Wiznet makers

1. Project Overview

This project demonstrates a real-time Ethernet-based JPEG streaming system built using the WIZnet Pico (RP2040/RP2350) board combined with the ArduCAM Quick-Bootup 3MP DVP Camera.

Two main components make up the system:

Camera side (Pico): Captures JPEG frames via an 8-bit DVP interface and transmits them over Ethernet by splitting the data into UDP packets.

PC side (Python): Receives packets, reassembles complete frames using header information, and decodes the JPEGs in real time with OpenCV.

Through this project, makers can gain a complete understanding of real-time embedded camera streaming over Ethernet, forming a foundation for IoT vision, robotics, and remote monitoring applications.

ArduCam X WIZnet Pico UDP Streaming Github Project Link

🧠 Core Keywords:
RP2040 DVP PIO DMA / UDP JPEG Streaming / Frame Reassembly / OpenCV Real-Time Decoding

2. Hardware Modules

WIZnet Pico (RP2040 / RP2350)

[WIZnet Pico RP2040 Based Link]

[WIZnet Pico RP2350 Based Link]

Supports both RP2040 and RP2350 (up to 200MHz system clock)

Embedded Ethernet options:

W5100S / W5500 / W6100 — SPI (40MHz)

W6300 — QSPI Quad (37.5MHz)

Fully compatible with Pico SDK 1.5.1

Supports simultaneous use of DVP Camera + Ethernet + SPI Flash + UART

Arducam Quick-Bootup 3MP DVP Camera for IoT

[Arducam Quick-Bootup 3MP DVP Camera for IoT]

💡 With its 300ms instant-on capability and ultra-low power design,
this module is ideal for “instant-response IoT vision applications.”

3. Performance Comparison (Sys Clock 200MHz)

✅ The W6300 QSPI variant delivers smooth HD streaming even in real-time applications.

These values represent typical JPEG-compressed frame sizes and throughput measured under 37.5 MHz QSPI operation on the W6300-EVB-Pico2 board.

4. Pin Mapping (Pico ↔ ArduCAM)

📸 Each PCLK rising edge samples one pixel (8-bit data),
while VSYNC HIGH defines the active frame transmission window.

5. JPEG Frame Capture Sequence (on Pico)

① VSYNC ↑ → Frame Start  
② HSYNC ↑ → New Line Start  
③ PCLK ↑ → Sample D0–D7 (8-bit pixel)  
④ DMA stores 32-bit chunks into line buffer  
⑤ HSYNC ↓ → Line End  
⑥ VSYNC ↓ → Frame End

The PIO handles the timing logic, while DMA continuously transfers 32-bit data blocks (512 bytes per line) into RAM.

📥 If the SOI marker (0xFFD8) isn’t detected within the first 4 lines (512B × 4),
the system retries until a valid frame is captured.

6. Final Streaming Process (UDP-based Transmission)

Since JPEG frames vary in size (a few KB to tens of KB),
a single UDP packet cannot contain the entire image.

Therefore, the Pico splits each frame into ≤1,400-byte payloads,
attaching a small 4-byte header to each.

Example: UDP Packetization of a 30KB JPEG Frame

A 30KB (≈30,000 bytes) JPEG image is split into 22 UDP packets,
each carrying up to 1,400 bytes (4-byte header + 1,396B data).

Example Sequence:

Frame #1 (30KB JPEG)
 ├─ Packet 0: [Frame=1, ID=0, Total=22, End=0]
 ├─ Packet 1: [Frame=1, ID=1, Total=22, End=0]
 ...
 └─ Packet 21: [Frame=1, ID=21, Total=22, End=1]

Receiver reassembles packets by Frame/Packet ID → complete JPEG → display.

📡 Transmission (Pico)

total_packets = (jpeg_size + PAYLOAD_SIZE - 1) / PAYLOAD_SIZE;

for (pkt_id = 0; pkt_id < total_packets; pkt_id++) {
    tx_packet[0] = frame_id;
    tx_packet[1] = pkt_id;
    tx_packet[2] = total_packets;
    tx_packet[3] = (pkt_id == total_packets - 1) ? 0x01 : 0x00;

    memcpy(tx_packet + 4, jpeg_data + offset, chunk_size);
    sendto(socket, tx_packet, chunk_size + 4, destip, destport);
}

🧩 Reassembly (Python)

On the receiver side, each packet is reassembled based on Frame ID and Packet ID.
The Assembler class reconstructs full JPEG frames as follows:

fid, pid, tot = pkt[0], pkt[1], pkt[2]
self.buf.setdefault(fid, {})[pid] = pkt[4:]
if len(self.buf[fid]) == tot:
    data = b"".join(self.buf[fid][i] for i in range(tot))
    return data  # Complete JPEG frame restored

OpenCV then decodes and displays the frame in real time.

🖥️ Even with UDP transport, this design achieves lossless frame assembly through intelligent ID-based reordering.

7. Key Takeaways