Cell-Free ISAC: Integrated Sensing & Communication

Updated 14 December 2025

Cell-Free ISAC is a distributed wireless architecture where multiple access points jointly deliver communication and radar sensing through coordinated beamforming.
Joint beamforming and resource allocation techniques balance communication throughput and sensing accuracy while ensuring uniform coverage and energy efficiency.
Advanced methods such as deep learning acceleration and secure beamforming enhance real-time optimization, privacy protection, and performance trade-offs.

Cell-Free Integrated Sensing and Communication (ISAC) systems constitute a class of distributed wireless architectures in which geographically distributed access points (APs) collaboratively provide both communication and radar-style sensing services over shared spectral and hardware resources. These systems are designed to achieve high spectral and energy efficiency, robust multi-static sensing, and uniform service quality by eliminating traditional cellular boundaries and fully leveraging spatial macro-diversity. Recent research advances demonstrate that cell-free ISAC architectures not only rival but often surpass conventional centralized and cellular solutions in key metrics such as communication throughput, radar accuracy, coverage, spectral efficiency, and privacy control, while introducing new technical challenges involving resource allocation, synchronization, trade-off balancing, and real-time decentralized implementation.

1. Architectural Principles and Signal Models

Cell-free ISAC systems deploy a large number of physically distributed APs—each featuring multiple antennas—across the service region. Unlike traditional cellular layouts, any AP can serve any user or perform sensing in any region, coordinated typically via a central processing unit (CPU) connected through high-capacity fronthaul links. ISAC unifies the transmission of communication data and radar signals through joint waveform design and beamforming, supporting both multi-user MIMO downlink and multi-static sensing paradigms.

For joint transmission, the APs broadcast superpositions:

$\mathbf{x}_m = \sum_{k=1}^K \mathbf{w}_{mk} q_k + \sum_{t=1}^T \mathbf{s}_{mt}$

where $q_k$ is communication data for user $k$ , $\mathbf{w}_{mk}$ is the communication beamformer, and $\mathbf{s}_{mt}$ is the dedicated sensing signal toward target $t$ (Galappaththige et al., 27 Feb 2025).

The sensing echoes from each target are received at designated AP clusters:

$\mathbf{y}_r[n] = \sum_{k=1}^{N_{\mathrm{Tx}}} \alpha_{r,k} \sqrt{\beta_{r,k}} \mathbf{a}(\phi_r) \mathbf{a}^T(\phi_k) \mathbf{x}_k[n] + \mathbf{n}_r[n]$

where $\alpha_{r,k}$ encapsulates RCS and propagation gains, and $\mathbf{a}(\cdot)$ provides array steering (Silva et al., 2023).

This decentralized structure enables joint multi-user communication and radar sensing—for instance, continuous target illumination, rapid location updates, and dynamic resource adaptation across heterogeneous users and targets.

2. Joint Beamforming and Resource Allocation

Optimal operation of cell-free ISAC systems hinges on joint beamforming and resource allocation methodologies that balance competing requirements in sensing and communications. The governing optimization problem is typically non-convex, featuring coupled quadratic constraints and objectives:

Pareto-Optimal Beamforming: Maximize a weighted combination of sensing-SNR ( $SSNR$ ) and communication-SINR ( $SINR$ ), under per-AP power constraints, ensuring minimum QoS for both services (Elrashidy et al., 24 Dec 2024, Demirhan et al., 2023):

$\max_{\{w_{lq}\}} SSNR,~\text{subject to}~\min_n SINR_n \geq \gamma_\mathrm{high}$

This is commonly solved via semidefinite programming (SDP) relaxations, block-coordinate descent, or alternating optimization.

Power Allocation under Sensing Constraints: Maximize communication SINR given CRLB-based constraints on sensing accuracy (location/velocity), using penalty-projected conjugate gradient or steepest descent algorithms (Xia et al., 26 May 2025).
Multi-objective Optimization: Jointly maximize sum communication rates and minimize aggregate sensing error, exploring the Pareto front via evolutionary algorithms (NSGA-II) or reinforcement learning (DQN) (Zeng et al., 2023).

The practical realization of these solutions ranges from centralized solvers operating at the CPU (requiring high fronthaul rates) to decentralized schemes in which lightweight deep learning models execute locally at each AP, trained offline to approach optimal performance with sub-millisecond inference latencies (Elrashidy et al., 24 Dec 2024, Farzanullah et al., 7 Jun 2025).

3. Sensing and Communication Performance Metrics

Cell-free ISAC introduces novel metrics to quantify system performance:

Communication SINR/Spectral Efficiency:

$SINR_k = \frac{|\sum_m \mathbf{h}_{mk}^H \mathbf{w}_{mk}|^2}{\sum_{i \neq k}|\sum_m \mathbf{h}_{mk}^H \mathbf{w}_{mi}|^2 + \sum_t |\sum_m \mathbf{h}_{mk}^H \mathbf{s}_{mt}|^2 + \sigma^2}$

$SE_k = \log_2(1 + SINR_k)$

(Mao et al., 2023, Galappaththige et al., 27 Feb 2025)

Sensing-SNR (SSNR), Sensing Spectral Efficiency (SE), and Cramér-Rao Bounds (CRLB):

$SSNR = \frac{\sum_{r,l} \sigma_{s,lr}^2 \| \mathbf{a}(\theta_l)^H \bar{W}_l \|^2}{\sum_{r} \sigma_{a,r}^2}$

$SE_\mathrm{sen} = \frac{1}{L} \log_2 \det \left( I_L + \frac{1}{\sigma^2} X^H G^H G X \right)$

$\mathrm{Var}(\hat{\theta}) \geq \mathrm{CRB}(\theta) = [F^{-1}(\theta)]_{nn}$

where $F(\theta)$ is the Fisher Information Matrix of the system (Galappaththige et al., 27 Feb 2025, Demirhan et al., 2023, Xia et al., 26 May 2025).

Integrated Trade-Off Regions: The communication–sensing (C–S) region characterizes achievable SINR/SE for both services under joint resource allocation, forming a multidimensional Pareto surface dependent on AP count, antenna dimension, and pilot allocation (Mao et al., 2023). This provides a unified framework for exploring fundamental limits and trade-offs.
AoS and Coverage Metrics: For dynamic scenarios such as drone detection, the Age of Sensing (AoS) quantifies timeliness of updates, while sensing coverage captures the fraction of area or hotspots achieving a prescribed detection threshold; trade-off optimization ensures high coverage with minimal latency (Li et al., 25 Jan 2025, Behdad et al., 7 Dec 2025).

4. Multi-Static Sensing, Cooperative Processing, and Synchronization

Cell-free ISAC leverages multi-static radar principles—distributed transmit/receive pairs probing targets from multiple spatial viewpoints—yielding pronounced diversity gains and improved robustness over mono-static (co-located) configurations:

The aggregate Fisher information in multi-static setups scales with the product of numbers of transmit and receive APs, leading to CRLB improvement as $1/(M N)$ (Galappaththige et al., 27 Feb 2025, Liu et al., 30 Jun 2025, Silva et al., 2023).
Symbol-level multi-view fusion algorithms, such as the proposed SL-MDTS (Symbol-Level Multi-Dynamic Target Sensing), combine tensor decomposition, MUSIC/IDFT, and weighted parameter search to significantly reduce localization and velocity estimation errors—by up to 44% and 41.4% respectively over grid/lattice fusion (Liu et al., 30 Jun 2025).
High-accuracy time/frequency/phase synchronization, often achieved over optical fiber fronthaul, is crucial for maintaining coherent combining and maximizing spatial resolution in multi-node cooperative sensing (Liu et al., 30 Jun 2025).

Centralized and distributed fusion strategies are implemented, depending on the fronthaul and computational capabilities of the infrastructure. The system must coordinate AP placement and dynamic resource assignment to optimize information geometry for sensing (Liu et al., 30 Jun 2025).

5. Deep Learning and Algorithmic Acceleration

The computational overhead inherent to convex optimization and iterative resource-allocation can be addressed effectively using deep learning and generative modeling:

Decentralized DNN-based Beamforming: Each AP is equipped with a local neural network that predicts beamforming vectors from local CSI and steering information. Training is performed offline in a teacher–student manner: “SSNR teacher” and “SINR teacher” models maximize individual objectives, while an adaptive student model balances both via normalization and gap-metric adaptation (Elrashidy et al., 24 Dec 2024). U-Net architectures outperform autoencoders and standard CNNs in balancing trade-offs.
Conditional Denoising Diffusion Models (CDDM): For channel estimation in cell-free ISAC, multimodal transformers are used to fuse location and radar-sensing returns as condition vectors; diffusion denoising is conditioned on this context, yielding substantial NMSE gains (up to 9 dB over MMSE baselines, and 27.8% over conventional diffusion) and resilience to pilot contamination (Farzanullah et al., 7 Jun 2025).

Algorithmic acceleration (from seconds to milliseconds per instance) and reduced fronthaul load are achieved, facilitating real-time beamforming, resource allocation, and channel estimation at large scale.

6. Security, Privacy, and Information Protection

Cell-free ISAC systems must address privacy threats at both the communication and sensing layers:

Physical-layer privacy-preserving beamforming: Joint alternate optimization of precoders and access point (AP) roles (transmit vs. receive) shapes beampatterns to minimize information leakage, quantified via mutual information and adversarial detection probability ( $P_D$ ) (Åkesson et al., 19 Sep 2024). AP selection exploits mutual information bounds to choose receiver clusters with minimal leakage.
Joint secure beamforming against eavesdroppers: Problem formulations include constraints on maximum tolerable SNR at information eavesdroppers and detection probability at sensing eavesdroppers, solved globally via semidefinite relaxation (SDR) with proven tightness. Coordinated jamming and covariance shaping further limit adversarial capabilities (Ren et al., 2023).
Adversary models: Internal adversaries may attempt target localization via aggregated beampattern inference, raising fundamental trade-offs between sensing fidelity and privacy protection (Silva et al., 2023).

Empirical results demonstrate up to 30% reduction in adversarial detection probability and maintenance of sensing SINR with minimal privacy loss for dynamically chosen AP roles and precoder configurations.

7. Trade-Offs, Applications, and Open Challenges

Cell-free ISAC design space is characterized by strong trade-offs among communication throughput, sensing accuracy, latency, coverage, and privacy:

Increasing communication QoS (SINR, SE) reduces degrees of freedom available for sensing, limiting coverage and timeliness (Li et al., 25 Jan 2025, Behdad et al., 7 Dec 2025).
Network configuration strategies—hotspot grouping, AP clustering, pilot assignment—must jointly optimize ambiguity (pilot reuse interference), AoS, and coverage, using mixed-integer programming and convex relaxations (Behdad et al., 7 Dec 2025).
Practical implementations address fronthaul limitations, hardware impairments, synchronization errors, pilot contamination, and mobility (Galappaththige et al., 27 Feb 2025, Zeng et al., 2023).
Application scenarios include drone detection, terrestrial target tracking, passive environmental mapping, joint localization, and dynamic spectrum management.

Emerging research directions include near-field beamforming, reconfigurable intelligent surfaces, hybrid analog/digital designs in mmWave bands, AI/ML-driven resource adaptation, and joint communication–sensing integration in perceptive, fluid, and mobile network architectures (Galappaththige et al., 27 Feb 2025).

For detailed methodologies, algorithms, and quantitative results, see (Elrashidy et al., 24 Dec 2024, Farzanullah et al., 7 Jun 2025, Silva et al., 2023, Ren et al., 2023, Åkesson et al., 19 Sep 2024, Liu et al., 30 Jun 2025, Zeng et al., 2023, Galappaththige et al., 27 Feb 2025, Xia et al., 26 May 2025, Demirhan et al., 2023, Li et al., 25 Jan 2025, Behdad et al., 7 Dec 2025), and related references.