Edge-AI and RF bench
A Raspberry Pi 5 with a Hailo-8 accelerator, software-defined radio, and a pile of sensors. A bench for learning edge inference and RF.
A corner of the lab is given over to small hardware: somewhere to learn edge inference, software-defined radio, and sensor plumbing without a rack in sight.
The core
A Raspberry Pi 5 paired with a Hailo-8 AI accelerator (26 TOPS) on NVMe. The intent is to run signal classification on it at the edge rather than shipping captures off-box; as the bring-up log below is honest about, that path isn’t working yet.
Around it
- Software-defined radio: an ADALM-PLUTO and a Pluto-class LibreSDR (Zynq-7020 + AD9361), aimed at turning the Pi and the Hailo into an RF signal-analysis box. Getting the radio enumerating over IIO was the easy part; the bring-up log covers where the open-source SDR toolchain and the Hailo stack fought back.
- Microcontrollers: ESP32 and Raspberry Pi Pico W nodes that publish over MQTT.
- Sensing: GPS and I²S audio.
Why
It’s the maker version of the security habit: take something you don’t understand (a radio, a tensor accelerator, a noisy analog sensor) and keep poking until you do. Most of it talks MQTT back to the home lab, which is where the data eventually lands.