Time for another update
I got the new Noise Killer boards back this morning (plus the second batch of each of the other boards).
To recap, this latest version changes three things:
1. Fixed an error with two signals (VDG and NRDS) being swapped on a header
2. Improve the design so that it's reliable with 74HCT parts
3. Remove the superfluous noise-killer disable input
Here's a photo of one built up and installed:
As you can see, it's populated entirely with 74HCT parts, and initial testing indicates that works just fine, and that there is plenty of timing margin.
One late change to the parts list: R2 had changed from 270R to 1K0 to give a ~75ns long write pulse, which should be good for operation at 4MHz (if the rest of your Atom is up to that). I've updated the schematic and BOM to reflect this.
Anyway, I was about to post a message to the forum saying everything is great when in some further testing I hit a small snaglet.
With certain text screens I was still seeing some kind of screen noise:
As this noise is appearing as red flecks, it looks like something is adding noise to one of the colour difference signals.
And indeed that's what was happening. Here's what PA looked like when it reached the HDMI adaptor board:
The top signal is PA (on a scale of 200mV/division). The nasty spikes are clearly the result of crosstalk from another near-by signal. Now, the HDMI adapter has some digital filtering in the CPLD to make it more resilient to noise like this. But in this case that's clearly not sufficient.
The bottom signal is what I think is the culprit: Video Data 7 (VD7) into the 6847 (on a scale of 2V/division). I think this must always have been present to some extent. But the device driving this signal is now a 74HCT part. That has quite fast edges and a full 5V rail-to-rail swing. Both of those things make crosstalk worse.
To verify this, I replaced the 74HCT573 with a 74ALS573 (which is a bit slower) and the result was less crosstalk:
In fact, as this stage the red noise on the test screen actually went away,
If you look at the Atom Schematic, it becomes apparent why VD7 is such a problem:
This data line, and only this data line, is output to PL4 (as the INV signal). So the crosstalk is happening on the cable connecting the Atom to the HDMI Adapter.
Anyway, the fix is really simple. This signal is unused by the HDMI Adapter, so cut pin 4 of the cable, as close to the Atom as possible:
The result is greatly reduced crosstalk, even with the HCT part:
As one final test, I've made up a 1M long cable between the Atom and the HDMI Adapter, and that seems to work perfectly.
I'm pretty happy with this outcome, so I'll be sending people who have requested board a PM shortly.