Sounds like those black bars (that mark where to stop stretching the image on a 640×480 3:4 set) are put there by the DC, but do not affect the pixel aspect ratio coming out of the DC by being there. Is it correct to think that the DC hangs those black sidebars over the 720×480 rendered graphics and then sends all of that to the TV as 720x480p without changing the pixel aspect ratio, and then the 3:4 TV squishes it, black bars and all, and that is the first time the pixel aspect ratio gets affected?
I’m no expert of DC internals, but the following is how I’ve understood the chain works:
* DC internal framebuffer is typically set at 640×480, and content rendered there is in 4:3 (1:1 PAR).
* PowerVR utilizes fixed 27MHz pixel clock, and with default register settings draws 525 scanlines of 858 dots each (=59.94Hz) in VGA mode.
* Border/background color can be drawn on ~720 dots, while rendered image is placed in the centermost 640 dots. Active content is drawn on 480 lines (240 for non-VGA 60Hz output) – CRT TVs typically display only ~224 lines due to overscan, resulting to top and bottom getting cropped.
In analog video, all the receving end sees is 525 scanlines of ~32us each – there is no indication on aspect ratio (WSS does not apply here) or whether actual source picture is 640×480, 720×480 or something else. A digitizer can either assume that the signal follows some standard (e.g. CEA or VESA) or try auto-adjusting correct size, both which are not guaranteed to give correct result (especially with DC).