HDCD technology in detales.
Dec 22, 2005 at 10:27 AM Thread Starter Post #1 of 9

_Ilya_

New Head-Fier
Joined
Dec 22, 2005
Posts
3
Likes
10
Hi, Guys.

Recently I investigated all information about HDCD, including data sheets for HDCD processors. Seems that HDCD operates as follows:

Data range from -30000 : 30000 is linear. Range above and below is non linear. Multiplier coefficients are recorded in sub-channel (part of audio CD format). Sub channel is not big enough to keep additional 4 bits, but it can keep multiplication coefficient.

In other words HDCD data format is a 2 pieswise interpolation to logarithmic scale. Full range is appropriate to 20 bit. Data precision at any point is 16 bit. (technology similar to voice A-law, M-law compression).

HDCD played on standard CD equipment will distort high levels.
HDCD played on separate transport and processor may be distorted because if sub channel is not transmitted, but processor is HDCD capable, processor will try to apply HDCD decoding because it will find HDCD patterns in upper levels, but without sub channel it does not know how to amplify it. The result is floating level (The same may be achieved if you copy HDCD using NERO but do not copy sub channel).
HDCD can not be saved in .wav because sub channel will be lost. HDCD processor will find patterns in peaks, led will flash.

In general HDCD is a clever solution, giving 20 bit dynamic range with 16 bit precision. Good thing it is not multichannel. I have a very good stereo but I do not know how to listen SACD when it is 5.1...

Why HDCD sounds better ? -
Only because producing labels do not apply compression ! If you simply record LP using 20 bit ADC you will find it can not be recorded on 16 bit CD !!! LP data range is about 20 bit and data recorded linear. After compression you get specific "CD" sound. In this terms all formats giving 20 and more bits dynamic range are all the same ! Data goes linear and no distortions. All other differences are negligible. DAC specific, processing... all this does not matter, actually.

SACD, HDCD, DVD-a - all of them seems to have audio material like it was originally recorded, ie linear.

Regards.
 
Dec 22, 2005 at 4:52 PM Post #2 of 9

Canman

Headphoneus Supremus
Joined
May 13, 2003
Posts
3,667
Likes
14
Quote:

Originally Posted by _Ilya_
Hi, Guys.
HDCD played on standard CD equipment will distort high levels.



My recordings that were mastered in HDCD are all among the best sounding CD's in my collection. This is without a CD player with an HDCD processor. I think it is due to the careful sound engineering/mastering.

Quote:

Why HDCD sounds better ? -
Only because producing labels do not apply compression ! If you simply record LP using 20 bit ADC you will find it can not be recorded on 16 bit CD !!! LP data range is about 20 bit and data recorded linear. After compression you get specific "CD" sound. In this terms all formats giving 20 and more bits dynamic range are all the same ! Data goes linear and no distortions. All other differences are negligible. DAC specific, processing... all this does not matter, actually.


From this perspective, if an analog signal was recorded digitally at 16 bit word length, there would be no compression necessary to transfer the data to CD.

It is taken for granted, however, that recording an analog signal at a higher word length and properly dithering to 16 bit results in superior dynamic range.
 
Dec 22, 2005 at 5:45 PM Post #3 of 9

_Ilya_

New Head-Fier
Joined
Dec 22, 2005
Posts
3
Likes
10
Usually yes, but if you campare as they sounds on HDCD player, you will find that standard player does not reproduce peaks in HDCD recording. But due to material was recorded without compression and peaks are only 5-10% from all recording time, you feel it sounds better, and it is, HDCD codes at peacks where choosen so they behave like soft limiter... It is a trick ^) You hear HDCD record on sandard player like it is high dynamic record processed by soft limiter to limit it by 16 bit, but on HDCD equipment you have all peaks reproduced.

Normally high dynamic original record is processed by compressor, and it is the question what is better... You can try to play in Cool editor with compression and limiting to understand what I mean...

If analog tapes would be digitized by 16 bit the quantizing noise would be terrible...

btw - dithering is a process of rounding last 1 or last 2 bits, it is not for reducing from 20 to 16...
 
Dec 22, 2005 at 6:13 PM Post #4 of 9

moj0

100+ Head-Fier
Joined
Sep 29, 2005
Posts
125
Likes
11
so.. listen to any hdcds lately?
 
Dec 22, 2005 at 8:27 PM Post #5 of 9

jpr703

Headphoneus Supremus
Joined
Sep 24, 2004
Posts
2,733
Likes
14
I have a handful of HDCDs in my collection. One that I've compared directly against it's non HDCD counterpart is Workingman's Dead. On my CD player, a NAD c542 which does support the HDCD format, the HDCD version sounds much better.
 
Dec 23, 2005 at 9:35 AM Post #6 of 9

_Ilya_

New Head-Fier
Joined
Dec 22, 2005
Posts
3
Likes
10
Sometimes HDCD on standard player is even better then non HDCD version on standard player, but this is not the rule, DOORS legacy edition for example sounds worse on standard CD player then standard records, but on HDCD player it is perfect.

I do not trying to say that HDCD is not better, it is, I just tried to explain how it works, when you know it you will not wonder that hdcd disk can not be played on transport+processor even if processor is supporting HDCD, or why you can not record .wav from HDCD keeping HDCD encoding... All other surprises becomes clear.
 
Jan 20, 2015 at 6:56 PM Post #7 of 9

wondras

New Head-Fier
Joined
Jan 20, 2015
Posts
2
Likes
10
[Moderator note: Before everyone reports the post, I realise this post is a necro-bump, but as the OP told me he was searching for SACD info and kept finding people referencing this thread, I'm allowing it.]
 
Quote:
 Multiplier coefficients are recorded in sub-channel (part of audio CD format). Sub channel is not big enough to keep additional 4 bits, but it can keep multiplication coefficient.

 
I'm fairly certain that HDCD does NOT use sub-channel information for coefficients or anything else. Everything it does in the 16-bit PCM data.
 
Apologies for resurrecting such an ancient thread, but it comes up frequently when searching for HDCD info, so I think it's important to be accurate. I, too, have read through various technical docs and datasheets about HDCD, but I can find no indication that it uses sub-channel data, and what I do find points to the opposite conclusion.
 
The best technical description of the process I've found is from the now-defunct HDCD.com website:
http://web.archive.org/web/20050404101503/http://www.hdcd.com/partners/proaudio/AES_Paper.pdf
 
Quote 1:
All parameters and processing choices made during the encoding phase are inserted into the data stream as pseudo-random noise encrypted control signals inserted into the least significant bit of the audio data as part of the dither, on an as-needed basis. The encrypted control signal allows the decoder to apply accurately timed conjugate processes during playback without using media format dependent sub-codes

Quote 2:
Using the LSB of the program to carry the side channel commands has several advantages. First, it is available without regard to which medium the audio is stored on or transmitted over, as long as the data is preserved. Also, the command data automatically follows the audio data from one storage medium to another over a digital audio transmission medium, without requiring special equipment which "knows about" the commands.

 
 
Also of interest is the datasheet for the PMD100 HDCD decoder chip:
http://www.datasheetarchive.com/dlmain/Datasheets-319/81124.pdf
 
The data input timing diagram shows that it only expects the PCM audio data, MSB-to-LSB. There is no provision for additional sub channel data. It does have commands to set an attenuation amount, but these appear to be intended for "soft-mute" and gain matching rather than dynamic control by the program material. "The delay between an attenuation command and the first sample it affects can be between 84 and 100 input samples." This doesn't sound like something you'd use for instantaneous adjustments.
 
 
Anyway...  All of this leads me to believe that the HDCD data stored entirely within the 16-bit PCM data, sub-channel data is not needed, and a lossless rip to WAV/FLAC can be fully decoded later on.
 
Am I missing something?
 
Jan 23, 2015 at 6:09 PM Post #9 of 9

wondras

New Head-Fier
Joined
Jan 20, 2015
Posts
2
Likes
10
   
Excuse me for this obvious question, but are you familiar with the foobar HDCD component, which has made this possible since 2010?


Yes, though personally I use CueTools to create decoded FLAC files that can be played anywhere. My motivation for posting was that the (mis-)information in this thread continues to propagate in posts like this one: http://forum.doom9.org/showthread.php?p=1612018#post1612018
 
There doesn't seem to be a definitive page or discussion on HDCD. I'm not trying to start one here, though...  (insert dead horse smiley)
 

Users who are viewing this thread

Top