This post is also available in: עברית (Hebrew)
China-based company Chengdu Aircraft Design Institute has reportedly created a sophisticated large language model (LLM) designed specifically for electronic warfare (EW) drones. This ChatGPT-like model is capable of rapidly disrupting enemy radar and radio communications.
According to a study published on October 24 in the peer-reviewed Journal of Detection & Control in China, the LLM’s decision-making capabilities exceed those of traditional artificial intelligence methods and even experienced human EW experts. This collaboration involved the Aviation Industry Corporation of China and Northwestern Polytechnical University in Xian.
According to Interesting Engineering, the new LLM enhances the operational efficiency of drones engaged in electronic warfare by accelerating their ability to conduct EW maneuvers. This includes disrupting enemy radar systems using targeted electromagnetic signals. The dynamic nature of electronic warfare means that defenders frequently change their signal patterns to evade attacks, prompting attackers to adapt in real time.
What sets this LLM apart is its ability to interpret sensor data and react in milliseconds—previously a challenge for such models. The development team trained the LLM using an extensive collection of literature on electronic warfare, including radar technology and combat records, primarily sourced in Chinese.
To further expedite decision-making, the LLM is paired with a raw data processor that converts real-time data into actionable insights. The LLM then generates instructions for EW jamming equipment based on its analysis. According to SCMP, Initial test results indicate that this system allows for rapid adjustments to attack strategies—up to ten times per second—far surpassing traditional methods.
Additionally, the LLM has proven more effective at creating false targets on enemy radar, a crucial tactic that enhances the effectiveness of electronic warfare beyond simple signal suppression.
Despite the promising results, SCMP reports that experts caution that challenges remain, such as hardware limitations, model size, and security concerns. Therefore, time will only tell the true results of this research.