Disentangling Overlapping Signals in Multi User Wireless Sensing

279 concepts227 papers2 proposals

Contactless sensing—using ambient radio frequency (RF) signals like WiFi, millimeter-wave (mmWave) radar, ultra-wideband (UWB) pulses, or even floor-embedded pressure mats—promises a privacy-preserving, device-free way to perceive human activity. In a vacuum, these systems work beautifully. A single individual walking through a room creates predictable, distinct disruptions in the wireless channel that can be mapped to precise poses, actions, or vital signs.

However, when multiple subjects occupy the same space, this elegant mapping collapses. Because radio waves and physical forces obey the principle of superposition, a receiver does not capture isolated data streams for each person. Instead, it records a single, highly tangled composite signal where individual reflections overlap, collide, and interfere.

Resolving this multi-user signal interference is the central hurdle to scaling wireless perception from controlled laboratory settings to the messy reality of shared human environments.

The Mechanism of Signal Collision

To understand why multi-user sensing is difficult, we must look at how physical bodies interact with ambient signals. When an RF signal propagates through a room, every human body acts as a complex, moving reflector. The signal received is the sum of multiple propagation paths (multipath reflections).

[WiFi Transmitter] ───┬───> (Subject A: Moving) ───> [Receiver: Overlapping CSI]
                      └───> (Subject B: Static) ───>

As noted by Zhao et al., RF signals possess an inherently lower spatial resolution than visible light. When two or more individuals are in close proximity, their individual reflection boundaries blur together, leading to identity confusion or merged, deformed reconstructions. This phenomenon manifests across three distinct granularities of sensing:

  1. Macro-movements (Poses and Actions): When tracking full-body skeletons or identifying actions, the presence of multiple actors introduces severe phase inconsistencies and signal cancellation. The authors of one study point out that without high-resolution visual cues, spatial decomposition becomes so complex that systems struggle to determine whether a signal represents independent actors or a single coordinated interaction.
  2. Micro-movements (Vital Signs): Extracting respiration or heart rates relies on detecting sub-millimeter chest displacements. Zhang et al. and Li et al. demonstrate that when multiple subjects are present, their weak, rhythmic physiological signatures mask one another, preventing standard filters from isolating distinct heartbeats or breathing rates.
  3. Passive Tag Collisions: In systems utilizing passive RFID arrays, the problem shifts from raw wave reflections to backscattered data packets. As explored by Wang et al., close-proximity social interactions cause simultaneous tag triggers, leading to high false-negative rates due to packet collisions at the reader.

Why Traditional Filtering Fails

In classical signal processing, unwanted signal components are often treated as noise and removed via frequency-selective filtering or static background subtraction. This approach fails in multi-user scenarios because the "interference" is structurally identical to the target signal.

If Subject A is walking and Subject B is sitting down, Subject B's micro-Doppler signature or Channel State Information (CSI) perturbations are not random noise—they are structured data. As shown by Tang et al., the micro-Doppler spectrogram of one individual can completely mask or warp the signature of another. Because human movements overlap dynamically in both the spatial and frequency domains, linear source-separation techniques cannot cleanly untangle the non-linear mixing that occurs at the receiver.

Emerging Paradigms for Disentanglement

To overcome this limitation, recent research has shifted away from trying to prevent signal mixing, focusing instead on architectures and mathematical frameworks designed to deconvolve the mixture.

Generative and Structural Priors

One of the most powerful tools for resolving ambiguity is the injection of human structural priors into the decoding process. If a wireless signal is ambiguous, a model trained on how the human body physically moves can constrain the solution space.

Fan et al. address this by framing multi-subject pose estimation as a conditional denoising task using diffusion models. By treating the overlapping signal interference as a form of structured noise, the model leverages learned temporal and skeletal priors to iteratively reconstruct distinct, anatomically valid human meshes from sparse, multi-user radar echoes.

Omni-Dimensional Scanning and State Space Models

Traditional deep learning models, like Transformers, scale quadratically when trying to compute attention across complex, multi-user signal streams. To maintain real-time tracking in high-density crowds, researchers are exploring alternative architectures.

Zhang et al. introduced a frequency-aware State Space Model (SSM) that utilizes omni-dimensional scanning to process both the amplitude and phase of radar echoes. By tracking the continuous evolution of the wireless channel state rather than computing pairwise token correlations, this approach efficiently isolates individual human signatures even as the environment becomes increasingly saturated with concurrent motion.

Combinatorial and Coding Frameworks

When dealing with discrete sensors, such as passive tag fields or localized tracking grids, researchers have borrowed combinatorial strategies from genomics and information theory to handle data collisions.

For instance, the mathematical logic of "overlapping pools"—originally designed to identify rare genetic variants among mixed DNA samples without sequencing every individual (Prabhu)—has inspired modern sensor coding. Lo et al. demonstrate how "multiset color coding" can resolve ambiguous sensor activations. By assigning sensors to overlapping, uniquely decodable ID multisets, the system can mathematically pinpoint individual trajectories even when multiple targets trigger the same tracking field simultaneously.

The Path Forward

As surveyed by Mu, transitioning from single-occupant tracking to robust multi-user deconvolution remains one of the defining challenges of the field. Early device-free tracking systems, such as the one proposed by Yao, proved that ambient radio waves could capture human presence without cameras or wearables. The current generation of work is proving that through a combination of high-order harmonic analysis, structural generative models, and advanced spatial filtering, we can see not just a crowd, but the individuals within it.

History & citationVersion 2 · 2026-10-04

Current · Permanent link

  1. Version 2 (latest) · 2026-10-04 · 215 papers
  2. Version 1 · 2026-10-04 · 292 papers
Cite this direction

Latest version (a living page):

@misc{lacuna_multi_user_signal_interference_034817a4,
  title = {multi-user signal interference},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/multi-user-signal-interference/rdr_034817a4721f45ce8996ecc0bd7f57e8},
  note = {Latest version of a living research direction; cite a version for a fixed text. Accessed 2026-11-19},
}

This exact version (version 2, frozen text):

@misc{lacuna_multi_user_signal_interference_034817a4_v2,
  title = {multi-user signal interference},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/multi-user-signal-interference/rdr_034817a4721f45ce8996ecc0bd7f57e8/v/2},
  note = {Version 2, frozen 2026-10-04. Accessed 2026-11-19},
}
About

These directions overlap with this one through shared papers.

See something wrong? Report a mistake