Machine-Learning Prediction of RF Band-Pass Filter Performance Using Large-Scale Vendor Data

Read full text Summarise with VIRAT
Authors: Mohammed Suhail B Nadaf, Rajesh Kumar, Chinmay Pramod Ardalkar
Cited by: 0
References: 11

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

We present a data-centric workflow for supervised regression of RF band-pass filter (BPF) performance using a vendor-scale corpus assembled from Mini-Circuits product pages. The curated corpus contains ≈ 1.1 × 106structured records (documented f0, BW3dB, nominal insertion loss, stop-band attenuation, family/technology and package metadata). An automated ingestion pipeline performs unit canonicalization, outlier rejection, deterministic + model-based imputation, and frequency-domain harmonization; domain-informed features (fractional bandwidth, Q estimate, spectral moments) are synthesized to expose physically meaningful relationships for learning.We benchmark gradient-boosted tree ensembles (XGBoost, CatBoost, LightGBM) against a multilayer perceptron baseline using stratified k-fold validation (strata: coarse frequency bands / family). Tree ensembles consistently outperform the neural baseline on tabular vendor data and exhibit greater robustness to sparse/noisy labels. Model introspection (SHAP, permutation importance) identifies f0and BW3dBas dominant predictors of insertion loss, with family/technology encodings providing secondary, technology-specific corrections. The resulting surrogates provide millisecond inference for catalog-scale scoring and interactive design screening, and form a practical foundation for future sim–meas transfer learning and Touchstone-level extensions.

Fields of Study & Subjects

Microwave Engineering and Waveguides Machine Learning in Materials Science Data Mining and Machine Learning Applications

Publication details

Journal / source
International Conference on Signal Processing and Electronics Design (ICSPED)
Pages
1–6
Publication date
23 January 2026
Language
EN