With the rapid development of Sixth-Generation (6G) mobile networks, there is an increasing demand for higher spectral efficiency, energy efficiency, low latency, and extended coverage. In this context, Unmanned Aerial Vehicle (UAV) communication has emerged as a key technology for 6G and beyond, leveraging its unique Line-of-Sight (LoS) propagation characteristics to ensure high data transmission rates. The JUYE UAV, with its controllable mobility, rapid deployment capability, low operational costs, and flexible operation, demonstrates significant advantages when deployed as an aerial base station. It enhances network performance by improving spectral efficiency and connection reliability, particularly in critical scenarios such as emergency communications, remote area coverage, and maritime operations.
To meet the growing quality-of-service requirements, integrating advanced multiple access technologies like Non-Orthogonal Multiple Access (NOMA) and Rate-Splitting Multiple Access (RSMA) is essential. However, as the number of ground users increases, the complexity of handling inter-user interference cancellation in NOMA systems also rises. In contrast, RSMA decomposes each user’s message into two parts: a common message decodable by all users and a private message decodable only by specific users or user groups. This message-splitting mechanism creates a balanced interference management strategy between fully decoding interference and treating it as noise. Thus, RSMA has become a research hotspot due to its lower implementation complexity. It unifies and generalizes traditional multiple access schemes like Orthogonal Multiple Access (OMA), Space Division Multiple Access (SDMA), and NOMA, showing significant advantages in spectral efficiency, energy efficiency, quality of service, and transmission reliability. Integrating RSMA technology with Unmanned Aerial Vehicle communication systems is considered a highly promising direction for future mobile networks.
UAV communication systems face dual threats of inter-user interference and malicious attacks due to the inherent broadcast and superposition nature of their channels. To address this Physical Layer Security (PLS) issue, the introduction of Intelligent Reflecting Surfaces (IRS) provides a novel solution for enhancing system performance. As an emerging wireless communication technology, IRS dynamically adjusts the electromagnetic wave propagation environment, enabling precise control and optimization of signals. While reducing system power consumption and costs, it significantly improves communication security and stability. In research on IRS-assisted Unmanned Aerial Vehicle communications, various optimization strategies have been proposed. For instance, some studies focus on minimizing private message transmission power in IRS-assisted networks using RSMA by jointly optimizing base station beamforming and IRS phase shifts. Others explore Simultaneously Transmitting and Reflecting Reconfigurable Intelligent Surface (STAR-RIS) assisted RSMA secure simultaneous wireless information and power transfer systems, maximizing the worst-case secrecy rate under imperfect Channel State Information (CSI) conditions. Additionally, research on RIS-assisted RSMA-enhanced integrated sensing and communication systems jointly optimizes communication/radar precoders and RIS phase shifts to maximize the minimum secrecy rate under transmit power constraints.
Despite these advancements, current research on security in IRS-assisted RSMA systems has limitations. First, existing works primarily focus on optimizing the minimum secrecy rate of legitimate users and transmit power, with less attention to overall system secrecy rate improvement. Second, most studies consider only a single eavesdropper, overlooking the practical threats posed by multiple eavesdroppers. Furthermore, while RSMA shows significant potential in physical layer security, existing research often employs NOMA technology, not fully exploring RSMA’s application in IRS-assisted Unmanned Aerial Vehicle communication systems. To address these gaps, this paper proposes a resource allocation algorithm for maximizing the secrecy rate in IRS-assisted UAV RSMA systems. The main contributions are as follows:
First, we propose a downlink RSMA multi-user communication scenario for an IRS-assisted Unmanned Aerial Vehicle system. In the presence of multiple external eavesdroppers, we maximize the system secrecy rate by jointly optimizing precoding vectors, common secrecy rate allocation, IRS phase shifts, and UAV positioning. Second, we employ a hierarchical optimization approach to decompose the original problem into two parts: an inner-layer optimization problem that alternately optimizes precoding vectors, common secrecy rate allocation, and IRS phase shifts given a fixed UAV position; and an outer-layer optimization problem that optimizes the UAV position given other variables. In the inner-layer optimization, non-convex problems are transformed into convex optimization problems using methods like Successive Convex Approximation (SCA), relaxation variables, first-order Taylor expansion, and Semidefinite Relaxation (SDR). For the outer-layer optimization, the Particle Swarm Optimization (PSO) algorithm is used to obtain the optimal UAV deployment position. Third, simulation results show that the proposed algorithm effectively improves the system secrecy rate and outperforms existing benchmark schemes.
We consider an IRS-assisted Unmanned Aerial Vehicle system with a downlink RSMA communication scenario. The system includes one UAV base station equipped with multiple antennas, one IRS mounted on a building, multiple legitimate users, and multiple eavesdroppers. The UAV’s position is denoted as \( \mathbf{q} = (x, y, H) \), the IRS position as \( \mathbf{r} = (x_r, y_r, z_r) \), the legitimate user \( U_k \)’s position as \( \mathbf{w}_k = (x_k, y_k, 0) \), and the eavesdropper \( E_j \)’s position as \( \mathbf{w}_j = (x_j, y_j, 0) \). The sets of legitimate users and eavesdroppers are defined as \( \mathcal{K} = \{1, 2, \dots, K\} \) and \( \mathcal{J} = \{1, 2, \dots, J\} \), respectively. The channel gains between the UAV and \( U_k \), UAV and \( E_j \), UAV and IRS, IRS and \( U_k \), and IRS and \( E_j \) are denoted as \( \mathbf{h}_{u,k} \in \mathbb{C}^{M \times 1} \), \( \mathbf{h}_{u,j} \in \mathbb{C}^{M \times 1} \), \( \mathbf{G} \in \mathbb{C}^{N \times M} \), \( \mathbf{h}_{r,k} \in \mathbb{C}^{N \times 1} \), and \( \mathbf{h}_{r,j} \in \mathbb{C}^{N \times 1} \), respectively. The IRS reflection phase shift vector is \( \mathbf{u} = [e^{j\theta_1}, e^{j\theta_2}, \dots, e^{j\theta_N}] \), and the reflection matrix is \( \mathbf{\Theta} = \text{diag}(\mathbf{u}) \in \mathbb{C}^{N \times N} \), where \( \theta_n \in [0, 2\pi) \) is the phase shift of the \( n \)-th IRS element.
The UAV base station employs RSMA to transmit messages to multiple users, sending \( K+1 \) data streams \( \mathbf{s} = [s_c, s_1, \dots, s_K]^T \in \mathbb{C}^{(K+1) \times 1} \) with \( \mathbb{E}\{\mathbf{s}\mathbf{s}^H\} = \mathbf{I} \). The superimposed signal transmitted by the UAV is \( \mathbf{x} = \mathbf{P}\mathbf{s} = \mathbf{p}_c s_c + \sum_{k \in \mathcal{K}} \mathbf{p}_k s_k \), where \( \mathbf{P} = [\mathbf{p}_c, \mathbf{p}_1, \dots, \mathbf{p}_K] \) is the precoding matrix, and \( \mathbf{p}_c, \mathbf{p}_k \in \mathbb{C}^{M \times 1} \) are the precoding vectors for the common stream \( s_c \) and private stream \( s_k \), respectively. The received signals at \( U_k \) and \( E_j \) are \( y_k = \mathbf{h}_k \mathbf{x} + n_k = (\mathbf{h}_{u,k}^H + \mathbf{h}_{r,k}^H \mathbf{\Theta} \mathbf{G}) \mathbf{x} + n_k \) and \( y_j = \mathbf{h}_j \mathbf{x} + n_j = (\mathbf{h}_{u,j}^H + \mathbf{h}_{r,j}^H \mathbf{\Theta} \mathbf{G}) \mathbf{x} + n_j \), where \( n_k, n_j \) are additive white Gaussian noise with zero mean and variance \( \sigma^2 \).
The achievable rates for the common and private streams at \( U_k \) are \( R_{c,k} = \log_2(1 + \Gamma_{c,k}) \) and \( R_k = \log_2(1 + \Gamma_k) \), where \( \Gamma_{c,k} = \frac{|\mathbf{h}_k \mathbf{p}_c|^2}{\sum_{k \in \mathcal{K}} |\mathbf{h}_k \mathbf{p}_k|^2 + \sigma^2} \) and \( \Gamma_k = \frac{|\mathbf{h}_k \mathbf{p}_k|^2}{\sum_{k’ \in \mathcal{K}, k’ \neq k} |\mathbf{h}_k \mathbf{p}_{k’}|^2 + \sigma^2} \). The eavesdropping rates at \( E_j \) for the common and private streams of \( U_k \) are \( R_{c,j} = \log_2(1 + \gamma_{c,j}) \) and \( R_{k,j} = \log_2(1 + \gamma_{k,j}) \), where \( \gamma_{c,j} = \frac{|\mathbf{h}_j \mathbf{p}_c|^2}{\sum_{k \in \mathcal{K}} |\mathbf{h}_j \mathbf{p}_k|^2 + \sigma^2} \) and \( \gamma_{k,j} = \frac{|\mathbf{h}_j \mathbf{p}_k|^2}{\sum_{k’ \in \mathcal{K}, k’ \neq k} |\mathbf{h}_j \mathbf{p}_{k’}|^2 + |\mathbf{h}_j \mathbf{p}_c|^2 + \sigma^2} \).
Considering the worst-case scenario with \( J \) eavesdroppers, the actual secrecy rate for \( U_k \) is \( \hat{R}_{\text{sec},k} = r_{\text{sec},c,k} + r_{\text{sec},p,k} \), where \( r_{\text{sec},c,k} \) is the non-negative common secrecy rate allocated to user \( k \), satisfying \( \sum_{k \in \mathcal{K}} r_{\text{sec},c,k} \leq R_c – \max_{j \in \mathcal{J}} \{R_{c,j}\} \), and \( r_{\text{sec},p,k} = [R_k – \max_{j \in \mathcal{J}} \{R_{k,j}\}]^+ \), with \( [x]^+ = \max(0, x) \). The system secrecy rate is \( \hat{R}_{\text{sec}}^{\text{tot}} = \sum_{k=1}^K \hat{R}_{\text{sec},k} \).
The optimization problem to maximize the system secrecy rate is formulated as:
$$
\begin{aligned}
\text{P1: } & \max_{\mathbf{p}, \mathbf{c}, \mathbf{q}, \mathbf{u}} \hat{R}_{\text{sec}}^{\text{tot}} \\
\text{s.t. } & \text{C1a: } \sum_{k \in \mathcal{K}} r_{\text{sec},c,k} \leq R_c – \max_{j \in \mathcal{J}} \{R_{c,j}\} \\
& \text{C1b: } r_{\text{sec},c,k} \geq 0, \forall k \in \mathcal{K} \\
& \text{C1c: } R_c \leq R_{c,k}, \forall k \in \mathcal{K} \\
& \text{C1d: } \text{tr}(\mathbf{P} \mathbf{P}^H) \leq P_{\text{max}} \\
& \text{C1e: } \theta_n \in [0, 2\pi), n = 1, 2, \dots, N \\
& \text{C1f: } x_{\text{min}} \leq x \leq x_{\text{max}}, y_{\text{min}} \leq y \leq y_{\text{max}}
\end{aligned}
$$
where \( \mathbf{c} = [r_{\text{sec},c,1}, r_{\text{sec},c,2}, \dots, r_{\text{sec},c,K}]^T \) is the common secrecy rate allocation vector, and \( P_{\text{max}} \) is the maximum transmit power of the Unmanned Aerial Vehicle.
Due to the non-convexity of the objective function and constraints C1a and C1c, and the coupling of optimization variables, the problem is challenging to solve directly. We introduce an auxiliary variable \( t \) and reformulate the problem as:
$$
\begin{aligned}
\text{P2: } & \max_{\mathbf{p}, \mathbf{c}, \mathbf{q}, \mathbf{u}} t \\
\text{s.t. } & \text{C2a: } \sum_{k \in \mathcal{K}} r_{\text{sec},c,k} + \sum_{k \in \mathcal{K}} r_{\text{sec},p,k} \geq t, \forall k \in \mathcal{K} \\
& \text{C1a-C1f}
\end{aligned}
$$
We adopt a hierarchical optimization approach, decomposing P2 into inner-layer and outer-layer optimization problems. For inner-layer optimization, given the UAV position \( \mathbf{q} \), we alternately optimize the precoding vectors and common secrecy rate allocation subproblem and the IRS phase shift subproblem.
For the precoding vectors and common secrecy rate allocation optimization, we handle non-convex constraints using SCA and first-order Taylor approximations. Introducing auxiliary variables \( \rho_k \) and \( \rho_{k,e} \), we approximate \( r_{\text{sec},p,k} \) as \( \tilde{r}_{\text{sec},p,k} = \log_2(1 + \rho_k) – \frac{\log_2(e)}{1 + \rho_{k,e}^{(t)}} (\rho_{k,e} – \rho_{k,e}^{(t)}) \), where \( \rho_{k,e}^{(t)} \) is the value at the \( t \)-th iteration. Similarly, we transform constraints on \( \Gamma_{c,k} \) and \( \Gamma_k \) into convex forms using difference-of-convex decomposition and first-order approximations. For constraint C1a, we introduce auxiliary variables \( R_{c,e} \), \( x_{c,j} \), \( a_{k,j} \), and \( \rho_{c,j} \), and derive affine approximations. The transformed problem becomes a second-order cone program (SOCP) that can be solved using CVX.
For IRS phase shift optimization, we define \( \mathbf{H}_q = [\text{diag}(\mathbf{h}_{r,q}^H) \mathbf{G}; \mathbf{h}_{u,q}^H] \) and \( \mathbf{v} = [\mathbf{u}; 1] \), so that \( |(\mathbf{h}_{u,q}^H + \mathbf{h}_{r,q}^H \mathbf{\Theta} \mathbf{G}) \mathbf{p}_n|^2 = \text{Tr}(\mathbf{V} \mathbf{H}_q \mathbf{p}_n \mathbf{p}_n^H \mathbf{H}_q^H) \), where \( \mathbf{V} = \mathbf{v} \mathbf{v}^H \). We relax the rank-one constraint using SDR and solve the resulting convex problem. For discrete IRS phase shifts, we quantize the continuous solution to the nearest discrete value in the set \( \mathcal{F} = \{0, \frac{2\pi}{2^B}, \dots, \frac{2\pi(2^B – 1)}{2^B}\} \), where \( B \) is the number of quantization bits.
For outer-layer optimization, we optimize the UAV position \( \mathbf{q} \) using the Particle Swarm Optimization algorithm. We initialize \( P \) particles with positions \( \mathbf{W}_p = [\mathbf{w}_1, \mathbf{w}_2, \dots, \mathbf{w}_p] \) and velocities \( \mathbf{v}_p \). For each particle position, we compute the utility function \( Q(\mathbf{w}_i) = \hat{R}_{\text{sec}}^{\text{tot}}(\mathbf{w}_i) \) by solving the inner-layer optimization. The velocity and position update equations are:
$$
\begin{aligned}
\mathbf{v}_i^{t+1} &= w \mathbf{v}_i^t + c_1 r_1 (\mathbf{w}_{i,l}^t – \mathbf{w}_i^t) + c_2 r_2 (\mathbf{w}_g^t – \mathbf{w}_i^t) \\
\mathbf{w}_i^{t+1} &= \mathbf{w}_i^t + \mathbf{v}_i^{t+1}
\end{aligned}
$$
where \( w \) is the inertia weight, \( c_1 \) and \( c_2 \) are learning factors, \( r_1 \) and \( r_2 \) are random variables uniformly distributed in [0,1], \( \mathbf{w}_{i,l}^t \) is the personal best position of particle \( i \), and \( \mathbf{w}_g^t \) is the global best position. The process iterates until convergence or a maximum number of iterations is reached.
The convergence of the inner-layer algorithm is guaranteed as the objective function is non-decreasing and bounded above due to the transmit power constraint. The PSO algorithm ensures convergence as the global best solution is non-decreasing and bounded. The computational complexity of the inner-layer algorithm is \( O(T_{\text{max}} \log(1/\epsilon_1) (O_1 + O_2)) \), where \( O_1 = O((MK + KJ)^{3.5}) \) for the SOCP and \( O_2 = O(N^{6.5}) \) for the SDP, \( T_{\text{max}} \) is the maximum iterations, and \( \epsilon_1 \) is the accuracy. The outer-layer PSO complexity is \( O(P T) \), where \( P \) is the population size and \( T \) is the maximum iterations. Thus, the overall complexity is \( O(P T \cdot T_{\text{max}} \log(1/\epsilon_1) (O_1 + O_2)) \).
We conduct simulations to validate the proposed algorithm. The channels are modeled as Rician fading. The channel between the UAV and legitimate user \( U_k \) is:
$$
\mathbf{h}_{u,k} = \sqrt{\frac{\rho_0}{\|\mathbf{q} – \mathbf{w}_k\|^{\kappa_1}}} \left( \sqrt{\frac{\beta_1}{\beta_1 + 1}} \bar{\mathbf{h}}_{u,k} + \sqrt{\frac{1}{\beta_1 + 1}} \tilde{\mathbf{h}}_{u,k} \right)
$$
where \( \rho_0 \) is the path loss at reference distance 1 m, \( \kappa_1 \) is the path loss exponent, \( \beta_1 \) is the Rician factor, \( \bar{\mathbf{h}}_{u,k} \) is the LoS component, and \( \tilde{\mathbf{h}}_{u,k} \) is the non-LoS component. Similar models apply to other channels. The UAV-IRS channel is LoS-dominated. We set \( K = 4 \) users, \( \rho_0 = -30 \) dB, \( \kappa_1 = \kappa_2 = \kappa_3 = \kappa_4 = 3.6 \), \( \beta_1 = \beta_2 = \beta_3 = \beta_4 = 10 \) dB, UAV height \( H = 100 \) m, PSO parameters \( w = 0.7 \), \( c_1 = c_2 = 2 \).
The following table summarizes key simulation parameters:
| Parameter | Value |
|---|---|
| Number of legitimate users \( K \) | 4 |
| Number of eavesdroppers \( J \) | 2 |
| UAV antennas \( M \) | 4, 6, 8 |
| IRS elements \( N \) | 16, 32, 64 |
| UAV max transmit power \( P_{\text{max}} \) | 20, 30 dBm |
| Path loss exponent \( \kappa \) | 3.6 |
| Rician factor \( \beta \) | 10 dB |
| UAV height \( H \) | 100 m |
We compare the proposed algorithm with three benchmarks: RSMA without IRS, NOMA with IRS, and NOMA without IRS. The results show that the proposed algorithm achieves higher secrecy rates across various scenarios.
First, we examine convergence. The system secrecy rate increases with iterations and converges within 20-40 iterations under different UAV antenna configurations and transmit powers. For example, with \( M = 6 \) and \( P_{\text{max}} = 30 \) dBm, the secrecy rate is 15.1% higher than with \( M = 4 \) in early iterations, and the gap narrows later. With \( M = 4 \) and \( P_{\text{max}} = 30 \) dBm, the rate is 14.5% higher than with \( P_{\text{max}} = 20 \) dBm.
Next, we analyze the impact of UAV transmit power. As \( P_{\text{max}} \) increases, the secrecy rate improves for all schemes because higher power enhances signal strength at legitimate users. At \( P_{\text{max}} = 30 \) dBm, the proposed scheme outperforms RSMA without IRS, NOMA with IRS, and NOMA without IRS by 7.3%, 168%, and 187.5%, respectively.
The effect of UAV antenna count is also significant. With more antennas, the secrecy rate rises due to increased spatial degrees of freedom and improved beamforming gain. At \( M = 8 \), the proposed scheme shows improvements of 5.3%, 151%, and 169.2% over the benchmarks.
Increasing the number of IRS reflecting elements \( N \) boosts the secrecy rate for IRS-assisted schemes by providing more spatial freedom to enhance legitimate channels and suppress eavesdropping. At \( N = 16 \), the proposed scheme achieves 11%, 144.6%, and 197.3% gains over the benchmarks.
Finally, the optimal UAV position is near the center of legitimate users and close to the IRS, minimizing path loss and leveraging IRS passive beamforming to focus energy on legitimate users while degrading eavesdropping channels.
In conclusion, this paper addresses secure communication in IRS-assisted Unmanned Aerial Vehicle RSMA systems with multiple eavesdroppers. We propose a resource allocation algorithm that jointly optimizes precoding vectors, common secrecy rate allocation, IRS phase shifts, and UAV positioning to maximize the system secrecy rate. Using hierarchical optimization, SCA, SDR, and PSO, we effectively solve the non-convex problem. Simulations confirm that the proposed algorithm significantly outperforms benchmark schemes in secrecy rate. Future work could explore robust beamforming under imperfect CSI, discrete IRS phase shift optimization, and integration with other advanced technologies for enhanced security in JUYE UAV applications.