Curriculum

Curriculum Outline¶

Analog Data | April 18–19, 2026 | Instructor: Rajath Kumar K S | edgeai.analogdata.io

Day 1 — ESP32-S3, IDF Foundations & Connectivity¶

Saturday, April 18, 2026

Module 0 — Toolchain Setup & Docker Dev Environment¶

What you'll do: Pull the pre-built Docker image, flash your first IDF project, and verify your entire toolchain works — all within the first hour.

Experiments:

• Flash the Hello World project to your XIAO ESP32-S3
• Explore the IDF project structure and build system

Hardware: XIAO ESP32-S3, USB-C cable

Module 1 — GPIO: Digital Input & Output¶

What you'll do: Control LEDs and read button presses using the ESP-IDF GPIO driver.

Experiments:

• LED Blink — Blink an LED at a configurable rate
• Button-Controlled LED — Toggle LED on button press with software debouncing
• RGB LED Color Sequencer — Cycle through 7 colors using 3 GPIO pins

Hardware: LED, RGB LED, 2× Push Buttons, resistors

Module 2 — PWM: LED Dimming with Potentiometer¶

What you'll do: Use the LEDC peripheral to generate PWM signals and map potentiometer position to LED brightness in real time.

Experiments:

• Fixed PWM — Set LED to 50% brightness using 13-bit resolution
• Pot → Brightness — Turn the knob, watch the LED dim and brighten live
• Buzzer Tone — Play a 440 Hz tone using PWM on the passive buzzer

Hardware: LED, Potentiometer, Passive Buzzer

Module 3 — ADC & Capacitive Touch¶

What you'll do: Read analog voltages accurately and use the ESP32-S3's built-in capacitive touch sensor without any external IC.

Experiments:

• ADC with Voltage Display — Read pot position as voltage (mV) with 64-sample oversampling
• Capacitive Touch Sensor — Touch a wire, trigger an event — no external hardware needed
• Volume Knob Simulator — Pot sets level (0–100%), touch pad toggles mute

Hardware: Potentiometer, bare jumper wire (touch pad)

Module 4 — I2C: OLED Display¶

What you'll do: Communicate with the SSD1306 OLED over I2C and render text, graphics, and live sensor data on the 128×64 display.

Experiments:

• I2C Bus Scan — Discover all I2C devices on the bus by address
• OLED Hello World — Display text and workshop branding on screen
• Live Bar Graph — Pot reading displayed as a real-time animated bar on the OLED

Hardware: SSD1306 I2C OLED Display, Potentiometer

Module 5 — UART: Serial Communication¶

What you'll do: Use UART to control hardware via serial commands and parse real GPS location data from the NEO-6M module.

Experiments:

• LED Control via Terminal — Type ON / OFF in the serial monitor to control the LED
• GPS NMEA Parser — Read live NMEA sentences from NEO-6M and extract latitude/longitude in decimal degrees

Hardware: LED, NEO-6M GPS Module

Module 6 — SPI: Flash Memory¶

What you'll do: Talk to the W25Q32 SPI flash chip — read its manufacturer ID, write a string to memory, and read it back to verify.

Experiments:

• JEDEC ID Read — Identify the flash chip manufacturer and capacity over SPI
• Write & Read Back — Write "AnalogData2026" to address 0x000000 and verify on readback

Hardware: W25Q32 SPI Flash (4MB)

Module 7 — FreeRTOS: Real-Time Multitasking¶

What you'll do: Build a multi-task firmware architecture using the five most important FreeRTOS primitives. Understand why tasks deadlock and how to prevent it.

Experiments:

• Tasks — Run sensor and display tasks in parallel, pinned to separate cores
• Queues — Pass structured sensor data safely between tasks without shared memory
• Mutexes — Prevent OLED corruption when two tasks try to write simultaneously
• Event Groups — Wait until WiFi AND sensor are both ready before starting the data pipeline
• Software Timers — Blink a heartbeat LED every 500ms without blocking any task

Hardware: LED, OLED Display

Module 8 — WiFi: Modes & Connectivity¶

What you'll do: Connect to a router, create your own hotspot, and scan nearby networks — all three WiFi modes in one session.

Experiments:

• Station Mode (STA) — Connect to workshop WiFi router and get an IP address
• Access Point Mode (AP) — Turn the ESP32-S3 into a hotspot named AnalogData-ESP32
• WiFi Scan — List all nearby networks with SSID, RSSI signal strength, and channel

Hardware: XIAO ESP32-S3 (WiFi built-in)

Module 9 — HTTP: Web Client & Web Server¶

What you'll do: Make HTTP requests to external APIs and serve a web page from the ESP32 itself.

Experiments:

• HTTP GET — Fetch live IST time from a public API and print it to serial
• HTTP POST — Send DHT11 readings as JSON to a server endpoint
• Embedded Web Server — Control the LED from any browser on the same network — no app needed

Hardware: LED

Day 2 — Cloud & AI on the Edge¶

Sunday, April 19, 2026

Module 10 — MQTT: Publish / Subscribe Messaging¶

What you'll do: Implement the full MQTT model — broker, publisher, and subscriber — and build a live sensor dashboard.

Experiments:

• MQTT Publish — Send DHT11 temperature and humidity to a broker every 5 seconds
• MQTT Subscribe + LED Control — Control LED from your phone using MQTT Explorer or IoT MQTT Panel
• Live Dashboard — DHT11 data streams to Node-RED or Home Assistant dashboard in real time

Hardware: DHT11 Sensor, LED

Module 11 — mDNS, TCP & UDP¶

What you'll do: Discover the ESP32 by name on the network, build a raw TCP echo server, and broadcast sensor data over UDP without any connection overhead.

Experiments:

• mDNS — Access your board at http://esp32sensor.local — no IP address needed
• TCP Echo Server — Connect via nc from your laptop, send text, get it echoed back
• UDP Broadcast — Broadcast sensor JSON to all devices on the network every 2 seconds

Hardware: XIAO ESP32-S3

Module 12 — Sleep Modes & Power Management¶

What you'll do: Squeeze months of battery life from a 1000 mAh cell by using the ESP32-S3's sleep modes correctly.

Experiments:

• Light Sleep (Timer Wakeup) — CPU sleeps 5 seconds, wakes seamlessly, execution continues
• Deep Sleep (Timer Wakeup) — Board sleeps at ~14µA, boots every 30 seconds, tracks boot count in RTC memory
• Deep Sleep (GPIO Wakeup) — Board sleeps until button is pressed — ideal for door/window sensor products
• Production Pattern — Wake → Read DHT11 → Publish MQTT → Sleep 60s — estimated 6 months on 1000 mAh

Hardware: Push Button, DHT11 Sensor

Module 13 — TinyML Basics: Data Collection & Training¶

What you'll do: Collect a labeled dataset from the DHT11 sensor, train a classifier in a Jupyter Notebook, and export a quantized TFLite model — ready to run on the chip.

Experiments:

• Data Logger — Firmware streams timestamp, temperature, humidity, label as CSV over serial
• Jupyter Training — Load dataset → normalize → train Dense NN → evaluate accuracy
• INT8 Quantization — Convert float32 model to INT8 TFLite (target: <5KB, suitable for OTA)

Hardware: DHT11 Sensor

Module 14 — TFLite Micro: On-Device Inference¶

What you'll do: Embed the trained model in firmware and run live inference on the ESP32-S3 — no cloud, no internet, no latency.

Experiments:

• Model Embedding — Convert .tflite → C array using xxd, include in firmware build
• Live Inference — Read DHT11 every 5 seconds → classify as normal / hot / humid → print result to serial

Hardware: DHT11 Sensor

Module 15 — Real-Time Audio Keyword Spotting (Demo)¶

What you'll watch & understand: Full pipeline from microphone to GPIO — say "Hey Analog", the LED triggers in under 50ms.

Pipeline: I2S Mic → PCM Audio → MFCC Features → TFLite CNN → Keyword Decision → GPIO

Topics covered:

• I2S microphone setup and DMA ring buffer
• MFCC feature extraction with ESP-DSP accelerated FFT
• CNN model architecture (Conv1D → MaxPool → Dense → Softmax)
• Sliding window inference for real-time detection
• Repo walkthrough: training dataset, Jupyter notebook, pre-trained model, full firmware

Module 16 — Camera + Person Detection (Demo)¶

What you'll watch & understand: End-to-end ML — from data collection with the OV2640 camera to on-device MobileNetV2 inference — with a face detection bonus.

Topics covered:

• Capture 96×96 JPEG images via HTTP from the ESP32 camera
• Train MobileNetV2 with transfer learning in Google Colab
• Fine-tune, quantize to INT8 (~900KB model)
• Run inference on-device — person detected → GPIO trigger + MQTT alert
• OV2640 camera configuration and frame capture
• Transfer learning and fine-tuning strategy
• PSRAM usage for frame buffer (8MB on XIAO ESP32-S3)
• Face detection with YuNet (~150ms inference)
• Full Colab notebook from raw images to deployed model

Module 17 — MQTT with AWS IoT Core¶

What you'll do: Connect the ESP32-S3 to AWS IoT Core over mutual TLS and build a full edge-to-cloud telemetry loop — sensor data flows all the way to DynamoDB.

Experiments:

• AWS Setup — Create a Thing, download device certificate and private key, attach IoT policy
• TLS Connection — Embed X.509 certificates in firmware, connect to AWS endpoint on port 8883
• Publish Telemetry — DHT11 readings publish as JSON to analog-data/sensors topic
• IoT Rule — Route incoming messages to DynamoDB table and CloudWatch metric — live in the console

Hardware: DHT11 Sensor

Your Hardware Kit¶

Every attendee receives the following kit (hardware is returned after the workshop):

What You Take Home¶

• All firmware source code for every experiment
• Jupyter notebooks for TinyML training (runs on Google Colab — free)
• Access to the workshop GitHub repository with full project structure
• Pre-built Docker image analogdata/esp32s3-devenv:latest — your dev environment forever

Prerequisites¶

Questions? Reach us at support@analogdata.io | build.analogdata.io | +91 96633 53992