Cell-Free ISAC Systems: Design & Advances

Updated 29 December 2025

Cell-Free ISAC systems are integrated architectures that combine distributed massive MIMO with joint radar sensing and wireless communications to enhance spectral efficiency and coverage.
They leverage dynamic AP role assignment, centralized coordination, and joint beamforming to balance conflicting sensing and communication objectives in real time.
Advanced techniques such as RF fingerprinting, tensor decomposition, and decentralized optimization tackle challenges like synchronization, interference mitigation, and resource allocation.

Cell-Free Integrated Sensing and Communication (ISAC) Systems unify distributed massive MIMO radio access architectures with joint radar sensing and wireless communication, aiming to maximize spectral efficiency, environmental perception, and coverage without strict cell boundaries. By leveraging widely distributed access points (APs)—such as remote radio units (RRUs) or user-centric APs—connected through low-latency fronthaul, these systems achieve multi-static sensing, robust multi-user communications, and flexible resource allocation. Key challenges include synchronization, signal processing for interference mitigation, joint beamforming and power control, and the balancing of conflicting communication and sensing objectives.

1. System Architectures and Operational Principles

Cell-Free ISAC systems employ a dense deployment of geographically distributed APs—typically equipped with multiple antennas—under the centralized coordination of an edge or cloud controller. The foundational structure consists of:

Dynamic AP assignment: Each AP can be configured as a downlink (transmit) or uplink (receive) node in each time slot, enabling flexible adaptation to communication and sensing needs (Yu et al., 2023). For example, in the NAFD (Network-Assisted Full Duplex) mode, separate sets of APs handle transmission and reception simultaneously (Zeng et al., 2023).
Waveform generation: Downlink APs transmit composite ISAC waveforms comprising both communication symbols and patterns that are specifically tailored for environmental sensing (e.g., distinct RF-fingerprint features) (Yu et al., 2023).
Fronthaul and distributed processing: An edge distributed unit (EDU) or a central processing unit (CPU) assigns roles, synchronizes APs, and aggregates/schedules data and control information.

This architecture enables two principal sensing modes:

Passive ISAC: Communication waveforms are reused for sensing; the echoes are processed by neighboring uplink RRUs/receivers for target/environmental estimation.
Active ISAC: Separate probing (radar) signals are transmitted alongside communication payload, with dedicated power/resources allocated for sensing (Zeng et al., 2023, Behdad et al., 2023).

2. Sensing and Communication Signal Models

The transmit signal at any downlink AP combines communication beamforming, power control, and (when active sensing is used) a distinct sensing component:

$x_\mathrm{AP} = \sum_{k}\sqrt{\rho_k}\, w_{k}\, s_k + \sqrt{\rho_0}\, w_0\, s_0$

where $w_k$ are the communication beamformers, $w_0$ the sensing/prioritized beamformer, and $\rho_k$ / $\rho_0$ the power-control coefficients (Li et al., 25 Jan 2025). For passive ISAC, the same waveform $s$ is broadcast from all downlink APs, with unique RF fingerprints $f_d$ per AP to enable post-hoc separation of echoes at the receivers (Yu et al., 2023).

At the receiver, the observed signal comprises:

Line-of-sight (LOS) component from all transmitting APs
Reflections from targets or scatterers (multi-static sensing)
Possible uplink user transmissions (handled or subtracted when idle)

Uplink RRUs or sensing APs employ multi-dimensional tensor and matrix decompositions, subspace parametric estimation, or compressive sensing methods to extract time delays, Doppler shifts, azimuth/elevation angles, and reflector positions from the received signals (Yu et al., 2023, Behdad et al., 2023).

3. Joint Resource and Beamforming Optimization

Integrated resource allocation encompasses power control, beamformer design, and sometimes AP role assignment. The fundamental optimization problems form bi-objective or multi-objective programs balancing communication SINR (Signal-to-Interference-plus-Noise Ratio) and sensing accuracy (e.g., minimizing Cramér-Rao Lower Bound (CRLB), sensing SNR, or maximizing detection probability) (Xia et al., 26 May 2025, Demirhan et al., 2023, Ren et al., 2023).

Typical formulations:

Example (CRLB-constrained SINR maximization):

$\begin{aligned} \max_{\bm{\rho}} & \quad \gamma_c(\bm{\rho}) = \delta\, \mathbf{g}^{\top} \bm{\rho} \ \text{s.t.} & \quad \mathrm{CRLB}_\text{loc}(\bm{\rho}) \leq \epsilon_\text{loc} \ & \quad \mathrm{CRLB}_\text{vel}(\bm{\rho}) \leq \epsilon_\text{vel} \ & \quad \sum_{n=1}^N \rho_n = 1, \ 0 \leq \rho_n \leq \rho_{\max, n} \end{aligned}$

(Xia et al., 26 May 2025)

Example (max-min fairness, JSC beamforming):

$\begin{aligned} \max_{\{\mathbf{w}_s\}} & \quad \mathrm{SNR}^{(s)}(\{\mathbf{w}_s\}) \ \text{s.t.} & \quad \mathrm{SINR}_u(\{\mathbf{w}_s\}) \geq \gamma_u, \ \forall u \ & \quad \sum_{s} \|\mathbf{w}_{m,s}\|^2 \leq P_m, \ \forall m \end{aligned}$

(Demirhan et al., 2023)

Solution methods include convex relaxations (SDR), successive convex approximation (SCA), alternating direction method of multipliers (ADMM), and learning-based surrogates (e.g., teacher-student DNN/GNN architectures) (Elrashidy et al., 24 Dec 2024, Zafari et al., 1 Aug 2025). Real-time deployment is facilitated by offloading iterative steps to CPUs or edge processors, or by designing distributed/consensus algorithms when full centralization is infeasible.

4. Sensing Algorithms, RF Fingerprinting, and Interference Mitigation

Notable ISAC-specific techniques include:

RF-Fingerprint Library Construction: In passive ISAC, each AP is characterized by a learned RF-fingerprint vector $f_d$ , which fingerprints its hardware-induced spectral or temporal signatures. These are extracted through tensor decomposition and supplied to CNN classifiers for source-echo tagging (Yu et al., 2023).
Path and Source Separation: Classification of multipath components (via extracted fingerprint features and minimum distance) enables discrimination among spatially co-located or temporally overlapping echoes, mitigating intra-network interference in the same time-frequency setting.
Channel Parameter Estimation and Multiview Fusion: Delays, angles, and Doppler shifts for each path are extracted using compressive sensing or high-resolution subspace techniques (MUSIC, ESPRIT); reflector/source positions are then estimated via nonlinear least squares and fused across APs for robust multi-static localization (Yu et al., 2023, Liu et al., 30 Jun 2025).
Interference Mitigation: After echo classification, MIMO beamspace spatial filtering (using estimated $\phi, \theta$ ) isolates individual paths, reducing crosstalk without power/spectrum overhead (Yu et al., 2023).

5. Performance Metrics, Trade-Offs, and System Analysis

Multiple metrics quantify system capabilities and trade-offs:

Classification Accuracy: For RF fingerprint identification with CNNs, accuracy ranges from 81.54% (with PSD_ratio=0.3, small symbol block) to 98% (PSD_ratio=0.4, long block), directly impacting sensing error (Yu et al., 2023).
Localization Error: Reflector estimation achieves sub-2 m RMSE at high fingerprint accuracy and moderate channel parameter errors ( $<0.5$ m, $<0.5^\circ$ ).
Communication Throughput: Passive ISAC imposes no additional power/bandwidth load; throughput and BER are identical to pure communication baselines.
Age of Sensing (AoS) and Coverage: AoS is critical in tracking dynamic targets (e.g., unauthorized drones); adaptive blocklength and power allocation can reduce AoS by up to 45% without sacrificing spatial coverage (Li et al., 25 Jan 2025).
Pareto Region: Varying power split between sensing and communication sweeps a frontier: maximizing one degrades the other. Joint optimization (e.g., via DQN, NSGA-II) yields strictly superior solutions over equal power or static resource splits (Zeng et al., 2023).
Sensing-Communication (C-S) Region: Traces fundamental limits as a 3-D boundary given AP number/geometry, with key shape transitions as coverage, minimal beampattern error, and maximal communication rate are jointly tuned (Mao et al., 2023).

6. Practical Implementations, Learning-Based Methods, and Privacy/Security

Pragmatic deployment of cell-free ISAC systems raises several system-level considerations:

Algorithmic Complexity: Tensor decomposition, compressive sensing, and CNN classification are edge-deployable with GPU acceleration; learning-based DNN/GNN models (both centralized and decentralized) enable near-real-time inference at orders-of-magnitude faster rates than convex program solves (Elrashidy et al., 24 Dec 2024, Demirhan et al., 26 Sep 2024).
Decentralized Resource Optimization: Recent approaches leverage local beamformer design and decentralized consensus-based power allocation (e.g., ADMM), which reduce fronthaul overhead and scale to large AP networks with low-per-iteration information exchange (Zafari et al., 1 Aug 2025).
Synchronization: Distributed APs require sub-microsecond time/frequency/phase alignment for coherent sensing—in practice, implemented via GPS-disciplined clocks or fronthaul-based synchronization (Galappaththige et al., 27 Feb 2025).
Privacy and Security: AP-transmit/receive role scheduling and precoder design can be jointly optimized to minimize mutual information leakage and adversarial target inference, significantly lowering detection probability even for legitimate communication users acting as eavesdroppers (Åkesson et al., 19 Sep 2024, Ren et al., 2023).
Adaptive Library Maintenance: RF-fingerprint libraries must be periodically recalibrated as hardware ages or in response to environmental/temperature drift (Yu et al., 2023).

7. Advanced Extensions and Research Directions

Current and emerging research in cell-free ISAC encompasses:

Symbol-Level Multiview Sensing Fusion: Tensor decomposition and data association at each AP, followed by symbol-level fusion and continuous 2D/velocity parameter estimation (e.g., using artificial bee colony optimization), now achieves up to 44% localization and 41.4% velocity RMSE improvement over grid-based benchmarks (Liu et al., 30 Jun 2025).
Learning-Based Beamforming: GNN and teacher-student DNN approaches yield robust mapping from random or partial CSI to high-quality beamformers, both scaling linearly with AP/user number and supporting dynamic topology changes (Wang et al., 13 Oct 2024, Elrashidy et al., 24 Dec 2024, Demirhan et al., 26 Sep 2024).
Joint Placement and Antenna Allocation: ADMM-based minimax CRLB optimization enables system-level planning of AP geometries and per-AP antenna provision for worst-case multi-target estimation (Liu et al., 30 Jun 2025).
Synchronization and Fronthaul Management: Efficient fronthaul usage and latency-aware processing remain open bottlenecks; decentralized ADMM and resource-efficient algorithm design address these issues (Zafari et al., 1 Aug 2025).
Security with Sensing/Information Eavesdroppers: Semidefinite relaxation yields globally optimal beamformers satisfying both communication secrecy and sensing privacy constraints, with multi-static diversity enhancing detection even in adversarial conditions (Ren et al., 2023).

Continued research directions in this area include real-world field trials, robust designs for hardware impairment/clutter/mobility, extensions to multi-cell or near-field (XL-MIMO) scenarios, robust distributed learning/fusion under partial or delayed CSI, and federated approaches for privacy-preserving ISAC deployments.

References

Passive Integrated Sensing and Communication Scheme based on RF Fingerprint Information Extraction for Cell-Free RAN (Yu et al., 2023)
Detecting Unauthorized Drones with Cell-Free Integrated Sensing and Communication (Li et al., 25 Jan 2025)
Joint Space-Time Adaptive Processing and Beamforming Design for Cell-Free ISAC Systems (Liu et al., 18 Oct 2024)
Unsupervised Learning Approach for Beamforming in Cell-Free Integrated Sensing and Communication (Elrashidy et al., 24 Dec 2024)
Multi-Static Target Detection and Power Allocation for Integrated Sensing and Communication in Cell-Free Massive MIMO (Behdad et al., 2023)
Integrated Sensing and Communication for Network-Assisted Full-Duplex Cell-Free Distributed Massive MIMO Systems (Zeng et al., 2023)
Power allocation for cell-free MIMO integrated sensing and communication (Xia et al., 26 May 2025)
Communication-Sensing Region for Cell-Free Massive MIMO ISAC Systems (Mao et al., 2023)
Cooperative Sensing in Cell-free Massive MIMO ISAC Systems: Performance Optimization and Signal Processing (Liu et al., 30 Jun 2025)
Cell-Free Integrated Sensing and Communication: Principles, Advances, and Future Directions (Galappaththige et al., 27 Feb 2025)
Coordinated Decentralized Resource Optimization for Cell-Free ISAC Systems (Zafari et al., 1 Aug 2025)