructure. Pfeiffer et al. leverage this concept by identifying lines of reflection from environment maps, treating those reflection points as virtual sources within an Extended Kalman Filter to track millimeter-wave (mmWave) emitters when direct paths are completely blocked.
This geometric mapping is not limited to perfect, flat laboratory mirrors. In real-world autonomous driving scenarios, Scheiner et al. show that deep neural networks can interpret chaotic radar reflections from irregular relay surfaces—such as rows of parked cars or textured building facades—to accurately track the positions and velocities of pedestrians hidden around blind street corners.
Decoding the Multipath Spectrum
To transform a chaotic jumble of echoes into distinct virtual anchors, systems must resolve the exact timing and angle of incoming signals. This is achieved by analyzing the fine-grained metadata of a wireless or acoustic channel.
Channel Impulse Response (CIR) and State Information (CSI)
When a signal is emitted, it arrives at the receiver as a series of delayed, attenuated spikes. This profile is captured via the Channel Impulse Response (CIR) or Channel State Information (CSI).
- Acoustic Rhythms: Wang et al. exploit this in their MultiResp system by scanning every individual path within an acoustic CIR for periodic variations. Because human breathing causes minute, rhythmic movements in the chest, the specific path bouncing off a person's back or chest carries a breathing signature. By treating environmental reflections as redundant data streams, their subsequent work, OmniResMonitor, maintains continuous physiological monitoring even when a subject faces away from the sensor or sits behind furniture (Wang et al.).
- Radio Frequency Fingerprints: In the radio domain, human movement alters the local multipath field. Yao et al. demonstrate that passive, battery-free RFID tags can map these environmental disturbances into unique signal fingerprints, enabling device-free tracking without direct LOS. For wider-bandwidth applications, Santoro et al. utilize Ultra-Wideband (UWB) CIR disturbances paired with particle filters to track tag-less targets at sub-meter accuracy.
Mathematical Un-Winding and Wavefronts
If the environment's geometry is known, algorithms can mathematically "un-wind" or back-propagate the received signals along their reflected trajectories. An et al. show that measuring the similarity of these back-propagated acoustic signals reveals their true point of origin. By incorporating these higher-order reflections into inverse ray-tracing models, the authors demonstrate a 40% improvement in 3D sound source localization accuracy over traditional LOS-only baselines (An et al.).
[Receiver] <---- (Direct Path Blocked by Obstacle) ----X [Hidden Source]
^ /
\ /
\___________ [Reflected Path via Wall] ____________/
As wireless systems move toward higher frequencies and massive antenna arrays, the physical nature of the signal wavefront changes. Gao et al. demonstrate that in 6G ultra-massive MIMO systems, signals operate in the "hybrid-field," where wavefronts are distinctly spherical rather than flat. By extracting geometric information directly from these spherical wavefronts, environmental scatterers are instantly transformed into high-precision virtual anchors, enabling sub-decimeter localization without requiring a dominant direct path.
Turning Noise into an Asset
The core philosophy of multipath exploitation is that complexity is a feature, not a bug. Confined or cluttered environments—traditionally considered worst-case scenarios for localization—actually provide the richest datasets.
Consider a highly reflective concrete tunnel. For standard automotive radar, a tunnel creates a chaotic cloud of "ghost points"—false detections born from signals bouncing off the ceiling and walls. However, He et al. show that by modeling the tunnel's reflection geometry, these ghost points can be deterministically remapped to their true points of origin. This turns the tunnel into a giant distributed sensor array, allowing a single mmWave radar to detect and localize vehicles hidden far beyond the line-of-sight of expensive LiDAR or camera systems.
Similarly, in through-wall radar imaging (TWRI), signals bouncing off multiple interior walls generate ghost artifacts. Rather than deploying complex filters to suppress them, Deng et al. construct a double-reflection imaging dictionary that treats secondary bounces as informative signal components. This leverages multiple perspectives from a single transmission, significantly boosting the overall detection energy and clarity of the layout map.
This concept extends naturally across diverse physical mediums:
- Underwater Acoustics: Weiss et al. introduce a semi-blind localization method that uses a multi-ray propagation model and basic depth knowledge to pinpoint underwater sources via surface-and-bottom bounces when the direct path is obstructed.
- Sonar Arrays: Liu et al. employ a "Virtual Station" geometric framework to convert complex underwater multipath propagation into a simplified line-of-sight problem, drastically reducing the number of signal snapshots required for high-precision target localization.
- Terahertz (THz) Communications: In next-generation 6G networks, Chaccour et al. show that non-line-of-sight THz reflections can be parsed via tensor decomposition to map physical obstacles in real time, creating a predictive blockage map that allows the network to proactively manage its directional beams.
Programmable Virtual Anchors
The ultimate evolution of this concept moves from passively exploiting existing reflections to actively controlling them. This is achieved through Reconfigurable Intelligent Surfaces (RIS)—metasurfaces composed of thousands of passive, software-controlled reflecting elements that can alter the phase, amplitude, and direction of incoming radio waves.
[Base Station] ------> [ RIS (Programmable Mirror) ]
\
\ (Controlled Reflection Angle)
v
[Hidden Target]
As detailed by Strinati et al. in the RISE-6G perspective, RIS technology shifts the paradigm from treating the physical world as a static "black box" to treating it as a programmable "Wireless Environment as a Service." By dynamically manipulating multipath propagation, an RIS acts as a customizable, high-precision virtual anchor, routing signals around obstacles and providing centimeter-level 3D positioning in dense industrial or urban settings.
To make this computationally efficient for fast-moving targets, such as vehicle platoons, Wang et al. model these non-line-of-sight reflections using clustered burst sparsity. This mathematical framework allows tracking systems to rapidly isolate and utilize RIS-assisted multipath signals, refining position estimates on the fly.
Once the signal interacts with the surface, high-resolution subspace processing can be applied. Liu et al. demonstrate that by applying algorithms like MUSIC (Multiple Signal Classification) to the temporal-domain reflections of a passive RIS, a base station can resolve and localize multiple targets in close angular proximity, even when they are completely hidden from the base station's direct view.
By treating the environment not as an adversary that degrades signals, but as a distributed network of virtual anchors, modern localization systems achieve what was once considered impossible: clear, high-fidelity spatial awareness without a direct line-of-sight.
Key Foundational Work
Dude, where's my card?: RFID positioning that works with multipath and non-line of sight (2013) — contributed 1 concept elements to this direction.
Discovering the Structure of a Planar Mirror System from Multiple Observations of a Single Point (2013) — contributed 1 concept elements to this direction.
Survey of Channel and Radio Propagation Models for Wireless MIMO Systems (2007) — contributed 1 concept elements to this direction.