Thanks to the cleanness of the CleanComp (get it :P), we can now see that some models of the 2600 inherently have bad PCB and chip design, leading to jailbars (vertical stripes).

So far I have observed this only in NTSC models, and never once in a single PAL so far.

For this diagnostics I was fixing the model REV B C010433. Many variations likely exist, but the overall problem is likely the same, and the fix will be similar.

The noise is caused by the chroma signal and phase signals bleeding into the LUMA pins.

The LUM0 pin gets its noise from the COL pin.

And the LUM1 pin gets its noise from the Phase 0 pin.

LUM2 gets its noise from the BLK signal if it is connected to the COL pin as it basically re-introduces that COL noise onto LUM2 as well. This results in distinctive jailbars.

Let's start with COL pin and how it adds noise to the LUMA channels.

The LUM0 pin on NTSC models picks noise ultimately from the COLOR (pin 9) as it is right next to it on the NTSC chip, whereas on PAL it is physically separated by PAL-I pin.

Unfortunately this physical closeness inside the die of the TIA chip cannot be fixed, but its direct effects are minimal (100mV approx).

The main cause of the cross-talk noise comes from the fact this COLOR trace comes out of the PCB itself both on the top and bottom.

The bottom trace adds noise to the LUM1 signal for two reasons.

Firstly, you can see how close the COL pin comes to the LUM1 trace. This adds lots of noise to LUM1.

Cutting this trace disconnects the signal from the part of the trace that is close to LUM1, thereby removing the noise.

On top of that, you will notice there is a resistor (on the top side of the board) that this trace originally connected to, sending COL to !BLK pin, which then is physically close to the LUM1 trace which would re-add this noise.

Cutting the trace resolved this issue at the same time by disconnecting !BLK from COL (something you will see is the correct thing to do in a moment.

By cutting the bottom trace to remove the noise, we also disconnected the pull-up resistor on the COL pin, so we must reinstate that, in a place that does not add noise back to the LUMA pins.

I chose to place the resistor on the top side, bending out the COL pin from the PCB so that the solder pads which are physically close to each other between COL and LUM0 are not joined, and then sending a new resistor directly on the TIA COL pin to 5V.

At this same time I also chose to weaken the pull-up strength of the COL pin by using a 3.3k resistor instead of the factory 1k resistor.

With this fix, we have removed most of the noise from LUM0 and LUM1 pins but not all.

On the top side we need to remove the LUM0 pull-up resistor to access the trace underneath, which is running close to the COL pins solder pad.

Now most of the noise is removed, but still present due to the TIA die package itself, the only choice we have left to further remove noise is to strengthen the pull-up resistors on the LUMA pins.

The TIA can sink around 8mA of current so the pull-ups can be as low as 1k safely.

By cutting the trace from LUM0 in the previous step its already missing the pull-up resistor, and for LUM1 you can remove the original pull-up resistor or stack another on top and calculate the parallel resistance. I find it easier to just remove both resistors, and add a new 1k resistor to LUM0 and LUM1 pins to 5V.

I do this here on the CleanComp pins as they are connected to the TIA pins.

If you do not want to strengthen the pull-ups you can just replace them with the same value for LUM0 if you cut the trace on top, otherwise the original pull-ups are already in place and nothing else is needed.

With the fixes done, all LUMA signals are down to around 100mV noise, which cannot be cleaned up much more as they are likely from the physical pins being close inside the TIA die.

The results however speak for themselves.

Because we lifted the COL pin completely off the PCB, the CleanComp is no longer connected, so remember to solder a wire from the COL on the CleanComp to the TIA.

Now we have COL restored, I have also noticed on the NTSC versions the COL signals is much stronger, so turn down the S-Video potentiometer anti-clockwise all the way until the image is black and white, then turn it back up just enough to bring the color back. This will give you the best visual.

Or use an oscilloscope to tweak the level to exactly 350mV when connected to the receiver.

Even though Atari officially recommended to link COL and !BLK together with a resistor, do not do this. It is 100% wrong.

Here is it disconnected.

And here is it joined.

As part of the entire Chroma circuit, I can tell you it should look like the disconnected image. The CleanComp completely reconstructs this signal, as does the Atari 2600 circuitry (just not as well).

By connecting !BLK you are massively reducing the chroma wave amplitude, causing the TV/receiver to super over-compensate colors, making them very wrong but over-saturated.

Remove this resistor if doing a CleanComp mod for sure.

If you still get jailbars it is important to know that the visual issue can come from many other places including:

All tests done above were using a genuine power supply, a CleanComp V2 (white board), an OSSC Add-on through OSSC to a UGreen HDMI capture card.

I have previously tested the UGreen HDMI capture card doesn't introduce noise on composite or S-Video using the rest pattern output from OSSC to HDMI, and tested the OSSC Composite Add-on board with a known clean Neo Geo AES to prove the entire signal path is clean from noise, before then testing the Atari 2600.