Dual-Target RIS-Assisted ISAC Scheme

• The paper introduces a dual target-mounted RIS-assisted ISAC scheme that integrates secure downlink beamforming with UAV sensing and adversarial interference management.
• It employs SDR-based optimization to jointly design BS beamforming and legitimate RIS phase shifts for enhancing secrecy rate and AoD estimation accuracy.
• Simulation results reveal that increasing legitimate RIS size improves secrecy rate, while malicious RIS interference and power scaling exhibit non-monotonic trends.

Dual target-mounted RISs-assisted ISAC denotes an integrated sensing and communication configuration in which a base station (BS) with ISAC capability simultaneously senses two unmanned aerial vehicle targets and communicates with legitimate users through a reconfigurable intelligent surface mounted on the legitimate UAV target, while a second RIS mounted on an eavesdropper UAV target acts maliciously by launching random interference attacks and by directly eavesdropping on downlink transmissions. In the formulation reported in "Dual Target-Mounted RISs-Assisted ISAC Against Eavesdropping and Malicious Interference" (Yigit et al., 2 Sep 2025), the architecture couples secure multiuser downlink transmission, target sensing, adversarial interference, and angle-of-departure estimation within a single optimization framework whose principal communication metric is secrecy rate and whose sensing metrics are sensing signal-to-interference-plus-noise ratio and the Cramér–Rao bound for AoD estimation.

1. System architecture and signal representation

The scheme considers a BS equipped with a uniform linear array of $T_x$ transmit antennas. Its two concurrent functions are sensing and communication. The sensing task targets two UAVs: a legitimate UAV target, denoted $L$-UAV, and an eavesdropper UAV target, denoted $E$-UAV. The communication task serves $K$ single-antenna legitimate users through a legitimate RIS mounted on the $L$-UAV. The hostile $E$-UAV carries a malicious RIS whose purpose is to disrupt BS-to-user communication by random interference, while the UAV itself also attempts to decode the transmitted data (Yigit et al., 2 Sep 2025).

The two target-mounted RISs are passive and are represented by diagonal reflection matrices

$$\mathbf{\Omega}_L=\mathrm{diag}\!\left(e^{j\psi_1},e^{j\psi_2},\ldots,e^{j\psi_{N_L}}\right),$$

$$\mathbf{\Omega}_M=\mathrm{diag}\!\left(e^{j\zeta_1},e^{j\zeta_2},\ldots,e^{j\zeta_{N_M}}\right).$$

Here, $\psi_l$ is the phase shift of the $l$-th element of the legitimate RIS, whereas $L$0 is the phase shift of the $L$1-th element of the malicious RIS. The paper assumes the malicious RIS phases are random, with i.i.d. $L$2, while the legitimate RIS phases are optimized (Yigit et al., 2 Sep 2025).

The BS emits a unified ISAC waveform,

$L$3

where $L$4 is the communication beamforming matrix, $L$5 is the sensing beamforming matrix, $L$6, $L$7, and the sensing and communication signals are orthogonal: $L$8 The overall transmit covariance is

$L$9

with power constraint $E$0.

The user-side received signal explicitly superposes the desired path through the legitimate RIS and the interference path through the malicious RIS: $E$1 where the BS-to-RIS and RIS-to-user links are Rician fading,

$E$2

The corresponding user SINR is

$E$3

The second interference term is the malicious RIS attack contribution.

The eavesdropper UAV is modeled as receiving the BS transmission directly,

$E$4

with decoding SINR for user $E$5,

$E$6

2. Sensing model, AoD estimation, and performance criteria

The sensing subsystem operates through line-of-sight round-trip echoes from both UAV targets. The BS-to-target steering vector is modeled by a two-dimensional AoD: $E$7 where $E$8 and $E$9 (Yigit et al., 2 Sep 2025).

Over $K$0 coherent blocks, the received echo from target $K$1 is

$K$2

with round-trip path attenuation $K$3. The sensing SINR for target $K$4 is

$K$5

This definition treats the other sensed UAV as an interference source, so the sensing task is intrinsically coupled across targets.

The scheme also derives a Cramér–Rao bound for estimating the AoD pair $K$6. After vectorizing the received signal,

$K$7

the Fisher information matrix is partitioned as

$K$8

The AoD-estimation CRB is then

$K$9

A central structural point is that the FIM entries depend on derivatives of the sensing steering matrix and on the transmit covariance $L$0. Consequently, beamforming design affects not only communication and secrecy, but also estimation accuracy. This is the technical basis for the paper’s joint treatment of secrecy rate, sensing SINR, and CRB (Yigit et al., 2 Sep 2025).

3. Dual security threats and secrecy-rate formulation

The scheme is defined by two simultaneous security threats. The first is direct eavesdropping: the $L$1-UAV listens to the BS transmission and attempts to decode legitimate user data. The second is a random interference attack: the malicious RIS mounted on the same $L$2-UAV reflects the BS signal with random phases, thereby producing interference at legitimate users and degrading user SINR and secrecy rate. The paper identifies this as a harder setting than standard eavesdropper-only secrecy models (Yigit et al., 2 Sep 2025).

User-$L$3 secrecy rate is defined as

$L$4

where

$L$5

and the total secrecy rate is

$L$6

This formulation makes secrecy depend jointly on the legitimate users’ SINRs, the eavesdropper’s decoding SINRs, the legitimate RIS configuration, and the malicious RIS attack contribution.

A common simplification in RIS-aided physical-layer security is to model the adversary exclusively as a passive eavesdropper. The dual target-mounted RIS setting rejects that simplification by allowing the hostile UAV to be both a listener and a disruption source. This suggests a broader adversarial model in which sensing geometry and interference structure are inseparable from secrecy performance, rather than being treated as separate design layers.

4. Optimization problem and SDR-based two-stage solution

The central design problem is posed as

$L$7

subject to

$L$8

$L$9

$E$0

The optimization variables are the BS beamforming matrix $E$1 and the legitimate RIS phase-shift matrix $E$2. The formulation is non-convex because secrecy rate is non-convex in $E$3 and $E$4, and because the legitimate RIS is subject to unit-modulus constraints (Yigit et al., 2 Sep 2025).

The paper decomposes the problem into two stages because the sensing channels of the UAVs to the BS are independent of the legitimate RIS configuration. In the first stage, the BS beamforming matrix is optimized through

$E$5

under the sensing and power constraints. Introducing

$E$6

the sensing SINR is rewritten as

$E$7

An SDR reformulation is then applied in terms of $E$8, after which the problem is solved with CVX/SeDuMi and the transmit beamforming matrix $E$9 is recovered by eigenvalue decomposition.

In the second stage, $$\mathbf{\Omega}_L=\mathrm{diag}\!\left(e^{j\psi_1},e^{j\psi_2},\ldots,e^{j\psi_{N_L}}\right),$$0 is held fixed and the legitimate RIS phases are optimized through

$$\mathbf{\Omega}_L=\mathrm{diag}\!\left(e^{j\psi_1},e^{j\psi_2},\ldots,e^{j\psi_{N_L}}\right),$$1

subject to unit-modulus constraints. The authors convert the SINR expression to a quadratic form by defining lifted matrix variables

$$\mathbf{\Omega}_L=\mathrm{diag}\!\left(e^{j\psi_1},e^{j\psi_2},\ldots,e^{j\psi_{N_L}}\right),$$2

and by introducing matrices such as

$$\mathbf{\Omega}_L=\mathrm{diag}\!\left(e^{j\psi_1},e^{j\psi_2},\ldots,e^{j\psi_{N_L}}\right),$$3

with analogous definitions for the interference terms. The user SINR becomes

$$\mathbf{\Omega}_L=\mathrm{diag}\!\left(e^{j\psi_1},e^{j\psi_2},\ldots,e^{j\psi_{N_L}}\right),$$4

The resulting QCQP/SDR problem relaxes the rank-one constraint, enforces $$\mathbf{\Omega}_L=\mathrm{diag}\!\left(e^{j\psi_1},e^{j\psi_2},\ldots,e^{j\psi_{N_L}}\right),$$5 and $$\mathbf{\Omega}_L=\mathrm{diag}\!\left(e^{j\psi_1},e^{j\psi_2},\ldots,e^{j\psi_{N_L}}\right),$$6, and recovers a feasible phase vector by Gaussian randomization or EVD if the returned $$\mathbf{\Omega}_L=\mathrm{diag}\!\left(e^{j\psi_1},e^{j\psi_2},\ldots,e^{j\psi_{N_L}}\right),$$7 is not rank-one (Yigit et al., 2 Sep 2025).

The paper reports the SDR complexities as

$$\mathbf{\Omega}_L=\mathrm{diag}\!\left(e^{j\psi_1},e^{j\psi_2},\ldots,e^{j\psi_{N_L}}\right),$$8

for beamforming optimization and

$$\mathbf{\Omega}_L=\mathrm{diag}\!\left(e^{j\psi_1},e^{j\psi_2},\ldots,e^{j\psi_{N_L}}\right),$$9

for RIS phase optimization, with solution accuracy $$\mathbf{\Omega}_M=\mathrm{diag}\!\left(e^{j\zeta_1},e^{j\zeta_2},\ldots,e^{j\zeta_{N_M}}\right).$$0. This establishes the scheme as algorithmically structured rather than heuristic, although the use of SDR and relaxation indicates that exact global optimality is not generally claimed.

5. Simulation regime and numerical characteristics

The simulation setup specified in the paper uses carrier frequency $$\mathbf{\Omega}_M=\mathrm{diag}\!\left(e^{j\zeta_1},e^{j\zeta_2},\ldots,e^{j\zeta_{N_M}}\right).$$1 GHz, $$\mathbf{\Omega}_M=\mathrm{diag}\!\left(e^{j\zeta_1},e^{j\zeta_2},\ldots,e^{j\zeta_{N_M}}\right).$$2 BS antennas, $$\mathbf{\Omega}_M=\mathrm{diag}\!\left(e^{j\zeta_1},e^{j\zeta_2},\ldots,e^{j\zeta_{N_M}}\right).$$3 users, $$\mathbf{\Omega}_M=\mathrm{diag}\!\left(e^{j\zeta_1},e^{j\zeta_2},\ldots,e^{j\zeta_{N_M}}\right).$$4 UAV targets, coherent block length $$\mathbf{\Omega}_M=\mathrm{diag}\!\left(e^{j\zeta_1},e^{j\zeta_2},\ldots,e^{j\zeta_{N_M}}\right).$$5, Rician factor $$\mathbf{\Omega}_M=\mathrm{diag}\!\left(e^{j\zeta_1},e^{j\zeta_2},\ldots,e^{j\zeta_{N_M}}\right).$$6 dB, path-loss exponent $$\mathbf{\Omega}_M=\mathrm{diag}\!\left(e^{j\zeta_1},e^{j\zeta_2},\ldots,e^{j\zeta_{N_M}}\right).$$7, and noise power $$\mathbf{\Omega}_M=\mathrm{diag}\!\left(e^{j\zeta_1},e^{j\zeta_2},\ldots,e^{j\zeta_{N_M}}\right).$$8 dBW. The BS transmit power is varied across simulations, as are the legitimate RIS size $$\mathbf{\Omega}_M=\mathrm{diag}\!\left(e^{j\zeta_1},e^{j\zeta_2},\ldots,e^{j\zeta_{N_M}}\right).$$9 and malicious RIS size $\psi_l$0. Geometrically, the BS height is $\psi_l$1 m, both UAVs are at $\psi_l$2 m, and users are on the ground; example target AoDs and distances are given in Table I of the source paper (Yigit et al., 2 Sep 2025).

The reported communication results exhibit several consistent trends. Increasing the malicious RIS size $\psi_l$3 worsens secrecy rate $\psi_l$4, because the random interference becomes stronger. Increasing the legitimate RIS size $\psi_l$5 significantly improves secrecy rate by strengthening the legitimate BS–RIS–user path. The proposed design is also reported to sustain secure communication even when $\psi_l$6 is large, which the authors use to characterize the SDR-based solution as robust under strong attacks.

The relationship between BS transmit power and secrecy is explicitly non-monotonic. Although larger $\psi_l$7 increases the intended signal strength, beyond a certain point it also increases sensing-related interference leakage to users and strengthens the eavesdropper’s reception, so the secrecy rate can decrease after a threshold. This is one of the paper’s most consequential numerical observations because it rules out a simple high-power design principle for secure ISAC in this setting.

The sensing results are likewise geometry-dependent. As the eavesdropper UAV moves farther away, its sensing SINR decreases due to path loss. The legitimate UAV may experience better sensing SINR when the eavesdropper is farther away, due to reduced inter-target interference. For AoD estimation, larger target distance worsens the CRB, whereas increasing BS transmit power improves the CRB. These findings show that the communication and sensing subsystems are coupled through both power allocation and target placement (Yigit et al., 2 Sep 2025).

When compared with a prior RIS-aided ISAC benchmark in which only an eavesdropper is present, the proposed scheme performs worse in secrecy rate because it faces two threats simultaneously: direct eavesdropping and malicious RIS interference. The comparison is used to demonstrate the increased difficulty of the dual-threat scenario rather than to claim dominance over simpler baselines.

6. Position within dual-RIS ISAC research

The expression “dual-RIS ISAC” covers multiple architectural families, and the target-mounted UAV scheme should not be conflated with all of them. A distinct line of work, exemplified by "Joint Active and Passive Beamforming with Sensing-Assisted Discrete Phase Shifts for Dual-RIS ISAC Systems" (Xue et al., 28 Oct 2025), studies a semi-passive dual-RIS-assisted mmWave ISAC system in which two RISs create virtual propagation paths for blocked links, the RIS sensing elements estimate user angles through 2D-MUSIC, and the main design objective is a max-min user SINR fairness problem under discrete phase shifts. That framework uses alternating optimization, SDR plus bisection for active beamforming, and sensing-assisted restriction of the RIS phase search space.

By contrast, the dual target-mounted RISs-assisted ISAC scheme centers on UAV-target sensing, secrecy-rate maximization, malicious random-phase interference, and CRB-based AoD estimation (Yigit et al., 2 Sep 2025). The RISs are not simply passive infrastructure components placed to overcome blockage; one is mounted on a legitimate target and assists secure downlink transmission, while the other is mounted on an adversarial target and actively disrupts communication.

A common misconception is therefore to treat all dual-RIS ISAC systems as variants of the same optimization problem. The literature summarized here indicates otherwise. In one case, dual RISs are cooperative sensing-and-reflection structures for fairness-oriented mmWave communication (Xue et al., 28 Oct 2025). In the target-mounted scheme, the dual-RIS structure is inseparable from an adversarial sensing-and-security model in which the legitimate RIS and malicious RIS have opposed operational roles (Yigit et al., 2 Sep 2025).

Within that narrower meaning, the practical significance claimed for the target-mounted scheme is that target-mounted RISs can function not only as communication enablers but also as active security entities in ISAC systems. A plausible implication is that future UAV-enabled wireless environments may require RIS design frameworks that explicitly combine secure downlink transmission, target tracking, interference resilience, and adversarial modeling, rather than optimizing any one of these objectives in isolation.