Multi-Cluster UAV-Aided SAGIN SemCom Architecture

• The paper introduces a joint resource and UAV placement optimization framework that achieves up to 30% higher throughput compared to baseline methods.
• It employs adaptive relay strategies with DeepSC-based semantic transmission to support heterogeneous users through efficient segmentation of semantic and bit-level communication.
• The study demonstrates that coordinated spectral and spatial optimization in a hierarchical space-air-ground network significantly improves end-to-end communication performance.

A multi-cluster UAV-aided SAGIN SemCom architecture integrates semantic communication (SemCom) with a hierarchical space-air-ground network, leveraging unmanned aerial vehicles (UAVs) as intermediate relays to enhance spectral efficiency and coverage in 6G and beyond. This system jointly supports both semantic users (SemUsers), capable of decoding semantic representations via DeepSC, and conventional users (ConUsers), limited to standard bit-level communication. By employing SemCom at the satellite-to-UAV link and an adaptive relay strategy at the UAVs, the architecture enables end-to-end communication efficiency, optimally allocating resources and dynamically positioning UAVs to meet diverse user demands and changing spatial conditions (Yin et al., 23 Nov 2025).

1. Network Model and System Entities

The architecture consists of three segments:

• Space segment: A single multi-antenna satellite base station (S) equipped with $K$ transmit antennas conducts space-to-air downlink.
• Air segment: $N$ single-antenna UAV relays $\{R^{(n)}\}_{n=1}^N$, each assigned to a distinct ground cluster for spatial reuse.
• Ground segment: Each UAV $R^{(n)}$ serves a local cluster of ground users (GUs), which are divided into:
  • Semantic users (SemUsers) $U^{(n)}_{\text{sem},i}$, $i=1\ldots I^{(n)}$, which decode semantic codewords with DeepSC.
  • Conventional users (ConUsers) $U^{(n)}_{\text{con},j}$, $j=1\ldots J^{(n)}$, which require bit-level streams.

Frequency Division Multiple Access (FDMA) is used for:

• Satellite-to-UAV: Orthogonal subbands per relay (no inter-UAV interference).
• UAV-to-GU: Inside each cluster, intra-cluster links share a band, and inter-cluster interference is negligible due to frequency and spatial isolation.

Channel models:

• Space-to-air (S$\to$R):

$$h^{(n)}_{S2R} = \sqrt{\varsigma^{(n)} \frac{\tilde{\sigma}^{(n)}}{4\pi d_{S2R}^{(n)}}}$$

with large-scale path loss, where $d_{S2R}^{(n)}$ is distance, $\tilde{\sigma}^{(n)}$ is wavelength, and $\varsigma^{(n)}$ is MRT beamforming gain.

• Air-to-ground (R$\to$X):

$$H^{(n)}_{R2X}(\ell_R^{(n)}, \ell_X) = \beta_0 \|\ell_R^{(n)}-\ell_X\|^{-\alpha}$$

for $X\in\{\text{SU},i\} \cup \{\text{CU},j\}$, with free-space LoS propagation.

2. Semantic and Bit-Level Communication Strategies

2.1 Semantic Transmission (Satellite-to-UAV)

The satellite executes DeepSC encoding, mapping raw inputs to $Q$ semantic symbols per block. Transmission is analog, at symbol rate $B^{(n)}_{S2R}$. The received semantic fidelity is a function $\varepsilon_Q(r)$ of SNR $r$.

• Effective semantic rate (suts/s):

$$\Psi^{(n)}_{S2R} = \frac{B^{(n)}_{S2R} M}{Q}\, \varepsilon_Q(r^{(n)}_{S2R})$$

• Equivalent bit rate (bits/s):

$$\Gamma^{(n)}_{S2R} = \mu_1 \frac{B^{(n)}_{S2R}}{Q}\, \varepsilon_Q(r^{(n)}_{S2R})$$

where $M$ is the semantic units/block and $\mu_1$ is bits/unit.

2.2 Adaptive UAV Relay

Upon reception, UAV $R^{(n)}$ bifurcates processing:

• To SemUsers: Re-encodes semantic symbols into bits using standard source and channel coding; transmits digitally to SemUsers.
  • Semantic rate:

  $$\Psi^{(n)}_{R2SU,i} = \frac{B^{(n)}_{R2SU,i} \eta^{(n)}_{R2SU,i} M}{\mu_2 Q}$$ - Bit-equivalent rate:

  $$\Gamma^{(n)}_{R2SU,i} = \frac{\mu_1}{\mu_2 Q} B^{(n)}_{R2SU,i} \eta^{(n)}_{R2SU,i}$$

• To ConUsers: Fully decodes the semantic symbols to recover the raw bit message onboard, then re-encodes and transmits.

  • Achievable bit rate:

  $$\Gamma^{(n)}_{R2CU,j} = B^{(n)}_{R2CU,j} \eta^{(n)}_{R2CU,j}$$

The UAV relay strategy accounts for the heterogeneous decoding capabilities and optimizes transmission modes accordingly.

3. Joint Resource and UAV Position Optimization

The primary objective is to maximize the aggregate end-to-end sum-rate to all ground users by optimally selecting:

• UAV 3D positions $\{\ell_R^{(n)}\}$

• Per-link power allocations $$\{P^{(n)}_{S2R}, P^{(n)}_{R2SU,i}, P^{(n)}_{R2CU,j}\}$$

• Per-link bandwidth allocations $$\{B^{(n)}_{S2R}, B^{(n)}_{R2SU,i}, B^{(n)}_{R2CU,j}\}$$

Optimization Objective and Constraints

The optimization is formally stated as: $$\max_{P,B,\ell} \sum_{n=1}^N \left(\sum_{i=1}^{I^{(n)}} \Gamma^{(n)}_{R2SU,i} + \sum_{j=1}^{J^{(n)}} \Gamma^{(n)}_{R2CU,j}\right)$$

Subject to:

• Relay flow constraint: Total GU outflow cannot exceed satellite-to-UAV link capacity.

• Total bandwidth and power budgets: Satellite and UAVs.

• Computation power at UAV: Related to CPU cycles required for relaying by user type.

• UAV altitude and flight range box.

• Non-negativity, semantic similarity, and other QoS constraints.

These jointly non-convex constraints create a tightly coupled, large-scale resource allocation problem.

4. Alternating Optimization Algorithm

To address the non-convex problem, an alternating optimization (AO) framework is employed, using auxiliary variables to decouple SNR, channel gains, spectral efficiencies, and CPU frequency constraints. The process iteratively solves three convex subproblems:

Stage	Variables	Solution Approach
Bandwidth Allocation ($P3$)	$$\{B^{(n)}_{S2R},B^{(n)}_{R2SU,i},B^{(n)}_{R2CU,j}\}$$	KKT/primal-dual
Auxiliary Variable Update	Bounds on gains, SNRs, spectral efficiencies, and CPU frequency	Closed-form (KKT)
Power & UAV Placement ($P5$)	$$\{P^{(n)}_{S2R},P^{(n)}_{R2SU,i},P^{(n)}_{R2CU,j}, \ell_R^{(n)}\}$$	KKT after linearization

UAV $n$'s position decouples and is computed as a weighted centroid of user locations, with separate weights for SemUsers and ConUsers. This structure guarantees convergence to a stationary point because the objective is non-decreasing and the feasible set is compact.

The iterative AO pseudo-code sequence is:

1. Initialize all resource and position variables.

2. Repeat:

   • Solve for optimal bandwidth allocation.
   • Update auxiliary bounds.
   • Solve for power and UAV positions.
3. Until convergence of sum-rate.

5. Performance Analysis and Empirical Results

5.1 Simulation Setup

Key physical and system parameters include:

• Satellite at 60 km orbital height; UAVs hover at 1 km.
• $N_0 = 10^{-20}$ W/Hz; $Q=4$ symbols/block; $\mu_1=48$ bits/unit; $\mu_2=4$ bits/symbol.
• Computation: $G_{sem}=2$ FLOP/sut, $G_{con}=4$ FLOP/bit, $z=2$ FLOP/cycle, $\zeta_0=10^{-3}$.
• Air-to-ground path loss: $\beta_0=-60$ dB at 1 m, $\alpha=2$; space-to-air: $|h|^2\approx-80$ dB.
• Budgets: $\tilde B_S=10$ MHz, $\tilde P_S=1$ kW; $\tilde B_R=10$ MHz, $\tilde P_R=10$ W.
• Users in each cluster randomly placed within a $1\times 1$ km$^2$ area.

5.2 Baseline Comparisons

• Fixed-bandwidth: Equal bandwidths, optimizes power and UAV placement.
• Fixed-power: Equal power per link, optimizes bandwidth and UAV position.
• Fixed-location: UAVs at geometric centroids, optimizes bandwidth and power.

5.3 Key Findings

• Bandwidth limitation: Rapid sum-rate increases at low UAV bandwidth, but plateaus when satellite bandwidth becomes the bottleneck. Proposed joint design shows 20–30% higher throughput than baselines.
• Power limitation: Diminishing rate gains at high UAV power, which are amplified when satellite power is also increased.
• User mix impact: All-SemUser clusters achieve the highest sum-rates, while all-ConUser clusters are lowest. Physically grouping users by capability yields higher efficiency than mixing.
• Cluster scaling: Increasing the number of UAVs (clusters) enables super-linear sum-rate growth via spatial reuse.
• Satellite movement: Dynamic spatial diversity is exploited as the satellite passes over clusters, and the AO algorithm reallocates resources efficiently in real time.
• UAV placement: Optimal positions shift toward ConUsers to offset their weaker links. The weighted centroid method outperforms naïve geometric centroids.

5.4 Practical Insights and Trade-offs

• Inter-hop resource competition necessitates careful end-to-end balancing of bandwidth and power.
• Semantic links yield approximately 2$\times$ spectral efficiency compared to bit-level, offset by additional onboard computation at the UAV.
• Coordinated resource and UAV position optimization unlocks spatial and semantic diversity, with up to 30% gains over decoupled baselines.
• Organizing clusters by user type simplifies allocation and can improve overall spectral efficiency.

6. Summary and Implications

The multi-cluster UAV-aided SAGIN SemCom system introduces a hierarchical, spatially distributed network for 6G-era communications, combining semantic-aware information transfer with adaptive relay strategies and coordinated resource management (Yin et al., 23 Nov 2025). The architecture supports hybrid user populations and leverages the spectral efficiency of SemCom while maintaining bit-level accessibility for legacy devices. The joint optimization framework and AO algorithm achieve significant sum-rate and efficiency improvements. This line of research demonstrates that intelligent placement, resource allocation, and semantic processing—when considered jointly—can overcome longstanding channel impairments in space-air-ground networks and maximize network performance under realistic physical and operational constraints.