Jon.Fairbairn@computer-lab.cambridge.ac.uk (Jon Fairbairn) (01/14/91)
A recent posting to this newsgroup (I'm preparing this off-line so I can't
check who) referred to DCC, quoting some of Philips' claims. Do we believe
Philips? Well, the cynic in me says that since they are involved in
cost-cutting exercises to "bring the company back to profitability," and the
failure of DCC would prove embarrassing, a few grains of salt are needed. So
here are the grains of salt as I see them.
Background info.
One point of DCC is to use a carrier essentially the same shape as compact
cassette. Why this is important is not clear to me, because the vaunted
advantage -- that the player can also play compact cassettes -- doesn't count
for much. If you've got lots of cassettes, then you've presumably also got a
cassette deck, so it doesn't matter if the new medium won't play them.
Worse, if you get rid of your cassette deck, then you'll have no way of
digitising old tapes to save them from wearing out, because you'll be playing
them on your digital recorder. That side of the argument doesn't appear in
the PR bumf, though.
Using something that size for digital sound poses a problem. Recording the
bits straight on the tape would require the tape to move insanely fast, and
you wouldn't get much on. RDAT uses a rotating head so that the effective
tape speed is high. DCC uses sixteen thin-film heads across the tape, but
even so requires a data-compression algorithm to get a respectable amount of
sound on a tape. The tape speed is the same as for compact cassette, so the
continuous playing time is 45 minutes, with 60 minutes (per side) promised
for thinner tape.
Warning! The next two paragraphs are Noddy level!
Now, it is possible to compress data with absolutely no loss of quality (not
like dynamic range compression, which sacrifices amplitude resolution for
noise reduction). Here's a hypothetical example as an illustration. The odd
numbered samples in the compressed version are exactly the same sixteen bit
values as in the original. The even numbered ones are stored as the
difference between the sample and the average of the two adjacent samples.
So the even numbered ones are relative to the linear interpolation between
their neighbours. Some of the time this number will be small enough to be
held in eight bits, and some of the time more. So if we use one bit to
store the size of the sample, the even numbered words will sometimes take
seventeen bits, but other times just nine. In the worst case (such as a
22.05kHz sine wave) this means a 1/16 _increase_ in the size, and in the best
case, where every other sample happens to fit in eight bits (such as silence,
or triangular waveforms at the right frequencies), there will be a 25/32
reduction in size. Music falls somewhere in between.
That's not a very good algorithm, but it illustrates two points:
1) For perfectly reversible compression the data rate can vary enormously;
there'll always be signals that can't be compressed, so the peak rate is the
same as uncompressed, and the minimum rate is less than that by the
compression factor.
2) Because the point of compression is to remove redundancy the resistance to
loss of data is reduced. If you lose one sample from a DAT or CD recording,
you can interpolate to get an approximation back. If you lose an odd numbered
sample from stuff compressed as above, you've lost its neighbours too, so
you'll have to interpolate three samples.
The crux.
Now, DCC doesn't even use a perfect compression algorithm (presumably because
the data rate fluctuations would be too high). PASC (Precision Adaptive
SubCoding, I think -- see "Opening shots fired in digital tape war", New
Scientist 12 Jan for a brief, no-nonsense description.) uses knowledge about
the ear's ability to distinguish sounds in order to _discard_ some of the
information. But people with well trained ears can hear more than people who
don't give a toss, and I wouldn't like to bet that PASC is tailored for the
best ears -- more likely the average. Even disregarding that, the data rate
for PASC is not constant ("an _average_ of four bits per sample"). This is
because sounds that are audibly more complex take more bits. What happens if
a complex sound goes on too long? The mechanism won't be able to get enough
bits off the tape, that's what. Of course, this won't happen when Philips
demonstrates DCC :-)
Well, maybe PASC is psychoacoustically perfect, so not even golden lugholes
can tell the difference. The data has been compressed by a factor of four on
average, so bugger-all redundancy. So as the tape wears out, we can expect
noticable degredation quite soon. And if your using a 120-minute tape, soon
is _soon_.
Final downer.
Suppose we swallow Philips' line? Well, in that case, we've still only got
a medium that will record 60mins max continuous. 45 if you want to keep it.
That's not enough for some CDs, and not enough for broadcast concerts. And
even if you use both sides the chance that the break will be in a convenient
place is remote. But if PASC works, it could be used with RDAT hardware to ||
give tapes capable of storing eight hours of continuous music. Not that I ||
know many pieces that long :-). ||
If I let the cynic in me talk again, the real reason for DCC is that it
_isn't_ as good as DAT. You can't record whole CDs on one side, and the
quality is degraded, so you can't make good copies. This is what the record
companies want, so that they can continue to sell CDs at absurdly inflated
prices. Don't believe them when they say that copying CDs would kill the
market. People buy pre-recorded compact cassettes. Lots of them. More than
CDs. They buy them because they are cheaper than CDs, and it makes no
difference to them that the cost price of a CD is about the same (if you take
into account the market sizes) as compact cassettes.
So if you're hanging on for DCC, don't!
It's worse than that. It's quite likely that non-high-enders (non-hi-fiers,
at least) will swallow the hype. That'll mean that DCC will be cheap, and the
mass market will be there, so DATs will stay expensive. So if you know
mundanes who are planning to get DCC, tell them the story :-|
Jon <Jon.Fairbairn@cl.cam.ac.uk>hans@smab.se (Hans C Larsson) (01/16/91)
Jon.Fairbairn@computer-lab.cambridge.ac.uk (Jon Fairbairn) writes: [About DCC] >sound on a tape. The tape speed is the same as for compact cassette, so the >continuous playing time is 45 minutes, with 60 minutes (per side) promised >for thinner tape. The digital mode uses special cassettes (with sliding covers ala DAT) that are one-sided *and* auto-reverse !. I.e. you get 45 + (clonk) + 45 minutes with DCC-90 tapes. >Now, it is possible to compress data with absolutely no loss of quality (not >like dynamic range compression, which sacrifices amplitude resolution for ... >or triangular waveforms at the right frequencies), there will be a 25/32 >reduction in size. Music falls somewhere in between. NICAM TV-stereo (currently in use in UK & Scandinavia) uses 10+4 bits, similar to the first 18 bits CD's (16+2). (Floating bits) >So if you're hanging on for DCC, don't! I second that.. -- ------------------------------------------------------------ Hans C Larsson Email: hans@smab.se Saab Missiles, Sweden Motto: "keep it short"