Fadecandy is firmware for the Teensy 3.0, a tiny and inexpensive ARM microcontroller board.

Fadecandy drives addressable LED strips with the WS2811 and WS2812 controllers. These LED strips are common and inexpensive, available from many suppliers for around $0.25 per pixel.

This firmware is based on Stoffregen's excellent OctoWS2811 library, which pumps out serial data for these LED strips entirely using DMA. This firmware builds on Paul's work by adding:

• A high performance USB protocol
• Zero copy architecture with triple-buffering
• Interpolation between keyframes
• Gamma and color correction with per-channel lookup tables
• Temporal dithering

These features add up to give very smooth fades and high dynamic range. Ever notice that annoying stair-stepping effect when fading LEDs from off to dim? Fadecandy avoids that using a form of delta-sigma modulation. It rapidly wiggles each pixel's value up or down by one 8-bit step, in order to achieve 16-bit resolution for fades.

Example videos:

• Smooth interpolation
• Multiple Fadecandy boards running in unison

• 512 pixels supported per Teensy board (8 strings, 64 pixels per string)
• Very high hardware frame rate (395 FPS) to support temporal dithering
• Full-speed (12 Mbps) USB
• 257x3-entry 16-bit color lookup table, for gamma correction and color balance

Fadecandy internally represents colors with 16 bits of precision per channel, or 48 bits per pixel. Why 48-bit color? In combination with our dithering algorithm, this gives a lot more color resolution. It's especially helpful near the low end of the brightness range, where stair-stepping and color popping artifacts can be most apparent.

Each pixel goes through the following processing steps in Fadecandy:

• 8 bit per channel framebuffer values are expanded to 16 bits per channel
• We interpolate smoothly from the old framebuffer values to the new framebuffer values
• This interpolated 16-bit value goes through the color LUT, which itself is linearly interpolated
• The final 16-bit value is fed into our temporal dithering algorithm, which results in an 8-bit color
• These 8-bit colors are converted to the format needed by OctoWS2811's DMA engine
• In hardware, the converted colors are streamed out to eight LED strings in parallel

The color lookup tables can be used to implement gamma correction, brightness and contrast, and white point correction. Each channel (RGB) has a 257 entry table. Each entry is a 16-bit intensity. Entry 0 corresponds to the 16-bit color 0x0000, entry 1 corresponds to 0x0100, etc. The 257th entry corresponds to 0x10000, which is just past the end of the 16-bit intensity space.

By default, Fadecandy interprets each frame it receives as a keyframe. In-between these keyframes, Fadecandy will generate smooth intermediate frames using linear interpolation. The interpolation duration is determined by the elapsed time between when the final packet of one frame is received and when the final packet of the next frame is received.

This scheme works well when frames are arriving at a nearly constant rate. If frames suddenly arrive slower than they had been arriving, interpolation will proceed faster than it optimally should, and one keyframe will hold steady until the next keyframe arrives. If frames suddenly arrive faster than they had been arriving, Fadecandy will need to jump ahead in order to avoid falling behind.

This keyframe interpolation is not intended as a substitute for other forms of animation control. It is intended to generate high-framerate video from a source that operates at typical video framerates.

The Fadecandy project includes an Open Pixel Control server which can drive multiple Fadecandy boards and DMX adaptors. USB devices may be hotplugged while the server is up, and the server uses a JSON configuration file to map OPC messages to individual Fadecandy boards and DMX devices.

Why use Open Pixel Control?

• You can keep your effects code portable among different lighting controllers.
• OPC includes an OpenGL-based simulator, allowing you to develop effects before the hardware is done.
• The OPC server manages USB hotplug and multi-device synchronization, so you don't have to.
• The OPC server loads color-correction data into each Fadecandy board.

For more information, see the Server README.

• To install a firmware image, you'll need the Teensy Loader or the command line version.
• To recompile the firmware (only needed if you want to modify it), please use the recommended ARM toolchain.

The pin assignment is the same as the original OctoWS2811 pinout:

Remember that each strip may be up to 64 LEDs long. It's fine to have shorter strips or to leave some outputs unused. These outputs are 3.3V logic signals at 800 kilobits per second.

It usually works to connect them directly to the 5V inputs of your WS2811 LED strips, but better signal integrity can be had quite easily by adding a series resistor (47 to 220 ohms) near the Teensy 3.0 board. And for the best signal integrity, you can use a level-shifting buffer (like the 74HCT245) to convert the 3.3V logic to 5V. For more information, see the OctoWS2811 web site.

Running these control signals over long distances is tricky, and not recommended unless you're using some kind of purpose-built line driver like an RS-485 transmitter/receiver pair. For large installations, it's usually easier to run the USB signals long distances and place the Teensy boards close to your LEDs.

To achieve the best CPU efficiency, Fadecandy uses a custom packet-oriented USB protocol rather than emulating a USB serial device. This simple USB protocol is easy to speak using cross-platform libraries like libusb and PyUSB. Examples are included. If you use the included Open Pixel Control bridge, you need not worry about the USB protocol at all.

The device has a single Bulk OUT endpoint which expects packets of up to 64 bytes. Multiple packets may be transmitted in one LibUSB "write" operation, as long as the buffer you provide is a multiple of 64 bytes in length.

Each packet begins with an 8-bit control byte, which is divided into three bit-fields:

• The 'type' code indicates what kind of packet this is.
• The 'final' bit, if set, causes the most recent group of packets to take effect
• The packet index is used to sequence packets within a particular type code

The following packet types are recognized:

In a type 0 packet, the USB packet contains up to 21 pixels of 24-bit RGB color data. The last packet (index 24) only needs to contain 8 valid pixels. Pixels 9-20 in these packets are ignored.

In a type 1 packet, the USB packet contains up to 31 lookup-table entries. The lookup table is structured as three arrays of 257 entries, starting with the entire red-channel LUT, then the green-channel LUT, then the blue-channel LUT. Each packet is structured as follows:

A type 2 packet sets optional device-wide configuration settings:

Control transfers supported on EP0:

Micah Elizabeth Scott <micah@scanlime.org>