Radar–Cellular Coexistence

• Radar–Cellular Coexistence is the practice of sharing spectrum between high-power radars and cellular networks by mitigating severe mutual interference.
• Key methodologies include null space projection, cognitive waveform design, and machine learning approaches to optimize detection and throughput under strict interference constraints.
• Practical implementations use robust channel estimation, joint beamforming, and adaptive resource allocation to meet stringent performance metrics and regulatory standards.

Radar–Cellular Coexistence is the discipline concerned with enabling simultaneous operation of high-power radar systems and terrestrial cellular (broadband wireless) networks in overlapping spectral bands. The central challenge is the severe mutual interference arising from dense spectrum usage, wide-area deployments, and stringent requirements for both radar detection performance and cellular link quality. Advances in MIMO, beamforming, scheduling, machine learning, and optimization have fueled rapid progress, with modern approaches integrating spatial, temporal, spectral, and cooperative mechanisms for robust spectrum sharing.

1. Interference Models and Performance Metrics

Radar–cellular coexistence is intrinsically a coupled electromagnetic environment, governed by bi-directional interference:

• Radar-to-cellular coupling: The principal concern is high-power radar transmissions saturating or interfering with cellular receivers. Signal models typically take the form

$r = \sum_{i=1}^K G_i s_i + H x + n$

where $s_i$ are cellular user signals, $G_i$ their channels, $H$ is the MIMO channel from radar to the base station, $x$ is the (potentially multi-antenna) radar waveform, and $n$ is receiver noise (Khawar et al., 2015).

• Cellular-to-radar coupling: Cellular downlink signals act as in-band or adjacent-band interference for radar receivers. Aggregate cellular interference often requires stochastic geometry or log-normal sum modeling to capture deployed base station layouts, power, shadowing, and macro-cell propagation (Krishnan et al., 2017, Krishnan et al., 2017). The interference-to-noise ratio (INR) or signal-to-interference-plus-noise ratio (SINR) at the radar is the primary regulatory and system performance metric, with regulatory INR thresholds commonly set at $-6$ to $-10$ dB at the radar receiver (Krishnan et al., 2017, Krishnan et al., 2017).
• Detection and throughput metrics: Radar performance is typically specified by the probability of detection $P_D$ at a fixed probability of false alarm $P_{FA}$, with

$$P_D = Q\left(Q^{-1}(P_{FA}) - \sqrt{2K \cdot \mathsf{SINR}_r}\right)$$

for $K$ integrated pulses (Labib et al., 2017). Cellular system performance uses conventional metrics (throughput, BLER, fairness indices, etc.), with focus on SINR and achievable rate degradation under radar interference (Labib et al., 2017, Rao et al., 2020).

2. Spatial, Spectral, and Temporal Mitigation Techniques

2.1 Null Space Projection (NSP) and Spatial Filtering

NSP is a cornerstone spatial interference mitigation technique:

• For a radar with $M$ transmit elements and a BS with $N$ antennas, the radar projects its waveform into the null-space of the interference channel $H\in \mathbb{C}^{N\times M}$ to force $Hx=0$ (Khawar et al., 2015). The projection operator is

$P_{\mathrm{null}} = I_M - H^{\dagger} H$

where $H^\dagger = H^H(HH^H)^{-1}$. High-dimensional radar arrays are necessary to guarantee a sufficiently large null-space, thereby minimizing main-lobe distortion even for targets angularly proximate to nulled directions (Khawar et al., 2015).

• When radar targets lie closely adjacent to the BS direction, main-lobe reduction is quantified by the beampattern reduction factor:

$$\eta = \frac{\| P_{\mathrm{null}} a(\theta_T) \|^2}{\| a(\theta_T) \|^2}$$

with $\eta \to 0$ dB (negligible loss) as $M$ increases for fixed $N$ (Khawar et al., 2015). Similar SSVSP (Small Singular Value Space Projection) methods generalize NSP by projecting onto low-interference subspaces rather than strict nulls, easing requirements where perfect nulling is infeasible (Abdelhadi et al., 2015, Mahal et al., 2015).

2.2 Cognitive and Joint Waveform Design

Cognitive approaches enable adaptation based on measured environment and interference:

• Cognitive communications systems: Cellular systems adapt their transmission in time (via dynamic spectrum access/DFS protocols), frequency (power control, spectral notching), and space (beamforming/null steering), often with protection regions and real-time sensing to avoid harmful interference to radar incumbents (Labib et al., 2017).
• Cognitive radars: Radar waveforms are actively shaped in space and/or frequency to minimize projected power toward communication receivers. Techniques include water-filling in the spectral domain, mutual-information-maximizing OFDM radar, and joint radar–comm waveform co-design (Labib et al., 2017, Zheng et al., 2019).
• Joint cognition: Both radar and cellular adapt cooperatively, including multi-objective optimization of precoders, joint OFDM subcarrier assignment, or coordinated spectrum allocation (Labib et al., 2017).

2.3 Temporal and Frequency Domain Approaches

• DFS and TDM: Cellular operation is temporally suspended when radar is detected (e.g., dynamic frequency selection in 5 GHz WiFi bands). Strict regulatory timings are imposed for channel check, channel move time, and non-occupancy (Labib et al., 2017).
• PRB (Physical Resource Block) blanking: Upon radar detection, affected PRBs (identified via ML-aided spectrogram analysis) are dynamically and temporarily deactivated, with additional link adaptation (e.g., radar-aware MCS control) to optimize throughput for unaffected resources (Chiejina et al., 5 Jan 2026).
• Guard zone and protection region enforcement: Minimum separation distances (20–50 km) between radars and cellular transmitters are set to enforce probability constraints on radar INR, subject to propagation and shadowing statistics (Krishnan et al., 2017, Krishnan et al., 2017, Rao et al., 2020). Power control and cell densification can shrink the required protection radius (Krishnan et al., 2017). In stochastic geometry models, spectrum sharing is quantified as a function of node density, guard zone radius, and spectrum usage parameters (Maeng et al., 2022).

3. Robust Channel Estimation and CSI Feedback

Robust interference-channel state information (ICSI) is crucial for effective nulling and beamforming:

• Interfering channel estimation: In practical (often uncoordinated) scenarios, the cellular system must estimate the radar interference channel blindly, by exploiting structure in radar waveforms and hypothesis testing approaches (GLRT, Rao test, etc.) under both LoS and NLoS propagation (Liu et al., 2018).
• CSI feedback under pulsed radar: Pilot-based CSI estimation is fundamentally limited by the radar’s pulsed nature; pilots and data experience different interference statistics, leading to bimodal SINR distributions and unreliable CQI. Semi-blind SINR estimation (combining pilot and blind metrics) and dual-CSI feedback (separated for interference-free and radar-impaired states) are recommended for robust link adaptation and scheduling (Rao et al., 2020, Rao et al., 2020).
• Coherence times: For LoS maritime/costal deployments, the coherence time of the radar–BS channel ($T_{\mathrm{coh}}$) is typically orders of magnitude greater than radar PRI, rendering channel tracking and feedback effective in most operational conditions (Khawar et al., 2015).

4. Optimization and Resource Allocation Frameworks

Spectrum sharing operates under explicit interference and QoS constraints, commonly solved by convex/SDR techniques:

• WSMMSE formulations: Weighted sum mean-squared error minimization (subject to linear power constraints) enables partial cooperation between radars and cellular BSs, supporting collaborative message delivery while bounding harmful interference (Abdelhadi et al., 2015).
• Robust joint beamforming: MIMO radar and multi-user MIMO/FD cellular transceivers are jointly designed to maximize radar $P_D$ or minimize communication MSE, under per-link SINR and INR constraints, power budgets, and error-uncertainty models for channel estimation (Liu et al., 2016, Liu et al., 2017, Biswas et al., 2018). Semidefinite programming, alternating convex optimization, and tractable SDP relaxations are the primary algorithmic strategies.
• Resource allocation under fairness objectives: Two-stage convex optimization embedding proportional fairness utilities protects radar-exclusive resources in band assignment and rebalances cellular QoS using standard distributed Lagrangian methods (Ghorbanzadeh et al., 2014).
• Constructive interference exploitation: For specific modulations (e.g., PSK), useful multi-user interference power may be exploited to reduce BS transmit power while maintaining SINR and radar INR thresholds, yielding substantial energy savings (Liu et al., 2017).

5. Machine Learning and Real-Time Spectrum Adaptation

Emerging solutions leverage data-driven methodologies for sub-second spectrum sharing:

• ML-based radar detection in RAN: Deep neural networks (DNNs) on low-overhead RAN telemetry, hybridized with real-time spectrogram analysis (YOLO-based object detection), achieve fast (sub-100 ms detection, $<$1 s evacuation) radar pulse identification and localization (Chiejina et al., 5 Jan 2026, Villa et al., 2023). ML predictors outperform classical methods in detection accuracy and response time, enabling tight coordination with RAN control plane for PRB blanking and MCS adjustment.
• Contextual multi-armed bandits for spectrum selection: Cognitive radar adapts transmission parameters sequentially using contextual Thompson Sampling, learning optimal sub-band/pulse configurations via observed reward (SINR, interference) and rapid Bayesian exploration/exploitation. This approach achieves sublinear regret and fast convergence in highly non-stationary environments compared to classical RL baselines (e.g., DQN) (Thornton et al., 2020).

6. System-Level Deployment and Regulatory Constraints

• Protection distances and aggregate interference: Detailed link-budget and log-normal aggregate interference models, using empirically validated propagation (e.g., eHATA), establish that 30–50 km exclusion zones are typically required in co-channel, full-power scenarios to satisfy ATC or naval radar INR protection ($-6$ to $-10$ dB with $>$ 90% reliability) (Krishnan et al., 2017, Krishnan et al., 2017). Adjacent-channel operation and adaptive power control can reduce this distance to 1 km or less. MAC-layer blanking and time/frequency “off-beam” methods extend coexistence to dynamic/rotating radar platforms (Labib et al., 2017).
• Massive MIMO and 3D beamforming: Arrays with high element counts at BSs and radars permit sharp spatial filtering, agile null-steering, and adaptive downtilt to suppress mutual interference without substantial throughput loss, as quantified by mean and distributional interference power, as well as radar ROC curves (Buzzi et al., 2018, Rao et al., 2020).
• Stochastic geometry and network-level analysis: Analytical frameworks for overlay and TDMA-based sharing (e.g., for UAV radar–cellular systems) deliver closed-form trade-off curves for successful ranging probability (SRP) and transmission capacity (TC), governing regime selection based on system densities and spectral division parameters (Maeng et al., 2022).

7. Open Challenges and Research Directions

Ongoing and unresolved challenges include:

• Accurate channel estimation under mobility and hardware nonidealities, including fast-moving platforms (UAV, airborne radar) and time-varying propagation (Khawar et al., 2015, Abdelhadi et al., 2015).
• CSI acquisition and sharing: Realistic strategies for low-latency, robust CSI exchange between uncoordinated radar and cellular entities.
• Regulatory harmonization: Alignment of dynamic spectrum access systems (SAS), real-time environmental sensing capability (ESC), and O-RAN/Near-RT RIC frameworks for compliance at scale (Chiejina et al., 5 Jan 2026).
• Integrated radar–communication design: Co-optimization of radar detection metrics (e.g., range/Doppler/CRB) and cellular QoS, especially under practical restrictions (finite alphabet, PAPR, incomplete CSI) (Liu et al., 2016, Liu et al., 2017, Biswas et al., 2018, Zheng et al., 2019).
• Machine learning for adaptive spectrum management: Development of robust, explainable, and interoperable ML-managed coexistence in open RAN and distributed architectures.
• Physical-layer security and resilience: Safeguarding mission-critical radar operations against adversarial interference or denial-of-service.

These aspects ensure that radar–cellular coexistence remains an essential and dynamic field at the confluence of statistical signal processing, information theory, wireless networking, optimization, and machine learning (Labib et al., 2017, Zheng et al., 2019, Chiejina et al., 5 Jan 2026).