Ideal components cannot predict the losses and resonances of a fabricated RF filter. Part 3 evaluates frequency-dependent Murata 0402 models, selects practical component values, and transfers them into the CST 3D EM model to analyze S₁₁, EM fields, and power loss as a basis for EMI optimization.
An ideal RF filter changes once PCB geometry is included. Part 2 imports the Altium layout into CST, uses EM co-simulation to expose trace, pad, and coupling parasitics, and re-optimizes the component values to restore the target S₁₁ response.
How we package a full-stack web app: React frontend, Node backend, PostgreSQL included, into self-contained Windows MSI and Linux .deb installers that run fully offline, survive upgrades without losing data, and build automatically in CI.
An ESP32-S3 RF filter that passes simulation can still fail on the fabricated board. Part 1 of this series builds the foundation: a CLC PI filter designed in CST Studio Suite, optimized with CMA-ES across the 2.4 GHz Wi-Fi channels, and carried into a controlled-impedance Altium layout.
How we taught a neural network to pinpoint people indoors from Bluetooth beacon signals. GANs and RNNs failed; a simple regression network with a Euclidean-distance loss got us to ~1.5 m accuracy — proof that feature engineering beats model sophistication.
Our journey migrating a serverless IoT platform from AWS to on-premise, adapting services with Node.js, Docker, and custom JWT authentication. We replaced cloud monitoring with Prometheus and logrotate, gaining full control and offline capability.