Particle swarm optimized deep learning for jamming detection and throughput enhancement in cognitive radio networks

Why protecting wireless signals matters

Our phones, sensors, and smart gadgets increasingly share the same crowded airwaves. To make room for everyone, next‑generation networks let some devices temporarily borrow unused radio channels. But this flexibility comes with a risk: a malicious transmitter can deliberately flood a channel with noise, a "jamming" attack that blocks nearby devices from talking at all. This paper introduces a new method, called DeepSwarm, that helps such flexible radios quickly recognize when they are being jammed and hop to safer channels, boosting both security and data throughput.

Smart sharing of invisible highways

Modern wireless systems use a concept known as cognitive radio, where "secondary" devices are allowed to transmit only when a licensed "primary" user is not using a given channel. These radios constantly sense which channels are free and share that information with a central controller. Because many devices and potential attackers can access the same frequencies, the network must decide, slot by slot, which channels each device should use to carry data while still avoiding licensed users. In this setting, a jammer can cause major disruption by targeting popular channels, forcing legitimate devices into repeated collisions, dropped packets, and wasted battery power.

A cat‑and‑mouse game over the air

The authors describe the struggle between normal users and a jammer as a strategic game. Each side chooses channels to either send data or cause interference, trying to maximize its own benefit. Normal users want high, reliable throughput at low power cost, while the jammer wants to disrupt as many transmissions as possible with limited energy. The paper develops a mathematical model of this interaction that accounts for how many channels are free, how many users are active, how likely they are to collide with each other, and how much extra interference a jammer adds. This model quantifies, in a single utility measure, how good a particular channel choice is for either side.

Teaching radios to recognize attacks

Instead of solving this complicated game purely with equations, the authors turn to data‑driven learning. They design a compact deep neural network that looks at simple measurements already available in the network: average throughput on a channel, how much that throughput fluctuates, received signal quality, measured interference power, and whether the channel is sensed as busy or idle. From these features, the network learns to tell apart normal users from jammers. To get the most out of a small, practical dataset, the team uses particle swarm optimization, a population‑based search method inspired by flocking behavior, to automatically tune how many layers and neurons the network has, as well as its learning rate and regularization settings. This tuned model, DeepSwarm, is trained offline but then runs quickly in real time.

Choosing better channels on the fly

Once DeepSwarm can accurately flag which transmitters behave like jammers, the network uses that information to clean up its view of the spectrum. Reports from suspected jammers are ignored; only trusted users influence the decision about which channels are really free. With a clearer picture, the system coordinates which idle channels secondary users should hop to in each time slot, spreading them out to avoid both each other and the jammer’s favorite targets. Simulations show that after DeepSwarm is deployed, users distribute themselves more evenly across channels, congestion drops, and they automatically steer away from heavily jammed frequencies, even as jammer tactics change.

Gains in reliability and speed

In extensive tests, DeepSwarm detects jammers with about 98% accuracy, precision, and recall, outperforming common machine‑learning baselines such as support vector machines, linear models, and ensemble stacking methods. More importantly for everyday performance, this improved awareness translates into much higher usable data rates. Compared with a static channel‑selection strategy that ignores jamming, the DeepSwarm‑guided hopping scheme can raise normalized throughput by up to 32% under a range of attack intensities. When compared with a previous game‑theoretic benchmark that relies only on trial‑and‑error learning, the new approach roughly boosts effective throughput by 70–80% while cutting the chance of being jammed in half.

What this means for future wireless devices

For a non‑specialist, the main takeaway is that the authors have built a kind of guardian for flexible radios: a lightweight learning system that spots foul play on the airwaves and helps devices quickly move to cleaner channels. By combining strategic modeling of attackers with a tuned neural network, DeepSwarm offers a scalable way to keep data flowing even in hostile environments. This could be especially valuable for dense Internet‑of‑Things deployments and machine‑to‑machine links, where many low‑power devices must share spectrum safely and efficiently without constant human oversight.

Citation: Imran, M., Ibrahim, K., Zhiwen, P. et al. Particle swarm optimized deep learning for jamming detection and throughput enhancement in cognitive radio networks. Sci Rep 16, 8715 (2026). https://doi.org/10.1038/s41598-026-41642-9

Keywords: cognitive radio, wireless jamming, deep learning, frequency hopping, IoT security