AW: [sdiy] rev eng -> fwd eng?

[Magnus Danielson]

From: Grant Richter <grichter at asapnet.net>
Subject: Re: AW: [sdiy] rev eng -> fwd eng?
Date: Thu, 30 Aug 2001 17:10:02 -0500

Grant,

> Digital audio programming, to this point, has been about data reduction.
> The standard FFT is optimized for computing efficiency and not audio
> performance.

Would you please elaborate some on that... what makes FFT less worth
than the DFT of the same size? I could possibly agree to a _different_
round-off pattern, but other than that?

> The problem is modeling continuous time functions on discrete time systems,
> which is very processor intensive, even at low data rates. SPICE is an
> example of this type of computation.

Indeed, but there is engines which can make the predictions pretty fast.

> We have now reached the stage where the excess computing power of the
> Pentium and G4 class machines make this feasible. I am aware of at least one
> group working on implementing advanced continuous time based Slow Fourier
> Transforms which provide a much finer structure than the Fast Fourier
> method.

Please elaborate. For me it sounds like it is either bogus talk or
their point needs some explanation.

> If this works, it opens the door to developing on the s-plane (continuous
> time), and reusing the large base of s-plane technology which has never been
> "ported" to the z-plane (discrete time).

One thing that happend was that they found out it was better to design
in z-plane directly then starting with s-plane and then transform the
design. The reason is that you encounter different problems in the
z-plane and a translated form will allways cary some assumption of a
solution, and this will never be well matched with the actual requirements.

> An entire book could be written about the waste caused by having a global
> level paradigm shift with no master plan. Switching the technology base from
> continuous time computing to discrete time computing cost the equivalent of
> about 1 year of the Gross Global Product (if you consider 10% over 10 years)
> 
> Hopefully, the next paradigm shift will be conducted with less clownish
> disorder. The human race can ill afford expenses on that order of magnitude.

Indeed.

The main trouble thought is that the sample frequency is too near the
signal, so the transformations will all be hevilly deformed in the
warping.

Cheers,
Magnus