JCP: Joint Communication & Indoor Positioning

• JCP is a paradigm that repurposes communication signals from 5G, Wi‑Fi, UWB, and VLC for both data transmission and indoor localization.
• It fuses observables like ToA, AoA, RSS, and CSI with methods such as ML inference, sparse recovery, and nonlinear optimization to yield geometric insights.
• Research in JCP targets system robustness, addressing challenges like multipath effects, hardware mismatches, and communication–positioning tradeoffs through hybrid architectures.

Searching arXiv for recent and foundational papers on joint communication and indoor positioning, including 5G, VLC, Wi‑Fi, and UWB. Joint communication and indoor positioning (JCP) denotes system designs in which the same communication infrastructure, waveforms, channel measurements, or protocol artifacts are exploited not only for data transfer but also for indoor localization. In the literature represented here, JCP spans 5G NR downlink and uplink processing, Wi‑Fi fingerprinting and beamforming feedback, UWB ranging and proximity analysis, visible-light communication (VLC) with RSS and spatial modulation, camera-based visible-light positioning, and hybrid multi-technology fusion. Across these variants, the central technical question is not merely whether indoor position can be estimated, but how communication-side observables—such as ToA, TDoA, AoA, DOA, carrier phase, CSI/BFM, RSS, optical impulse responses, or proximity traces—can be converted into robust geometric information under multipath, blockage, synchronization error, dimming, feedback constraints, and deployment-specific geometry (Henninger et al., 2022, Chen et al., 2021, Leblebici et al., 6 Aug 2025, Tao et al., 13 Feb 2026).

1. Conceptual Scope and System Classes

JCP is most directly realized when a communication signal already present in the network is repurposed as a positioning signal. In 5G, this appears as the extraction of ToA, AoA, DOA, and carrier phase from NR uplink or downlink reference structures, without introducing a separate positioning-only transmitter chain (Chen et al., 2021, Pan et al., 2022). In Wi‑Fi, the same principle appears in two distinct forms: radio location fingerprinting that reuses existing IEEE 802.11 infrastructure, and feedback-centric methods that treat the standardized beamforming feedback matrix as a compressed sensing feature for localization (Kjærgaard, 2010, Tao et al., 13 Feb 2026). In VLC, the same LEDs can support both illumination and data transmission while their channel gains or identifiers support positioning; in camera-based visible-light positioning, ceiling cameras and light sources define an infrastructure-side localization architecture with millimeter-level experimental accuracy (Leblebici et al., 6 Aug 2025, Pan et al., 3 Jun 2025).

A broader interpretation includes systems in which positioning data are used to infer communication structure rather than to improve radio-link performance. A UWB study of a standing social reception treated close physical proximity as evidence of contact, then used weighted networks and Infomap communities to visualize interpersonal communication. In that setting, positioning is not an auxiliary service to communication; it is the measurement substrate from which communication patterns are inferred (Shinto et al., 8 Oct 2025). This suggests that JCP has both a waveform-level meaning—shared infrastructure for communications and localization—and an interaction-level meaning in which localization becomes a proxy for communication behavior.

A recurrent architectural distinction concerns where the measurement burden lies. Some JCP systems are infrastructure-centric, such as ceiling-mounted cameras localizing LED-equipped targets or ceiling photodetectors extracting optical channel fingerprints (Pan et al., 3 Jun 2025, Hosseinianfar et al., 2018). Others are receiver-centric, such as SDR-based 5G NR downlink ranging using commercial gNB broadcasts (Chen et al., 2021). Still others distribute the burden across both ends, as in Wi‑Fi active scanning or 5G uplink SRS estimation at picocell gNBs (Kjærgaard, 2010, Pan et al., 2022). This division strongly affects scalability, latency, privacy, and coexistence with communication traffic.

2. Measurements, Signal Models, and Estimation Principles

Indoor JCP systems use several recurring classes of observables. The first class is geometric timing and angle information. In a probabilistic 5G formulation, ToA contributes range likelihood terms and AoA contributes directional likelihood terms, with joint estimation written as

$$(\hat{\mathbf x}_{\cap}, \hat{\tau}_{\cap}) = \arg\max_{\mathbf{x},\tau} \left\{ \mathcal{L}_T(\mathbf{x},\tau)+\mathcal{L}_{\angle}(\mathbf{x}) \right\},$$

so that delay and angle constraints are fused directly in a maximum-likelihood objective (Henninger et al., 2022). A second class is RSS-based geometry, common in VLC, where received optical power is converted into transmitter–receiver distances and lateration or radical-axis constructions (Gu et al., 2015, Leblebici et al., 6 Aug 2025). A third class is high-dimensional channel representation, including CSI, BFM, or full impulse-response fingerprints, which are then processed by sparse recovery, nearest-neighbor matching, or learned embeddings (Tao et al., 13 Feb 2026, Gligoric et al., 2018, Hosseinianfar et al., 2018).

Estimation-theoretic analysis appears explicitly in multipath-assisted radio localization. With a known floor plan, specular multipath components can be interpreted as signals arriving from virtual anchors, and the position error bound is expressed via the equivalent Fisher information matrix:

$$\mathcal{P}(p)=\sqrt{\operatorname{tr}\left\{[J_p^{-1}]_{2\times2}\right\}}.$$

In this framework, the information contributed by each resolvable multipath component depends on effective bandwidth, extended SINR, geometry, and overlap with other paths (Leitinger et al., 2014). This result is structurally important because it establishes that localization quality is governed not only by SNR or anchor count, but by the resolvability and geometry of the individual propagation components.

Optical JCP uses analogous but modality-specific signal models. A dimming-aware VLC JCP system models the received vector as

$$\mathbf{y}=\mathbf{H}\tilde{\mathbf{x}}+\mathbf{w},$$

where the transmit vector contains both a sparse spatial-modulation symbol and a DC-bias/dimming term. The same estimated channel coefficients are used for both symbol detection and RSS-based localization, so channel estimation, dimming estimation, communication detection, and positioning become tightly coupled (Leblebici et al., 6 Aug 2025). In passive multi-camera VLP, by contrast, each image observation defines a projection ray, and localization proceeds from linear least squares to nonlinear reprojection-error minimization over all cameras and targets (Pan et al., 3 Jun 2025).

These formulations imply that JCP is less a single algorithmic family than a shared estimation doctrine: observables created for communication are mapped into geometric constraints, then combined using lateration, triangulation, fingerprint matching, sparse recovery, nonlinear least squares, or ML/Bayesian inference depending on the modality.

3. 5G and 5G-Advanced Indoor JCP

The 5G literature in this set treats JCP as a natural consequence of large bandwidth, synchronization structure, and multi-antenna reception, but it also shows that indoor multipath and hardware mismatch remain dominant obstacles. One line of work uses commercial 5G NR downlink broadcasts. An SDR receiver built around a USRP B210 samples live NR signals, performs coarse synchronization from the SSB, extracts PBCH DM-RS, acquires and tracks multipath delays, and then refines ToA ranging using the carrier phase of the first-arrived path. In office tests with a commercial gNB, the static ToA accuracy measured by the $1$-$\sigma$ interval was about $0.5$ m, and in pedestrian motion the probability of range accuracy within $0.8$ m was reported as $95\%$ (Chen et al., 2021).

A second line uses joint ToA/AoA localization from 5G uplink measurements in an indoor factory. There, poor propagation conditions generate outlier ToA and AoA measurements, so robust initialization becomes central. An iterative reweighting method evaluates every locator as reference once, computes residuals, applies Andrews’ sine reweighting, and then initializes a gradient search for joint ML estimation. In a 3.75 GHz, about 100 MHz indoor-factory proof of concept with 6 synchronized locators and 28 test points, the initialized joint method reached a mean error of $0.24$ m, a 95th percentile of $0.51$ m, and a 99th percentile of $$\mathcal{P}(p)=\sqrt{\operatorname{tr}\left\{[J_p^{-1}]_{2\times2}\right\}}.$$0 m; in the central area the mean error was $$\mathcal{P}(p)=\sqrt{\operatorname{tr}\left\{[J_p^{-1}]_{2\times2}\right\}}.$$1 m and the maximum error $$\mathcal{P}(p)=\sqrt{\operatorname{tr}\left\{[J_p^{-1}]_{2\times2}\right\}}.$$2 m (Henninger et al., 2022).

A third line addresses picocell hardware realism. Uplink SRS processing at 5G TRPs suffers from direction-dependent array modeling error, particularly when small-scale arrays are used. A calibrated JADE pipeline first removes RF-chain error, then fits antenna phase error as a function of angle, resolves multipath in the TOA domain via IAA, and finally retrieves LOS DOA by a conventional beamformer. The field result most directly relevant to deployment is a triangulation positioning error of $$\mathcal{P}(p)=\sqrt{\operatorname{tr}\left\{[J_p^{-1}]_{2\times2}\right\}}.$$3 m for 90% of cases using only DOAs estimated at two separated receiving points (Pan et al., 2022).

At the architectural level, 5G-Advanced work argues that time- and angle-based measurements alone are insufficient for centimeter-level indoor performance in dense factory channels. Carrier-phase measurements, discussed in 3GPP Rel-18, are presented as a complementary layer that can refine both trilateration- and triangulation-based positioning. The same work emphasizes that centimeter-level performance depends on phase-continuous reference signaling, robust ambiguity resolution, strong synchronization, and LOS-sensitive operation rather than on carrier phase alone (Nikonowicz et al., 2022).

Band selection further complicates the JCP narrative. Ray-tracing-assisted OTDoA evaluation in a Bosch production hall compared C-band at 3.775 GHz and mmWave at 26.85 GHz using first-arriving multipath components only. In the static setup, average 2D positioning errors were $$\mathcal{P}(p)=\sqrt{\operatorname{tr}\left\{[J_p^{-1}]_{2\times2}\right\}}.$$4 m for C-band and $$\mathcal{P}(p)=\sqrt{\operatorname{tr}\left\{[J_p^{-1}]_{2\times2}\right\}}.$$5 m for mmWave; in the dynamic forklift setup, they became $$\mathcal{P}(p)=\sqrt{\operatorname{tr}\left\{[J_p^{-1}]_{2\times2}\right\}}.$$6 m and $$\mathcal{P}(p)=\sqrt{\operatorname{tr}\left\{[J_p^{-1}]_{2\times2}\right\}}.$$7 m, respectively. Both met the 3GPP Release 16 indoor requirement of $$\mathcal{P}(p)=\sqrt{\operatorname{tr}\left\{[J_p^{-1}]_{2\times2}\right\}}.$$8 m for 80% of cases, but neither met the stricter Release 17 target of $$\mathcal{P}(p)=\sqrt{\operatorname{tr}\left\{[J_p^{-1}]_{2\times2}\right\}}.$$9 m for 90% of cases (Muthineni et al., 2024). This directly contradicts the simplistic assumption that a communication-favorable band or larger bandwidth automatically yields the best positioning performance indoors.

4. Visible-Light and Optical JCP

Optical JCP exhibits a particularly tight coupling between channel modeling and localization, because illumination, communication, and sensing often share the same LED/photodiode geometry. A basic RSS-based VLC positioning system with four ceiling LEDs in a $$\mathbf{y}=\mathbf{H}\tilde{\mathbf{x}}+\mathbf{w},$$0 room shows how sensitive such systems are to multipath reflections. When reflections were neglected, the RMS positioning error over the whole room was $$\mathbf{y}=\mathbf{H}\tilde{\mathbf{x}}+\mathbf{w},$$1 m; with reflections included, it increased to $$\mathbf{y}=\mathbf{H}\tilde{\mathbf{x}}+\mathbf{w},$$2 m. The degradation was especially severe near corners and edges, while the center point remained almost unaffected (Gu et al., 2015). This establishes that in VLC, as in RF, multipath can transform an apparently well-conditioned RSS model into a strongly biased estimator.

More integrated VLC designs use the optical link explicitly for both functions. A dimming-aware JCP system combines spatial modulation with $$\mathbf{y}=\mathbf{H}\tilde{\mathbf{x}}+\mathbf{w},$$3-PAM, pilot-aided joint least-squares estimation of channel and dimming coefficients, RSS-based 2D and 3D positioning, and a radical-axis-based geometric refinement. In the reported simulations, localization reached sub-centimeter accuracy at high SNR, BER for $$\mathbf{y}=\mathbf{H}\tilde{\mathbf{x}}+\mathbf{w},$$4 dropped below $$\mathbf{y}=\mathbf{H}\tilde{\mathbf{x}}+\mathbf{w},$$5, and both communication and positioning performed best near the geometric center of the LED layout (Leblebici et al., 6 Aug 2025). The centrality effect is not incidental: it reflects stronger and more symmetric LOS components, more informative RSS gradients, and weaker relative dominance of NLOS reflections.

Other optical JCP systems move away from pure RSS. A compressed-sensing VLC scheme assumes many LEDs simultaneously transmit their positional identifiers; the receiver solves a sparse recovery problem to detect the active LED set and then estimates position by proximity. In a $$\mathbf{y}=\mathbf{H}\tilde{\mathbf{x}}+\mathbf{w},$$6 m square open-plan office with $$\mathbf{y}=\mathbf{H}\tilde{\mathbf{x}}+\mathbf{w},$$7 LEDs, the minimum proximity-method MPE in one configuration was about $$\mathbf{y}=\mathbf{H}\tilde{\mathbf{x}}+\mathbf{w},$$8 m, and higher SNR or longer ID sequences brought performance closer to that bound (Gligoric et al., 2018). An uplink optical wireless positioning system instead treats the line-of-sight peak, second power peak, and LOS–SPP delay of the impulse response as location fingerprints. In a $$\mathbf{y}=\mathbf{H}\tilde{\mathbf{x}}+\mathbf{w},$$9 m room, this yielded about $1$0 cm RMS accuracy with one photodetector and about $1$1 cm with four photodetectors at high SNR (Hosseinianfar et al., 2018). In both cases, the communication channel is not merely a carrier of data; it is the fingerprint source itself.

The highest reported precision in this optical subset comes from passive multi-target VLP based on Multi-Camera Joint Optimization. Multiple pre-calibrated ceiling cameras observe point light sources on multiple targets, first estimate positions from linear least squares over projection rays, then refine them by nonlinear joint optimization minimizing reprojection error. The method achieved about 19% MPE improvement over the baseline in simulation and an experimental 3-D average position error of $1$2 mm (Pan et al., 3 Jun 2025). This indicates that optical JCP can move from coarse RSS geometry to millimeter-scale infrastructure-based localization when imaging geometry is sufficiently well calibrated.

5. Wi‑Fi, UWB, and Multi-Technology JCP

Wi‑Fi-oriented JCP has long been shaped by a coexistence problem: the same channel measurements needed for positioning often interfere with communication. Radio location fingerprinting reuses existing infrastructure such as IEEE 802.11, but active scanning across channels takes the radio off the communication channel. ComPoScan addresses this by switching between active scanning when movement is detected and monitor sniffing when the user is still. In the reported validation, this adaptive policy increased throughput by a factor of 122, decreased delay by a factor of 10, and reduced dropped packets by 73%, while preserving positioning capability (Kjærgaard, 2010). The significance is methodological: JCP in Wi‑Fi is not only about extracting location from communication data, but about managing the protocol-level conflict between acquiring that data and maintaining service continuity.

A more recent Wi‑Fi formulation uses the standardized beamforming feedback matrix rather than raw CSI. FPNet treats BFM as a compressed CSI representation already native to IEEE 802.11ac/ax/be, learns a codeword that supports both beamforming feedback reconstruction and indoor position classification, and adds an AP-side anomaly detector trained on normal reconstructed BFMs. Using 2.4 GHz hardware, FPNet reported 97.52% positioning accuracy at 100 feedback bits, a 22.92% net-throughput gain relative to standard T0 feedback at the same bit budget, anomaly-detection accuracy over 99%, and a false alarm rate below 1.5% (Tao et al., 13 Feb 2026). This is a distinct JCP model from fingerprinting: it is not passively reusing RSS measurements but jointly redesigning a standards-compliant feedback representation for communication efficiency, positioning accuracy, and reliability.

UWB appears in two different JCP roles. As a ranging technology, it forms part of the HYMN dataset, a synchronized industrial-hall corpus that includes UWB, BLE, Wi‑Fi, 5G, and GNSS across indoor, outdoor, and indoor–outdoor transition regions. HYMN logs raw and processed measurements with millisecond-level timestamp alignment and ground truth from a Leica TS16 total station; it reports mean ranging residuals of 1.03 m for UWB, 3.30 m for Wi‑Fi, and 8.07 m for BLE (Ammad et al., 22 Apr 2026). As a social-sensing technology, UWB tags can be used to infer interaction networks from spatial proximity. In a study of 26 participants over approximately two hours, static aggregation over the full session produced only one community at all thresholds, whereas 5-minute temporal segmentation exposed meaningful splits, merges, and participant transitions that matched visually observed groupings (Shinto et al., 8 Oct 2025). A plausible implication is that JCP observables may support either physical localization or latent interaction inference, but the correct temporal granularity is application-dependent.

6. Multipath, Geometry, Tradeoffs, and Research Directions

A central controversy in indoor positioning is whether multipath is fundamentally an impairment or a resource. The answer in the cited literature is conditional rather than absolute. In RSS-based VLC, reflections distort power maps and bias distance estimates, causing meter-scale errors near corners (Gu et al., 2015). In EFIM-based UWB theory, however, specular multipath from a known floor plan can be reinterpreted as virtual anchors whose delays contribute usable localization information, provided the components are resolvable and not lost to path overlap (Leitinger et al., 2014). In uplink optical fingerprinting, the second power peak and LOS–SPP delay are not nuisances but explicit features that improve discriminability (Hosseinianfar et al., 2018). The misconception that multipath is always harmful is therefore inaccurate; its value depends on the signal model, bandwidth, prior geometry, and estimator structure.

Deployment geometry is equally decisive. In PASS-based integrated positioning and communications, the determinant of the least-squares geometry matrix is proportional to the area of the triangle formed by three pinching-antenna positions, so larger triangle area and smaller GDoP reduce error amplification. The same work also shows that non-parallel waveguide deployment improves positioning accuracy in the multi-PA case, even though more PAs can worsen localization because of symmetry points and sidelobes (Zhang et al., 26 May 2026). In 5G-Advanced, angle accuracy depends strongly on array geometry, while synchronization uncertainty and clock offsets directly reduce the available position information (Nikonowicz et al., 2022, Leitinger et al., 2014). These results suggest that anchor count alone is a weak design descriptor; geometry, diversity, and resolvability dominate.

Communication–positioning tradeoffs are explicit in several systems. In PASS, noise has a “serious double-impact on data rate”: it worsens uplink positioning and then degrades downlink rate through suboptimal PA relocation (Zhang et al., 26 May 2026). In dimming-aware VLC, brightness control affects both the offset that must be removed for communication and the RSS map used for positioning (Leblebici et al., 6 Aug 2025). In Wi‑Fi, scan-heavy positioning can collapse throughput unless measurement acquisition is adapted to user motion (Kjærgaard, 2010). In 5G industrial halls, mmWave may be attractive for communication yet still underperform C-band in first-arrival OTDoA positioning under clutter and dynamic blockage (Muthineni et al., 2024). These studies collectively reject the idea that a JCP system can be optimized by maximizing only throughput, only bandwidth, or only positioning accuracy.

Current research directions in this subset therefore converge on four themes. The first is robustness to outliers, hardware mismatch, and out-of-distribution conditions, addressed by IRLS initialization in 5G and anomaly detection in Wi‑Fi (Henninger et al., 2022, Tao et al., 13 Feb 2026). The second is richer environmental modeling, including floor-plan-aware multipath, ray tracing, dimming-aware optical channels, and heterogeneous datasets (Leitinger et al., 2014, Muthineni et al., 2024, Ammad et al., 22 Apr 2026). The third is infrastructure reuse with minimal sensing overhead, whether through standardized feedback, commercial gNB broadcasts, or existing LED and camera deployments (Chen et al., 2021, Tao et al., 13 Feb 2026, Pan et al., 3 Jun 2025). The fourth is hybridization across modalities, especially at indoor–outdoor boundaries where no single technology remains reliable (Ammad et al., 22 Apr 2026). Taken together, these works define JCP not as a single finished architecture, but as an indoor-systems paradigm in which communication signals, hardware, and protocol states are treated as native sources of position information, with performance ultimately bounded by geometry, resolvability, and the realism of the channel model.