Antenna Forge (YAF) is an AI-driven platform for inverse antenna design, automating the exploration, generation, optimization, and verification of novel antenna topologies. It targets engineers and researchers seeking superior antenna designs by integrating real physics solvers with advanced optimization, validated against benchmarks.

How It Works

The platform integrates real physics solvers (NEC2, openEMS) directly into an AI-driven optimization loop. It employs closed-loop inverse design where optimization algorithms drive simulations, ensuring results are based on actual physical computations, not surrogates or analytical fallbacks. This guarantees high fidelity and enables discovery of novel designs, with simulation data available for surrogate model training.

Quick Start & Requirements

• Install: Clone repo, copy .env.example to .env, run docker compose up -d.
• Prerequisites: Docker, Docker Compose. Optional NEC2 (necpp) and openEMS backends must be installed separately.
• Access: Frontend at http://localhost:5173, API health check at http://localhost:8000/health.
• Docs: Key docs: docs/HONEST_STATUS.md, docs/next-steps.md, docs/case_study_yagi.md.

Highlighted Details

• Yagi-Uda Design: Achieved a 9-parameter Yagi-Uda design using differential evolution and NEC2, outperforming canonical Viezbicke designs in gain (+1.60 dB) and front-to-back ratio (+1.21 dB) with identical element counts.
• Honest Simulation: Solvers raise SolverUnavailable errors if missing, preventing fabricated results and ensuring simulation integrity.
• Open-Source Core: Offers a self-hostable MIT-licensed core for wire/planar antennas using classical optimization, serving as a foundation for research and surrogate model training.
• Reproducible Demos: Provides command-line demos and benchmarks, including a 3-panel dipole simulation and Yagi-Uda case study, verifiable with real NEC2 or openEMS runs.

Maintenance & Community

Developed by a single engineer using AI coding assistants. No specific community channels listed. Commercial inquiries should be directed to open a GitHub issue.

Licensing & Compatibility

Core source code is MIT licensed. Optional NEC2 (necpp) and openEMS backends are GPL-licensed. Users integrating GPL components must manage combined-work obligations per NOTICE. MIT generally permits commercial use, but GPL dependencies introduce copyleft restrictions for derivative works.

Limitations & Caveats

Features planned but not yet available in the open-source core: broader full-wave coverage (microstrip arrays, metasurfaces, 3D structures), commercial solver integration (HFSS, CST, FEKO), generative AI geometry design (diffusion/VAE), multi-objective/multi-band optimization, RIS design, in-browser visual platform, cloud compute, and team collaboration tools. Existing generative models are experimental and not yet integrated into a simulation loop.