Outdoors, global positioning systems (GPS) provide seamless navigation because satellite signals travel mostly unimpeded to a receiver. Indoors, however, physical structures block these satellite signals entirely. As Tian Wang et al. note in Interoperable localization for mobile group users, this forces indoor tracking systems to rely on terrestrial radio frequencies (RF) like WiFi, Bluetooth, or Ultra-Wideband (UWB).
But indoor spaces are cluttered with walls, furniture, and metal pillars. When an RF transmitter emits a signal, the wave does not travel along a single, clean path. Instead, it bounces off every surrounding surface, creating multiple delayed, attenuated copies that arrive at the receiver at slightly different times. This phenomenon is known as multipath propagation, and it is the primary obstacle to achieving high-precision indoor localization and sensing.
Why Multipath Destroys Traditional Ranging
To understand why multipath propagation is so destructive, we must look at how traditional wireless localization algorithms calculate distance. Most systems rely on one of three geometric metrics:
- Received Signal Strength Indicator (RSSI): Measuring how much the signal attenuated to estimate distance.
- Time of Arrival (ToA): Measuring the exact time flight takes from transmitter to receiver.
- Time Difference of Arrival (TDoA): Measuring the relative delay between two synchronized receivers.
In an ideal vacuum, a signal's power drops predictably over distance. Indoors, however, reflected waves collide with the direct-path wave. If a reflection arrives exactly in phase with the direct path, they combine constructively, making the signal appear artificially strong. If it arrives out of phase, they combine destructively, causing a severe drop in power (multipath fading). As Brian Ferris et al. demonstrate in WiFi-SLAM Using Gaussian Process Latent Variable Models, these unpredictable fluctuations break the monotonic relationship between signal strength and physical distance, rendering simple path-loss models highly inaccurate.
Ideal (Line-of-Sight):
Transmitter ───────────────────────────────> Receiver (Single Clean Pulse)
Real Indoor Environment (Multipath):
Transmitter ─────────── Direct Path ───────> Receiver (Overlapping,
└─ Bounce ──> Wall ── Bounce ──> Smeared Waveform)
Time-based systems (ToA and TDoA) suffer from a different manifestation of the same problem. If a reflected path arrives a fraction of a nanosecond after the direct path, the two waveforms overlap in the receiver's circuitry. This smearing hides the true "leading edge" of the direct-path signal. Xiqin Wang et al. emphasize this challenge in A TDOA technique with Super-Resolution based on the Volume Cross-Correlation Function, explaining that these overlapping multipath components obscure the direct-path arrival, making precise timing and positioning incredibly difficult.
To overcome this environmental noise, modern research splits into three core strategies: isolating the direct path mathematically, treating the interference as a unique spatial fingerprint, or engineering the physical environment to bypass the reflections entirely.
Strategy 1: Isolating the Direct Path via CSI and Super-Resolution
If the goal is to preserve geometric trilateration, the receiver must find a way to peer through the echoes and isolate the true, first-arriving direct path.
One way to achieve this is by shifting from coarse, single-value metrics like RSSI to Channel State Information (CSI). While RSSI merely reports the total aggregated power across an entire channel, CSI reveals how the wireless channel alters the amplitude and phase of individual frequency subcarriers. Essentially, CSI provides a fine-grained matrix representing the environment's unique transfer function.
By meticulously modeling how hardware imperfections distort this matrix, Navid Tadayon et al. demonstrate in Decimeter Ranging with Channel State Information that sub-meter ranging can be extracted from standard, off-the-shelf WiFi hardware. Their work proves that the ultimate limits of indoor positioning are often dictated by manageable hardware artifacts rather than insurmountable environmental noise; if you can cleanly calibrate the hardware, you can resolve the underlying multipath components.
Complementing this, advanced signal processing frameworks use super-resolution algorithms to mathematically decouple closely spaced arrivals. By applying geometric optimization techniques—such as the volume cross-correlation functions proposed by Xiqin Wang et al.—systems can separate overlapping wave packets in the time domain, identifying the exact nanosecond the direct path hit the antenna array.
Strategy 2: Turning Interference into a Spatial Fingerprint
An alternative philosophy discards the struggle to isolate the direct path. Instead, it asks: What if we treat multipath interference as a feature rather than noise?
Because the physical layout of an indoor space (concrete walls, metal pillars, heavy furniture) is static, the way a radio wave bounces around a specific room is highly deterministic. A receiver standing at Coordinate A will experience a fundamentally different pattern of constructive and destructive interference than a receiver standing at Coordinate B. This complex interference pattern acts as a unique spatial signature—a radio fingerprint.
Physical Space Radio Feature Space (CSI Matrix)
┌───────────────┐ ┌───────────────────────────────┐
│ [Room A] │ ───Transformer──>│ █ ▄ █ ▄ ▄ █ (Unique Pattern │
│ User Position│ Pre-training │ ▄ █ ▄ █ █ ▄ for Position A) │
└───────────────┘ └───────────────────────────────┘
Recent breakthroughs leverage deep learning to map these intricate signatures directly to physical coordinates without requiring explicit geometric calculations. For example, Jonathan Ott et al. developed a framework detailed in Radio Foundation Models: Pre-training Transformers for 5G-based Indoor Localization. By training self-supervised Transformers on massive amounts of unlabeled radio signals, the model learns to internalize the underlying "radio-geometry" of an environment. It recognizes the complex multipath structures of the room, allowing the system to achieve state-of-the-art localization accuracy while requiring ten times less labeled reference data than traditional fingerprinting methods.
This fingerprinting logic extends beyond tracking devices to tracking human movement (device-free sensing). In Action Recognition Based on CSI Signal Using Improved Deep Residual Network Model, Jian Zhao et al. show that while static furniture creates a complex baseline of multipath reflections, a moving human body introduces dynamic variations into the CSI matrix. By using deep residual networks equipped with attention mechanisms, their system filters out the static multipath background and isolates the subtle, time-varying phase shifts caused exclusively by human gestures and postures.
Strategy 3: Spatial and Structural Engineering
The final strategy moves the battleground from software to hardware, altering how signals are physically transmitted or directed through space.
One hardware-driven solution is to use Ultra-Wideband (UWB) pulses, which are exceptionally short in duration (often less than a nanosecond). This extreme brevity allows UWB systems to naturally separate the direct path from reflections in the time domain, as the echoes arrive long after the initial pulse has cleared the receiver. However, even UWB can struggle in severe non-line-of-sight (NLOS) conditions where the direct path is completely blocked by dense obstacles.
To circumvent this, Weiguo Wang et al. introduced BIFROST: Reinventing WiFi Signals Based on Dispersion Effect for Accurate Indoor Localization. Instead of relying on standard omnidirectional antennas, the authors utilize leaky wave antennas (LWAs) to artificially induce a frequency-dispersion effect. This structural modification ensures that different frequencies travel along different spatial paths, effectively synthesizing a usable line-of-sight component even when the direct physical path is obstructed by environmental clutter.
Passive Multipath:
Transmitter ───> Bounces wildly off random walls ───> Chaotic Receiver Signal
Active RIS Control:
Transmitter ───> Reconfigurable Intelligent Surface ───> Focused Beam to Receiver
(Phase-shifted reflections)
Similarly, the advent of Reconfigurable Intelligent Surfaces (RIS) allows systems to actively manipulate the indoor environment. As explored by Xiaowei Qian et al. in Sensing-based Beamforming Design for Joint Performance Enhancement of RIS-Aided ISAC Systems, an RIS consists of an array of passive scattering elements that can independently shift the phase of incoming radio waves. Instead of passively suffering from chaotic multipath degradation, the system uses coarse location sensing to dynamically program the surface, forcing the reflections to combine constructively at the user's exact location. This transforms a hostile, unpredictable indoor channel into a highly structured, programmable beamforming pipeline.