You
I am doing AI for signal processing, especially AFCI, is there any Highly mentioned or popular introduction book.
I foucus on AI solution side, so i wanna control depth in dsp but just helpful engoih to help me undertsand data therefore develop algorhtn for it.
Then what sbout about the edge deployment and robustness to different loads. I learned AI in unveristy but have few industry dsp experience. My current EK-RA2A2
Evaluation Kit for RA2A2 MCU Group
Claude
Searched the web
Books (right depth for an AI person)
- Steven W. Smith, The Scientist and Engineer’s Guide to DSP is free online and the most often recommended “practical, light math” intro. Read the chapters on sampling, FFT, filters, and windowing; skip the rest.
- Richard Lyons, Understanding Digital Signal Processing is the classic industry favorite. It’s intuitive and engineer-focused, and good as a reference when something confuses you.
- Allen Downey, Think DSP is free and Python-based. It’s the fastest bridge from ML habits to spectrograms and filters.
- Warden & Situnayake, TinyML and Situnayake & Plunkett, AI at the Edge cover deployment workflow, quantization, and how to evaluate on-device models.
There’s no popular AFCI textbook. Your real “book” is the standards: UL 1699 (AC), UL 1699B (DC/PV), and IEC 62606 (AFDD). Their test procedures for masking loads and unwanted tripping effectively define your dataset and evaluation plan.
Check your hardware first
The RA2A2 is a metering chip. Its 24-bit sigma-delta ADC samples at about 7.8–8.3 kHz, and it has 48 KB of SRAM on a 48 MHz Cortex-M23 with no FPU and no DSP instructions. Much of the useful arc signature is broadband, high-frequency noise that is well above 4 kHz. That means you’ll mostly see low-frequency cues like “shoulders” near zero crossings and cycle-to-cycle randomness. That works for a learning prototype, but it limits accuracy against difficult loads. Commercial references typically pair an analog front end with a bigger MCU. Renesas’s own AFCI approach, for example, uses GreenPAK to filter and shape the high-frequency arc signature before the MCU. A partner design built with Reality AI runs on an RA6M4 and claims detection in under 4 ms using less than 100 kB of Flash/RAM. Renesas RA2A2 Arm Cortex-M23 microcontroller offers high-resolution 24-bit ADC, up to 512KB dual-bank flash +3
See 2026-10-05 DSP resources for AI-based AFCI development - Load robustness and deployment.
See 2026-10-05 DSP resources for AI-based AFCI development - Smart meter chips.
See 2026-10-05 DSP resources for AI-based AFCI development - Samples per mains cycle.
See 2026-10-05 DSP resources for AI-based AFCI development - Period and frequency.