bit-perfect is so yesterday anyway: http://www.gearslutz.com/board/tips-techniques/334385-intersample-peaks.html
Ideally you'd want to attenuate by -3dB to overcome this problem.
Why would bit-perfect be "so yesterday", and why attenuating by -3dB?
This is not a problem but somewhat nonintuitive phenomenon of the reconstuction process (even 10dB overshoot is mentioned).
The link provides info on somewhat unexpected behavior of reconstructed continuous signal at the intersample time/points, somewhere between the nearest left and right samples. It overshoots the values of these samples and it's perfectly within the scope of, and completely predicted by the sampling theory.
This is not some artifact resulting from an incomplete or inadequate theory - in fact, it would be alarming if this was not the case.
As I uderstood it, the whole point of the article is to bring this fact to our, and especially attention of the DA converter designers, which can not assume the sample values give some sort of "upper bound" for the values of reconstructed signal. IMHO.
"Lavry's sinc function theorem"... didn't know this elementary definite integral from every basic calculus table has its own name.There's a link to a PDF from Dan Lavry, he's the guru of digital audio...whatever he says is true.
With all due respect to mr Lavry.
Thanks for the interesting link (as always ).
Cheers.