Unified ISAC Framework Overview

• Unified ISAC framework is a comprehensive model that integrates sensing and communication using shared waveform, spectrum, and control resources.
• It employs analytical methodologies like mutual information, broadcast-channel signaling, and CRLB formulations to optimize performance.
• The framework addresses resource allocation trade-offs and security challenges, paving the way for adaptive and efficient ISAC systems.

Unified ISAC framework denotes a family of analytical, architectural, and operational formulations in which integrated sensing and communication are modeled as one system rather than as two loosely coupled subsystems. Across recent work, the unifying premise is that sensing and communication share waveform, frequency band, spectrum, hardware, and control resources, so they must also share abstractions for performance evaluation, channel generation, resource allocation, security, or deployment. In that sense, unification has been expressed through mutual information, broadcast-channel signaling, sensing QoS–aware resource allocation, extended geometry-based stochastic channel models, metasurface transceiver abstractions, and full-stack virtualization of signal generation, propagation, and acquisition (Ouyang et al., 2022, Nikbakht et al., 13 Sep 2025, Liu et al., 3 Dec 2025, Pu et al., 23 Apr 2026).

1. Conceptual scope of unification

A unified ISAC framework begins from the observation that the same transmissions may simultaneously carry data and probe the environment. In the mutual-information formulation, sensing and communication are supported within the same time-frequency-power-hardware resources, rather than under conventional frequency-division sensing and communications. In the multi-domain transportation architecture, the same transmissions also feed back into beam alignment, tracking, and resource adaptation, so sensed information is no longer external to communication control (Ouyang et al., 2022, Keskin et al., 20 Nov 2025).

The same idea can be stated in signaling terms. In the broadcast-channel formulation, the transmitter serves an actual communication user and a virtual sensing user, so sensing enters the system as a destination of broadcast signaling rather than as a separate radar chain. This makes superposition coding, dirty-paper coding, and related multiplexing schemes directly applicable to ISAC. The communication-centric and sensing-centric operating modes differ only in which layer is synthesized first and which function is treated as the higher-priority user (Nikbakht et al., 13 Sep 2025).

This suggests that “unified” is not a single canonical model. In the current literature, the term may refer to a common metric space, a common signaling abstraction, a common resource-allocation layer, a common channel-modeling methodology, or a common control loop. The underlying theme is stable: once sensing and communication share physical resources and system state, their design variables cannot be treated as independent.

2. Analytical foundations

The earliest explicit analytical unification is the mutual-information framework. Communication is quantified by conditional mutual information such as $I(X_{\mathrm D};Y_{\mathrm C}\mid H)$ or $I(X_{\mathrm C};Y_{\mathrm U}\mid H)$, while sensing is quantified by $I(G;Y_{\mathrm S}\mid X_{\mathrm D})$ or $I(G;Y_{\mathrm U}\mid X_{\mathrm S})$. This yields a common information-theoretic language, a common unit, and a common set of optimization tools. Within that framework, the principal system-level objects are the sensing-communication rate region and the high-SNR slope. The resulting conclusions are regime dependent: in single-antenna downlink ISAC, one dual-functional signal can simultaneously maximize communication rate and sensing rate; in multi-antenna downlink, the rate region is not rectangular because the sensing-optimal and communication-optimal beamformers differ; in uplink, sensing-centric and communication-centric SIC generate two extreme points connected by time-sharing (Ouyang et al., 2022).

A different analytical unification appears in the broadcast-channel framework for MIMO-OFDM ISAC. There, one transmit signal is decomposed as $$\mathbf{x}(t)=f_{\mathrm{sp}}(\mathbf{x}^{(s)}(t),\mathbf{x}^{(c)}(t))$$, and the sensing and communication functions are interpreted as layered BC signaling toward actual and virtual users. The value of this abstraction is not only conceptual. It permits direct use of dirty-paper coding, frequency-division multiplexing, and superposition coding, and it accommodates the cases where the sensing waveform is known or unknown at the communication receiver. In this view, ISAC tradeoffs are naturally expressed through feasible regions analogous to BC operating regions (Nikbakht et al., 13 Sep 2025).

A third analytical strand unifies sensing accuracy bounds rather than rates. The CRLB framework starts from a generic received signal model and derives a waveform-agnostic FIM and EFIM for delay, Doppler, and angle. Its central technical point is the explicit treatment of the delay–Doppler coupling terms $C_0$ and $C_1$, together with the conditions under which they vanish or become negligible. This produces a common comparison basis across FMCW, PMCW, OFDM, OTFS, and virtual-array sensing, and shows that several waveform-specific CRLBs are special cases of one generic framework. The same logic extends to TDM, FDM, and CDM multiplexing in virtual-array sensing, so waveform choice and multiplexing choice can be compared within one information-bound formalism (Su et al., 27 May 2026).

3. Resource-allocation formulations

At the network level, a unified ISAC framework treats sensing as a service with explicit sensing QoS, not merely as an incidental by-product of communication. In perceptive networks, sensing QoS is defined by the probability of detection for detection tasks, the Cramér–Rao bound for localization, and the posterior CRB for moving target indication. These metrics are embedded in one optimization template: maximize sensing QoS subject to communication QoS and total power and bandwidth constraints. The framework is further organized by two generic policies: fairness, which maximizes the worst sensing QoS among targets, and comprehensiveness, which maximizes aggregate sensing utility under proportional importance constraints (Dong et al., 2022).

In OFDM-ISAC, the same unification is realized directly on the time-frequency lattice. A single OFDM frame of $N$ subcarriers and $M$ OFDM symbols is shared by sensing and multi-user communication through binary RE-allocation variables and continuous power loading. The framework derives closed-form expressions for delay resolution, Doppler resolution, delay-Doppler peak sidelobe level, and received sensing SNR as explicit functions of the sensing RE selector $\mathbf{u}_0$ and the power vector $I(X_{\mathrm C};Y_{\mathrm U}\mid H)$0. Communication performance is measured by multi-user sum-rate on the same grid. Two optimization problems then follow naturally: a resolution-oriented design that minimizes weighted delay-Doppler widths under PSL, sensing SNR, rate, and power constraints, and a sidelobe-oriented design that minimizes PSL under resolution, sensing SNR, and communication constraints. Both are solved through Dinkelbach’s transform and majorization-minimization (Li et al., 9 Apr 2025).

These formulations clarify a recurrent ISAC tradeoff. Power, bandwidth, and occupancy do not affect sensing and communication symmetrically. In sensing-as-a-service formulations, increasing the communication threshold forces a reallocation of power and bandwidth away from sensing targets, which lowers detection probability, worsens localization CRBs, and enlarges tracking PCRBs. In OFDM-ISAC, concentrating sensing REs near time-frequency edges sharpens nominal resolution, but can increase ambiguity or sidelobes unless additional REs are spent on suppression. The important point is that a unified framework turns these tradeoffs into explicit constraint sets rather than informal design tensions.

4. Channel, target, and environment models

A major line of work uses channel modeling as the unifying layer. In the extended GBSM formulation, the sensing channel is decomposed into a target channel and a background channel. The target channel is modeled as a concatenated $I(X_{\mathrm C};Y_{\mathrm U}\mid H)$1 link, with explicit target scattering points, angle-dependent RCS, and delay formation by convolution of the two sub-links. The background channel remains close to legacy 3GPP communication GBSM, but is modified by a power control factor that captures the target’s impact on background clutter. This structure applies to both mono-static and bi-static sensing and preserves 3GPP-style large-scale and small-scale parameter generation (Zhang et al., 14 Apr 2025).

The Release 19 survey generalizes that architecture into a standardization-oriented methodology. Its central definition is

$I(X_{\mathrm C};Y_{\mathrm U}\mid H)$2

with additional support for physical objects, environment objects, shared clusters, spatial consistency, macro-Doppler, micro-Doppler, and monostatic background generation through a multi-reference-point model. The resulting E-GBSM is implemented in a standardized simulator and calibrated in two phases against Release 19 reference assumptions. At this layer, “unified” means one channel-modeling methodology for communication links, target echoes, background clutter, deterministic environment objects, and partially shared scatterers. The open tension between concatenated target modeling and simpler substitutes is explicitly treated as a fidelity–complexity tradeoff rather than a settled standard (Liu et al., 3 Dec 2025).

Target scattering itself has also been given a unified abstraction. The RCS framework for 3GPP standardization decomposes target scattering as

$I(X_{\mathrm C};Y_{\mathrm U}\mid H)$3

where $I(X_{\mathrm C};Y_{\mathrm U}\mid H)$4 is a large-scale power factor, $I(X_{\mathrm C};Y_{\mathrm U}\mid H)$5 is a small-scale angular-dependent component, and $I(X_{\mathrm C};Y_{\mathrm U}\mid H)$6 is a random component. This form is intended to support both system-level and link-level simulation, and it is validated for UAV, human, and vehicle targets across five frequency bands. In channel-generation terms, the decomposition is attractive because $I(X_{\mathrm C};Y_{\mathrm U}\mid H)$7 maps naturally to large-scale target-channel strength, while $I(X_{\mathrm C};Y_{\mathrm U}\mid H)$8 and $I(X_{\mathrm C};Y_{\mathrm U}\mid H)$9 shape delay spread, angular spread, and target-dependent multipath structure (Zhang et al., 27 May 2025).

5. Architectural realizations and deployment layers

At the transceiver layer, unification has been pursued through modular hardware abstractions. In intelligent-metasurface-enabled ISAC, a broad class of transceivers is represented by three components—baseband processing $I(G;Y_{\mathrm S}\mid X_{\mathrm D})$0, RF feeding $I(G;Y_{\mathrm S}\mid X_{\mathrm D})$1, and metasurface reconfiguration $I(G;Y_{\mathrm S}\mid X_{\mathrm D})$2—combined through the cascaded operator $I(G;Y_{\mathrm S}\mid X_{\mathrm D})$3. This yields a common communication model,

$I(G;Y_{\mathrm S}\mid X_{\mathrm D})$4

and a common sensing beampattern model,

$I(G;Y_{\mathrm S}\mid X_{\mathrm D})$5

RIS, SIM, DMA, and RHS then appear as special cases distinguished by the structure of $I(G;Y_{\mathrm S}\mid X_{\mathrm D})$6, $I(G;Y_{\mathrm S}\mid X_{\mathrm D})$7, $I(G;Y_{\mathrm S}\mid X_{\mathrm D})$8, and the number of layers $I(G;Y_{\mathrm S}\mid X_{\mathrm D})$9 (Li et al., 16 Jun 2025).

At the sensing-algorithm layer, wideband near-field XL-MIMO localization is unified by a joint spatial-frequency steering manifold that retains both spherical-wave curvature and frequency dependence. The full WB-NF model supports a coherence-aware compressed-sensing dictionary in the angle-distance domain, while two derived metrics define the effective boundaries of the NB-NF and WB-FF regions. The result is a regime-adaptive framework: NB-NF MUSIC is appropriate when wideband effects are negligible, WB-FF MUSIC is appropriate when near-field curvature is negligible, and a full WB-NF CS-based method is required in the intermediate “gray zone” where neither approximation is reliable (Zhang et al., 29 Mar 2026).

A deployment-oriented extension virtualizes the entire sensing stack on commercial cellular infrastructure. Signal generation is virtualized by coordinating distributed BSs, fragmented carriers, and separated time slots into a space–time–frequency synthetic ISAC network. Signal propagation is virtualized by treating specular multipath, together with digital maps, as virtual radars or virtual arrays. Signal acquisition is virtualized by sub-Nyquist sampling and structured recovery, so a low-rate ADC can function as a “virtual high-rate ADC.” The framework is explicitly full stack because aperture, bandwidth, coherent integration time, multipath geometry, and acquisition bandwidth are all synthesized computationally rather than by hardware replacement (Pu et al., 23 Apr 2026).

Beyond transceivers and channels, unification also appears at the observation-processing layer. Human sensing and environmental sensing have been described by one propagation-to-inference pipeline: wireless infrastructure emits or carries communication signals, physical phenomena perturb propagation, these perturbations appear in CSI, Doppler, delay, AoA, RSSI, RSRP, or channel statistics, and inference models map those observables to application outputs such as localization, activity recognition, rainfall intensity, soil moisture, or water level. This does not yield a single protocol stack, but it does identify a common sensing substrate across domains that are usually studied separately (Wu et al., 18 Jul 2025).

6. Security, limitations, and open directions

Security-oriented work extends the idea of unification beyond performance and modeling into trust and control. In transportation-focused 6G ISAC, security is explicitly multi-domain: cyber-physical, physical-layer, and protocol/security-management. The proposed architecture is organized around a four-function security cycle consisting of authentication fusion, cross-layer key generation, cross-layer anomaly detection, and dynamic security adaptation. At this level, the claim is that secure communication protocols alone are insufficient because attacks can propagate through shared sensing, waveform, hardware, and control loops without violating digital credentials (Keskin et al., 20 Nov 2025).

Near-field secure ISAC pushes that logic into hierarchical control. A Bayesian–Stackelberg framework couples belief-driven sensing and beamforming with adaptive hybrid node role switching between secure transmission and cooperative jamming. The unifying state variable is a posterior belief over adversary directions, which drives sensing intensity, protected beam angles, leakage caps, and follower role selection. The reported performance gains—up to a 35% increase in secrecy rates and a success rate exceeding 98%—are tied to that dual-loop integration of sensing, beamforming, and security (Iqbal et al., 10 Feb 2026).

Several misconceptions follow from treating “unified ISAC framework” too narrowly. It does not necessarily mean one universal metric; the literature alternates among mutual information, rate regions, CRLBs, PCRBs, ambiguity-function metrics, channel decompositions, and belief-state utilities. It also does not necessarily mean a standards-ready implementation. Some frameworks are deliberately conceptual or architectural, some are standardization-oriented but not finalized standards, and some rely on strong assumptions such as Gaussian or linear sensing models, ideal SIC, static and accurate digital maps, or diagonal metasurface reconfiguration that neglects strong mutual coupling (Ouyang et al., 2022, Li et al., 16 Jun 2025, Pu et al., 23 Apr 2026, Liu et al., 3 Dec 2025).

Open problems are correspondingly layered. Analytical work still lacks complete Pareto-boundary characterizations in several multi-antenna settings. Channel modeling still faces unresolved issues in target-background power normalization, forward scattering, multiple-target interactions, and monostatic spatial consistency. Hardware-oriented frameworks must incorporate continuous apertures, strong coupling, and broader architecture classes. Security architectures still lack standardized trust scores, interfaces, and formal state machines for cross-domain adaptation. The overall trajectory suggests that a unified ISAC framework is best understood as an umbrella research program: to expose sensing and communication as coupled functions of one waveform-hardware-environment-control system, and to choose the unifying layer—metric, signal model, channel model, resource allocator, or security loop—according to the design problem at hand.