ML on Microcontrollers & Real-Time Anomaly Detection

Deploy and debug custom DSP and classical machine learning algorithms directly on bare-metal and RTOS-based microcontrollers.

4 Weeks / 32 Hours Hands-on Labs

Certificate of TinyML Specialty Awarded

Training Covering Area

Go beyond deep learning. Learn how to write classical ML algorithms (K-Means, SVM, Random Forest, decision trees) and signal processing code in standard C++ for immediate execution on microcontrollers without heavy dependencies.

Module 1: DSP & ADC Sampling Foundations

ADC configuration, sampling rates, Nyquist frequency limits, and hardware buffers
Digital filter design: Implementing FIR and IIR filters in standard C++ code
Fast Fourier Transform (FFT) implementations and spectral frequency analysis
Windowing algorithms (Hamming, Hanning) and signal feature extraction

Module 2: Classical Machine Learning Conversion

Feature selection for microcontrollers: Minimizing floating-point math overhead
Training classical ML classifiers (Decision Trees, Naive Bayes, SVMs) in Scikit-Learn
Using Python micro-compiler scripts to auto-generate plain C structures from trained models
Deploying and evaluating K-Means clustering for real-time unsupervised anomaly detection

Module 3: Interrupts & Execution Optimization

Triggering classification cycles using DMA double buffering
Pinning calculations to specific processor cores, configuring compiler optimization flags (-O3)
Profiling clock cycles using hardware debug pins, timers, and oscilloscope captures
Managing task scheduling and resource allocation within RTOS environments

Module 4: Real-World Case Studies

Case Study 1: Real-time electrocardiogram (ECG) heartbeat anomaly detection
Case Study 2: Industrial rotating motor vibration diagnostics and predictive warning triggers
Case Study 3: Touch capacitive sensor swipe gesture detection and pattern classification
Model updating: Over-The-Air parameter tuning without firmware re-compilation