Source-linked AI summary
Learning-Based Beam Adaptation for Active/Passive-Aware Monopulse ISAC with Leakage Suppression
Jafar Norolahi, Alireza Vahid
TL;DR
Single-node ISAC beam control must adapt to changing spatial activity while preserving communication and sensing and limiting unintended illumination. The paper combines monopulse sensing, fused radar and uplink evidence, and SAC-based beam adaptation with protected-zone awareness. Across one to twelve passive targets, SAC stays within the aggregate leakage budget and, from two targets onward, achieves the highest leakage-compliant sum rate among the compared baselines.
Problem
Single-node ISAC studies less often combine radar spatial awareness, active/passive direction identification, real-time beam adaptation, and protection of a designated angular sector in one control loop.
Source: §I · p. 1Method
A single-gNB monopulse ISAC framework fuses radar detections and uplink evidence to distinguish active from passive directions and adapt the downlink beam with SAC.
Source: §I · pp. 1–2Results
Across one to twelve passive targets, SAC maintains aggregate passive leakage below budget and from two targets onward achieves the highest leakage-compliant sum rate among Fixed MRT, heuristic nulling, and DDPG.
Source: §V · p. 6, §IV · p. 6Takeaways & Limitations
The results show that the controller can preserve communication service while limiting unintended illumination toward passive directions and the protected zone in the evaluated sweep.
Source: §V · p. 6
Abstract
from arXiv · showhide
This paper proposes a learning-based beam adaptation framework for active/passive-aware monopulse integrated sensing and communication (ISAC) with spatial leakage suppression. A single gNB uses a shared array and orthogonal frequency division multiplexing dual-functional radar-communication signaling to jointly support downlink transmission and monopulse sensing. Radar detections and uplink spatial evidence are fused to distinguish cooperative active directions from passive directions, while a protected angular zone limits unintended illumination. Soft Actor-Critic deep reinforcement learning adapts the transmit beam using communication, sensing, leakage, and previous-action feedback. Numerical results show accurate active/passive spatial classification and robust leakage control as the number of passive targets increases from one to twelve. The proposed SAC controller maintains aggregate passive leakage below the prescribed budget across the full sweep and, from two passive targets onward, achieves the highest leakage-compliant sum rate among the considered baselines.
I. Introduction
The paper addresses the challenge of adapting a single-node ISAC beam to changing scenes while preserving communication and sensing and limiting unintended illumination. It proposes a closed-loop monopulse framework that fuses spatial evidence with learning-based beam adaptation and incorporates protected-zone leakage suppression.
I. Introduction
- The paper identifies a gap in single-node ISAC control that jointly integrates radar spatial awareness, active/passive identification, real-time beam adaptation, and protected-sector protection.Monopulse sensing is highlighted as supporting angle tracking and 3-D localization with a single node and pulse.Source: §I · p. 1
- A single-gNB monopulse ISAC framework uses sensing feedback to distinguish cooperative uplink-supported directions from radar-only directions and update the downlink beam policy.The approach links monopulse sensing, communication-aware adaptation, and spatial leakage suppression in a closed loop.Source: §I · pp. 1–2
- The proposed framework jointly supports communication, spatial sensing, localization, tracking, and adaptive beam control in one closed-loop architecture.Its beam adaptation uses fused uplink and radar evidence.Source: §I · p. 2
- Protected-zone leakage suppression is incorporated into the control objective to reduce unintended illumination while maintaining communication and localization performance.Source: §I · p. 2
II. System Model and Problem Formulation
- The system-model section defines the single-gNB monopulse ISAC setting, scene geometry, protected zone, and beam adaptation objective.Source: §II · p. 2, Figure 1 · p. 2
A. Scenario and Geometry
The scenario uses a single gNB for downlink communication and monopulse sensing, with active users and passive radar-observed directions represented in a shared spatial model. The signal and sensing model combines radar detections with uplink spatial evidence to classify directions and support leakage-aware beam adaptation.
A. Scenario and Geometry
- The gNB shares an antenna platform for downlink communication and monopulse sensing while serving legitimate users and monitoring directions to support or suppress.Source: A. Scenario and Geometry · p. 2
- Active directions correspond to legitimate uplink-supported users, while passive directions are radar-observed without legal uplink support.Azimuth and elevation provide the common spatial representation for beamforming, sensing, and leakage evaluation.Source: A. Scenario and Geometry · p. 2
- The protected zone is an angular region where radiation must remain low, and the beam policy suppresses leakage toward it and passive directions.Source: A. Scenario and Geometry · p. 2
B. Signal and Sensing Model
- An OFDM-based DFRC transmission uses a shared aperture and radio-frequency chain for downlink communication and monopulse sensing.The transmitted signal includes a downlink precoder, user symbols, and a sensing component.Source: B. Signal and Sensing Model · p. 2
- Radar detections and uplink spectrum are fused to identify active and passive directions; passive directions have radar detections without sufficient uplink support.The monopulse processing chain produces angle estimates and 3-D localization, while the resulting direction sets feed leakage-aware beam adaptation.Source: B. Signal and Sensing Model · p. 3
C. Design Objective
The design objective serves legitimate users while limiting downlink radiation toward passive directions and a protected zone. It treats sensing support as implicit in the retained DFRC signal and its power allocation.
C. Design Objective
- The objective seeks reliable service for legitimate users while suppressing unintended downlink radiation toward passive directions and the protected zone.Source: C. Design Objective · p. 3
- Passive-direction exposure and protected-zone exposure are the two undesired spatial-radiation mechanisms measured by leakage terms LE and LZ.Source: C. Design Objective · p. 4
- Sensing support is enforced implicitly through the retained DFRC sensing signal and its power allocation, rather than an explicit localization term in the objective.Source: C. Design Objective · p. 4
- Problem (22) is nonconvex because multiuser interference couples user SINRs and communication and sensing share transmit resources.Source: C. Design Objective · p. 4
- SAC-DRL learns an adaptive beam-control policy from communication and sensing feedback because the coupled objectives and changing environment make online solution challenging.Source: §III · p. 4
A. MDP Formulation
Beam adaptation is formulated as a frame-by-frame Markov decision process in which the controller observes a state, selects an action, and receives a reward and next state. The sequential formulation balances communication performance with leakage suppression.
A. MDP Formulation
- At each frame, the controller observes state st, selects action at, applies it at the gNB, and receives reward rt and next state st+1.Source: A. MDP Formulation · p. 4
- The formulation casts beam adaptation as a sequential decision problem balancing communication performance and leakage suppression.Source: A. MDP Formulation · p. 4
- The discount factor in the formulation is denoted by γ.Source: A. MDP Formulation · p. 4
B. State Design
The controller state uses quantities available to the controller, including spatial, communication, leakage, and prior-action feedback. It excludes ground-truth passive directions, which are used only to score held-out evaluation.
B. State Design
- The state includes active and sensed passive angle pairs, protected-zone center and SNR, user-rate feedback, leakage violations, and the previous action.Source: B. State Design · p. 4
- The state allows up to four normalized active azimuth/elevation pairs, four sensed passive pairs, and four normalized user-rate feedback entries.Source: B. State Design · p. 4
- No ground-truth passive direction enters the state; true directions are used only to score held-out evaluation.Source: B. State Design · p. 4
C. Action Design
The SAC-DRL policy outputs a continuous action devoted to gNB beam adaptation. A mapping enforces a feasible transmit configuration while keeping the action space tractable.
C. Action Design
- The SAC-DRL policy outputs a continuous action vector devoted entirely to gNB beam adaptation.Source: C. Action Design · p. 4
- The action is mapped to a feasible transmit configuration through Ψ(·), which enforces transmit-power constraints and beamformer structure.Source: C. Action Design · p. 4
- The lower-dimensional parameterization keeps the action space tractable while allowing adaptive control of the effective gNB beam and, if needed, communication-versus-sensing power allocation.Source: C. Action Design · p. 4
D. Reward Design
The reward balances communication performance against leakage-budget violations, localization error, and changes from the previous action. Its penalties target violations without rewarding unnecessary over-nulling.
D. Reward Design
- The reward combines sum rate with penalties for passive-direction and protected-zone leakage, 3-D sensing error, and changes from the previous action.The action-change term discourages abrupt beam-control updates.Source: D. Reward Design · p. 4
- The design penalizes leakage-budget violations without rewarding unnecessary over-nulling, while keeping failure-rate and violation statistics separately visible.The example weights are (λE, ρE, λZ, ρZ, λloc, ηa) = (0.40, 0.30, 0.35, 0.25, 0.002, 0.01).Source: D. Reward Design · p. 4
- Algorithm 1 forms the state from communication and sensing outputs, applies feasible beams from the SAC action, then stores the transition and updates the actor, critics, and target networks.The target critics are soft-updated after the SAC updates.Source: D. Reward Design · p. 5
E. SAC-DRL Update Mechanism
SAC-DRL learns a stochastic beam-control policy in continuous action space using replay-buffer transitions, soft Bellman targets, twin critics, and an entropy-regularized actor update.
E. SAC-DRL Update Mechanism
- SAC-DRL uses a stochastic actor, twin critics, target critics, and a replay buffer to learn beam control in continuous action space.Each frame contributes a state, action, reward, and next-state transition to replay.Source: E. SAC-DRL Update Mechanism · p. 5
- For each sampled transition, the algorithm constructs a soft Bellman target, then minimizes squared Bellman error to update both critics.The actor maximizes the entropy-regularized critic value.Source: E. SAC-DRL Update Mechanism · p. 5
- The implementation fixes the entropy weight at 0.05; evaluation freezes the actor and uses its deterministic mean action.Training uses stochastic actions and five independent seeds, with 140 episodes of 20 frames per seed.Source: E. SAC-DRL Update Mechanism · p. 5
- The actor and twin critics use two 64-unit ReLU layers, a 105-transition replay buffer, batch size 64, discount 0.98, and one update per frame.The target critics use Polyak factor 0.01; each seed yields 2800 transitions per policy.Source: E. SAC-DRL Update Mechanism · p. 5
F. Closed-Loop Beam Adaptation Procedure
The frame-wise procedure feeds communication and sensing outputs to SAC, maps its action to feasible beams, and uses the resulting transition for learning. Numerical results report active/passive classification, leakage suppression, legal-user rates, and scalable leakage-compliant sum rates.
F. Closed-Loop Beam Adaptation Procedure
- At each frame, communication and sensing outputs form the state, SAC generates a beam-control action, and Ψ maps it to a feasible transmit configuration for application and learning.The resulting transition is stored in replay and used to update the actor, twin critics, and target networks.Source: F. Closed-Loop Beam Adaptation Procedure · p. 5
IV. Numerical Results
- The scalability tests vary passive targets from one to twelve at 5-dB SNR; Fixed MRT, heuristic nulling, and DDPG serve as baselines.The reference geometry is used for Figs. 2–5, while Figs. 6–7 use randomized geometry and passive-target populations.Source: §IV · p. 5
- With the 2° association gate, detections 0 and 1 are classified as active and detections 99 and 100 as passive; reported 3-D localization errors are 0.499, 0.104, 0.479, and 0.067 m.Figure 3 reports radar-side DOA estimation for active users and passive targets.Source: §IV · p. 5, Figure 3 · p. 5
- Passive-direction leakages decrease from −81.3/−79.9 dB to −143.7/−133.1 dB, while protected-zone leakage decreases from −80.0 to −135.7 dB over the frames.Figure 4 plots passive-direction and protected-zone leakage during proposed beam adaptation.Source: §IV · p. 5, Figure 4 · p. 6
- UE0 reaches 8.05 × 10^-4 bits/s/Hz while UE1 stabilizes near 3.76 × 10^-4 bits/s/Hz under leakage-aware adaptation.Figure 5 shows the two legal users' achieved rates over adaptation frames.Source: §IV · p. 6, Figure 5 · p. 6
- From two passive targets onward, SAC provides the highest leakage-compliant rate in Fig. 6, remaining near 2.39 bit/s/Hz for two-six targets and reaching 2.283 bit/s/Hz at twelve.At twelve targets, SAC's rate is approximately 96% and 177% higher than DDPG and heuristic nulling, respectively; it maintains aggregate passive leakage below −45 dB throughout the sweep, from −57.40 to −49.28 dB.Source: §IV · p. 6
V. Conclusion
The framework combines uplink-supported spatial classification, radar-based 3-D sensing, protected-zone awareness, and SAC beam adaptation. Across one to twelve passive targets, SAC stays below the passive-leakage budget and, from two targets onward, achieves the highest leakage-compliant sum rate among the compared baselines.
V. Conclusion
- The framework combines uplink-supported spatial classification, radar-based 3-D sensing, protected-zone awareness, and SAC-based beam adaptation.The controller limits unintended illumination toward passive directions and the protected zone while preserving useful communication service.Source: §V · p. 6
- Across one to twelve passive targets, SAC maintains aggregate passive leakage below budget and achieves the highest leakage-compliant sum rate from two targets onward.The conclusion also reports a comparatively stable communication rate over the sweep.Source: §V · p. 6, Figure 7 · p. 6, Figure 6 · p. 6
- Numerical results confirm accurate active/passive discrimination as the passive-target population increases from one to twelve.The conclusion notes that additional calibration, missed-detection, and mobility tests expose remaining operating limits.Source: §V · p. 6