Introduction - If you have any usage issues, please Google them yourself
The idea to create 16 parallel 800kBit/s data streams is the following:
Use a Timer to create an 800kHz time base and a DMA request every 1.25us.
Use 2 compare modules to create DMA requests at the low bit time (350ns) and the high bit time (700ns)
The 1.25us DMA request sets all bits of the GPIO port high
The 350ns DMA request transfers the data from the frame buffer to the GPIO port. If the bit is a 0, the GPIO pin will go low, otherwise it will stay high.
The 700ns DMA request sets all GPIO pins low.
Repeat steps 1 to 3 until all bits have been transmitted.
This creates a stream of pulses with a pulse period of 1.25us and a pulse width of either 350ns or 700ns depending on the bit value the pulse represents.
Transferring the data via DMA to the GPIO port means that per 16 LEDs one half word (two bytes) is needed per bit. At 24 bits per LED that makes 24 half words (48 bytes) per 16 LEDs.