[rec.audio.high-end] Theta DSP digital preamp

kube@kokoro.UCSD.EDU (Paul Kube) (01/10/91)

I've been reading a bit about the Theta DSPre.  This is a $4000
"digital preamp" containing an 8x oversampling 16-bit D/A that has got
some pretty breathless reviews-- comments of the "I always thought
analog was better than digital but I like CD's through the Theta
better than vinyl through my $10000 phono section" type.  The moral
is, apparently, that you can get very nice performance from a
conventional 44KHz linear 16-bit format and off-the-shelf 176KHz
DAC--- if you have the right interpolation algorithm for generating
the additional 3 samples per tick on playback.  (Maybe 24 bit 88KHz
isn't needed after all...)

But like typical audiophile reviews, these leave a lot of unanswered
questions.  Such as, just what is the hardware here?  The designer
Mike Moffat says two RISC processors, one per channel, each at 38
MIPS.  Has anyone looked inside to see, or otherwise heard, if these
are iPSC/860's?  How much ROM does he use for each?

Also, according to the press releases, the big difference between the
Theta oversampling algorithm and the usual 16-bit algorithm is that it
is optimized for both phase and amplitude fidelity, instead of just
flat amplitude response.  So that leads to more questions.  I haven't
seen a description of the "usual" oversampling algorithm, but why not
just use a symmetric (noncausal, but so what) impulse response in the
interpolation filter, and therefore a real transfer function, so it
will be guaranteed phase invariant?  Then you can just worry about how
to truncate a sinc to give you bandpass ripple and rolloff you can
live with, given how much computer you have.  Or is Theta really
talking about trying to reconstruct phase relationships destroyed by
"brick wall" analog filters used in 44KHz recorders?  That would be
trickier, but then I would think it would sound lousy on heavily
oversampled, digitally filtered CD's like Cheskys, but there's no
mention of this in any of the reviews I've read.

Can anyone shed some light on this for me?  

	Gratefully,

	Paul Kube@ucsd.edu,  ...!ucsd!kube

jas@proteon.com (John A. Shriver) (01/17/91)

The processors are Digital Signal Processors.  There are special
microprocessors with instruction sets designed for signal processing
applications.  In particular, incredibly fast integer multiply and
divide.  (But no floating point!)  The Texas Instruments TMS340 is one
example of a DSP, and I think it was used in some of the early
products of this sort.  I think most of these folks are now using some
Motorola part instead, which is faster and/or has more address space.
I suspect you can predict that the ROM is as big as the chip allows,
these chips do not have a lot of bits in the program counter.

What Theta and Wadia seem to be doing is giving up some frequency
response for better phase and impulse response.  This is apparently
pretty routine signal processing stuff, well understood in the 60's
and 70's.  The Philips engeineers either weren't aware of the prior
art, or decided that you HAD to have flat frequency response, because
as anyone who reads "Stereo Review" knows, you MUST have flat
frequency response.

I don't think anyone other than Spectral is trying to "correct" for
the errors in the record process, and they have a switch for it.

Some listeners don't think that even these DSP based CD systems are
perfect yet.  They alleviate some of the flaws, and are much better
than records in some ways.  Not in all ways.  It may depend on what
You value in sound reproduction.

I think what the DSP systems (and the one bit systems) show is that
Digital Audio could be near-perfect, if only it was 18 bits at 100
KHz!  If only they would at least start using that sort of technology
for the master tapes, so that our current musical heritage would be
useful when we get a better delivery media than the current CD!

At least Philips is not so stupid as to throw away the Mercury 35MM
master tapes.  Some part of that company is smart enough to know that
16 bit 44.1 HKz is not perfect enough to make a copy of them.  (There
is a continuing fear that some record company may be stupid enough to
throw out master tapes in favor of digital copies.  EMI did this a
generation ago when they threw out the master stampers of the 78's
that they reissued as the "Great Performances of the Century" series,
in favor of analog tape copies that are now considered pretty
substandard.)