author: Tyson Liddell
tags: capturing-composite-video-with-a-microcontroller
With the MCU now able to understand PAL video timing, the ADC can be configured to capture the active video of the scanlines. It is this part of the signal that contains the picture information, a monochrome image. These scanlines are grouped into fields using the equalising and long-sync pulses described in a previous post and a full PAL image frame is made up of two interlaced fields.
For each scanline, an interrupt routine enables the ADC at the appropriate time and continuously moves samples from the ADC FIFO into a storage array. Sampling continues for the length of a scanline's active video signal and then stops until the next scanline. The ADC has a maximum sample rate of 500 ksps and the processor is running at 80 MHz. The interrupt code is simple enough that the FIFO should never overflow under these conditions.
Storage space is something to be careful of here. Capturing 1000 scanlines at 30 samples per scanline requires 30000 samples to be stored in SRAM. Storing these as 32-bit or 16-bit integers would require 120,000 or 60,000 bytes of storage, respectively. However, the SRAM only has a 32 KB byte capacity. The solution is to truncate the 12-bit samples to 8-bit and store them as 8-bit (unsigned) integers. This will allow for 256 shades of grey, provided the samples are normalised appropriately.
{% include github_commit.html repo='tysonliddell/bare-metal-tiva' sha='5013a92d67e200b026ef54e5edadd53f289c94ff' %}
The ADC produces 12-bit samples in the range 0 V to 3.3 V, while the monochrome PAL signal uses values in the range 0.3 V (full black) to 1.0 V (full white) to represent luminance. To make full use of the 8-bits of data that will be stored, the values 0.3 V and 1.0 V should map to 0x00 and 0xFF in storage, respectively. This can be achieved by applying the 12-bit -> 12-bit affine transformation
sample -> (sample - SAMPLE_300mV) * (SAMPLE_3300mV / SAMPLE_700mV) = (sample - SAMPLE_300mV) * (0xFFF / SAMPLE_700mV),
which maps SAMPLE_300mV -> 0x000 and SAMPLE_1000mV -> 0xFFF, from which we can then truncate the 4 least significant bits.
{% include github_commit.html repo='tysonliddell/bare-metal-tiva' sha='08ca800cf3e6f4378e76d5be6f719c04e036b8ec' %}
The implementation so far allow samples to be obtained from the ADC and printed to the UART. Unfortunately, due to the high volume of data, UART data corruption was observed:
..., 136, 136, 135, 135, 13136, 136, 136, ...
Note the value 13136 above, which is not a valid 8-bit integer. A reliable communication protocol needs to be implemented on top of the UART to deal with this.