Scaler II vs Scaler problems

Hi Istvan,

The 'missing' pixels you talk about are most likely an expected effect caused by downscale (you should not see any such effect for upscale or pass through). What are the input/output resolutions where you notice it the most? Basically, when the scaler downscales a line by a factor of N it starts at pixel 0 in the line and uses every Nth pixel as the center of point for the lanczos kernel to produce each output pixel. Hence, the final output pixel in each line was created from a kernel centered N-1 pixels from the edge of the input frame, so it might appear as if you lose a pixel or two for larger downscales. A similar affect occurs when downscaling the number of lines. The Scaler I and Scaler II use exactly the same algorithm so you will not currently see any improvement in the Scaler II.

While this effect is expected of the algorithm used, it might be greater than it should be due to a potential error in the coefficients generated by the Megawizard. When you run the Megawizard for the Scaler I you can preview the coefficients it is going to generate. If you select, for example, 9 taps and look at the coefficients you will see that the center (highest value) for the kernel in phase 0 is not centered on tap 4, which should be the center tap, but is somewhere between 2 and 3. In fact, all the phases are centered slightly to the left of where the probably should be, meaning your whole image could be shifted up and left by one or two pixels more than it should. It is my understanding that this issue has been picked up at Altera and should be resolved in the Scaler II in the 11.0 release.

I think the reason the Scaler II was created was because it uses a new line-based approach internally that will eventually allow some of the internal components to be exposed to users to generate more flexible systems than the current frame-based approach allows.

Hope this helps.

Regards,

Kieron