Real-Time Arrhythmia Detection at the Edge: Deploying TinyML on ESP32 for Raw ECG Analysis

A developer built a real-time arrhythmia detector using TinyML on an ESP32 microcontroller, processing raw ECG data locally with TensorFlow Lite for Microcontrollers. The system uses a quantized 1D-CNN model trained on the MIT-BIH Arrhythmia Database to classify heart rhythms, enabling low-latency, privacy-preserving health monitoring on wearable devices.

In the world of wearable health technology, the holy grail has always been moving intelligence from the cloud to the edge. Waiting for a cloud server to analyze your heart rhythm is not just a latency issue—it's a privacy and battery life concern. Today, we are diving deep into TinyML , Edge AI , and ECG signal processing to build a real-time abnormality detector. By leveraging TensorFlow Lite for Microcontrollers and the versatile ESP32 , we can process raw electrocardiogram ECG data locally. This approach ensures low-latency detection of arrhythmias while keeping sensitive medical data on-device. If you've been looking to bridge the gap between high-level deep learning and low-level embedded systems, you're in the right place The pipeline involves capturing a high-frequency analog signal, cleaning it, and feeding it into a quantized Convolutional Neural Network CNN . Here is how the data flows through our ESP32: php graph TD A Raw ECG Signal/Sensor -- |ADC Sampling| B Preprocessing: Bandpass Filter B -- C{Buffer Management} C -- |Windowed Segment| D TFLite Micro Inference Engine D -- E{CNN Model Classification} E -- |Normal| F Log: Sinus Rhythm E -- |Abnormal| G Trigger Alert: Arrhythmia G -- |Bluetooth/Wi-Fi| H Mobile Dashboard To follow this advanced guide, you'll need: Before we touch the C++ code, we need a model. Typically, we use the MIT-BIH Arrhythmia Database to train a 1D-CNN. The crucial step is Post-Training Quantization . Since the ESP32 doesn't have a dedicated NPU, we convert our 32-bit float model into an 8-bit integer INT8 model. This reduces the size by 4x and speeds up inference significantly without a massive drop in accuracy. We need to load the model as a byte array and set up the TFLite interpreter. Here is a simplified implementation of the inference loop. include <TensorFlowLite.h include "model data.h" // Your exported INT8 model include "tensorflow/lite/micro/all ops resolver.h" include "tensorflow/lite/micro/micro interpreter.h" // Constants for the ECG window const int kTensorArenaSize = 30 1024; // 30KB Arena uint8 t tensor arena kTensorArenaSize ; // TFLite globals tflite::MicroInterpreter interpreter = nullptr; TfLiteTensor input = nullptr; void setup { Serial.begin 115200 ; // 1. Load the model static tflite::MicroMutableOpResolver<5 resolver; resolver.AddConv2D ; resolver.AddMaxPool2D ; resolver.AddFullyConnected ; resolver.AddSoftmax ; resolver.AddReshape ; static tflite::MicroInterpreter static interpreter tflite::GetModel g model data , resolver, tensor arena, kTensorArenaSize ; interpreter = &static interpreter; interpreter- AllocateTensors ; input = interpreter- input 0 ; } void loop { // 2. Simulate/Read ECG Data Sampled at 125Hz or 250Hz float sample = analogRead 34 ; // ... Preprocessing Logic: Normalization & Filtering ... // 3. Fill the Input Tensor for int i = 0; i < input- dims- data 1 ; i++ { input- data.f i = processed samples i ; } // 4. Run Inference TfLiteStatus invoke status = interpreter- Invoke ; if invoke status = kTfLiteOk { Serial.println "Inference failed " ; return; } // 5. Analyze Results float normal score = interpreter- output 0 - data.f 0 ; float abnormal score = interpreter- output 0 - data.f 1 ; if abnormal score 0.8 { Serial.println "⚠️ Warning: Abnormal Heart Rhythm Detected " ; } } When moving from a prototype to a production-grade wearable, you'll encounter challenges like signal noise from muscle movement EMG and power consumption. For those looking to implement robust noise-cancellation algorithms or more efficient memory management in TinyML deployments, I highly recommend checking out more production-ready examples. Pro-Tip: For a deep dive into advanced signal processing patterns and optimizing TensorFlow Lite for mission-critical edge applications, visit the WellAlly Tech Blog . They provide excellent resources on scaling these localized AI models to enterprise-grade health solutions. Raw ECG data is messy. You'll need a digital Bandpass Filter usually 0.5Hz to 40Hz to remove baseline wander and high-frequency noise. In C++, this can be implemented as a simple IIR filter to keep the computational overhead low on the ESP32. // Example: Simple Low-pass filter component float lowPass float input, float prevOutput, float alpha { return prevOutput + alpha input - prevOutput ; } Deploying a CNN on an ESP32 for real-time ECG analysis isn't just a "cool project"—it's a glimpse into the future of decentralized healthcare. By processing data locally, we respect user privacy and reduce the load on our infrastructure. What's next for your TinyML journey? Happy hacking 🚀💻 If you enjoyed this tutorial, don't forget to ❤️ and bookmark it For more advanced Edge AI content, keep an eye on my profile or visit the official WellAlly Blog.