Regarding AA-filters or not, here the purely scientific point-of-view (I'll hope my English doens't cause more irritation than before;-):
- We have a 6µm-pixel-pitch-sensor (like todays Dalsa, Kodak and Sony-sensors) which means the max. frequency this kind of sensor is capable to resolve is 83 linepairs/mm (Nyquist: you need two pixels to resolve one linepair or one pixel for each line) - there is absolutely no way such a sensor will be able to resolve more than that, it's simply impossible
- Whenever a higher frequency (>83lp/mm) "hits" this sensor, false information is created (alaising) which may look like real information due to simple test charts and tricky processing (jsut look how often dpreview claims lpph beyond Nyquist!) but it isn't, with a bayer-filtered/color-interpolating system it's even worse, it becomes this really nasty "colorful" moire we all fear!
- To avoid alaising, the contrast at the max. frequency (83lp/mm) has to be zero! So the S2-sensor with 7500pixels wideness will never be able to actually resolve 7500lines/3750linepairs!
- AA-filters blur the image, the actual design/manufacturing might be tricky, but the principle/goal is simple: reduce contrast at Nyquist-limit to zero, while not affecting lower frequencies (steep). Sadly, that's impossible with any low-pass-filter in the world, they ALWAYS reduce contrast below nyquist and that's their downside! This information is lost forever and - despite common believe - cannot be restored with any kind postprocessing/sharpening (only simple structures as lines/borders may appear that way)!
In fact, that's why all AA-filters are a compromise, they don't suppress contrast to zero (at Nyquist) to reduce the negative effect on lower frequencies, while still allowing a small amount of moire/alaising. Propably they've become better/steeper over time but the smaller pixel-pitches made it easier anyway, because the lenses have problems to deliver enough contrast up to Nyquist anway.
- What is the solution to this problem?
Oversampling! Build a sensor with many little photosites which is capable of resolving very high frequencies beyond the performance of the lenses, than downsample the image to the actual performance of the lenses. Professional video-cameras/scanners work that way.
- Why is no AA-filter better regarding IQ?
The high frequencies slightly below nyquist are not affected, enhancing microcontrast and information - impossible to retrieve by any post-processing! But when moire appears, it can be removed by the software/photographer WHERE it appears, of course much more work than just pressing one button but you can see and select moire and than remove it / make it invisible - an AA-filter always effects all images, everwhere! Of course the lost information cannot be retrieved but there isn't any information there in the AA-filtered-image, too!