Week 7

Oscilloscope Operation: Measuring Clock and Data

Demonstration of capturing a stable clock and synchronous data line, including probe reference selection and verifying edge alignment.

Combined capture shows stable clock and synchronous data suitable for decoding

Load Cell: Direct Analog Output and HX711 Digital Output

We compare the raw Wheatstone bridge differential output to the HX711-amplified, clocked digital interface.

Direct analog output

Probing V1 (green) and V2 (white) on press

Millivolt-level differential changes at the bridge confirm expected polarity and sensitivity.

The raw Wheatstone bridge output is in the mV range; direct probing between V1 and V2 shows small but measurable changes on load. Mechanical force perturbs the bridge imbalance, producing a differential voltage proportional to applied weight.

Microphone: Digital output with 16 kHz clock

We capture the probe points, the ~16 kHz clock, and the synchronous data output.

Probe points for clock and data relative to module ground

We identify stable ground and tap CLK and DOUT to assess timing integrity before decoding. Proper referencing minimizes measurement artifacts and preserves true edge timing.

~16 kHz clock observed as a clean square wave

The mic’s clock provides the sampling reference for synchronous data capture. Its frequency and duty cycle help verify correct configuration and signal quality.

Digital output edges align to the clock for bit capture

Data toggles on defined edges enabling reliable sampling and decoding. This confirms synchronous serial behavior suitable for I²S/PDM-style acquisition.

IR Phototransistor: Analog output

We power the IR emitter, probe the phototransistor node, and observe analog voltage changes with illumination.

Probe across sense node and reference

Illumination modulates collector current, changing the sensed voltage across the load. This configuration establishes a clear baseline for assessing the light-dependent response.

IR diode power supply setup

IR diode powered on

The emitter is powered and positioned to control incident IR on the phototransistor for response testing. This enables repeatable stimulus while varying distance and angle.

Analog output changes with IR intensity

Increasing light increases conduction, lowering the sensed voltage across the load resistor. The transfer is monotonic within the tested range, enabling thresholding or analog filtering.

1. Remilling Development Board with ESP32S3

Using the development board designed in Week 4, I remilled the board to address an SDA/SCL short that prevented proper I²C communication for sensor visualization. The redesigned board incorporates an ESP32S3 with a presoldered camera module header in addition to the LED-button interface.

2. Simple Input Device: Button-LED System

Implementing the button-LED system using code from Week 5 to verify board functionality. The system provides immediate visual feedback confirming proper GPIO configuration and signal handling.

Button LED system successfully operational on custom development board

3. Complex Input Device: Camera with Edge AI Inference

Plugging in the camera module to the ESP32S3 requires careful soldering of a small pin header under the microcontroller using solder paste and hot air, so we purchased and used a pre-soldered ESP32S3 camera module to avoid issues with the header. The camera integration enables real-time image capture and edge AI inference on-device.

18:42:41.622 -> Captured 160x120 (19200 bytes)
18:42:41.982 -> TTD: 12.103 years

18:42:41.622 -> E (158988) i2c.master: 
  I2C transaction unexpected 
  nack detected
18:42:41.622 -> E (158991) i2c.master: 
  I2C transaction failed

While troubleshooting the display issue on the development board, I tested the camera inference system on the ESP32S3 camera subsystem from my final project, confirming successful edge AI inference functionality.

Camera Inference Edge AI Model

The Edge Impulse library contains the complete inference engine for running neural network models on the ESP32S3. The library includes the Edge Impulse SDK with support for various sensor inputs, neural network inference execution, signal processing utilities, and classification output handling.

The model was trained on Edge Impulse with ~60 training images and ~60 testing images. You can view and interact with the model online at this Edge Impulse Studio link.

Camera Inference Microcontroller Script

The camera inference script captures images from the ESP32S3 camera module, processes them through the Edge AI model, and displays results on both the OLED screen and Serial Monitor.

BEGIN Setup
  Initialize Serial communication
  Configure LED and OLED display
  Initialize camera module with grayscale format (160x120)
  Load Edge AI inference library
END Setup

BEGIN Loop
  Check for touch trigger OR 60-second timeout
  
  IF trigger detected THEN
    Turn LED ON to indicate capture
    Capture image frame from camera
    
    IF capture successful THEN
      // Display on OLED with dithering
      Scale image to 128x64
      Apply contrast enhancement
      Apply Floyd-Steinberg dithering
      Display dithered image on OLED
      
      // Process for Edge AI inference
      Convert grayscale to RGB888
      Prepare input buffer for neural network
      Run Edge Impulse classifier
      Get inference result (TTD value)
      
      // Display results
      Display TTD overlay on OLED
      Print TTD value to Serial Monitor
    ELSE
      Show error message on OLED
    ENDIF
    
    Turn LED OFF
    Return frame buffer to camera driver
  ENDIF
  
  Small delay (30ms)
END Loop

AI-Assisted Week 7 Page Creation and Group Assignment Documentation

Cursor AI assisted with structuring this page, refining the group assignment documentation, creating the two-row highlights, and polishing the individual assignment text. The transcript below captures the session focused on Week 7 updates and guidelines.

AI-Assisted Camera Code Development

ChatGPT assisted in developing the camera inference code and troubleshooting the development board camera implementation. The AI provided guidance on camera module initialization, image capture optimization, Edge AI model integration, and I²C display communication debugging.

This work is licensed under a Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License