This circuit is not my idea but has been taken from CD32 design. By the way, that's pretty strange why this never went into regular production for the other models?
At the bottom of window you can see a piece of the CD32 mainboard enclosed in the metal case - this has been built for A500 as the standalone PAL encoder.

Overview

Actually, there are three phases to achieve finally composite signal. 1. RGB >> YUV. 2. YUV >> YC. 3. YC >> Composite. If the last phase known as YC mix process will omit a black and white information which is most responsible for quality will have "more space" in a separate wire. This is because PAL/NTSC systems have limited video bandwidth to about 4-6MHz and if a B&W info (commonly referred as lumination or Y) is mix together with a coded color info (commonly referred as chromination or C) then they must share the bandwidth and in the effect there are some visible degradations on the final picture.
Another thing is poor delay line used in the A1200/A600 designs that makes only worse, too. The S-Video signal is formed right after the RGB matrix and color coding process from the pins 15 and 16 - see the functional block diagram. Such a way saves more upper band frequencies and picture remain still sharp. With this circuit a separate YC video bands look much better than the Composite video and they are even comparable to the source RGB component.

There's one more interesting thing about CXA1145 - It has RGB outs. A few additional discrete parts can be video sources at the same time as all the others outputs. However, in A1200/A600 it's sensless because of the direct RGB output stage from DACs.

Principles

• all needed signals can be found on CXA1145 (U12) pins - see the pic
• no software needed
• no need any regulations
  

Implementation

For the best results all parts should be smd. The 75ohm resistors (1% metal-film are recommended) should be located as close as possible to the capacitors to minimize noise pickup and reflections caused by impedance mismatch. The 100uF capacitor should be a tantalum type or a low inductance electrolytic type for good and stable frequency response as well the 1nF cap should be a ceramic smd rather than any other type. S-video socket should be connected by the 75ohm wires and the traces between bases T1, T2 and the CXA1145 pins should be kept as short as possible for maximum performance.
One more tip to add... as you can see on the circuit diagram, there's no low post filter for EMI protection. The reference circuit from CD32 had this anyway, but it has been omited for best quality in this solution, again.
Btw the same LPF is mounted in A1200/A600 for RGB and Composite video. You can freely remove them. The LPF LC parts in the A1200 have the following marks: E258R, E259R, E260R and E263R - these are ferrite beads, after remove simply short remaining places by a solder, but the capacitors which are: E258C, E259C, E260C, E231C - only remove and do nothing more.
Also note that the original RF modulator must be removed for additional connector.

This is the schematic.

 

Here are the video frames captured and saved with lossless compresion for objective compare from both Composite and S-Video outputs. The differences can be easily visible because quality of sharpness and color reproduction is somewhat different. These frame grabs were taken from A500 and the standalone PAL encoder, so.. this is not the best result, especially when encoder wasn't made with the highest quality parts... Unfortunately, my old A1200 have damage sync line and this is all I could put on this page.

Composite B&W: visible loss of luma bandwidth (horizontal resolution), there's less details and the subsequent loss of sharpness

S-Video B&W:

Composite color: here's very nice noticeable cross-color interference and color ringing caused by crosstalks from mixed Y-C bands

S-Video color:


Standalone PAL encoder