Fixed Point DCT

CONTEXT:
I have implemented DCT's in fixed point (DCT-IV and DCT-II). Currently
I allow one extra bit, each time there is an addition, to avoid
overflow. For example for a size 64 transform there are 6 stages of
decomposition in radix-2 and each stage involves addition, which means
I need log2(64)= 6 extra bits to losslessly transform the worst case
input.

QUESTION:
Is there some way of producing worst case test vectors for the DCT?
Has work been done on the dynamic range of the DCT output (peak to
average power ratio, crest factor, etc?)

Something tells me I am using too many bits in my transform...

"porterboy" <porterboy76@yahoo.com> wrote in message 
news:c4b57fd0.0501270815.54455227@posting.google.com...
> Is there some way of producing worst case test vectors for the DCT?

A DCT coefficient will take on its maximum value when the input alternates 
between the maximum and minimum input values, according to the sign of that 
DCT coefficient's analysis vector.

--
Matt