Hi,

Yesterday, Roh and I replaced the video chip on my board with the broken
video input and replaced the LM4550 codec on a RC1 board with a WM9707
chip. After hours of struggles with (de)soldering problems, copper
planes spreading and dissipating the heat from the air gun (solution:
use a larger nozzle to heat more pins at once) and the tendency of
lead-free solder to make shorts between the chip body and the pins
(lesson learned: use lead for hand soldering of *QFP packages), we
managed to properly replace the two chips.

VIDEO INPUT
Replacement of the ADV7181 chip on my board restored the video input
functionality. This supports the theory that the chip was damaged
because of a voltage surge coming from my camera when it died.
Let's just add the Littelfuse transient voltage suppression devices on
the video input connectors of the RC3.

AUDIO
After replacement of the codec, when playing the test tone with the test
program, the audio output was remarkably noise-free.
However, when running Flickernoise, the demo firmware, or the recording
tests of the test program, one or more of the following symptoms appear
(they are very marked with Flickernoise):
1. "white" noise comparable to that of the LM4550 codec
2. very loud noises, beeps and screeches, sometimes periodic, happening
especially with Flickernoise and the demo firmware. They depend on the
activity of the software, e.g. moving windows, typing text, activating
certain functions affect the produced noises very significantly.
3. recorded audio data is garbage

My guess is that there is some subtle and probably timing-related FPGA
design bug that causes the AC97 link to become randomly corrupted
depending on other SoC activity. This would give a simple explanation to
the audio problems we have seen so far:
1. "white" noise on the ML401: same bug, same noise
2. loud noises with WM9707: the two different codecs react differently
to the AC97 corruption (maybe because of slightly different timing
specs)
3. audio input apparently not working in Flickernoise on Norman's board.
With the help of Murphy's law, process variations could have hit the
sweet spot that made the audio input pass the factory test and fail in
Flickernoise, because of different SoC activity with the test program
and Flickernoise.

Other (improbable) suspects could be:
1. crosstalk, i.e. strong SDRAM signals aggressing the AC97 link. The
PCB layout has a good separation between the two, and increasing the
AC97 drive to 24mA on the FPGA does not change anything to the problems.
So this is unlikely.
2. strong power supply noise - we did not measure any
3. strong EM emissions - we did not measure any

I'd say go ahead with the RC3 board run without any change to the audio.

S.

_______________________________________________
http://lists.milkymist.org/listinfo.cgi/devel-milkymist.org
IRC: #milkymist@Freenode
Twitter: www.twitter.com/milkymistvj
Ideas? http://milkymist.uservoice.com

Reply via email to