SpecVue

SpecVue

A 4-layer PCB for smart glasses featuring an ESP32-S3 with camera live streaming and on-device object classification over Wi-Fi.

Overview

This board is a compact (90×17 mm) smart glasses controller designed around the ESP32-S3. It captures video from a parallel-interface camera module, runs a lightweight object classifier (e.g., TensorFlow Lite Micro / ESP-DL) on the ESP32-S3, and streams results (bounding boxes, labels) over Wi-Fi — optionally with live video. Li-Po battery powered with built-in charging.

Block Diagram

USB-C ─┬─► MCP73831 (Charger) ──► Li-Po Battery
        │                              │
        └──► VSYS ◄────────────────────┘
               │
        ┌──────┼──────────┐
        │      │          │
   AMS1117  AP2121AK  LP2985
    3.3V      1.3V      2.8V
        │      │          │
   VDD_CPU  Core Rail  Camera IO
     & I/O    (1.3v)    (2.8v)

┌──────────────── Data Flow ────────────────┐

 Camera ──► ESP32-S3 ──┬─► TFLite Micro ──► Wi-Fi ──► Detected objects
 (DVP)      (capture)   │   (classifier)    (MQTT/UDP)
                        └─► Wi-Fi ──► Live video stream (optional)

Specifications

Power Tree

Power Path

1. USB VBUS feeds the MCP73831-2-OT (U3) Li-Po charger IC.
2. Battery connects through AO3401A (Q1) P-FET for reverse-polarity protection.
3. The FET's output (VSYS) is shared between USB and battery — no external power mux needed (Q1's body diode provides OR-ing).
4. VSYS powers all three LDOs in parallel.
5. SS14 (D4) Schottky diode provides additional input protection.

Microcontroller

ESP32-S3 (U1), QFN-56 7×7 mm, 0.4 mm pitch.

• Dual-core Xtensa LX7 @ up to 240 MHz
• Built-in Wi-Fi 802.11 b/g/n + BLE 5.0
• 512 kB SRAM, supports external PSRAM
• 40 MHz main crystal (Y2, 2520 package, 10 ppm) with 15 pF load caps
• CHIP_PU (pin 4) pulled high with a resistor, with reset button (Reset)

GPIO Allocation

Note: Several GPIOs (21, 33, 37, 41–46) are available for future expansion or current sensing.

Memory

W25Q128JVS (U2) — 128 Mbit (16 MB) SPI flash, SOIC-8 package.

The ESP32-S3 uses its dedicated SPI2 controller for flash, with four I/O lines for Quad-SPI operation. The flash operates at 3.3 V from the VDD_CPU rail.

Camera Interface

J4 — Molex 5034802400, 24-pin 0.5 mm pitch FFC connector.

The pinout is compatible with OV2640/OV5640/GC2035 8-bit parallel (DVP) camera modules. The 1.3 V rail (core) and 2.8 V rails (analog + digital IO) match typical CMOS sensor requirements.

Audio

MP34DT05TR-A (U7) — MEMS digital microphone, PDM output.

The MP34DT05TR-A uses a 4-pin custom footprint. The PDM clock and data connect to ESP32-S3 GPIOs for audio capture alongside video streaming.

USB-C

J1 — USB-C receptacle, 14-pin (USB 2.0 only).

No SBU lines or e-marker chip — this is a straightforward USB 2.0 sink configuration.

Battery Charger

MCP73831-2-OT (U3) — Single-cell Li-Po linear charger.

Status LEDs

All LEDs have ~470 Ω series current-limiting resistors.

Antenna

AE1 — 2.4 GHz PCB antenna (TI SWRA117D reference design), left-edge placement.

This is a PCB trace antenna based on Texas Instruments application note SWRA117D. It uses an inverted-F or meandered PCB trace pattern on the board edge with a ground clearance keep-out area beneath it. The RF feed uses a pi-network (C20, C21, L6) for impedance matching. No separate antenna component is needed — the antenna is part of the PCB itself.

PCB Stackup

Component Count

Getting Started

Prerequisites

• KiCad 9.x or later (files saved with generator v9.0)
• A 24-pin 0.5 mm pitch FFC camera module (OV2640/OV5640 compatible)

Opening the Project

git clone <repo-url>
cd SpecVue
# Open SpecVue.kicad_pro in KiCad

First-Time Setup — Footprint Libraries

Two custom footprints live in the project root as .kicad_mod files. The first time you open the project, KiCad may show warnings like "Footprint library 'FFC24' not found". This is expected — the footprint geometry is already embedded in the PCB (the board displays correctly), but the schematic-to-footprint link needs a library entry.

To resolve (one-time setup):

1. In KiCad, open Preferences → Manage Footprint Libraries (or PCB Editor → Preferences → Footprint Libraries)
2. Switch to the Project tab
3. Click Add empty row (folder+star icon) and configure each:

Alternatively, use a single entry pointing to the project root:

Then update the footprints via Tools → Update Footprints from Library and reassign FFC24: and footprints: references to the new library.

(If you skip this step, the schematic will show footprint warnings, but the PCB opens and edits fine — the geometry is stored in the .kicad_pcb file.)

Generating Gerbers

File → Fabrication → Gerbers in KiCad (no pre-built gerbers are committed).

Bill of Materials

See SpecVue.csv — all components listed with DigiKey part numbers. Key ICs:

Firmware Development

Use ESP-IDF or Arduino-ESP32 with the ESP32-S3 target. Key framework integration points:

• Camera: esp32-camera library (8-bit DVP mode)
• Classification: TensorFlow Lite Micro or ESP-DL for on-device inference; output is label + confidence per frame
• Communication: Stream detected objects via MQTT or UDP to a companion app/server; optionally embed bounding boxes in a live video stream
• Audio: I2S peripheral in PDM mode for the MEMS mic
• Wi-Fi: ESP32 native Wi-Fi stack for video streaming and detection telemetry

Board Dimensions

  ┌──────────────────────────────────┐
  │  90.00 mm                        │
  │                                  │
  │                                  │ 17.00 mm
  │                                  │
  └──────────────────────────────────┘

The 90×17 mm aspect fits within a standard glasses temple arm. No mounting holes — the board is intended to be secured by enclosure pressure or adhesive within the frame.

License

This hardware design is released under the CERN Open Hardware Licence Version 2 - Permissive (CERN-OHL-P-2.0).

Copyright (c) 2026 Rujul Rumale.

You may use, study, modify, manufacture, and distribute this design and products made from it; you must retain the copyright and licence notices. See LICENSE for the full text.