In article m, Rob
reply@ng wrote:
On 18/06/2012 11:05, Jim Lesurf wrote:
[snip]
What insofar as you can tell is the thinking behind HDCD - the
assumptions about say CD that it tries to redress?
Well, according to the Patent and the other info the creators have written
that I've read, their argument is as follows.
That Audio CD is limited in both
A) Dynamic range.
B) Bandwidth.
The argument for (A) is that we know people can - in fairly ideal
situations - 'hear' a range of sound levels covering about 120dB. From the
smallest sounds that can just be heard, up to the loudest that don't cause
pain or damage even on a short hearing. Whereas CD has a nominal range of
just over 90dB. So part of HDCD claims to be a menu of methods to 'expand'
the range of a CD.
This expansion is said it use two methods. One is to apply a nonlinear
curve to the max signal levels. In effect, squash the top 9dB or so into
3dB on a sample-by-sample basis. Bit like the old 'mu-law' and 'A-law'
nonlinear sampling of methods used in comms, etc, to get a wider range from
a limited number of bits per sample.
In this case the general result on a *non* HDCD player is to squash any
peaks, distort sustained loud sounds, and lift the quieter parts by about
6dB. i.e. pretty much like the kind of idiotic 'loudness' treatments
inflicted by those obsessed with 'louder is better'.
FWIW I have a feeling some people making CDs view HDCD as a sort of 'up
market' version of plain old 'make it louder to sell better' with the bonus
that some audio enthusiasts will be drawn by the label.
On a genuine HDCD player, the player knows the details of the nonlinearity
applied, and can expand it. Ideally, that removes the distortion and level
compression. So you can argue that the result is 'compression for the
masses, but a decent result for the elite who buy into HDCD'.
However that only gets you about 6dB greater range - at the cost of
distortion for non-users. i.e. just one more bit per sample. 17bits, not
the claimed 20.
For long 'quiet' passages HDCD is also said to gradually wind up the gain
level, moving the sound further above any noise of sample-level errors.
This should improve the range for long quiet passages. And data 'encoded'
in the LSB patterns tells an HDCD player how to correct this gain shift. So
correcting this adjustment as well.
That could provide the other (claimed) 13 bits per sample. i.e. if the gain
is wound up and down by up to 18dB during 'quiet parts' of the music.
The big *however* here is that pop/rock music hardly ever has passages like
those described in the HDCD info I've seen. This says that you need music
that stays below about -40dBFS for a long time for this to kick in. But
that isn't what most pop/rock/jazz is like.
And when I scan the pop/rock HDCDs I have, as yet I've not found the codes
or any sign of their spread in the LSB. (I'll keep looking, though.)
Of course, you can then point out that if we really need 120dB ranges then
we might not want to lose the LSB to use as a control channel. :-) The
documents claim we can't hear this. But if so, why do we need such a big
range? And how many domestic situations have any hope of hearing over a
120dB range? Most homes have a background noise of around 30dBA or more,
and won't welcome having to play out at 150dBA. 8-]
And of course essentially *all* the older HDCD rock/pop/jazz material that
has been remastered would have been recorded on systems with a range well
below 120dB anyway. You have to wonder what a 120dB range has to do with,
say, analogue Joni Mitchell tapes from the 1960s or 1970s.
In addition to all that, if you noise shape conventional LPCM CD you get an
audible available dynamic range that is somewhat higher than the basic 'bit
more than 90dB' value generally quoted by shifting quantisation noise to HF
where people don't hear low level noises so easily.
Now on to (B).
There the claims are fairly wooly. The argument is that the ADC and DAC
examine the incoming data and decide to adopt a given filter function from
a menu of options. For smooth music you might use a conventional time
symmetric filter. But for transient spikes you might go for a more 'join
the dots' one to avoid side ringing. This is then communicated to the DAC
via the LSB control channel.
The effect is said to be to let you mimic a system with a wider bandwidth
and make the result sound more like you'd been able to convey details above
22kHz. The snag is that a 'spikey' filter will alias in ways that the
decoder may not be truly able to correct as the information recorded
becomes ambiguous. (This isn't true for a filter that simply shifts the
ringing to after the peak. But what I've read doesn't seem to say this is
what HDCD uses.)
In some ways this is a bit like the 'spectral replication' trick in the
AAC+ (lossy) encoder which notes down how much HF was discarded and
makes a guess at what should replace it when the file is played.
(Essentially by assuming you put in 'harmonics' of the biggest tones in
what you kept.)
However I can't find any sign of this, or its control codes. So how much of
this is simply flim-flam I can't tell. At some point I'll look for excess
folded components near 22kHz as these would be symptoms of an attempt to
have a 'spiky' encode filter which tends to generate things at (22.05 - f)
kHz when fed an input at f kHz.
Hence the Patents, etc, make a number of sweeping claims about improved
dynamic range and bandwidth. But in practice I can't say that as yet I'm
confident that it does much more than give you an (optional) reversable
'soft limiting' to get about 6dB back.
Slainte,
Jim
--
Please use the address on the audiomisc page if you wish to email me.
Electronics
http://www.st-and.ac.uk/~www_pa/Scot...o/electron.htm
Armstrong Audio
http://www.audiomisc.co.uk/Armstrong/armstrong.html
Audio Misc
http://www.audiomisc.co.uk/index.html
HDCD
Hybrid Mode
