Automated Synthesis of Analog and Digital Filter Descriptions using Large Language Models

We present a Python-based user-friendly automated framework to synthesize analog and digital filters using Large Language Models (LLMs). For analog filters, the framework enables automated synthesis of SPICE-compatible circuit netlists of active filters (Butterworth and Sallen-Key configurations), based on user-defined English specifications such as order, cut-off frequency, closed-loop gain, and quality factor, and automated synthesis of Cauer passive analog networks from s-domain transfer function specifications. For digital filters, the framework enables automated generation of Verilog code of Finite Impulse Response (FIR) and Infinite Impulse Response (IIR) filters (Butterworth and Chebyshev configuration), based on input specifications such as passband and stopband frequency, ripple and attenuation. We consider a filter description to be accurately generated if its frequency characteristics (e.g., cut-off frequency) are within \(±\) 5% of the target, and all other target specification metrics are satisfied. Our technique achieves a maximum accuracy of \(≈\) 93.33% when evaluated with three different state-of-the-art LLMs, and is useful both as an educational tool and to practitioners.