JCAS: Joint Communication & Adaptive Sensing

• JCAS is a unified wireless system design that integrates communication and sensing by sharing RF resources, waveforms, and adaptive control variables.
• Adaptive sensing in JCAS dynamically adjusts pilot density, time, power, and beamforming to balance data throughput with accurate environmental detection.
• JCAS applications span autonomous driving, smart cities, and UAV networks, offering enhanced detection performance and resource efficiency through joint optimization.

Searching arXiv for papers on Joint Communication and Adaptive Sensing / JCAS to ground the article. Joint Communication and Adaptive Sensing (JCAS) denotes wireless systems in which communication and sensing are not merely co-located, but are jointly designed so that shared spectrum, waveforms, hardware, beamformers, and control variables can be adapted to changing communication demands, sensing tasks, interference conditions, and operating constraints. In the 6G literature, JCAS is also presented under the broader label of joint communications and sensing or integrated sensing and communication, with the common premise that the radio channel is both a transport medium for data and a source of environmental information (Khan et al., 10 Feb 2026). The “adaptive” qualifier is warranted when the system does not keep sensing resources fixed, but instead varies quantities such as pilot density, time allocation, power allocation, beam selection, waveform structure, or surface coefficients in response to state, geometry, clutter, mobility, or long-term utility objectives (Guven et al., 13 Mar 2026).

1. Conceptual scope and architectural regimes

JCAS merges two historically separate wireless functions into one unified system. Instead of designing communication and sensing independently with separate radios, spectra, waveforms, and signal-processing chains, JCAS reuses the same RF resources to transmit data and sense the environment simultaneously (Khan et al., 10 Feb 2026). This integration is motivated by use cases that require both connectivity and environmental perception, including autonomous driving, industrial automation, UAV operation, smart cities, XR, and non-terrestrial networks (Khan et al., 10 Feb 2026). A central point across the literature is that JCAS is fundamentally a multi-objective design problem: communication seeks reliable, high-throughput delivery, while sensing seeks accurate detection and estimation of targets or environmental parameters, and these objectives are often in tension (Khan et al., 10 Feb 2026).

The literature distinguishes several architectural axes. At the integration level, systems are described as coexistence-based, cooperative, or fully joint/unified, depending on whether communication and sensing merely share spectrum, exchange information across subsystems, or are implemented by one hardware platform and one waveform (Khan et al., 10 Feb 2026). Geometrically, JCAS appears in monostatic, bistatic, and multistatic forms, with the distinction determined by whether the transmitter and receiver are co-located, spatially separated, or distributed (Khan et al., 10 Feb 2026). This geometric classification is operationally important because it changes the propagation paths, the interference structure, and the observability of sensing parameters.

Adaptive sensing enters at the architectural level when the system is allowed to vary how much time, power, or waveform structure is devoted to sensing relative to communication. In cellular bistatic JCAS, this appears as the contrast between time-division mode (TDM), where sensing and communication are separated in time, and concurrent mode (CM), where they overlap in time and are separated by multibeam or spatial allocation (Vinogradova et al., 2024). In downlink systems with dedicated sensing and communications precoders, adaptation can also be temporal: precoders are redesigned slot by slot so that sensing quality is optimized in the long run while communication quality-of-service is enforced only in an average sense (Zakeri et al., 18 Mar 2025). In multi-UAV networks, adaptation is even more explicit: each UAV jointly controls mobility and OFDM pilot density, so that sensing reliability, communication throughput, battery evolution, charging, and carbon cost are co-optimized within one decision process (Guven et al., 13 Mar 2026).

2. System models, signal structure, and performance criteria

JCAS papers use a broad set of signal models, but several recurring structures are evident. In cellular sensing with clutter, a bistatic model is used in which a transmit base station serves a user equipment for downlink communication while a sensing receiver base station observes reflections from passive targets, clutter clusters, and temporally correlated noise (Vinogradova et al., 2024). In uplink JCAS, known OFDM preamble symbols are exploited simultaneously for channel estimation and sensing-parameter estimation such as angle of arrival, range, and Doppler (Chen et al., 2022). In multi-user MIMO JCAS, the base station transmits information-bearing signals to communication users while concurrently estimating the parameters of a sensing object from the echo signal (Perera et al., 13 Jan 2026). In RIS- and STAR-RIS-assisted settings, the propagation environment itself becomes reconfigurable through transmission, reflection, or simultaneous transmission-and-reflection coefficients (Khan et al., 10 Feb 2026).

The metrics reflect the dual nature of the problem. Communication objectives are quantified in terms of rate, sum rate, weighted sum rate, ergodic rate, spectral efficiency, signal-to-interference-plus-noise ratio, outage, secrecy rate, or fairness-related measures (Khan et al., 10 Feb 2026). Sensing objectives are quantified using Cramér–Rao bounds, sensing SINR, radar SINR, signal-to-clutter-plus-noise ratio, beampattern criteria, detection probability, detection error, or estimation accuracy for angle, range, velocity, or location (Khan et al., 10 Feb 2026). Some works place these utilities into a single Pareto or weighted-sum formulation, using mutual information for communication and Fisher information for sensing, then tracing the tradeoff boundary by sweeping a scalar weight (Perera et al., 13 Jan 2026). Other works select one sensing metric as the primary objective and treat communication as a hard or average constraint, as in long-term sensing-SNR maximization with minimum average communication SINR (Zakeri et al., 18 Mar 2025).

Spatial and network-scale analyses also introduce distributional performance measures. Rather than reporting only average coverage probability, stochastic-geometry work derives the meta distribution of the SIR for JCAS networks, thereby quantifying how conditional communication and sensing reliability varies across random network geometries (Ma et al., 2024). This line of analysis is directly relevant to adaptive sensing because it exposes spatial heterogeneity that average metrics conceal, suggesting that waveform allocation, scheduling, dwell time, or beam selection should depend on local geometric and interference conditions (Ma et al., 2024).

3. Adaptive control of time, power, beams, and waveforms

A defining feature of JCAS is that the tradeoff between communication and sensing is controlled by tunable resources. In cellular JCAS, TDM uses a time-allocation parameter $\alpha$, with sensing duration $T_s=\lfloor \alpha T \rfloor$, while CM uses a power-allocation parameter $\delta$, with sensing power $P_s=\delta P$ and communication power $P_c=(1-\delta)P$ (Vinogradova et al., 2024). Simulation results indicate that, in general, TDM gives somewhat better target-detection performance than CM, although both outperform existing approaches when their tradeoff parameters are tuned properly (Vinogradova et al., 2024). The associated intuition is that CM must divide power between sensing and communication at the same time, whereas TDM dedicates full transmit structure to sensing during sensing slots, improving the separation between target-induced spectral outliers and the clutter/noise bulk (Vinogradova et al., 2024).

Power adaptation also appears in cell-free massive MIMO. In a centralized cloud radio access architecture with multi-static sensing, the transmit access points jointly serve user equipments and optionally steer a beam toward a target. The design variable is the vector of power coefficients over communication and sensing streams, optimized to maximize sensing SNR under minimum per-user SINR and per-AP power constraints (Behdad et al., 2022). This formulation represents adaptive sensing in the precise sense that sensing performance is not incidental to communication beamforming; rather, power is redistributed in a sensing-aware way while preserving communication feasibility. The reported outcome is that, compared with a fully communication-centric power allocation, detection probability at fixed false alarm probability can be increased significantly, both with dedicated sensing symbols and when using only existing communication symbols (Behdad et al., 2022).

Dynamic precoding extends the same idea to stochastic control over time. In a downlink system with separate communication and radar precoders, the objective is to maximize long-term average sensing SNR subject to a minimum average communications SINR and a power budget (Zakeri et al., 18 Mar 2025). A virtual queue enforces the average communication constraint, and Lyapunov drift-plus-penalty transforms the problem into a sequence of per-slot non-convex programs (Zakeri et al., 18 Mar 2025). Two solvers are given: a successive convex approximation procedure and a closed-form zero-forcing design. A notable structural result is that, under the zero-forcing policy, only one subsystem is active at a time: either all power goes to radar or all power goes to communication, depending on which yields the larger per-slot objective (Zakeri et al., 18 Mar 2025). This suggests that “adaptive sensing” in JCAS can involve intentional temporal concentration of resources rather than continuous simultaneous sharing.

Waveform adaptation can be finer-grained than beam or power allocation. In a multi-UAV OFDM setting, pilot density is the explicit JCAS control variable, with $p_{\text{pilot}}(t)\in[0.01,0.30]$ and $p_{\text{comm}}(t)=1-p_{\text{pilot}}(t)$ (Guven et al., 13 Mar 2026). Increasing pilot density improves sensing but reduces communication payload capacity; decreasing it has the opposite effect (Guven et al., 13 Mar 2026). Simulation results show that adaptive pilot-density control learned by the agents outperforms a static $p_{\text{pilot}}=0.3$ baseline, with improvements around $10\%$ for larger fleets and around $5\%$ for 10 UAVs in some settings (Guven et al., 13 Mar 2026). At the waveform-design level, MIMO-OFDM data symbols themselves can also be optimized so that their time-domain realizations exhibit lower cyclic autocorrelation sidelobes and lower cyclic cross-correlation, reducing the SNR required for sensing by $T_s=\lfloor \alpha T \rfloor$0–$T_s=\lfloor \alpha T \rfloor$1 dB while incurring only $T_s=\lfloor \alpha T \rfloor$2–$T_s=\lfloor \alpha T \rfloor$3 dB uncoded BER loss (Wu et al., 2022).

4. Detection, estimation, and sensing-aided communication

JCAS adaptive sensing is not limited to resource allocation; it also includes adaptive inference strategies that exploit the structure induced by shared communication-sensing signaling. In cluttered cellular sensing, an eigenvalue-ratio detector based on random matrix theory estimates the number of detectable targets from the eigenvalues of the sensed sample covariance matrix, explicitly allowing clutter clusters and temporally correlated noise (Vinogradova et al., 2024). The target term is modeled as a fixed-rank perturbation that produces outlier eigenvalues detached from the clutter-plus-noise bulk, and the estimated target count is determined by consecutive eigenvalue ratios (Vinogradova et al., 2024). Existing model-order criteria such as MDL and AIC are reported to work in white noise but to fail in correlated-noise or cluttered settings (Vinogradova et al., 2024).

Cooperative and concurrent schemes show how communication and sensing can aid one another. In concurrent downlink and uplink JCAS, the base station transmits dedicated sensing symbols during the uplink slot and uses a successive-interference-cancellation pipeline that first decodes uplink data, reconstructs and subtracts it, and then estimates sensing range and velocity from the remaining echoes (Chen et al., 2022). The reported effect is more than a $T_s=\lfloor \alpha T \rfloor$4 dB performance improvement over traditional JCAS methods, and, in the detailed simulation comparison, the proposed method reduces the transmit power required to reach the same minimum range MSE by $T_s=\lfloor \alpha T \rfloor$5 dB or $T_s=\lfloor \alpha T \rfloor$6 dB depending on uplink power, while remaining only about $T_s=\lfloor \alpha T \rfloor$7 dB worse than an idealized no-uplink baseline (Chen et al., 2022). In downlink-uplink cooperative JCAS, a unified MUSIC-based sensing module is applied in both directions, and cross-slot fusion uses reciprocity to identify the communication user among detected reflectors and to improve both sensing and CSI estimation (Chen et al., 2022). The minimum location and velocity estimation mean square errors are reported to be about $T_s=\lfloor \alpha T \rfloor$8 dB lower than those of separated downlink and uplink JCAS schemes (Chen et al., 2022).

A complementary direction treats sensing as prior information for communication estimation. In uplink JCAS with OFDM preambles, angle-of-arrival estimates from MDL-based MUSIC are used to construct a sensing-aided Kalman filter that denoises least-squares CSI estimates across the array domain (Chen et al., 2022). The key observation is that AoA is more robust than range or Doppler to timing and frequency offsets because the offset-dependent exponentials cancel in the covariance structure used for AoA estimation (Chen et al., 2022). The resulting filter achieves BER close to MMSE with complexity $T_s=\lfloor \alpha T \rfloor$9 rather than $\delta$0, requiring about $\delta$1 dB less SNR than LS for the same 4-QAM BER and remaining only about $\delta$2 dB worse than MMSE (Chen et al., 2022). This suggests that adaptive sensing in JCAS can also mean adaptive use of sensed structure as a communication prior.

5. Learning-based JCAS and multi-agent adaptation

Learning is increasingly used when closed-form optimization is too slow, too model-sensitive, or incompatible with partial observability. In multi-UAV hotspot detection, the JCAS problem is formulated as a Dec-POMDP with centralized training and decentralized execution, where each agent observes local position, state of charge, pilot density, communication/SNR estimates, hotspot knowledge, neighboring UAV positions, carbon intensity, depot distance, and progress statistics (Guven et al., 13 Mar 2026). The action consists of motion and pilot-density control, and the reward combines mission completion, informed consensus, coverage gain, communication throughput, energy penalties, carbon penalties, revisit penalties, and knowledge-spread terms (Guven et al., 13 Mar 2026). Information propagates over a dynamic communication graph through iterative logical-OR consensus, so the learned policy must manage sensing, mobility, connectivity, energy, and carbon cost jointly (Guven et al., 13 Mar 2026).

Beamforming adaptation has likewise been cast as reinforcement learning. In massive-MIMO JCAS, a causality-driven TD3-INVASE framework augments standard actor-critic learning with a selector network that identifies action dimensions causally relevant to beamforming gain (Roy et al., 2024). The purpose is to avoid inefficient exploration in large antenna-phase action spaces, particularly when many action dimensions are redundant (Roy et al., 2024). Experiments based on DeepMIMO show that the proposed method reaches higher beamforming gain faster than TD3 and DDPG, with communication-beam true positive rate often reaching $\delta$3 and sensing-beam true positive rate around $\delta$4 (Roy et al., 2024). The paper also notes an important limitation: causal relevance is environment-dependent, so performance can degrade under large unseen antenna tilts when more action dimensions become relevant (Roy et al., 2024).

The broader STAR-RIS survey frames learning as essential for dynamic environments with uncertain channel statistics or prohibitive optimization latency (Khan et al., 10 Feb 2026). The surveyed methods include TD3, DDPG, SAC, and meta-learning variants for optimizing STAR-RIS coefficients, beamforming, trajectory, scheduling, and waveform adaptation (Khan et al., 10 Feb 2026). This suggests that adaptive sensing in JCAS is moving from static tradeoff analysis toward closed-loop, state-dependent control policies that jointly manipulate active beamforming, passive surfaces, waveform structure, and mobility.

6. Enabling hardware, programmable environments, and unresolved issues

JCAS adaptation is strongly conditioned by hardware architecture. A low-complexity receiver architecture combines wideband analog beamforming for time-of-arrival estimation with narrowband digital beamforming for angle-of-arrival estimation, then associates delay and angle across multiple non-coherent frames without requiring 2D or 3D joint estimation (Bedin et al., 2023). For $\delta$5 and $\delta$6 MHz, the reported total sampling rates are $\delta$7 MS/s for classical analog beamforming, $\delta$8 MS/s for the proposed architecture, and $\delta$9 GS/s for fully digital MIMO, with corresponding power figures of $P_s=\delta P$0 mW, $P_s=\delta P$1 mW, and $P_s=\delta P$2 W (Bedin et al., 2023). The method achieves performance similar to a fully digital high-bandwidth system while requiring a fraction of the aggregate sampling rate and much lower complexity (Bedin et al., 2023). Multibeam analog-array designs pursue a related goal by separating a fixed communication subbeam from packet-varying scanning sensing subbeams, thereby decoupling the angular requirements of communication and sensing within one steerable array (Zhang et al., 2018).

Programmable surfaces further enlarge the adaptive design space. STAR-RIS extends conventional reflective RIS by allowing full transmission, full reflection, or simultaneous transmission-and-reflection under the energy-conservation constraint $P_s=\delta P$3, yielding energy-splitting, mode-switching, and time-switching protocols with different tradeoff regions (Khan et al., 10 Feb 2026). In holographic JCAS, exact Cramér–Rao bounds for azimuth and elevation are derived for reconfigurable holographic surfaces with sub-wavelength antenna spacing, and communication rate is optimized jointly with the sum of the two CRBs through an MM-based alternating scheme (Sheemar et al., 21 Feb 2025). In RIS-assisted MIMO JCAS under mutual coupling, the physically consistent RIS response is modeled as $P_s=\delta P$4 rather than a diagonal phase-shift matrix, and numerical results indicate that incorporating mutual coupling improves both mutual information and Fisher information relative to conventional RIS-JCAS models (Wijekoon et al., 13 Jan 2026). These results suggest that adaptive sensing increasingly includes surface-level electromagnetic control, not only digital resource scheduling.

Several unresolved issues recur across the literature. Clutter, temporally correlated noise, and small clutter-rank regimes can raise false alarms or obscure target-induced spectral outliers (Vinogradova et al., 2024). CSI acquisition remains difficult in STAR-RIS systems because both transmitted and reflected cascaded channels must be estimated, particularly under mobility and large surface sizes (Khan et al., 10 Feb 2026). Hardware realism remains a gap because many analyses assume continuous, independent amplitude-phase control while practical surfaces exhibit coupled responses, quantization, mutual coupling, insertion loss, and leakage (Khan et al., 10 Feb 2026). At the network level, privacy concerns are substantial: JCAS introduces location tracking, identity disclosure, profiling, and misuse of sensed data, prompting proposals for an enhanced architecture with Sensing Policy, Consent, and Transparency Management, a dedicated sensing store, and interface-level threat analysis via STRIDE and LINDDUN (Dass et al., 2024). A plausible implication is that future JCAS systems will have to adapt not only to channels, clutter, and traffic, but also to policy, consent, disclosure, and trust-boundary constraints.

The current body of work therefore presents JCAS as a layered adaptive system. At the physical layer, it adapts beams, waveforms, precoders, pilots, and programmable surfaces. At the inference layer, it adapts detection and estimation algorithms to clutter, correlated noise, and reciprocity structure. At the network layer, it adapts to geometry, mobility, multi-agent coordination, and deployment heterogeneity. And at the architectural layer, it increasingly incorporates hardware constraints and privacy-governance mechanisms. Taken together, these strands indicate that “adaptive sensing” in JCAS is not a single algorithmic technique, but a design principle that spans waveform generation, resource allocation, channel inference, programmable propagation, and system control.