Transforming Satellite Communications Design with Generative AI

Effective application of generative AI solutions to engineering problems needs domain-specific insight. This project required the integration of expertise from both ADSP as AI experts and satellite communications expert partners, working collaboratively to ensure rapid iteration and clear communication throughout.

Additionally, satellite communications present particularly high demands for precision -solutions must be both innovative and rigorous, as only fully accurate, verifiable results are able to be implemented. We took pride in not just pushing the boundaries of what generative AI can achieve, but in reliably delivering robust, real-world outcomes.

💠 Novel Choke-Ring Antenna Design Optimisation: Neural networks were trained on programmatically generated datasets, encoding both the design and key performance metrics. These neural networks were used to generate novel results that would improve performance.

💠 Text-to-CAD Web App: A web application powered by large language models (LLMs) and OpenSCAD, enabling the rapid conversion of text-based specifications into accurate, parametric 3D models. The tool leverages Retrieval Augmented Generation to further enhance design sophistication.

💠 Fine-Tuned Stable Diffusion Model: Trained on a curated set of approximately 80 satellite images, this model generates and adapts satellite imagery, demonstrating strong learning capabilities and practical adaptability.