Unmodulated Visible Light Positioning (uVLP)

• Unmodulated VLP is an indoor positioning method that uses standard, unmodified light sources to infer location from ambient illumination without added modulation hardware.
• It exploits intensity-, imaging-, or spectral-based techniques to estimate positions by processing RSS, visual features, or characteristic frequencies for source inference.
• Applications range from smartphone localization and robotics to 6G indoor services, achieving accuracies from meter to centimeter levels when combined with sensor fusion.

Searching arXiv for the cited uVLP and adjacent VLP papers to ground the article. Search 1: uVLP survey and core papers. Unmodulated Visible Light Positioning (uVLP) denotes a class of indoor positioning systems that uses unmodulated or unmodified illumination sources—especially conventional LEDs, but also fluorescent lamps and, in some cases, ambient light such as sunlight/daylight—for localization without adding LED modulation hardware or changing the lighting infrastructure (Alijani et al., 17 Jul 2025). Instead of assigning explicit IDs through active modulation, uVLP estimates position by post-processing either total light intensity or irradiance measurements, or image data, and in some approaches exploits characteristic frequency, spectral signatures, or spatial appearance (Alijani et al., 17 Jul 2025). Within the broader VLP literature, this places uVLP in a distinct position: conventional VLP is usually framed as VLC-assisted and LED-ID-driven, whereas uVLP relies on light signals of opportunity and receiver-side inference rather than transmitter-side communication (Zhu et al., 2024).

1. Conceptual scope and relation to conventional VLP

The defining design choice in uVLP is that the light source is used as is. Conventional VLP typically relies on modulated LEDs that transmit identifiable signals through schemes such as OOK or PWM, and the receiver demultiplexes the intentionally encoded signal to recover IDs, distances, or angles. In uVLP, the receiver instead extracts position-relevant information from naturally available optical signals, preserving lighting output and avoiding added modulation hardware (Alijani et al., 17 Jul 2025).

This distinction is operational as well as architectural. In conventional VLP, the signal type is modulated or coded; in uVLP it is constant unmodulated light. Conventional VLP can transmit IDs or location information; uVLP generally cannot. Conventional VLP is described as having higher cost and complexity because of extra modulation hardware, whereas uVLP is described as lower in both because it reuses existing infrastructure. The survey literature also states that modulation may reduce effective illumination and radiant flux, whereas uVLP preserves lighting output (Alijani et al., 17 Jul 2025).

The broader VLP survey literature does not define uVLP as its main category, but it explicitly notes “unmodified LED-based VLP” and says that some works have verified the feasibility of using unmodified LEDs for positioning (Zhu et al., 2024). This places uVLP not outside the VLP field, but at its modulation-free boundary: it shares the same indoor optical infrastructure, yet departs from the default assumption that localization requires explicitly coded luminaires.

2. Physical observables and mathematical foundations

The physical observables in uVLP depend on the receiver. Intensity-based systems use received signal strength, total illuminance, spectral intensity over wavelength bands, characteristic-frequency peaks, or multi-sensor signatures. Imaging-based systems use lamp position in the image, corners, radiance textures, hidden visual features, and fixture constellations (Alijani et al., 17 Jul 2025).

For intensity-based formulations, the survey gives a generic optical model in which the received optical power from transmitter $k$ is

$$P_{r,k}=P_{t,k}\left[h_{c,\mathrm{LoS}}^{(k)}+\sum_A h_{c,\mathrm{NLoS}}^{(k,dA)}\right].$$

Under a Lambertian source model,

$$R_E(\phi_k)=\frac{m_k+1}{2\pi}\cos^{m_k}(\phi_k), \qquad m_k=\frac{-\ln(2)}{\ln(\cos \phi_{1/2})},$$

and the LoS gain becomes

$h_{c,\mathrm{LoS}}^{(k)}= \begin{cases} \dfrac{(m_k+1)A_R}{2\pi d_k^2}\cos^{m_k}(\phi_k)\cos(\psi_k)T_s(\psi_k)\tau(\psi_k), & 0\le \psi_k \le \psi_c,\[6pt] 0, & \text{otherwise}. \end{cases}$

When sources are not separated, the observable reduces to the composite field

$RSS_{\mathrm{total}}=\sum_{k=1}^{M} RSS_k,$

which is the central measurement model in undemultiplexed uVLP (Alijani et al., 17 Jul 2025).

The optical ambiguity induced by aggregated measurements is one of the central technical problems of uVLP. “Lightitude” addresses this by modeling received light strength (RLS) as a function of radiation angle, incidence angle, and distance, with an empirical Gaussian for the incidence-angle term to account for the shading and recessing of COTS light sensors: $$L_s = L_0 \cdot \frac{\cos(\phi)\cdot e^{-\left(\frac{\theta-\theta_0}{\delta}\right)^2}}{d^2}.$$ For multiple lamps,

$L_a=\sum_{i=1}^{n} L_{s_i}.$

Instead of solving ambiguity from one scalar intensity sample, the method uses user mobility, predicts RLS sequences along candidate paths, and scores them via Dynamic Time Warping with particle-filter-style weighting (Hu et al., 2015).

Imaging-based uVLP replaces or augments photometry with projective geometry. In “Non-Point Visible Light Transmitter Localization based on Monocular Camera,” a single known rectangular ceiling luminaire is treated as an extended planar object rather than a point emitter. With a calibrated monocular camera, known lamp geometry, and image points $a,b,c,d$ corresponding to lamp corners $A,B,C,D$, the method estimates the vertical separation from apparent lamp span: $L = \frac{d_{AD}\,f}{|x_a-x_d|},$ then computes distances to selected lamp points and solves trilateration: $$P_{r,k}=P_{t,k}\left[h_{c,\mathrm{LoS}}^{(k)}+\sum_A h_{c,\mathrm{NLoS}}^{(k,dA)}\right].$$0 with analogous equations for $$P_{r,k}=P_{t,k}\left[h_{c,\mathrm{LoS}}^{(k)}+\sum_A h_{c,\mathrm{NLoS}}^{(k,dA)}\right].$$1 and $$P_{r,k}=P_{t,k}\left[h_{c,\mathrm{LoS}}^{(k)}+\sum_A h_{c,\mathrm{NLoS}}^{(k,dA)}\right].$$2, and $$P_{r,k}=P_{t,k}\left[h_{c,\mathrm{LoS}}^{(k)}+\sum_A h_{c,\mathrm{NLoS}}^{(k,dA)}\right].$$3 (Zhao et al., 2021). This is a paradigmatic example of geometry-driven uVLP: localization is inferred from luminaire shape and perspective deformation rather than optical waveform measurements.

Many practical uVLP systems are naturally expressed as nonlinear state-space estimators rather than one-shot inverse problems. The survey therefore gives the generic filtering form

$$P_{r,k}=P_{t,k}\left[h_{c,\mathrm{LoS}}^{(k)}+\sum_A h_{c,\mathrm{NLoS}}^{(k,dA)}\right].$$4

used by EKF, IEKF, and particle-filter systems that fuse light measurements with IMU, PDR, magnetometer, BLE, LiDAR, or odometry (Alijani et al., 17 Jul 2025).

3. Taxonomy of uVLP techniques

The most explicit recent taxonomy organizes uVLP along two axes: receiver technology and signal treatment. Receiver technology splits the field into intensity-based and imaging-based methods. Signal treatment splits it into undemultiplexed and demultiplexed methods (Alijani et al., 17 Jul 2025).

Receiver family	Undemultiplexed	Demultiplexed
Intensity-based	Aggregate RSS, fingerprints, PDR/IMU fusion	Characteristic-frequency separation, RSS-based ranging
Imaging-based	Visual landmarks, image fingerprints, SLAM/EKF	Rolling-shutter CF extraction, source-level identification

Undemultiplexed intensity-based uVLP works from the aggregate light field. It is therefore dominated by fingerprinting, temporal sequence matching, and probabilistic filtering. Representative sensor platforms include photodiodes, ambient light sensors, solar cells, and spectral sensors (Alijani et al., 17 Jul 2025). “Lightitude” is a canonical instance: it uses existing fluorescent lamps and the ambient light sensors of unmodified Google Nexus 4, Nexus 7, and Moto 360 devices, together with inertial sensing, to localize from RLS sequences gathered while walking (Hu et al., 2015).

Demultiplexed intensity-based uVLP attempts to recover per-source contributions without active modulation, most commonly by exploiting naturally occurring characteristic frequencies introduced by lamp driver circuitry. Once sources are separated, classical multilateration- or model-based localization becomes possible again, and reported performance can move from meter-level toward decimeter- and centimeter-level regimes (Alijani et al., 17 Jul 2025).

Undemultiplexed imaging-based uVLP uses ceiling lights as visual landmarks or as image fingerprints. The discriminative cues are corners, lamp shapes, radiance patterns, hidden visual features, and ceiling constellations (Alijani et al., 17 Jul 2025). The monocular non-point-transmitter method belongs here in its estimation core, although its infrastructure description still allows a separate lamp-identification mechanism (Zhao et al., 2021).

Demultiplexed imaging-based uVLP is largely associated with rolling-shutter cameras that recover aliased characteristic frequencies from unmodified luminaires. The survey models camera exposure as

$$P_{r,k}=P_{t,k}\left[h_{c,\mathrm{LoS}}^{(k)}+\sum_A h_{c,\mathrm{NLoS}}^{(k,dA)}\right].$$5

with rectangular exposure window $$P_{r,k}=P_{t,k}\left[h_{c,\mathrm{LoS}}^{(k)}+\sum_A h_{c,\mathrm{NLoS}}^{(k,dA)}\right].$$6, and derives the rolling-shutter frequency response

$$P_{r,k}=P_{t,k}\left[h_{c,\mathrm{LoS}}^{(k)}+\sum_A h_{c,\mathrm{NLoS}}^{(k,dA)}\right].$$7

With frame period $$P_{r,k}=P_{t,k}\left[h_{c,\mathrm{LoS}}^{(k)}+\sum_A h_{c,\mathrm{NLoS}}^{(k,dA)}\right].$$8, $$P_{r,k}=P_{t,k}\left[h_{c,\mathrm{LoS}}^{(k)}+\sum_A h_{c,\mathrm{NLoS}}^{(k,dA)}\right].$$9 sensor rows, and frame rate $$R_E(\phi_k)=\frac{m_k+1}{2\pi}\cos^{m_k}(\phi_k), \qquad m_k=\frac{-\ln(2)}{\ln(\cos \phi_{1/2})},$$0,

$$R_E(\phi_k)=\frac{m_k+1}{2\pi}\cos^{m_k}(\phi_k), \qquad m_k=\frac{-\ln(2)}{\ln(\cos \phi_{1/2})},$$1

which underlies LiTell-style extraction of aliased characteristic frequencies from unmodified fluorescent lights (Alijani et al., 17 Jul 2025).

4. Representative systems and methodological lineages

A large fraction of uVLP research uses fingerprints or trajectory-dependent light signatures rather than explicit geometric source separation. NaviLight, LiLo, DeepML, LiMag, IDyLL, Spectral-Loc, HueSense, HueLoc, Ambilight, and LuxTrace are all cited as representative systems in this class, spanning ALS, spectral sensors, solar cells, and fused multi-sensor platforms (Alijani et al., 17 Jul 2025). Their common structure is that localization is inferred from a location-dependent but not source-resolved optical field.

“Lightitude” remains one of the clearest primary-source demonstrations of this paradigm. It explicitly uses ubiquitous visible lights rather than special LEDs, does not require redeploying infrastructures or special devices, and models the light field over a six-dimensional variable space comprising three-dimensional position and three-dimensional orientation. It then uses mobility-induced RLS sequences and DTW-scored candidate trajectories to resolve the ambiguity that “a unique RLS value may correspond to multiple possible positions” (Hu et al., 2015).

A second major lineage is geometry-driven camera uVLP. The non-point-transmitter monocular method localizes from one rectangular ceiling light whose geometry is known a priori. Its core contribution is to reject the point-source approximation and treat the luminaire as a finite-area planar object that supplies multiple geometric constraints in a single frame. In the reported simulation, the room size is $$R_E(\phi_k)=\frac{m_k+1}{2\pi}\cos^{m_k}(\phi_k), \qquad m_k=\frac{-\ln(2)}{\ln(\cos \phi_{1/2})},$$2, the transmitter corners are $$R_E(\phi_k)=\frac{m_k+1}{2\pi}\cos^{m_k}(\phi_k), \qquad m_k=\frac{-\ln(2)}{\ln(\cos \phi_{1/2})},$$3, $$R_E(\phi_k)=\frac{m_k+1}{2\pi}\cos^{m_k}(\phi_k), \qquad m_k=\frac{-\ln(2)}{\ln(\cos \phi_{1/2})},$$4, $$R_E(\phi_k)=\frac{m_k+1}{2\pi}\cos^{m_k}(\phi_k), \qquad m_k=\frac{-\ln(2)}{\ln(\cos \phi_{1/2})},$$5, and $$R_E(\phi_k)=\frac{m_k+1}{2\pi}\cos^{m_k}(\phi_k), \qquad m_k=\frac{-\ln(2)}{\ln(\cos \phi_{1/2})},$$6, the maximum offset on the $$R_E(\phi_k)=\frac{m_k+1}{2\pi}\cos^{m_k}(\phi_k), \qquad m_k=\frac{-\ln(2)}{\ln(\cos \phi_{1/2})},$$7 plane is $$R_E(\phi_k)=\frac{m_k+1}{2\pi}\cos^{m_k}(\phi_k), \qquad m_k=\frac{-\ln(2)}{\ln(\cos \phi_{1/2})},$$8, and the maximum RMSE is $$R_E(\phi_k)=\frac{m_k+1}{2\pi}\cos^{m_k}(\phi_k), \qquad m_k=\frac{-\ln(2)}{\ln(\cos \phi_{1/2})},$$9 (Zhao et al., 2021). The method is therefore strongly informative for uVLP, even though the paper does not fully resolve transmitter identification without auxiliary information.

A third lineage is demultiplexed uVLP based on naturally occurring characteristic frequencies. The survey identifies Bastiaens et al., LiTell, LiTell2/Pulsar, and related work as the core examples. These systems attempt to recover some of the source-specific structure that conventional VLP obtains through active modulation, but they do so from driver-induced signatures already present in unmodified illumination (Alijani et al., 17 Jul 2025).

5. Performance regimes and application domains

The reviewed literature spans room-level classification, meter-level navigation, sub-meter tracking, decimeter-level localization, and centimeter-level benchmark cases (Alijani et al., 17 Jul 2025). This range is not incidental: it reflects the strong dependence of uVLP performance on whether the system is undemultiplexed or demultiplexed, intensity-based or imaging-based, and whether it is fused with inertial or other exteroceptive sensors.

In undemultiplexed intensity-based systems, reported performance is often meter-level to sub-meter. NaviLight reports 85% within $h_{c,\mathrm{LoS}}^{(k)}= \begin{cases} \dfrac{(m_k+1)A_R}{2\pi d_k^2}\cos^{m_k}(\phi_k)\cos(\psi_k)T_s(\psi_k)\tau(\psi_k), & 0\le \psi_k \le \psi_c,\[6pt] 0, & \text{otherwise}. \end{cases}$0 m in an office, $h_{c,\mathrm{LoS}}^{(k)}= \begin{cases} \dfrac{(m_k+1)A_R}{2\pi d_k^2}\cos^{m_k}(\phi_k)\cos(\psi_k)T_s(\psi_k)\tau(\psi_k), & 0\le \psi_k \le \psi_c,\[6pt] 0, & \text{otherwise}. \end{cases}$1 m in a mall, and $h_{c,\mathrm{LoS}}^{(k)}= \begin{cases} \dfrac{(m_k+1)A_R}{2\pi d_k^2}\cos^{m_k}(\phi_k)\cos(\psi_k)T_s(\psi_k)\tau(\psi_k), & 0\le \psi_k \le \psi_c,\[6pt] 0, & \text{otherwise}. \end{cases}$2 m in a parking environment. LiLo reports average localization error $h_{c,\mathrm{LoS}}^{(k)}= \begin{cases} \dfrac{(m_k+1)A_R}{2\pi d_k^2}\cos^{m_k}(\phi_k)\cos(\psi_k)T_s(\psi_k)\tau(\psi_k), & 0\le \psi_k \le \psi_c,\[6pt] 0, & \text{otherwise}. \end{cases}$3 m. GraphSLAM, which fuses light, magnetic field, WiFi, and PDR, reports $h_{c,\mathrm{LoS}}^{(k)}= \begin{cases} \dfrac{(m_k+1)A_R}{2\pi d_k^2}\cos^{m_k}(\phi_k)\cos(\psi_k)T_s(\psi_k)\tau(\psi_k), & 0\le \psi_k \le \psi_c,\[6pt] 0, & \text{otherwise}. \end{cases}$4 m median and $h_{c,\mathrm{LoS}}^{(k)}= \begin{cases} \dfrac{(m_k+1)A_R}{2\pi d_k^2}\cos^{m_k}(\phi_k)\cos(\psi_k)T_s(\psi_k)\tau(\psi_k), & 0\le \psi_k \le \psi_c,\[6pt] 0, & \text{otherwise}. \end{cases}$5 m at the 90th percentile (Alijani et al., 17 Jul 2025). “Lightitude” reports mean accuracy $h_{c,\mathrm{LoS}}^{(k)}= \begin{cases} \dfrac{(m_k+1)A_R}{2\pi d_k^2}\cos^{m_k}(\phi_k)\cos(\psi_k)T_s(\psi_k)\tau(\psi_k), & 0\le \psi_k \le \psi_c,\[6pt] 0, & \text{otherwise}. \end{cases}$6 m in office; the abstract also states $h_{c,\mathrm{LoS}}^{(k)}= \begin{cases} \dfrac{(m_k+1)A_R}{2\pi d_k^2}\cos^{m_k}(\phi_k)\cos(\psi_k)T_s(\psi_k)\tau(\psi_k), & 0\le \psi_k \le \psi_c,\[6pt] 0, & \text{otherwise}. \end{cases}$7 m in library, while the detailed section reports $h_{c,\mathrm{LoS}}^{(k)}= \begin{cases} \dfrac{(m_k+1)A_R}{2\pi d_k^2}\cos^{m_k}(\phi_k)\cos(\psi_k)T_s(\psi_k)\tau(\psi_k), & 0\le \psi_k \le \psi_c,\[6pt] 0, & \text{otherwise}. \end{cases}$8 m for pure light in the library and $h_{c,\mathrm{LoS}}^{(k)}= \begin{cases} \dfrac{(m_k+1)A_R}{2\pi d_k^2}\cos^{m_k}(\phi_k)\cos(\psi_k)T_s(\psi_k)\tau(\psi_k), & 0\le \psi_k \le \psi_c,\[6pt] 0, & \text{otherwise}. \end{cases}$9 m for the WiFi-assisted version, a discrepancy explicitly noted in the technical synthesis (Hu et al., 2015).

Spectral and fused methods can be markedly tighter. Spectral-Loc is reported as sub-meter at the 90th percentile in both meeting-room and open-office tests; BLELight reports mean localization accuracy $RSS_{\mathrm{total}}=\sum_{k=1}^{M} RSS_k,$0 m; HueLoc reports mean localization error $RSS_{\mathrm{total}}=\sum_{k=1}^{M} RSS_k,$1 cm, and its fused version reports $RSS_{\mathrm{total}}=\sum_{k=1}^{M} RSS_k,$2 cm 2D mean and $RSS_{\mathrm{total}}=\sum_{k=1}^{M} RSS_k,$3 cm 3D mean (Alijani et al., 17 Jul 2025).

Demultiplexed characteristic-frequency methods approach the decimeter- and centimeter-level regimes more commonly associated with conventional VLP. Bastiaens et al. report, with model-based filtering, $RSS_{\mathrm{total}}=\sum_{k=1}^{M} RSS_k,$4 cm and $RSS_{\mathrm{total}}=\sum_{k=1}^{M} RSS_k,$5 cm in a $RSS_{\mathrm{total}}=\sum_{k=1}^{M} RSS_k,$6 m room. LiTell reports 10 cm accuracy in 90% of stationary cases under a lamp, 15 cm median accuracy during motion, and 90% of errors within 25 cm (Alijani et al., 17 Jul 2025).

Imaging-based uVLP can also reach high precision when ceiling geometry is stable and feature extraction is reliable. The survey reports that iLAMP achieved $RSS_{\mathrm{total}}=\sum_{k=1}^{M} RSS_k,$7 3D location error 3.5 cm and heading error $RSS_{\mathrm{total}}=\sum_{k=1}^{M} RSS_k,$8, while Chen et al. report maximum position errors $RSS_{\mathrm{total}}=\sum_{k=1}^{M} RSS_k,$9 m and $$L_s = L_0 \cdot \frac{\cos(\phi)\cdot e^{-\left(\frac{\theta-\theta_0}{\delta}\right)^2}}{d^2}.$$0 m with peak orientation error $$L_s = L_0 \cdot \frac{\cos(\phi)\cdot e^{-\left(\frac{\theta-\theta_0}{\delta}\right)^2}}{d^2}.$$1 in a robot setting (Alijani et al., 17 Jul 2025). A plausible implication is that geometry-rich camera methods are the most natural path to high-accuracy uVLP when transmitter identification can be solved without active coding.

Application domains follow directly from these performance regimes. The general VLP survey highlights smartphone indoor localization, robotics, navigation, museum and supermarket services, and asset tracking (Zhu et al., 2024). The uVLP survey places the field explicitly in the context of low-infrastructure indoor positioning and 6G-oriented location services (Alijani et al., 17 Jul 2025).

6. Limitations, boundary cases, and research directions

The principal technical limitations of uVLP are repeatedly identified as source ambiguity, ambient-light interference, receiver tilt sensitivity, calibration burden, device heterogeneity, and the fragility of simple RSS-distance models under reflections, blockage, or repetitive lighting layouts (Alijani et al., 17 Jul 2025). ALS-based systems are additionally constrained by low sampling rates; camera-based systems face high power consumption, latency, and privacy concerns (Alijani et al., 17 Jul 2025).

A persistent misconception is that any visible-light localization method that does not emphasize communication throughput is automatically uVLP. The literature is more precise. Some systems are explicitly not unmodulated because they depend on LED IDs, coded stripe patterns, or pilot-assisted source separation. The multi-mobile robot navigation demonstration decodes LED stripe-pattern IDs and is therefore modulated image-sensor VLP rather than uVLP (Chen et al., 2021). The multi-target tracking paper uses rolling-shutter LED-ID recognition by modulation frequency before subsequent image tracking and geometric positioning, so it is better understood as a hybrid or supporting paper for camera-based uVLP rather than a pure uVLP system (Huang et al., 2021). Likewise, the integrated event-camera VLC/VLP vehicle system relies on Barker synchronization, Walsh-Hadamard pilots, and 10 kHz blinking, making it adjacent to uVLP rather than a strict instance of it (Soga et al., 20 Oct 2025).

A second boundary case concerns methods whose estimation core is modulation-free but whose infrastructure assumptions still permit auxiliary identification or timing support. The monocular non-point-transmitter method is highly relevant to uVLP because its estimator uses only known lamp geometry and image projection, but the paper still assumes that the camera can recognize which transmitter it photographed, possibly via ID information (Zhao et al., 2021). The single-VCSEL scanning system is methodologically relevant because it replaces multiple anchors with known beam direction and RSS, yet it is not a canonical visible-light unmodulated LED system: it uses a 950 nm VCSEL, depends on scan-state knowledge, and may use pilot signals for synchronization (Dong et al., 26 Jan 2026).

The same caution applies to waveform-based or multiplexed optical localization theory. Hybrid TDOA/RSS VLP, RGB TOA/RSS estimation, and cooperative Lambertian RSS localization provide important models, CRLB analyses, and fusion principles, but they assume known transmitted waveforms, multiplexing, or source separability that strict uVLP does not automatically have (Kazikli et al., 2018, Demirel et al., 2021, Keskin et al., 2018). These works are best read as adjacent theoretical foundations rather than as direct uVLP implementations.

The main future directions emphasized in the uVLP survey are hybrid modulated/unmodulated systems, tighter fusion with IMU and PDR, advanced signal processing and machine learning for source identification and demultiplexing, better exploitation of spectral information, RIS-assisted coverage enhancement, and broader benchmarking under realistic sunlight, blockage, and large-scale deployment conditions (Alijani et al., 17 Jul 2025). This suggests that the long-term structure of the field may be plural rather than singular: undemultiplexed uVLP for ultra-low-infrastructure operation, demultiplexed uVLP for higher precision, and hybrid systems where a small number of modulated anchors stabilize a larger unmodified lighting field.