UAV swarm spectrum allocation is becoming a frontline issue for both civil and defence users. UAV swarms rely on wireless links for control and data. However, in contested environments, interference changes fast. As a result, swarms often lack complete channel information.
A research team led by Professor Lin at Harbin Engineering University has proposed an intelligent decision framework. The method aims to allocate spectrum even when interference data is incomplete and time-varying. In addition, it treats communications security as a core objective, not an afterthought.
Key Facts
- Problem: UAV swarms must share spectrum with uncertain and partially observed interference.
- Approach: Fuzzy logic models interference using qualitative inputs and updates constraints in real time.
- Optimisation: A dynamic, constrained, multi-objective model balances link performance and security.
- Acceleration: Transfer learning reuses prior solutions to speed re-optimisation after changes.
- Outlook: The team plans joint optimisation of spectrum, compute, and storage for 6G space–air–ground networks.
Why Spectrum Management Is Hard for Swarms
UAV swarms must coordinate at speed. Yet their open wireless links expose them to interference, jamming, and eavesdropping. Traditional spectrum schemes often assume full and accurate data. However, real deployments rarely offer that clarity.
Interference can come from many sources. For example, nearby emitters may appear and disappear. Meanwhile, weather, terrain, and mobility can reshape signal paths. Therefore, a swarm needs a method that works with partial observations.
Fuzzy Logic Turns Vague Interference into Usable Constraints
The framework uses fuzzy logic to describe external interference without precise measurements. It represents interference intensity, range, and distance through fuzzy sets and membership functions. Then it applies inference rules to translate those inputs into spectrum constraints.
Crucially, the constraints update as conditions change. That way, the swarm can act on “good-enough” situational awareness. Even so, it avoids waiting for perfect sensing that may never arrive.
A Multi-Objective Model Balances Performance and Security
At the centre of the method is a dynamic constrained multi-objective optimisation model. It seeks to reduce self-interference inside the swarm. At the same time, it limits the throughput that an eavesdropper could achieve.
The model also enforces practical constraints. For instance, it respects spectrum utilisation limits. It also prevents frequency conflicts. Finally, it maintains minimum link quality for mission traffic.
Because the objectives can conflict, the framework produces trade-off solutions. In other words, it offers options that keep links stable while reducing leakage risk.
Transfer Learning Speeds Decisions When the Environment Shifts
Interference landscapes can change in seconds. So the team designed a transfer learning component to avoid restarting from scratch. The algorithm, TrS-DCMOEA, maps prior solutions into a latent space. It then projects them into the new environment.
As a result, the new optimisation run starts with a higher-quality initial population. That speeds convergence. Therefore, it supports near-real-time spectrum updates for a moving swarm.
Why This Matters for Contested Spectrum and 6G Networks
Secure swarm links matter for both defence and civil missions. For example, disaster assessment swarms must keep links up amid congestion. Meanwhile, in hostile settings, swarms need to resist interference and limit interception.
The team also points toward larger system goals. Next, it plans to expand the framework to larger swarms. In addition, future work targets joint optimisation of spectrum, computation, and storage. This aligns with integrated space–air–ground 6G concepts.
Limitations and Next Steps
The framework is a strong step toward autonomy under uncertainty. Still, implementation details will matter. For example, real-time compute budgets and onboard power limits can shape performance. Likewise, contested-spectrum test ranges will be essential to validate behaviour.
Even so, the core idea is clear. When interference data is incomplete, the swarm can still make robust choices. It can do so by combining fuzzy perception, multi-objective optimisation, and transfer learning.
