Understanding the APA102 “Superled”

A couple of weeks ago I reported about a new type of RGB-LED with integrated controller, the APA102. One of the interesting new features of this device is a two-wire SPI interface instead of the proprietary one-wire protocol of the more common WS2812. Many microcontrollers have hardware SPI functions, which allow easy control of these LEDs, as opposed to timing critical bit banging. But it turned out this was not the end of the story. As pointed out by Bernd in a comment, there is some discrepancy between the datasheet and the actual behavior of the devices when it comes to the “end frame”. Reason enough to subject the APA102 to more scrutiny.

The diagram below summarizes the APA102 protocol as found in the data sheet.


To investigate the functionality of the AP102, I connected an ATtiny85 to the clock and data input lines and used a logic analyzer to probe both the input and output lines. The microcontroller was programmed to output various bit patterns as described below.

I focused my investigation on four areas:

  1.   Behavior of the “Start Frame”
  2.   Function of the “111” bits in the “LED Frame”
  3.   How data is forwarded to the next device
  4.   Impact of the “End Frame”

The Start Frame

I varied the number of zero bits in the start frame from below to above 32. It turns out that a minimum of 32 zeroes are required to initiate an update. Increasing the number of zeroes does not have any impact. The LED frame is identified by the first one bit following the start frame.

The LED output color is updated immediately after the first valid LED frame. This is quite interesting, since it means that almost arbitrary update rates of the APA102 are possible. However, this may lead to a “staggered” update for longer strings, where the first LEDs in a string are updated earlier than the later ones. The best way to work around this is to use a sufficiently high SPI clock rate.

The LED Frame

As noted above, the most significant bit of the LED frame has to be “1”, since it is used to identify the start of the frame. It appears that the next two bits serve no function and can have arbitrary values. To stay compliant with the data sheet, it makes sense to set them to “1”, though.

Data forwarding

The APA102 receives a valid SPI signal and outputs a valid SPI signal to the next devices. By definition, the data line is valid only during the rise edge of the clock signal. This resulted in an interesting problem for the APA102 designers. Simply feeding the input signal to the output would not leave enough time to evaluate and possibly alter the incoming data and could create all kinds of race conditions.


To work around this issue, the APA102 delays the data on the output by half a cycle. As shown in the figure above, this is accomplished by inverting the incoming clock signal at the output. The data output is forwarded during the rising edge of the incoming clock, but only becomes valid for the next device at the rising edge of the outgoing clock.

This design is quite ingenuous as it does not require any internal clock source. It does, however, have implications for the protocol: Since the data for each subsequent LED is delayed by half a clock cycle, but the clock is not, additional clock cycles have to be fed to the string even after all data has been sent. This is the sole reason for the existence of the “End frame”, as discussed below.

The diagram above shows how entire LED frames are forwarded from one device to the next one. Once a device detects a start frame (more than 31 zero bits), it will interpret the next “1” bit as start of its own LED frame. 32 bits are clocked into the PWM registers, while zeroes are pushed to the output. After the entire LED frame as been read, any subsequent data is simply forwarded until another start frame is detected.

The End Frame

As we have learned above, the only function of the “End frame” is to supply more clock pulses to the string until the data has permeated to the last LED. The number of clock pulses required is exactly half the total number of LEDs in the string. The recommended end frame length of 32 is only sufficient for strings up to 64 LEDs. This was first pointed out by Bernd in a comment. It should not matter, whether the end frame consists of ones or zeroes. Just don’t mix them.

Furthermore, omitting the end frame will not mean that data from the update is discarded. Instead it will be loaded in to the PWM registers at the start of the next update.


In summary, each update of an APA102 based LED string should consist of the following:

  1. A start frame of 32 zero bits (<0x00> <0x00> <0x00> <0x00>)
  2. A 32 bit LED frame for each LED in the string (<0xE0+brightness> <blue> <green> <red>)
  3. An end frame consisting of at least (n/2) bits of 1, where n is the number of LEDs in the string.

Unlike the WS2812 protocol, no waiting period is required before the next update. As discussed before, I strongly suggest to only use the full brightness setting (31) to reduce flicker.

I have no recommendation for a maximum or minimum SPI clock speed. There is no specification for this in the datasheet. So far, it seems that the LED is able to handle any clock setting that is thrown at it. I had no issues with 4 MHz and others have successfully tested 10 MHz and above.

Light weight APA102 library

I uploaded an initial release of the líght_apa102 library based on above findings, a companion to the light_ws2812 lib. You can find the code on github.

34 thoughts on “Understanding the APA102 “Superled”

  1. Just thought I’d let you know, I’ve had APA102 strip respond just fine to a BeagleBone pushing 60MHz SPI data.

  2. Thank you for sharing your work.
    I have chosen APA102 LED for my experiments with music lights and I am a bit confused about all available strip and library options.
    My planned system is Raspberry Pi B+, APA102 strip with 60 LEDs and, to be safe, some level shifter.
    Could you please clear some questions?

    1) So, I’ll need to enable SPI in kernel (Raspbian has a good utility for this) and connect APA102 to RPi’s SPI pins MOSI and CLK (physical pins 19 and 23, according to this http://maxembedded.com/wp-content/uploads/2014/07/Raspberry-Pi-GPIO-Layout-Model-B-Plus.png ) through a level shifter. Level shifter will have common ground with both RPi and LED strip’s power source, and will receive +3.3V from RPi and +5V from the LED strip power supply, right?

    2) Are all APA102 strips logically and electrically more-less equal? Can I freely replace this one:
    with this:
    or this:

    3) How your library differs from these:
    Have you discovered some important bits about APA102 which are integrated in your library and might be missing from other libraries?

    I myself am looking for some way to control LED strip through node.js. I’m better with C++ and Javascript and not Python.

    Thanks again,

    • Hi Martin,

      the strips look good, in general. Unfortunately I am not really familiar with Raspi solutions to drive these strips, so I can not help you in that regard. Sorry..

      I have not done a comprehensive comparison between libraries. I believe that the adafruit library was published after mine, so this is a moot point. I am also not really fond of their rebranding and publishing things without proper reference…

      Hyperion is an awesome project. I have never tried it, but since it works with other SPI based RGB-LEDs, I suppose it will also work with the APA102.

  3. thanks for your research.. got my code working much faster than without it… I am having some issues with the global field though… I would like to try out the dimming feature but the results are not at all as expected… have you played with that field at all?

    • I have not used the global dimming feature too much, due to above mentioned flickering issue, however when I tried it, it seemed to work normally.

      What are your observations?

      • Mostly that the amount actually dimmed didn’t seem to correlate to the 0-31 value in the global. 16 worked, most of the lower values were either off or randomly dim. This may still be a code error on my part (very rare :)) so still looking at that. If it worked for you, may well be a bug on my part.

        Also, not sure where you are, but I plan to install these lights (like 1100 of them) in a cofferred ceiling. have you had any experience with potential building code issues? how best to distribute power? I live in California so those concerns scare me a little.
        Thanks again for your help.

    • The data sheet shows an internal 4.5V regulator, so driving VDD at 3.3V would be way out of spec. One thing to watch out for when you don’t drive the voltage high enough is a design that works OK on certain batches of parts but looks horrible when you mix devices from different production lots. This can cause sadness when you go from your small-scale prototype to your full design.

  4. Hi !

    would you recommand APA102 over WS2812 ?
    I want to create a project but if i go with a long LED strip I am afraid of the 800KHz limitation on WS2812 which can lead me to a low fps…

    Thanks !

  5. Thanks for taking the time to write this up. The manufacturer’s data sheets are really bad. This cleared my questions right up. Thanks!

  6. Hey guys, I need a little help on how to run a very long string of these together .. like around 1200. I am not familiar enough with electrical stuff to figure out the best way to power them. I know I will have to inject a power supply every few meters, and I want to provide enough power that I can go full white. Here are the thoughts through my head so far… looking for someone to help out!

    1. do I distribute 110AC around the run and provide 5V supplies? Or do I distribute say 24VDC and run 5V buck converters ( found some for auto use with fairly high current). leaning towards 110AC.

    2. When I insert a power supply, I was thinking of breaking Vin, but letting ground run continuous…. If I have all my 110AC outlets on the same run from my breaker, should I also cut the ground run? seems like if I don’t, there would be massive current on that wire, but SPI needs ground too. Or do I just tap ground at each power supply?

    3. I found a pretty cool part (to me) called a SPIsolator: http://www.analog.com/en/products/interface-isolation/isolation/spisolator.html
    would be a lot more work, but from my limited understanding, I could insert this across all four wires of 2 strips of LEDS, add a new power and ground in to one side, and everything is isolated… but is this needed?

    sorry for the wordiness, and appreciate any help.. thanks!

  7. Kevin,
    What type of power supplies you use is up to you. Distributing 24Vdc if you don’t already have it has the disadvantage that you have two regulators in series: one from 110V to 24V, and one from 24V to 5V. This means that you will have two times a conversion loss.

    You don’t need to cut the ground wire of the strip.
    All current that you deliver to one strip to the local 5V will return through the corresponding ground. Therefore, the connecting ground will only carry the (very small) return current of the data signal. Just make sure that you connect each ground directly next to the 5Vdc on the strip.
    This means that you won’t need the SPIisolators either.

    • Thanks for the reply… So 110VAC sounds like the way to go. So it sounds like you are recommending that only the 5V be cut for each strip in series? And the ground will be uncut and tapped into the supply ground at each power supply?
      Lets say I want to turn on full white for an extended period… How much current can a strip take over time before it starts heating up? lets say 3M so 180 leds…. that’s 10.8A. Those traces look pretty small on the strip… but again, I don’t know much about this subject… that’s why I am so nervous!

      thanks again for the response.

  8. Many thanks, Tim – this post was very helpful!

    I may have found another variant on the APA102 – I got a 144 LED/meter strip from ebay, and the LEDs responded correctly to the above protocol, but appeared to be latching the colour – i.e. if I sent (255,0,0) then (0,255,0) and nothing more, only the first red would appear. I stepped through colour changes with a debugger and definitely saw that the colour change expected only occurred when the next frame of data was sent.

    I got hold of some Adafruit DotStar LEDs for comparison, and these definitely make the colour change immediately. I even wired up three of each type into one chain to really confirm it!

    Other than that, these ebay sourced LEDs look and behave exactly the same. All my online research hasn’t turned up anything similar; the other 2-wire LED protocols

  9. Very very close… chip, wires and pads are the same configuration and the pads are very nearly the same shapes.

  10. Managed to take a magnified look – the IC on the ebay LEDs is smaller and darker – can’t see the fine gold coloured detail.

  11. Hi!

    I just got a APA102 LED strip and have been playing with it with my Tiva C board for the last few days.

    Everything has been working great except two things.

    First, today i found a peculiar thing. I was reading my code and noticed that i had set up the SPI hardware as mode 0 (CPOL 0, CPHA 0) just like you, however, when looking at the timming diagram on the datasheet, it looks that it should be mode 3 (CPOL 1, CPHA 1). How can the strip work in mode 0? Is it ignoring the CLK polarity and only looks for the rising edge (CPHA)?

    Secondly, when changing the global brigthness value, it seems to work only up to value 7 or 8. After that it dosen´t seem to change anymore. Could it be because my Tiva C cannot supply more current? (Sorry if this looks like a noob question but i’m a software engeneer and only recently started working with electronics :S)

    Have you noticed anything like this?

    Anyway thank you for sharing your work. If you didn´t i would never even bought this great LEDs!

  12. Thanks for your research. I was having last led white all times. After removing the 32 bit frame it worked fine.

    Adafruit string used.

  13. Is anyone having any luck sourcing reels of components (Not an LED strip, but the individual components ready for pick and place?). I’ve already gotten burnt twice by counterfeits in Shenzhen (SK9822s sold as APA1002C. I would be happy, except they seem to have a significant input impedance that wrecks havoc with control signals, so I think they’re fake/surplus SK9822s on top of that :/) I’m happy to put in for 10 reels minimum order, but not if they are shady.

    Great article! Very helpful.

  14. Have tested the tls 3001 ic ? If so would you share your findings and possibly some code .

    I have a mass of tls 3001 pixels that I would like to have code for but am not at that level .

    This is an excellent read thank you .

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s