1ove9/antenna-forge

Antenna Forge, also called YAF, is a platform that uses AI to design radio antennas automatically. Instead of a person manually working out antenna shapes and dimensions, the system explores, generates, and tests antenna designs on its own, then verifies the results using real physics simulation engines rather than guesses or shortcuts. The core idea is that every design gets checked against actual electromagnetic simulation tools, specifically NEC2 and openEMS, both of which are established methods for calculating how an antenna behaves. If these simulation tools are not installed, the project intentionally raises an error instead of quietly making up plausible looking numbers, so the results you see are always real measurements. One example included in the project uses an optimization method called differential evolution to redesign a classic five element antenna shape, running the real simulator on every single attempt, thousands of times, and it ends up matching or beating a well known published antenna design in both signal strength and directionality. The project is built with several connected pieces. A web interface built with React and Three.js lets you view and edit antenna designs in 3D. Behind that sits an API built with FastAPI, which passes work to a task processing system so that simulations and optimizations can run in the background. There is a geometry engine for building antenna shapes, a physics module supporting more advanced antenna concepts, and a solver layer that can call NEC2, openEMS, and several other established simulation programs. An AI engine offers multiple approaches for generating and optimizing designs, including generative models and Bayesian optimization. To try it, you clone the repository, copy the example environment file, and start everything using Docker Compose, which brings up the database, task queue, and API together. Once running, you can create a new antenna design and simulate it through the API, or run any of the included demo scripts to reproduce example results directly. The project is written primarily in Python.