Secure Full Duplex Integrated Sensing and Communications

I. INTRODUCTION

• 6G trustworthiness motivates integrating security with sensing and communication, while CSI can expose personal information and enable eavesdropping.The paper argues for physical-layer security alongside cryptographic approaches.Source: §I · p. 1

A. Existing work

• Prior secure-ISAC work leaves a gap in FD settings that must jointly optimize beamformers and AN for sensing, communication, and secrecy.The paper contrasts its scope with work on half-duplex ISAC and prior FD ISAC designs.Source: A. Existing work · pp. 2–3

B. Contributions and Insights

• The model includes multiple legitimate UL and DL users and multiple eavesdroppers that may intercept both communication directions.The DFRC base station also receives UL signals and sensing echoes while transmitting, including self-interference.Source: B. Contributions and Insights · p. 3, A. Full Duplex Radar and Communication… · p. 5
• The power-efficient framework jointly designs AN, beamformers, and UL power to meet UL/DL secrecy requirements while illuminating eavesdroppers for sensing.AN is used for secrecy in both directions and also aids in forming radar beams toward eavesdroppers.Source: B. Contributions and Insights · p. 3, A. Full Duplex Radar and Communication… · p. 4
• SCA converts the non-convex design into iterative convex problems; the method settles within as few as 20 iterations and has rank-one DL beamforming vectors at each iteration.The rank-one result removes the need for post-processing such as Gaussian randomization.Source: B. Contributions and Insights · p. 3
• Simulations report secrecy–sensing and power–secrecy trade-offs, and say secrecy can remain stable as UL/DL user counts vary for a fixed number of eavesdroppers.The proposed approach is compared with benchmarks on secrecy rates and total power consumption.Source: B. Contributions and Insights · p. 3

B. Key Performance Indicators

• Key performance indicators are defined to assess the overall system and inform the secure FD ISAC optimization framework.The passage introduces the KPI definitions but does not specify their individual formulas.Source: B. Key Performance Indicators · p. 5

1) Communication Achievable Rates:

• The downlink user’s SINR is defined from the received downlink signal under the independent-symbol assumption.The downlink achievable rate is then given using Shannon theory.Source: 1) Communication Achievable Rates · pp. 5–6
• The downlink user’s achievable transmission rate is derived from the defined SINR using Shannon theory.Source: 1) Communication Achievable Rates · p. 6
• The uplink SINR measures the received performance of a user at the DFRC base station, which applies a receive beamforming vector for each UL user.The paper also defines the corresponding UL achievable rate.Source: 1) Communication Achievable Rates · p. 6
• The uplink user’s achievable transmission rate is defined from the uplink SINR.Source: 1) Communication Achievable Rates · p. 6

2) Secrecy Rates:

• Worst-case secrecy rates are defined separately for downlink and uplink because eavesdroppers receive both DL and UL signals.The definitions account for the eavesdroppers’ received SINRs for each direction.Source: 2) Secrecy Rates · p. 6

3) Radar Metrics:

• ISMR measures sidelobe energy relative to mainlobe energy, with mainlobes centered on eavesdropper directions to focus radar power.The metric seeks to lower sidelobe power and can accommodate uncertainty in eavesdropper angle of arrival through mainlobe width.Source: 3) Radar Metrics · pp. 6–7
• The design faces a push-pull: it seeks radar energy focused on eavesdroppers while limiting communication information revealed in those directions.The paper describes AN covariance design as part of managing this tension.Source: 3) Radar Metrics · p. 7

• The design minimizes total power across uplink power allocation, downlink beamforming, and artificial-noise generation while meeting minimum uplink and downlink secrecy rates and a maximum ISMR.The secrecy-rate thresholds are ρUL and ρDL, and the radar constraint is ISMR ≤ ISMRmax.Source: §III · p. 7
• The secrecy-rate requirements are worst-case bounds intended to limit information leakage to eavesdroppers while the ISMR constraint sustains radar beams toward them.Source: §III · p. 7
• The secrecy constraints are recast as SINR bounds for legitimate users and eavesdroppers, yielding the equivalent formulation in problem (28).The passage describes analogous transformations for the downlink and uplink secrecy-rate constraints.Source: §III · p. 7

A. Algorithmic Derivation

• The method first optimizes the uplink beamformer to maximize each user's uplink SINR, which has a generalized Rayleigh-quotient form.Source: A. Algorithmic Derivation · pp. 7–8
• Successive convex approximation linearizes a non-convex constraint and solves the resulting convex problems iteratively, providing a locally optimal solution.Source: A. Algorithmic Derivation · p. 8
• The relaxed subproblem is convex after the stated constraint reformulations, and dropping the rank-one constraint does not introduce sub-optimality for this problem's structure.Source: A. Algorithmic Derivation · pp. 8–9
• The design assumes perfect CSI; an ellipsoidal CSI-error model is outlined, but the imperfect-CSI extension is deferred to future work.The proposed extension is left for future work due to space constraints.Source: A. Algorithmic Derivation · p. 9

B. Optimality

• At each SCA iteration, the relaxed problem's solution is rank one for every downlink beamformer covariance matrix.Source: B. Optimality · p. 9, §1.3 · p. 9

1.3 ) satisfies

• The rank-one relaxation is exact for this problem, so eigenvalue decomposition suffices to recover the downlink beamforming vectors.Source: §1.3 · p. 9

V. SIMULATION RESULTS

• Monte Carlo simulations evaluate the proposed beamforming and power-allocation method under perfect CSI, with imperfect CSI left for future work.Source: A. Parameter Setup · p. 9

A. Parameter Setup

• The simulations use Rayleigh and Rician channel distributions, 20 dB log-normal shadowing, and free-space path loss.Source: A. Parameter Setup · p. 9
• The setup uses a 28 GHz carrier, 12-antenna transmit and receive ULAs at the base station, and 40 dBm base-station transmit power.Source: A. Parameter Setup · pp. 9–10

B. Benchmark Schemes

• ISO distributes power uniformly across UL, DL, and AN generation, then uses remaining power for eigenbeamforming while protecting legitimate users from interference.The design can still harm eavesdroppers’ reception.Source: B. Benchmark Schemes · p. 10
• The no-AN benchmark sets W = 0, uses zero-forcing precoding and uniform UL power allocation, and represents a communications-only focus.It is not optimized to form radar beams toward desired sensing locations.Source: B. Benchmark Schemes · p. 10
• SRCS minimizes eavesdropper SNR under communication SINR, radar, and power-budget constraints, but applies only to half-duplex scenarios.SCA without AN also sets W = 0 and is evaluated for half-duplex DL secure communication and sensing with no UL users (K = 0).Source: B. Benchmark Schemes · p. 10
• Each Monte Carlo trial uses the proposed method’s converged output power as the budget for the benchmarks, enabling a fair comparison despite secrecy rates increasing with transmission power.The benchmarks do not optimize radar-beam formation toward desired sensing locations for the isotropic and no-AN designs.Source: B. Benchmark Schemes · p. 10

Convergence and Power Consumption

• The algorithm converges in about 20 iterations; tightening the eavesdropper SINR target from 1 to 0.032 raises total power by 0.8 dB, and setting it to 0.001 raises it by 6.16 dB.The simulations also report 20 dB less required power at −80 dBm than at −60 dBm noise for the same SINR requirement.Source: C. Simulation Results · pp. 11–12

DL Secrecy Behaviour

• 13.28 bits/sec/Hz: the proposed method maintains this DL secrecy rate across channel conditions and noise levels, outperforming the isotropic and no-AN methods.The proposed rate is also stable across UL SINR requirements; SRCS settles around 11.96 bits/sec/Hz for L = 2 and 11.07 bits/sec/Hz for L = 5.Source: Figure 3 · p. 11, Figure 5 · p. 11, C. Simulation Results · pp. 12–13
• Above −30 dBm eavesdropper noise, the no-AN method can outperform the proposed method, which targets a minimum acceptable DL SINR across noise levels.Without AN, all available power goes toward communication; with high eavesdropper noise, maximizing DL SINR may suffice.Source: Figure 4 · p. 11, C. Simulation Results · p. 13

UL Secrecy Behaviour

• In the second node-position distribution, the proposed method achieves 1.38 bits/sec/Hz more UL secrecy rate than the no-AN and isotropic schemes at the same total power.When UL users lie on a 150 m-radius circle, all three methods perform alike; SRCS is absent because it is designed for half-duplex ISAC.Source: Figure 6 · p. 12, C. Simulation Results · p. 13

ISMR impact on Beampattern Design

• With ISMRmax = −30 dB, sidelobe power is about 10 dB above the −20 dB setting and 10 dB below the −40 dB setting.The paper notes that a less strict ISMR constraint allows higher achievable secrecy.Source: Figure 7 · p. 12, C. Simulation Results · p. 13

Sensing-Secrecy ISAC Trade-off

• For P = 3, relaxing ISMR^-1 below −20 dB raises secrecy rate from 0.74 to 2.76 bits/sec/Hz; an additional eavesdropper lowers it to 2.35 bits/sec/Hz.The reported trade-off is between higher secrecy and increasing ISMR; the ISMR constraint can be used to target a desired secrecy rate while performing radar tasks.Source: Figure 8 · p. 12, C. Simulation Results · p. 13

Importance of AN

• At a target secrecy rate of 3.79 bits/sec/Hz, AN reduces total consumed power from 8.35 dB to −1.92 dB, a gain of about 10.27 dB.The paper attributes the power saving to the additional subspace created by AN.Source: Figure 9 · p. 12, C. Simulation Results · p. 14
• At a feasibility rate of 0.93, AN raises the achievable secrecy rate from 3.79 to 7.35 bits/sec/Hz; below 2.73 bits/sec/Hz, both schemes have feasibility rates near 1.The paper attributes the higher-rate feasibility advantage to degrees of freedom introduced by AN.Source: C. Simulation Results · p. 14

VI. CONCLUSIONS AND RESEARCH DIRECTIONS

• The proposed successive convex approximation method addresses a non-convex power-minimization problem while jointly targeting uplink and downlink secrecy rates and radar ISMR levels.The framework also minimizes transmitted uplink/downlink power and artificial-noise generation power.Source: §VI · p. 14
• The authors report that simulations support the method’s effectiveness compared with classical secure-beamforming approaches and secure-ISAC methods.They cite its low iteration count and ability to satisfy the secrecy-rate and ISMR targets together.Source: §VI · p. 14
• Future work includes energy-efficient optimization and extensions to imperfect channel-state information.The conclusion frames energy efficiency as relevant to sustainable and green wireless networks.Source: §VI · p. 14
• The authors also propose exploring edge learning to distribute computation for the SCA-based algorithm, motivated by its relatively large number of required iterations.The discussion describes distributed implementation as a possible way to lower computation.Source: §VI · p. 14

APPENDIX A RANK-1 OPTIMALITY

• The appendix expresses the iteration’s optimization problem through a Lagrangian dual and uses the convex subproblem’s zero duality gap to solve it via the dual.The dual optimum determines the downlink beamforming matrices, artificial-noise covariance, and uplink power allocation.Source: Appendix A · pp. 14–15
• The rank-one proof assumes a higher-rank optimum, decomposes it into eigenvectors, and upper-bounds its cost to identify a rank-one solution attaining the bound.That construction contradicts the higher-rank assumption, establishing rank-one optimality.Source: Appendix A · p. 15