JCAS: Integrated Communications and Sensing

• JCAS is a unified system that converges data communications and environmental sensing by reusing spectrum, antennas, and RF hardware to boost efficiency.
• It employs diverse operating modes—communication-centric, sensing-centric, and dual-function—with techniques like STAR-RIS, full-duplex, and hybrid beamforming to optimize performance.
• JCAS presents inherent trade-offs, challenging designers to balance high-rate connectivity with accurate sensing amid constraints in channel estimation, waveform design, and hardware limitations.

Searching arXiv for recent and foundational JCAS papers to ground the article. Searching arXiv for STAR-RIS-enabled JCAS and full-duplex JCAS references. Joint Communications and Sensing (JCAS), also called Integrated Sensing and Communications (ISAC), unifies data communications and environmental sensing within a single system that reuses spectrum, waveforms, antennas, RF chains, and often the same hardware. In 6G research, it is treated as a native network capability whose objectives are simultaneously high-rate, reliable, and low-latency communication and accurate estimation of range, velocity, and angle, while improving spectral efficiency, reducing system complexity, and lowering hardware cost through dual-use RF signaling (Khan et al., 10 Feb 2026). Hexa-X-II frames JCAS as a cross-layer system in which physical-layer measurements such as delays, angles, Doppler, and power are produced, exposed, fused, and acted upon by higher-layer functions, thereby turning the network into a provider of both connectivity and geometric context (Wymeersch et al., 2024).

1. Conceptual scope and terminology

JCAS emerged from the observation that communications and sensing had historically evolved as separated functions, with distinct hardware and spectrum, despite sharing much of the same RF substrate. In separated systems, this duplication incurs infrastructure redundancy and inefficient spectrum utilization. The integrated alternative admits several operating styles. Communication-centric JCAS reuses communication waveforms such as OFDM for opportunistic sensing. Sensing-centric JCAS modulates radar waveforms to carry data. Dual-function JCAS jointly designs signals to balance both tasks. The distinction matters because the optimization variables, interference structure, and estimation fidelity depend strongly on which function remains primary (Khan et al., 10 Feb 2026).

The terminology is not entirely uniform across subfields. In the STAR-RIS survey, JCAS and ISAC are closely related, with JCAS emphasizing joint operation and co-optimization at physical, waveform, and network layers (Khan et al., 10 Feb 2026). In Hexa-X-II, JCAS and ISAC are treated as equivalent concepts, but the emphasis shifts from a purely physical-layer viewpoint to a continuum of integration levels, from loose integration with auxiliary sensors to tight concurrency on the same radio resources (Wymeersch et al., 2024). This suggests that the field is best understood not as a single waveform family, but as a systems paradigm with different degrees of coupling.

The architectural vocabulary is equally important. Monostatic sensing uses co-located transmitter and receiver and offers a simpler echo model, but it tightens self-interference requirements when full duplex is used. Bistatic and multistatic sensing distribute transmit and receive roles across different nodes, improving spatial diversity and coverage. Downlink, uplink, and sidelink sensing are all represented in the literature, and the sensing function may be infrastructure-based, device-based, or cooperative across cells and access points (Khan et al., 10 Feb 2026, Wymeersch et al., 2024).

2. Signal, channel, and performance models

The narrowband JCAS baseband model in STAR-RIS-assisted MIMO form is written as

$$\mathbf{y} = \Big( \mathbf{H}_d + \mathbf{H}_{BR}\boldsymbol{\Theta}_r\mathbf{H}_{RU} + \mathbf{H}_{BT}\boldsymbol{\Theta}_t\mathbf{H}_{TU} \Big)\mathbf{W}\mathbf{s} + \mathbf{n},$$

where $\mathbf{H}_d$ is the direct BS–user channel, $\mathbf{W}$ is the BS precoder, $\mathbf{s}$ is the transmit symbol vector with $E[\mathbf{s}\mathbf{s}^H]=\mathbf{I}$, and $\mathbf{n}\sim\mathcal{CN}(0,\sigma^2\mathbf{I})$. In wideband operation, the effective channel becomes subcarrier dependent,

$$\mathbf{y}[k]=\mathbf{H}_{\mathrm{eff}}[k]\mathbf{W}[k]\mathbf{s}[k]+\mathbf{n}[k],$$

with $\boldsymbol{\Theta}_t[k]$ and $\boldsymbol{\Theta}_r[k]$ capturing frequency-dependent STAR-RIS coefficients caused by meta-atom dispersion (Khan et al., 10 Feb 2026).

Communication performance is usually quantified by achievable rate, SINR, outage probability, latency, and energy efficiency. Under Gaussian signaling, a standard rate expression is

$$R = \log_2\det\!\left( \mathbf{I} + \frac{1}{\sigma^2} \mathbf{H}_{\mathrm{eff}} \mathbf{W}\mathbf{W}^H \mathbf{H}_{\mathrm{eff}}^H \right).$$

Sensing performance is quantified by estimation accuracy for range $\mathbf{H}_d$0, velocity $\mathbf{H}_d$1, and angle $\mathbf{H}_d$2; by Cramér–Rao bound (CRB) or squared position error bound (SPEB); by detection probability $\mathbf{H}_d$3 and false-alarm probability $\mathbf{H}_d$4; by radar SINR or SCNR in clutter; and by beampattern matching error. The core trade-off is explicit: increasing communication rate typically raises multi-user interference and can lower sensing SNR, whereas tightening sensing constraints such as beampattern or CRB thresholds restricts beamforming and reduces rate (Khan et al., 10 Feb 2026).

For monostatic sensing, the echo after matched filtering is modeled as proportional to

$\mathbf{H}_d$5

with Doppler $\mathbf{H}_d$6. The corresponding radar SNR scales approximately as

$\mathbf{H}_d$7

A far-field beampattern is written as

$\mathbf{H}_d$8

which makes the beam-domain coupling between BS and programmable surface explicit (Khan et al., 10 Feb 2026).

The CRB formalism is central throughout the literature. With parameter vector $\mathbf{H}_d$9, mean signal $\mathbf{W}$0, and AWGN variance $\mathbf{W}$1, the Fisher information matrix is

$\mathbf{W}$2

In full-duplex JCAS, range and velocity resolutions are commonly expressed as

$\mathbf{W}$3

and waveform ambiguity is characterized through

$\mathbf{W}$4

which directly links waveform choice to range–Doppler coupling and sidelobe behavior (Sheemar et al., 2023).

3. Waveforms, beamforming, and optimization

Waveform design spans shared OFDM, dedicated radar-like signals such as FMCW or chirp, and fully dual-function constant-modulus, pulsed, or space–time coded designs. OFDM remains particularly prominent because pilots and data subcarriers can be reused for sensing, while delay–Doppler maps can be formed from the same symbols used for communication. At the same time, OFDM sidelobes, PAPR, and wideband beam squint impose sensing penalties, so many JCAS formulations modify waveform support, symbol structure, or per-subcarrier beamforming rather than adopting communications waveforms unchanged (Sheemar et al., 2023, Khan et al., 10 Feb 2026).

Representative design objectives are explicitly multi-objective. A common formulation is

$\mathbf{W}$5

while beampattern synthesis often uses weighted error minimization,

$\mathbf{W}$6

Because these problems couple digital precoders, analog networks, and sensing metrics non-convexly, the algorithmic repertoire includes alternating optimization, block coordinate descent, semidefinite relaxation, fractional programming, ADMM, successive convex approximation, and manifold optimization (Khan et al., 10 Feb 2026).

Several recent works show how these abstractions are instantiated. In hybrid beamforming for massive MIMO JCAS, deep unfolding of a modified projected gradient ascent was reported to achieve up to $\mathbf{W}$7 higher communications sum rate, $\mathbf{W}$8 dB lower beampattern error, and up to a $\mathbf{W}$9 reduction in run time and computational complexity relative to a conventional design based on successive convex approximation and Riemannian manifold optimization (Nguyen et al., 2023). In dynamic precoding, a Lyapunov drift-plus-penalty formulation maximizes long-term average radar SNR subject to a minimum average communications SINR and a power budget, with the classical $\mathbf{s}$0 optimality gap and $\mathbf{s}$1 average queue size trade-off, and a closed-form zero-forcing policy that was reported to achieve nearly the same average radar SNR as the iterative SCA method (Zakeri et al., 18 Mar 2025).

Multi-carrier JCAS also exhibits a resource-structure trade-off. Rather than enforcing sensing on every subcarrier, one proposal assigns JCAS functionality to only a subset of subcarriers while keeping communications on all subcarriers. In an $\mathbf{s}$2 MIMO system with $\mathbf{s}$3 subcarriers, using $\mathbf{s}$4 JCAS subcarriers yielded a $\mathbf{s}$5 improvement in achievable communications rate at $\mathbf{s}$6 dB while generating a sensing beampattern with the same quality as the conventional full-band design (Nguyen et al., 2023). At the symbol level, optimized MIMO-OFDM data symbols were shown to reduce the sensing SNR requirement by $\mathbf{s}$7–$\mathbf{s}$8 dB while incurring only $\mathbf{s}$9–$E[\mathbf{s}\mathbf{s}^H]=\mathbf{I}$0 dB SNR loss for the uncoded BER, using a majorization–minimization algorithm that explicitly suppresses cyclic auto- and cross-correlation sidelobes (Wu et al., 2022).

Learning-based control has become a parallel design line rather than a replacement for model-based optimization. The literature described in the STAR-RIS survey includes model-free DRL with state variables comprising CSI, user QoS targets, target geometry, and protocol state; actions comprising continuous $E[\mathbf{s}\mathbf{s}^H]=\mathbf{I}$1, precoders, and possibly discrete mode selection; and rewards of the form

$E[\mathbf{s}\mathbf{s}^H]=\mathbf{I}$2

with TD3, DDPG, SAC, recurrent DRL, and meta-learning all represented (Khan et al., 10 Feb 2026).

4. Architectures and enabling hardware

A large portion of JCAS research concerns architectural choices that reshape the feasible rate–sensing region before any optimization is run. Full-duplex JCAS is one such choice. A full-duplex base station with separate transmit and receive arrays can transmit downlink, receive uplink, and sense simultaneously in the same band. This configuration supports monostatic sensing from downlink waveforms and multi-bistatic sensing from uplink users, and can provide a two-fold spectral-efficiency gain relative to half-duplex JCAS by enabling concurrent downlink and uplink transmissions (Sheemar et al., 2023). The same paper stresses the cost of that gain: self-interference can be $E[\mathbf{s}\mathbf{s}^H]=\mathbf{I}$3–$E[\mathbf{s}\mathbf{s}^H]=\mathbf{I}$4 dB stronger than desired signals, so practical systems require isolation, analog cancellation, digital cancellation, and SI-aware beamforming to push residual SI near the noise floor (Sheemar et al., 2023).

Not all concurrent designs require full-duplex communications in the classical sense. In concurrent downlink–uplink JCAS, a base station with two spatially separated arrays performs mono-static active sensing using echoes of its own downlink dedicated signals while simultaneously receiving uplink data. A successive interference cancellation pipeline jointly estimates uplink symbols and sensing parameters, and simulations reported more than $E[\mathbf{s}\mathbf{s}^H]=\mathbf{I}$5 dB sensing improvement over traditional downlink JCAS under uplink interference while maintaining reliable uplink communication (Chen et al., 2022). A related downlink–uplink cooperative design uses a unified MUSIC-based sensing module across consecutive TDD slots and reciprocity-aware fusion; the reported minimum location and velocity estimation SMSEs were about $E[\mathbf{s}\mathbf{s}^H]=\mathbf{I}$6 dB lower than those of separated downlink and uplink JCAS schemes (Chen et al., 2022).

Programmable surfaces are now a major JCAS enabler. Conventional RIS are reflective-only programmable metasurfaces. STAR-RIS extends this model by enabling simultaneous transmission and reflection, thereby providing full-space coverage from one surface. Its element coefficients satisfy

$E[\mathbf{s}\mathbf{s}^H]=\mathbf{I}$7

and in the ideal energy-splitting model

$E[\mathbf{s}\mathbf{s}^H]=\mathbf{I}$8

Energy splitting, mode switching, and time switching define the main operating modes. The principal benefit for JCAS is full-space beamforming, which can illuminate and serve both sides of the surface, mitigate blockage, and enlarge the feasible trade-off region between rate and sensing accuracy (Khan et al., 10 Feb 2026).

Other hardware platforms push in different directions. Holographic JCAS replaces discrete phased arrays with reconfigurable holographic surfaces having arbitrary element spacing. For a square RHS with $E[\mathbf{s}\mathbf{s}^H]=\mathbf{I}$9 elements and $\mathbf{n}\sim\mathcal{CN}(0,\sigma^2\mathbf{I})$0 RF chains, the analog beamformer is modeled as

$\mathbf{n}\sim\mathcal{CN}(0,\sigma^2\mathbf{I})$1

and exact CRBs are derived for azimuth and elevation estimation under arbitrary $\mathbf{n}\sim\mathcal{CN}(0,\sigma^2\mathbf{I})$2 and $\mathbf{n}\sim\mathcal{CN}(0,\sigma^2\mathbf{I})$3. The resulting weighted rate–CRB optimization is solved by a majorization–maximization procedure with alternating updates of $\mathbf{n}\sim\mathcal{CN}(0,\sigma^2\mathbf{I})$4 and $\mathbf{n}\sim\mathcal{CN}(0,\sigma^2\mathbf{I})$5, and the reported simulations show monotone objective improvement and lower $\mathbf{n}\sim\mathcal{CN}(0,\sigma^2\mathbf{I})$6 and $\mathbf{n}\sim\mathcal{CN}(0,\sigma^2\mathbf{I})$7 than a benchmark using dominant-eigen digital beamforming with random holographic weights (Sheemar et al., 21 Feb 2025).

Low-power front-end design has likewise become a JCAS topic in its own right. Analog multi-beam antenna arrays based on fixed passive transforms such as RF lens or Butler matrices implement DFT beams without per-element tunable phase shifters. The reported savings for $\mathbf{n}\sim\mathcal{CN}(0,\sigma^2\mathbf{I})$8 antennas and $\mathbf{n}\sim\mathcal{CN}(0,\sigma^2\mathbf{I})$9 RF chains are approximately $$\mathbf{y}[k]=\mathbf{H}_{\mathrm{eff}}[k]\mathbf{W}[k]\mathbf{s}[k]+\mathbf{n}[k],$$0 W relative to phase-shifter-aided hybrid arrays, while a base-2 wide beam synthesized from two adjacent DFT beams yields an approximately $$\mathbf{y}[k]=\mathbf{H}_{\mathrm{eff}}[k]\mathbf{W}[k]\mathbf{s}[k]+\mathbf{n}[k],$$1 dB gain advantage at the intersection angle for AoA estimation (Wu et al., 2022). In a different direction, a low-complexity architecture combining wideband analog and narrowband digital beamforming was shown to achieve performance similar to a fully-digital high-bandwidth system while using a fraction of the total aggregate sampling rate; with $$\mathbf{y}[k]=\mathbf{H}_{\mathrm{eff}}[k]\mathbf{W}[k]\mathbf{s}[k]+\mathbf{n}[k],$$2 and $$\mathbf{y}[k]=\mathbf{H}_{\mathrm{eff}}[k]\mathbf{W}[k]\mathbf{s}[k]+\mathbf{n}[k],$$3 MHz, the proposed architecture used $$\mathbf{y}[k]=\mathbf{H}_{\mathrm{eff}}[k]\mathbf{W}[k]\mathbf{s}[k]+\mathbf{n}[k],$$4 MS/s versus $$\mathbf{y}[k]=\mathbf{H}_{\mathrm{eff}}[k]\mathbf{W}[k]\mathbf{s}[k]+\mathbf{n}[k],$$5 GS/s for a fully-digital design (Bedin et al., 2023). Sparse waveform design follows the same philosophy: the MaRS waveform and hybrid-duplex architecture achieve high sensing resolution with only $$\mathbf{y}[k]=\mathbf{H}_{\mathrm{eff}}[k]\mathbf{W}[k]\mathbf{s}[k]+\mathbf{n}[k],$$6 OFDM resources, reducing the overhead of conventional methods to less than one tenth (Ma et al., 2022).

5. Network realizations and application domains

JCAS is no longer confined to a single-cell downlink abstraction. In cell-free massive MIMO, distributed access points jointly serve users and perform multi-static sensing. A cloud-RAN formulation with centralized power allocation maximizes a sensing SNR surrogate subject to minimum downlink SINR constraints for all UEs, and numerical results show that the detection probability at a given false-alarm probability can be increased significantly relative to a fully communication-centric power allocation, both when additional sensing symbols are used and when sensing relies only on existing communication symbols (Behdad et al., 2022).

Multi-user inference has also been explored beyond classical beamforming. In an uplink SCMA system assisted by an IRS, JCAS is posed as a joint inference problem over two sparsity structures: sparse codebooks on the communications side and sparse voxelized reflectivity on the sensing side. The resulting algorithm alternates SCMA-IRS message passing with GAMP-based environment reconstruction over a sliding window, so that decoded symbols improve the sensing dictionary fit and improved sensing sharpens subsequent multi-user detection (Tong et al., 2021). This line of work is conceptually distinct from radar-style JCAS, but it reinforces a recurring theme: the same measurements can support data detection and environment inference when sparsity, geometry, or protocol structure are exploited jointly.

At the network layer, uncoordinated JCAS has been studied using stochastic geometry. In CSMA-based dual-functional networks, a closed-form framework characterizes radar performance through maximum unambiguous range under a target false alarm probability and communication performance through medium access probability, success probability, and aggregated throughput density. The analysis shows that higher radar duty cycle and larger radar-node fraction increase radar-to-communication interference, while throughput increases and then decreases with network density because contention and interference eventually dominate (Keshtiarast et al., 2023). A different stochastic-geometry line formulates sensing coverage probability and sensing ergodic rate by treating mutual information about target parameters as the sensing analog of a communication rate. In the mmWave downlink setting studied there, network densification improves sensing SINR performance, in contrast to communications (Olson et al., 2022).

Cross-layer and non-terrestrial realizations broaden the concept further. Hexa-X-II introduces dedicated sensing network functions such as the Sensing Management Function and Sensing Processing Function, plus a new data plane for raw sensing measurements, thereby making JCAS an architectural rather than purely signal-processing construct (Wymeersch et al., 2024). In satellite networks, JCAS is motivated by shared payload constraints and by use cases including Earth observation augmentation, maritime monitoring, aviation support, disaster response, positioning augmentation, and space domain awareness. The satellite literature emphasizes that JCAS in LEO is distinguished by extreme Doppler: at Ka band with $$\mathbf{y}[k]=\mathbf{H}_{\mathrm{eff}}[k]\mathbf{W}[k]\mathbf{s}[k]+\mathbf{n}[k],$$7 GHz and relative speed about $$\mathbf{y}[k]=\mathbf{H}_{\mathrm{eff}}[k]\mathbf{W}[k]\mathbf{s}[k]+\mathbf{n}[k],$$8 km/s, one-way Doppler is approximately $$\mathbf{y}[k]=\mathbf{H}_{\mathrm{eff}}[k]\mathbf{W}[k]\mathbf{s}[k]+\mathbf{n}[k],$$9 kHz and monostatic Doppler approximately $\boldsymbol{\Theta}_t[k]$0 MHz (Sheemar et al., 9 Jan 2025). This makes Doppler-resilient waveform design and synchronization first-order design variables rather than implementation details.

Application-specific studies in the STAR-RIS survey cover vehicular JCAS, UAV-assisted designs, indoor localization and SLAM, and NTNs. Vehicular JCAS uses STAR-RIS to assist V2I and V2V while sensing pedestrians or vehicles. UAV-assisted JCAS jointly optimizes trajectory or hovering and STAR-RIS coefficients. Indoor localization and SLAM benefit from richer transmission and reflection paths, while sensor-embedded STARS provide bi-directional sensing. In NTNs and mmWave/THz systems, blockage mitigation across both sides of a surface and near-field focusing with large apertures become central deployment themes (Khan et al., 10 Feb 2026).

6. Open challenges, controversies, and research directions

The most persistent issue in JCAS is that its trade-offs are structural rather than incidental. Improving communication rate can reduce sensing SNR through increased multi-user interference. Tightening CRB or beampattern constraints can reduce rate. Full duplex can double spectral efficiency, but only if residual self-interference is driven near the noise floor. RIS and STAR-RIS can enlarge the feasible region, but they introduce hard constraints such as unit modulus, amplitude–phase coupling, insertion and leakage loss, calibration error, and finite-resolution quantization (Khan et al., 10 Feb 2026, Sheemar et al., 2023).

Channel knowledge is a second bottleneck. Large surfaces induce cascaded channels on both transmission and reflection sides, and joint estimation of $\boldsymbol{\Theta}_t[k]$1 is costly. Mobility compounds the difficulty, since coherence time, queue stability, and adaptive control all compete with training overhead. The literature therefore emphasizes overhead-aware estimation, sparse or parametric pilot design, compressed sensing, predictive control, and task-driven channel inference rather than brute-force CSI acquisition (Khan et al., 10 Feb 2026, Zakeri et al., 18 Mar 2025).

Near-field and wideband operation create a third frontier. Large apertures invalidate plane-wave approximations, requiring spherical-wave and distance-domain models. THz bands add beam squint, frequency-selective surface responses, and three-side beam split. Holographic arrays, STAR-RIS, and hybrid analog–digital front ends each expose different combinations of these effects. This suggests that future JCAS formulations will increasingly optimize over geometry-aware signal models rather than rely on far-field narrowband surrogates (Sheemar et al., 21 Feb 2025, Khan et al., 10 Feb 2026).

Privacy, security, sustainability, and standardization are no longer peripheral topics. The open problems listed in the STAR-RIS survey include secure and covert JCAS under eavesdropping, jamming, and sensing-aided attacks; holistic energy models including controller and circuit power; and placement optimization for green operation (Khan et al., 10 Feb 2026). Hexa-X-II frames reliability, security, and ethical operation as trustworthiness KPIs, while 3GPP, IEEE 802.11 TGBF, and ETSI ISG are all identified as relevant standardization venues for sensing channel models, PHY/MAC features, and architectural impacts (Wymeersch et al., 2024).

The cumulative literature points toward a specific 6G vision: intelligent, flexible, perceptive networks that combine shared waveforms, programmable propagation, distributed inference, and cross-layer orchestration. What remains unsettled is not whether communication and sensing can be co-located on the same platform, but how far integration can be pushed before estimation accuracy, synchronization, hardware realism, privacy, and run-time constraints dominate the design space.