USB Data Protocol

Applies to: USB HVS cameras (APX003 / APX014 sensors). This page documents the low-level wire protocol of the camera's USB stream — device identification, binary framing of APS/EVS data, ep0 control requests, and timestamp domains — for driver-level development that reads the USB stream directly (libusb / pyusb, etc.), bypassing the SDK. For regular development, start with the Toolkit SDK and the Programming Guides; for recorded-file formats (RAW8 / EVT2 / CSV) see Data Formats Reference.

Device Identification

VID: 0x1D6B PID: 0x0105
Interface 0, 3 bulk endpoints: EP1 IN = APS image, EP2 IN = EVS events, EP3 OUT = firmware upgrade

Sensor Identification

APS first-frame length	Sensor
700,424 B	APX003
983,048 B	APX014

APS Frame (EP1 IN)

┌─────────────────────────────┐
│  NV12 raw frame             │  003: 768×608 = 700,416 B
│  (Y plane + interleaved UV) │  014: 640×1024 = 983,040 B
├─────────────────────────────┤
│  Trailing timestamp (8B     │  Same source and domain as
│  little-endian u64)         │  the EVS frame timestamp
└─────────────────────────────┘

	APX003	APX014
Total frame length	700,424 B	983,048 B
Trailing timestamp unit	microseconds	raw ticks (not microseconds)

EVS Frame (EP2 IN)

APX003 (chunk = 524,288 B = 16 subframes × 32,768 B)

Each subframe covers one spatial quadrant (384×304 @2bit); four quadrants tile into the 768×608 event grid.

Subframe header (16 B):
  +0x00  u64  bits[0:15]=0xFFFF magic, bits[24:63]=40-bit timestamp (5ns/tick, ÷200=µs)
  +0x08  u64  bits[44:47]=quadrant number (0-3)
  +0x10  ...  event payload 29,184 B

Pixel encoding: each u64 holds 32 pixels (2bit/px, LSB-first); code>0 marks an event, polarity=code>>1. Coordinates: x = quadrant x offset + column×2, y = quadrant y offset + row×2.

APX014 (chunk = constant 164,864 B = 32 subframes × 5,152 B)

Each subframe covers 320×64 @2bit; 32 subframes tile into a 640×1024 packed grid, ×2 interpolated to 1280×1024 output.

Subframe header (24 B):
  +0x00  u32  low 24 bits = 0x0000FFFF magic
  +0x04  u32  timestamp (raw ticks)
  +0x0C  u32  frame number (low 24 bits; spread across a frame's 32 subframes ≤1)
  +0x16  u16  SubFrame=bits[5:10], Vflip=bit11, Hflip=bit12
  +0x18  ...  event payload 5,120 B
  tail   8B   padding

Pixel encoding is the same as 003 (2bit LSB-first). The subframe position is derived from the SubFrame number through the flip mapping table onto the 16-row × 2-column grid; rows interleave as y = m×16 + row, interpolation x_out = 2x + offset.

Validation: if any subframe magic is invalid or the frame-number spread >1 → discard the whole frame.

ep0 Control Requests (vendor request)

bRequest	Name	Direction/payload	Description
0x0A	CLEAR_SHM	OUT, 1B (`0x01`)	Clear shared memory, restore stream alignment
0x14	SET_EVS_FPS	OUT, 2B (u16 big-endian)	Switch frame rate (triggers a device reboot)
0x15	GET_EVS_FPS	IN, 2B (u16 big-endian)	Read frame rate
0x07	REBOOT	OUT	Reboot the device (older SDKs used it to clear shared memory; newer firmware changed it to REBOOT)

FPS options: 003 = 120/240/500/750/1000; 014 = 240/300/480/600.

Frame Boundaries

APS: the trailing 8-byte timestamp is not a multiple of 512 → forms a short packet, so bulk returns exactly at the frame end
EVS: chunks are always multiples of 1024, no short packets → must read by exact chunk length to stay aligned

Timestamp Units

Location	APX003	APX014
Sensor raw domain	40-bit, 5ns/tick	32-bit tick (~24MHz)
EVS event t	µs (÷200)	µs (self-calibrated)
APS trailing 8B	µs	raw ticks