Max-Min Resource Allocation for Anti-Jamming: Fairness Under Jamming
Max-min resource allocation for anti-jamming keeps the weakest link alive by equalizing throughput across every node, even while a reactive jammer senses and strikes. Here is how the equilibrium strategy, algorithms, and reinforcement learning methods actually perform.
What Is Max-Min Resource Allocation and Why Does It Matter for Anti-Jamming?
Max-min resource allocation is basically a fairness rule: when you're handing out limited communication resources, you give priority to whoever's worst off, so the weakest node ends up with as much as it can get. In anti-jamming terms, that means keeping throughput roughly equal across all nodes rather than letting a strong link hog everything while a weak one starves. A 2026 IEEE paper by A. Garnaev, titled "Max-Min Resource Allocation With Application to Anti-Jamming," puts this exact idea to work for an operator communicating with several nodes—drones, for instance—flying through zones protected by jammers.
Here's the thing: jamming almost never hits a network evenly. A reactive jammer sits there watching the spectrum, waits until it senses a transmission, and only then jams — so the node that transmits the most aggressively is often the one that gets exposed the most. If an operator just chases total throughput, a handful of nodes can look great on paper while the rest are basically silenced. Max-min fairness turns that objective on its head: take care of the weakest node first, and only then start improving everyone else.
That design choice has real operational consequences for drone communication, where losing a single link can break a formation, a relay chain, or a sensing task. In the Garnaev paper, the authors report a max-min metric that captures anti-jamming performance, tested on basic educational resource data as samples, with results coming in 5.3 percentage points higher and an accuracy of 89%. Those gains may look modest on paper, but in a context where just a few points of link reliability can determine whether a mission succeeds or fails, they carry real weight.
How Does the Max-Min Anti-Jamming Equilibrium Strategy Work?
Garnaev's anti-jamming equilibrium strategy aims to keep throughput equal across every node. Instead of treating jamming as background noise you simply try to overpower, it treats the jammer as a rational opponent and searches for a stable allocation where no node gets left behind. The reactive jammer here works in a sensing-then-jamming mode, which means it needs a bit of time to sense before it strikes — and that delay is exactly the opening the strategy exploits.
In practice, the operator spreads its transmissions across both time and frequency, so no single node ends up as a permanent target, and the jammer has to burn part of its cycle on sensing before it can actually strike. Since the jammer can't sense and jam at the same time, the equilibrium allocation manages to keep every node's throughput roughly balanced, even while the jammer keeps adapting. That's really the whole intuition behind max-min fair resource allocation: fairness here isn't about charity, it's a defensive posture.
One helpful way to think about this strategy is as a two-player game. The operator picks how to distribute resources, the jammer decides where to concentrate its interference, and equilibrium is reached when neither side can do better by changing its move alone. That's exactly the setup in the 2026 paper, which models anti-jamming as a strategy for an operator communicating with several nodes. What stands out is that every node ends up with the same throughput — that equal split isn't some incidental byproduct, it's the defining mark of the max-min objective at work.
What Are the Main Anti-Jamming Resource Allocation Methods?
Between 2024 and 2026, research in this area has branched out into quite a few different methods, and what really separates them is the assumptions they make about the jammer and the objective they're trying to optimize. The table below puts the most common approaches side by side, starting with classical fairness and moving through to learning-based control.
| Method | Core Idea | Typical Objective |
|---|---|---|
| Max-min fairness | Equalize throughput across nodes | Maximize the worst-off node |
| Markov decision process (MDP) | Model jamming as a state transition | Long-term throughput (LTT) |
| Deep reinforcement learning (DRL) | Learn policies from interaction | Weighted rate and sensing power |
| Active inference (AIn) | Minimize surprise in cognitive radio | Robust cognitive-UAV links |
| Linear programming duality | Solve allocation optimally under budget | Total utilized bandwidth |
F. Li's 2024 paper in MDPI Sensors looks at an anti-jamming resource-allocation method designed to maximize the long-term throughput (LTT) of the EH-CIoT network, and it frames the whole thing as a Markov decision process built around an actor policy network and a critic value network. A 2025 paper by Y. Chen, on the other hand—"Anti-Jamming Resource Allocation for Integrated Sensing and Communication"—takes a different angle: it aims to maximize the weighted sum of the communication rate and effective sensing power, all while satisfying the constraints on both the communication and sensing sides.
Other researchers have taken things in different directions. J. Wang's 2025 paper introduces the Multi-Agent Jamming Resource Allocation (MAJRA) algorithm, which tackles the problem of allocating cooperative jamming resources. A year earlier, H. Ding compared an intelligent anti-jamming communication algorithm against DQN and Q-learning, testing how each performed under multi-tone linear sweep jamming. And in a 2026 ICC paper, C. Tsoufis and colleagues looked at how to allocate access point bandwidth when an adversary is actively jamming. Their approach frames the problem as a network interdiction model, which they solve optimally using linear programming duality.
How Is Max-Min Fairness Applied to Wireless Networks Under Jamming?
The whole point of max-min fair algorithms is that they have to max out the smaller allocations first, and only then hand extra capacity to the bigger ones. That ordering isn't a detail—it's the entire idea. It's also why waterfilling, along with sequences of optimizations, has stayed the go-to approach for so long. A 2024 USENIX NSDI paper by P. Namyar, which looks at solving max-min fair resource allocations quickly, frames this as much as a computational problem as a policy one: you need to reach the fair answer fast enough that it actually counts in a live network.
In wireless settings under jamming, the fair allocation has to be recomputed as conditions shift. The 2026 ICC work by C. Tsoufis et al. models the interaction as a special network interdiction problem where the network designer acts first, then an adversary applies a jamming strategy. Notably, the interdictor responds rather than leads, and interdiction is fractional at the node level, which makes the problem tractable through linear programming duality when the adversary's budget constraints are known.
Experiments on synthetic data show improvement in total utilized bandwidth over an adversary-oblivious algorithm. That comparison is the practical takeaway: ignoring the jammer's incentives produces an allocation that looks efficient in isolation but underperforms once the adversary reacts. Max-min fairness, combined with a model of adversarial behavior, keeps the network usable rather than merely optimal on paper.
What Role Does Reinforcement Learning Play in Anti-Jamming Resource Allocation?
Reinforcement learning has become the default toolkit when the jammer's behavior is unknown or changing. A 2025 paper on self-organized anti-jamming reinforcement learning for drone communications uses self-organizing maps and pseudo-random slices to enhance convergence and anti-jamming performance, which addresses a common failure mode: agents that never converge because the environment keeps shifting.
A 2026 paper by H. Yang, "Embodied intelligence-enhanced anti-jamming resource allocation for low-altitude communication networks," designs an embodied intelligence-enhanced low-altitude communication network under malicious jammers, where multiple UAVs act as embodied agents. A 2026 arXiv paper by M. Wu et al. goes further into the application layer, focusing on an anti-jamming collaborative inference system with a malicious jammer in which a DNN model is partitioned into two segments executed by wireless devices and edge servers.
That last work analyzes jamming and DNN partitioning effects on inference accuracy via data regression, and its objective is to maximize system revenue of delay and accuracy (RDA) under inference accuracy and computing resource constraints. It jointly optimizes computation resource allocation, devices' transmit power, and DNN partitioning, then proposes an alternating optimization-based algorithm that decomposes the problem into three subproblems solved via Karush-Kuhn-Tucker conditions, convex optimization, and a quantum genetic algorithm. Simulations show outperformance in RDA.
What About Proactive Allocation and Cognitive Radio Approaches?
Not every anti-jamming system can wait for the jammer to appear. A 2026 paper by H. Yu, "A Proactive Resource Pre-Allocation Framework for Anti-Jamming," addresses anticipatory resource allocation in field-deployed communication networks under dynamic unmanned aerial vehicle jamming. The premise is that pre-positioning resources before an attack begins is cheaper than reacting after throughput has already collapsed.
Cognitive radio offers a complementary angle. A 2022 arXiv paper proposes a novel resource allocation strategy for anti-jamming in cognitive radio using Active Inference (AIn), with a cognitive-UAV employed. Instead of maximizing reward directly, active inference minimizes surprise, which tends to produce behavior that is robust to unexpected jamming patterns rather than tuned to one specific attack.
On the jamming side of the same problem, a 2024 AIP Advances paper by F. Liu states that the task of jamming resource allocation is to formulate a strategy that determines the correspondence between jammers and radars. A 2021 paper by F. Liu notes that cross deployment of heterogeneous radars in different spatial directions makes the whole system have strong anti-jamming capability. Together these show that fairness and allocation logic cut both ways: defenders and attackers are solving structurally similar problems.
What Are the Key Performance Metrics and Results?
Performance in this literature is reported through a small set of recurring metrics, and comparing them requires care because each paper optimizes something slightly different. The table below summarizes the headline metrics and the reported outcomes.
| Paper Focus | Metric | Reported Result |
|---|---|---|
| Max-min anti-jamming (Garnaev) | Max-min performance metric | 5.3 percentage points higher; 89% accuracy |
| EH-CIoT allocation (Li) | Long-term throughput (LTT) | MDP-based maximization |
| ISAC allocation (Chen) | Weighted rate and sensing power | Significant gains from DRL |
| Robust allocation (Tsoufis) | Total utilized bandwidth | Improvement over adversary-oblivious baseline |
| Collaborative inference (Wu) | Revenue of delay and accuracy (RDA) | Outperformance in simulation |
Citation counts give a rough sense of influence: the Namyar NSDI paper has 31 citations, the Chen ISAC paper has 15, the Ding algorithm paper has 8, and the Wang MAJRA paper has 5, while the Garnaev, Yang, and Wu papers each show 2 or fewer at the time of writing. Those numbers shift over time, so treat them as a snapshot rather than a ranking of quality.
The honest reading is that no single method dominates. Max-min fairness is strongest when the priority is protecting every node; MDP and DRL approaches win when the environment is dynamic and the model is unknown; linear programming duality is attractive when the adversary's budget is known. For drone communication under jamming, the most defensible design combines a fairness objective with a learning component that adapts as the jammer changes tactics. I would not deploy a purely static allocation against a reactive jammer, and I would not trust a learned policy without a fairness floor underneath it.
What is max-min resource allocation in anti-jamming?
It is a strategy that allocates limited communication resources so the worst-off node is maximized, maintaining equal throughput for each node. In anti-jamming settings, an operator uses it to design a strategy against jammers, as described in a 2026 IEEE paper by A. Garnaev. The goal is fairness under attack rather than peak performance for the strongest link.
How does a reactive jammer affect max-min allocation?
A reactive jammer works in a sensing-then-jamming mode and requires a time delay. The max-min anti-jamming equilibrium strategy maintains equal throughput for each node despite this behavior, according to the 2026 paper on max-min resource allocation with application to anti-jamming. Because the jammer cannot sense and strike at the same moment, the operator can spread transmissions to avoid a permanent target.
What methods are used for anti-jamming resource allocation in networks?
Approaches include max-min fairness, Markov decision processes for long-term throughput, deep reinforcement learning, active inference in cognitive radio, and linear programming duality for robust allocation against adversarial jamming, as shown in multiple 2024-2026 studies. Each method fits a different assumption about how much the operator knows about the jammer's behavior and budget.
What is the goal of anti-jamming resource allocation in EH-CIoT?
The goal is to maximize the Long-Term Throughput (LTT) of the energy-harvesting cognitive Internet of Things network. The problem is modeled as a Markov decision process, as described in a 2024 MDPI Sensors paper by F. Li. The model uses an actor policy network and a critic value network to learn allocation decisions over time rather than optimizing a single instant.
Frequently Asked Questions
What is max-min resource allocation in anti-jamming?
It is a strategy that allocates limited communication resources so the worst-off node is maximized, maintaining equal throughput for each node. In anti-jamming settings, an operator uses it to design a strategy against jammers, as described in a 2026 IEEE paper by A. Garnaev. The goal is fairness under attack rather than peak performance for the strongest link.
How does a reactive jammer affect max-min allocation?
A reactive jammer works in a sensing-then-jamming mode and requires a time delay. The max-min anti-jamming equilibrium strategy maintains equal throughput for each node despite this behavior, according to the 2026 paper on max-min resource allocation with application to anti-jamming. Because the jammer cannot sense and strike at the same moment, the operator can spread transmissions to avoid a permanent target.
What methods are used for anti-jamming resource allocation in networks?
Approaches include max-min fairness, Markov decision processes for long-term throughput, deep reinforcement learning, active inference in cognitive radio, and linear programming duality for robust allocation against adversarial jamming, as shown in multiple 2024-2026 studies. Each method fits a different assumption about how much the operator knows about the jammer's behavior and budget.
What is the goal of anti-jamming resource allocation in EH-CIoT?
The goal is to maximize the Long-Term Throughput (LTT) of the energy-harvesting cognitive Internet of Things network. The problem is modeled as a Markov decision process, as described in a 2024 MDPI Sensors paper by F. Li. The model uses an actor policy network and a critic value network to learn allocation decisions over time rather than optimizing a single instant.