EE496

CPRPal — Real-Time CPR Feedback and Training Device

Table of Contents

• Project Overview
• Repository Structure
• System Architecture
• Hardware Components
• Wiring and Pin Assignments
• PCB Design
• Firmware
• Operating Modes
• User Operation
• Signal Processing
• Power System
• Cloud and Web Application
• Setup and Installation
• Library Dependencies

Project Overview

CPRPal is a portable, puck-shaped CPR coaching device designed to help untrained bystanders and CPR trainees perform chest compressions with correct force, rhythm, and body positioning during a cardiac emergency or training session.

Sudden cardiac arrest is responsible for approximately 93% of out-of-hospital cardiac arrest deaths. Survival probability drops roughly 10% per minute without CPR or AED intervention, yet only about 3% of the U.S. population is CPR certified annually. CPRPal addresses this gap by providing real-time, easy-to-interpret feedback that requires no prior training to use.

The device is placed between the rescuer's hands and the patient's sternum during chest compressions. It measures compression force directly using a load cell, tracks body posture using a 9-DOF IMU, monitors for heartbeat presence using a PPG optical sensor, and delivers feedback through a color-coded NeoPixel LED ring, a 2.8" TFT LCD display, and a 100 BPM audio metronome buzzer.

Emergency Mode operates fully standalone with no Wi-Fi required, providing immediate real-time feedback the moment the button is pressed.

Training Mode connects to Wi-Fi, runs a 2-minute guided session, and uploads averaged session metrics to an Azure-hosted cloud dashboard for review.

The final prototype measures 4.5" x 4.5" x 2.25" and runs for over 15 hours on four AA batteries at a normal operating draw of approximately 155 mA.

CPRPal Demo: https://drive.google.com/file/d/1d1auYhc-_jhG4WK9BlywpUfyi4_rIxLO/view?usp=sharing (Note video isn't perfectly sync with operation)

Rendering	Final Build

Repository Structure

CPRPal/
├── 3DModel/
│   └── Mechanical enclosure and CAD files
│       - Top shell with lattice structure and LCD/button/LED cutouts
│       - Bottom shell with battery pocket, support beams, and hard stops
│       - Full assembly files for PLA 3D printing
│
├── ArduinoCode/
│   └── CPRPal.ino — Main ESP32 firmware (FreeRTOS-based)
│       - Sensor integration (HX711, BNO055, PPG)
│       - Real-time signal processing and CPM detection
│       - LED and display feedback systems
│       - Emergency and Training mode logic
│       - Power management and idle sleep
│       - HTTP cloud upload for training sessions
│
├── ArduinoWebsite/
│   └── Azure Static Web App source code
│       - Frontend dashboard (HTML/CSS/JS)
│       - Azure Function API endpoint
│       - Receives HTTP POST from ESP32
│       - Displays training session summary
│
├── PartsList/

System Architecture

CPRPal uses a dual-core FreeRTOS architecture on the Arduino Nano ESP32-S3.

Core 0 — Sensor Acquisition and Processing

All time-critical sensor tasks run on Core 0 to prevent display rendering overhead from introducing jitter into sensor sampling.

Core 1 — User Interface and Control

The Arduino loop() function runs on Core 1 and handles all output rendering:

• TFT display refresh at 10 Hz
• NeoPixel force bar graph refresh at 40 Hz
• Mode selection and session control logic
• Training session timer and metric accumulation
• Serial calibration commands

Shared State and Synchronization

All shared sensor variables use the g_ prefix, are declared volatile, and are protected by the FreeRTOS mutex xDataMutex. A second mutex xWireMutex serializes I2C bus access between taskIMU and any other bus user. This two-mutex architecture prevents data races while minimizing lock hold time.

Hardware Components

Wiring and Pin Assignments

ESP32 GPIO Map

I2C Bus

• BNO055 IMU address 0x28
• 10 kOhm pull-up resistors on SDA and SCL
• Bus speed 400 kHz (fast mode)
• External 32.768 kHz crystal for improved BNO055 sensor fusion accuracy

SPI Bus

• ILI9341 TFT LCD: DC on GPIO9, CS on GPIO10
• microSD: CS on GPIO4
• Both share the hardware SPI bus

Load Cell Wiring (Half-Bridge)

Two 1 kOhm resistors on the HX711 board complete the Wheatstone bridge by providing the two fixed-impedance legs of the half-bridge configuration.

PCB Design

CPRPal uses a custom panelized PCB fabricated by JLCPCB, separated into six press-fit sub-boards using mouse-bite breakaway tabs. Each sub-board can be removed and replaced independently without additional soldering.

Sub-Board Summary

Component Footprints

3.3V and 5V BuckBoost Boards

Power Hub Board

HX711 Board

BNO055 IMU Board

ESP32 Board

Bit Shift Board

Power Architecture

Two independent TPS63070 buck-boost converters are used instead of a single converter with an LDO post-regulator. This improves efficiency across the full 4xAA discharge range (approximately 6V fully charged to 4.8V near end of life), since the buck-boost topology maintains tight output regulation without the linear losses an LDO would introduce at shrinking headroom.

• 3.3V rail powers: ESP32-S3, BNO055 IMU, PPG sensor, piezo buzzer
• 5V rail powers: ILI9341 LCD, NeoPixel strips (18 px total), HX711 amplifier

Firmware

The firmware is a single Arduino sketch located in ArduinoCode/CPRPal.ino. It is written in C++ using the Arduino framework for ESP32 and relies on FreeRTOS task management, mutex-protected shared state, and the ESP32's LEDC hardware PWM peripheral for the buzzer metronome.

Key Constants

// Force feedback thresholds (lbs)
const float FORCE_THRESHOLD_LB      = 30.0f;   // Rising edge detection
const float FORCE_LOW_OK            = 75.0f;   // Below = TOO LIGHT (yellow)
const float FORCE_HIGH_OK           = 110.0f;  // Above = TOO STRONG (red)

// CPM thresholds
const float CPM_LOW_OK              = 100.0f;  // Below = TOO SLOW
const float CPM_HIGH_OK             = 120.0f;  // Above = TOO FAST
const unsigned long CPM_DEBOUNCE_MS = 180UL;   // Anti-bounce window
const int CPM_INTERVAL_COUNT        = 6;       // Rolling average window size

// Tilt feedback
const float TILT_WARN_DEG           = 8.0f;    // Above = ADJUST

// Timing
const unsigned long IDLE_TIMEOUT_MS     = 20000UL;  // 20s inactivity -> idle
const unsigned long TRAINING_SESSION_MS = 120000UL; // 2-minute session
const unsigned long TRAINING_SAMPLE_MS  = 3000UL;   // Sample every 3s
const unsigned long TRAINING_HOLD_MS    = 3000UL;   // Button hold -> Training

// Buzzer metronome
#define BUZZER_FREQ_HZ    2000  // 2 kHz tone
#define BUZZER_BEEP_MS    30    // On-time per beat in ms
#define BUZZER_PERIOD_MS  600   // Full period = 100 BPM

Live Serial Calibration Commands

The calibration factor can be adjusted without reflashing firmware using the Serial Monitor at 115200 baud:

EMA Filter

Raw load cell readings are smoothed before any threshold detection:

w_filtered[n] = 0.75 * w_filtered[n-1] + 0.25 * w_raw[n]

Smoothing factor alpha = 0.25 was chosen empirically to balance noise rejection against compression onset detection latency.

CPM Detection Algorithm

1. EMA-filtered force crosses above FORCE_THRESHOLD_LB (30 lb) — rising edge detected
2. 180ms debounce window prevents double-counting from signal ringing near threshold
3. Inter-edge interval converted to instantaneous CPM: CPM = 60000 / delta_t_ms
4. Stored in 6-entry circular buffer; rolling CPM = arithmetic mean of buffer entries
5. If no compression detected for 2000ms, CPM resets to zero and buffer clears

IMU Baseline Calibration

On startup, 100 BNO055 orientation samples are captured (after 50 warm-up discards) and averaged as the resting baseline. All subsequent readings are deltas from this baseline so tilt feedback is relative to device placement on the patient, not absolute gravity direction.

• Tilt magnitude = sqrt(delta_roll² + delta_pitch²)
• Above 5 degrees triggers ADJUST on LCD
• Above 10 degrees triggers LEAN on LCD

PPG Detection Algorithm

1. ADC sampled at ~50 Hz, 12-bit resolution, 3.6V full-scale attenuation
2. 10-sample moving average applied for noise reduction
3. Adaptive baseline tracks slow DC component of signal
4. Adaptive peak envelope tracks signal upper bound
5. Beat detected on rising edge exceeding 55% of amplitude above baseline
6. Flag resets when signal falls below 35% of amplitude (hysteresis)
7. BPM = average of 4 most recent valid inter-beat intervals
8. BPM zeroed after 3.5s without a detected beat
9. Finger contact confirmed when ADC reading exceeds FINGER_ON threshold (1700)

Operating Modes

Boot Sequence

On power-on the device performs the following sequence automatically:

1. Serial initialization and reset reason diagnostic printout to serial monitor
2. I2C bus initialization at 400 kHz on pins A4 and A5
3. BNO055 initialization and external crystal enable
4. SPI bus and TFT display initialization
5. Boot animation — 8 BMP frames loaded sequentially from microSD, 1.25s per frame
6. NeoPixel strip initialization at 40/255 brightness
7. HX711 initialization, power-up, and tare to zero
8. FreeRTOS sensor tasks spawned on Core 0
9. Mode selection screen displayed
10. User selects mode via button
11. IMU orientation baseline captured in current resting position

Emergency Mode

Activated by a short press of the start button.

• Wi-Fi fully disabled — no radio activity during emergency
• 100 BPM metronome starts immediately
• Live CPM, force, tilt, and PPG status shown on TFT display
• NeoPixel force bar updates at 40 Hz
• After 20s of inactivity (no load cell change exceeding 10 lb), all sensor tasks are suspended in RAM and device enters software idle mode at approximately 95 mA
• To wake: hold button for 2 seconds; all peripherals reinitialize and device returns to mode selection

Training Mode

Activated by holding the button for 3 seconds.

• Wi-Fi connection attempted with 10s timeout
• If no network found: graceful offline fallback, session runs locally, no upload at end
• 2-minute timed session with countdown timer displayed on LCD bottom-right
• Sensor metrics sampled every 3s into running sum accumulators
• At session end: averages computed, JSON serialized manually, HTTP POST sent to Azure endpoint
• Session summary displayed on LCD regardless of upload success
• Wi-Fi fully shut down after upload; device returns to mode selection

User Operation

Emergency Mode Step-by-Step

1. Remove CPRPal from first aid kit and slide power switch on. Keep device still on a flat surface during the ~5 second boot and IMU calibration sequence.
2. Call 911 immediately or direct a bystander to call. Do not delay emergency services.
3. Place CPRPal face-up on the center of the patient's chest, between the nipple line, with the LCD visible to you.
4. Short-press the start button once. Emergency Mode begins and the metronome starts at 100 BPM.
5. Place both hands on top of the device and compress in time with the beep and amber LED ring.
6. Watch the side LED bar: yellow = push harder, green = good depth, red = ease up.
7. Watch LCD TILT row: ADJUST or LEAN means recheck your body position and ensure force is going straight down.
8. Watch LCD PPG row: Heartbeat detect means consider pausing to reassess for pulse.
9. Continue compressions until EMS arrives or AED is ready.

Training Mode Step-by-Step

1. Power on and hold button for 3 seconds until "Training selected" appears on LCD.
2. Device connects to Wi-Fi and displays IP address on successful connection.
3. Scan QR code on device or navigate to the dashboard URL on a phone or computer.
4. Begin compressions following the same LED and LCD feedback as Emergency Mode.
5. 2-minute countdown timer shown at bottom right of LCD throughout session.
6. At session end, summary shown on LCD and data uploaded to dashboard if Wi-Fi available.

Feedback Reference

Signal Processing

Force Classification

CPM Classification

Tilt Classification

Power System

Rail Assignments

Power Budget

At 155 mA normal draw with a 2700 mAh 4xAA pack, theoretical runtime is approximately 17 hours. Measured battery life exceeded 10 hours under mixed-use testing.

Cloud and Web Application

Dashboard URL

https://gentle-stone-0c20e761e.6.azurestaticapps.net/

The dashboard is a static web application hosted on Microsoft Azure Static Web Apps using the free tier only. No charges are incurred.

Data Flow

ESP32 collects sensor data every 3s during Training Mode session
  -> Averages computed at 2-minute session end
  -> JSON payload manually serialized (no external library)
  -> HTTP POST to Azure Function API endpoint
  -> Azure processes and stores payload
  -> Dashboard displays session summary

JSON Payload Format

{
  "deviceId": "nano-esp32-test",
  "timestamp": "123456789",
  "mode": "TRAINING",
  "sessionType": "summary",
  "ppg": 1800,
  "imuY": 2.5,
  "imuZ": -1.2,
  "compressionRate": 110,
  "compressionDepth": 85.3,
  "recoil": 1,
  "status": "Good compressions",
  "sampleCount": 40,
  "sessionDurationMs": 120000
}

Status String Logic

Offline Fallback

If Wi-Fi is unavailable when Training Mode is entered:

1. Device displays offline warning on LCD
2. Full 2-minute session runs locally with all normal feedback active
3. Session summary displayed on LCD at session end
4. HTTP POST upload skipped

Setup and Installation

Requirements

• Arduino IDE 2.x
• Arduino ESP32 board package installed via Boards Manager
  • Search for esp32 by Espressif Systems
  • Select board: Arduino Nano ESP32

Step 1 — Install Required Libraries

Open Arduino IDE and go to Sketch -> Include Library -> Manage Libraries. Search for and install each of the following:

WiFi and HTTPClient are included with the ESP32 Arduino core and do not require separate installation.

Step 2 — Prepare the microSD Card

The boot animation requires 8 BMP image files placed in the root of a FAT32-formatted microSD card. Name the files exactly as follows:

/Step0.bmp
/Step1.bmp
/Step2.bmp
/Step3.bmp
/Step4.bmp
/Step5.bmp
/Step6.bmp
/Step7.bmp

Each image should be a 24-bit BMP at 240x320 pixels to match the ILI9341 display resolution. If the SD card fails to mount, the firmware will halt and display an error on the TFT screen.

Step 3 — Configure Wi-Fi Credentials

Open ArduinoCode/CPRPal.ino and locate the following lines near the top of the file:

const char* WIFI_SSID     = "YOUR_SSID_HERE";
const char* WIFI_PASSWORD = "YOUR_PASSWORD_HERE";

Replace with your Wi-Fi network credentials. Wi-Fi is only used during Training Mode. Emergency Mode runs fully offline regardless of these settings.

Note: The ESP32 posts directly to the Azure cloud endpoint over any internet-connected Wi-Fi network. The viewing device does not need to be on the same network as the ESP32.

Step 4 — Select Board and Port

In Arduino IDE:

1. Go to Tools -> Board -> esp32 -> Arduino Nano ESP32
2. Go to Tools -> Port -> select the correct COM port for your device
3. If the board is not recognized, hold the BOOT button on the ESP32 while connecting USB

Step 5 — Flash Firmware

Click Upload in Arduino IDE. Open the Serial Monitor at 115200 baud to view boot diagnostics, reset reason history, sensor readings, and calibration output.

Step 6 — Calibrate the Load Cell

1. Power on the device and let it boot fully
2. Open Serial Monitor at 115200 baud
3. Place a known weight on the device
4. Use + and - keys in Serial Monitor to adjust calibration factor until the displayed value matches the known weight
5. Note the final calibration factor and update calibration_factor in the sketch to save it permanently

Library Dependencies

CPRPal is a prototype device developed for educational purposes. It is not a certified medical device. Always call emergency services first in a cardiac emergency.