HDCD: an investigation

Discussion in 'Audio Hardware' started by SiriusB, Aug 23, 2007.

Thread Status:
Not open for further replies.
  1. SiriusB

    SiriusB New Member Thread Starter

    Location:
    New York
    At the turn of the millennium HDCD was touted as a multi-benefit audio technology -- not only was the A/D and D/A filtering supposed to be top-flight, but it also offered what might be called 'virtual 20 bit' playback of 16-bit CDs. The advent of native high-bit formats (e.g. 24 bit DVD-A) in music and now, in hi-rez video, has pretty much rendered HDCD obsolete, but there are still plenty of HDCDs out there and at least one player manufactere (Oppo) still includes HDCD decoding as a feature on its DVD players.

    And as HDCD encoding offers a number of options to the mastering engineer -- only some of which actually exploit the 'extra bits' -- I've always wondered which HDCDs used which options.

    Researching HDCD awhile back, I found

    this thread on

    hydrogenaudio.org where there was chatter about a new method of direct digitally capture of decoded HDCD output from Windows Media Player (WMP). Posters on the thread showed that indeed not all HDCDs are encoded alike...considerable 'peak extension' was seen for a Yes HDCD track ('Sound Chaser'), rather less for a Mahavishnu Orchestra track off 'THe Lost Trident Sessions' (see the HA link for specifics). Others reported that all HDCD did on their discs, was move track peak level downward by 6 dB.

    Very intrigued, I downloaded the WMP plugins mentioned in that thread and tried some rips myself, with the WMP HDCD decoding on and off (this is an option buried several layers deep in WMP). Comparing the resulting wavs , I've confirmed that HDCD effects can range from simply reducing a track's peak level by 6 dB, to adding almost 6 db of actual dynamic range to a track...with other options in between. I'll give examples here using Audition stats, starting from the most trivial effect to the most 'profound'


    Category One: Level Reduction Only. Examples: Beck 'Midnite Vultures', Van Halen '1984' If you rip these discs the usual way -- without HDCD decoding --then take a gander at the waveforms, you'll see some seriously 'bricky', highly compressed hot mastering. This is what your nonHDCD player's output stage is handed too -- and some players, that don't like lots of peaks up near 0 dBFS, won't like that too much. Unfortunately, even if you decode the discs first, you only get some relief -- all the decoding appears to do is lower the track level by 6 dB. That of course doesn't change the dynamic range. It will however reduce strain on players/systems that don't handle hot peaks well--the hottest peak now only hits -6dB FS before D/A converson. Notice too that the effective bit depth is only 17bits in the decoded version. This begs the question: what was the point of releasing these as HDCD??

    Code:
    SEXX LAWS
    
    undecoded
    Peak Amplitude:		-0.08	-0.09 dB
    Average RMS Power:		-11.14	-11.51 dB
    'dynamic range' (peak - avg):   11.06	11.42 dB
    Actual Bit Depth: 16 bits
    
    decoded
    Peak Amplitude: -6.1	-6.11
    Average RMS Power: -17.16	-17.54
    DR: 11.06	11.43
    Actual bit depth:  17 bits
     
    Gain in DR: 0	0.01

    Category Two: Level Reduction + Increased Bit Depth. Examples: King Crimson 30th Anniversary Series. The KC series isn't smashed to hell in the first place -- even undedcoded, the waveforms look good. As with the Beck and Van Halen discs, decoding these merely brings their level down by -6dB, but this time, the HDCD mastering has also increased the 'available' bit depth to 20 bits. Which is still pointless. The waveform is exactly the same,just quieter. Why add 'resolution' if it isn't used? I checked several Kc discs, including the the most recent in the series -- USA -- and they were all in this category.

    Code:
    LAMENT (from Starless & Bible Black)
    
    undecoded
    Peak Amplitude:		-0.1	-0.1
    Average RMS Power:		-13.2	-13.86
    DR:  15.6	15.26 
    Actual Bit Depth:		16 Bits
    
    decoded
    Peak Amplitude:	-6.92	-6.92
    Average RMS Power:	-22.52	-22.18
    DR:  15.6	15.26
    Actual Bit Depth:	20 Bits
    
    Gain in DR: 0	0
    Category Three: Level Reduction + Increased Bit Depth + Modest DR Increase. Example: Yes 'Big Generator'. Now we start getting into the use of 'peak extension' options, which extends the peaks and thus the dynamic range too (or at least the crest factor -- the difference between peak and RMS average level). The HDCD reduction in level is actually *necessary* here as a correlate of greater DR (it's why early CDs sometimes sounded quieter thena their LP counterparts...you were finally getting the real dynamic range of the source, without compression). This example is still very bricky-looking even after decoding --it hardly looks different at all, just 'smaller' -- so there hasn't been much peak extension (and level reduction is moderately less than 6dB). And while 4 more bits became 'available', less than 2 dB of dynamic range were actually gained. But things can be better!

    Code:
    BIG GENERATOR (title track)
    
    undecoded
    Peak Amplitude:		-0.38	-0.38
    Average RMS Power:		-10.86	-11.07
    DR:		10.48	10.69
    Actual Bit Depth:		16 Bits
    
    
    
    decoded
    Peak Amplitude:	-4.49	-4.49
    Average RMS Power:	-16.75	-16.96
    DR:	12.26	12.47
    Actual Bit Depth:		20 bits
    
    Gain in DR	1.78	1.78

    Category four: Level Reduction + Increased Bit Depth + Nice DR Increase. Examples: Yes 'Relayer', 'Tales', 'Yessongs'. 'Fragile' (and probably the rest of the first HDCD set); Joni Mitchell 'Mingus'. Now we're getting serious. The decoded waveforms look significantly different than the undecoded ones. Obvious peak extension, with several dBs increase in DR.


    Code:
    THE DRY CLEANER FROM DES MOINES (from 'Mingus')
    
    undecoded
    Peak Amplitude:		-0.11	-0.14
    Average RMS Power:		-16.51	-16
    DR:		16.4	15.86
    Actual Bit Depth:		16 Bits
    
    decoded
    Peak Amplitude:	-1.52	-1.95
    Average RMS Power:	-22.5	-22
    DR:	20.98	20.05
    Actual Bit Depth:	20 Bits	
    		
    Gain in DR:	4.58	4.19
    
    
    
    THE ANCIENT (from 'Tales from Topographic Oceans')
    
    undecoded
    Peak Amplitude:		-0.13	-0.09
    Total RMS Power:		-14.38	-13.36
    DR:		15.24	14.19
    Actual Bit Depth:		16 Bits
    
    decoded
    Peak Amplitude:	-1.72	-1.13
    Average RMS Power:	-21.38	-20.28
    DR:	19.66	19.15
    Actual Bit Depth:	20 Bits
    		
    Gain in DR:	4.42	4.96
    
    		
    			
    SIBERIAN KHATRU (from 'Yessongs')
    
    undecoded
    Peak Amplitude:		-0.06	-0.13
    Average RMS Power:		-11.26	-10.45
    DR:		11.2	10.32
    Actual Bit Depth:		16 Bits
    
    decoded
    Peak Amplitude:	-0.72	-1.83
    Average RMS Power:	-17.25	-16.43
    DR:	16.53	14.6
    Actual Bit Depth:	20 Bits	20 Bits
    	
    Gain in DR:	5.33	4.28
    
    
    SOUND CHASER (from 'Relayer')
    
    undecoded
    Peak Amplitude:		-0.14	-0.04
    Average RMS Power:		-12.57	-11.62
    Total RMS Power:		-11.7	-10.8
    DR:		12.43	11.58			
    Actual Bit Depth:		16 Bits
    
    decoded
    Peak Amplitude:	-1.93	-0.53
    Average RMS Power:	-18.56	-17.58
    DR:	16.63	17.05
    Actual Bit Depth:	20 Bits
    
    Gain in DR:	4.2	5.47
     
  2. SiriusB

    SiriusB New Member Thread Starter

    Location:
    New York
    Here's an example from Cat 1 -- Sexx Laws by Beck, from 'Midnite Vultures'
    Undecoded on top, decoded on bottom
     

    Attached Files:

  3. SiriusB

    SiriusB New Member Thread Starter

    Location:
    New York
    And now one from Cat 4 -- Siberian Khatru from 'Yessongs'
     

    Attached Files:

  4. Jeff Carney

    Jeff Carney Fan Of Specifics (No Koolaid)

    Location:
    SF
    Now if you can just decode the smiley-face EQ right out of those Yes HDCDs, we'll have something.

    Seriously, this is very interesting stuff. I will read it more attentively tomorrow.
     
  5. Onward

    Onward Forum Resident

    Location:
    Oslo, Norway
    Very interesting, thanks!

    Where does the extra data for the peak extension come from? AFAIK the extra data in a HDCD is encoded in the lsb, which can't contain enough information to restore peaks? (or can it?)

    Or are the peaks compressed during encoding and then "extended" during hdcd decoding? If so that would mean that the more of hdcd's features you use, the worse the cd will sound on a regular player (?).

    I haven't read the HA thread yet though...
     
  6. Dr. Merkwürdigli

    Dr. Merkwürdigli Active Member

    Location:
    Oslo, Norway
    Great work!:righton:

    If I understad this correct; HDCD equipment want make the King Crimson 30th Anniversary Series CDs sound any better.
     
  7. tot

    tot Forum Resident

    Location:
    Mougins, France
    That is how I understand it. I doubt the normal CD playback suffers too much since the instructions are in LSB and not in all samples.
     
  8. Capt Fongsby

    Capt Fongsby Music is the best. ... And cats.

    Location:
    Norway
    Nice job, SiriusB!

    As a result of both wanting a new CD player and curiousity about the HDCD format, I bought a NAD player w/HDCD capability last year, and I've been very happy with it. Given the sad state of SACD, I've found HDCD to be a more interesting format.

    Some members on this forum have consistently labeled HDCD as a pure hoax, which I found hard to believe, given my own experiences.

    So thanks for doing the investigation and presenting some results! :righton:
     
  9. tas

    tas Forum Resident

    Location:
    Canberra ACT
    I wish I could find the article I had back when HDCD was launched. The article indicated that the LSB contained a flag but the actual HDCD data was encoded into the unsed space in the CD standard - much like the short lived CD-Graphics.

    Anyone else know if this is so?
     
  10. tot

    tot Forum Resident

    Location:
    Mougins, France
    Someone somewhere reported that downloaded flac from HDCD also worked and illuminated HDCD indicator in the DAC (or whatever it was). That would mean that everything is in samples.
     
  11. Rick B.

    Rick B. Senior Member

    Location:
    Toronto
  12. Onward

    Onward Forum Resident

    Location:
    Oslo, Norway
    I just read the thread on Hydrogen Audio posted by SiriusB. Lots of good info on hdcd. I find it strange that there has been so little talk about the software hdcd decoder, thats a big step forward and means that the full dynamic range of those hdcd's with peak extension can be decoded and reencoded to flac or mp3's.

    I look forward to having decoded hdcd's on my ipod :righton:
     
  13. ElevatorSkyMovie

    ElevatorSkyMovie Senior Member

    Location:
    Oklahoma
    The David Lee Roth era Van Halen remasters are HDCD encoded. When I ran them through the WMP to decode and capture them a few months ago, I was very surprised (and unhappy) to see that it only dropped the level by 6db, just like the Beck track above. I also decoded Lucinda William's "Car Wheels on a Gravel Road", and it was the same way.
     
  14. benintune

    benintune New Member

    Location:
    Plano, TX
    Do HDCD encoded discs that don't have any peak limiting when played without decoding benifit when played decoded? In other words does HDCD offer any other benefits other than peak extending?
     
  15. PMC7027

    PMC7027 Forum Hall Of Fame

    Yes, the HDCD chip has various filter settings which are changed "on the fly" by the HDCD codes. The differences in the filters are designed to maximize performance of different parameters, such as impulse response, frequency response, phase linearity, etc. The HDCD encoding process determines when each filter will provide the best sonic performance and the control codes to select that filter are included in the HDCD data stream.
     
  16. jdmack

    jdmack Forum Resident

    Location:
    Silver Spring, MD
    How is the decoding accomplished? If I put a HDCD in my computer, and use Windows Media Player 11 to rip the disc, will it automatically rip the audio as 20-bit decoded HDCD, or do I have to make a settings adjustment?

    J. D.
     
  17. benintune

    benintune New Member

    Location:
    Plano, TX
    Ok, thanks. I didn't understand all that but I get the idea.

    I just ripped Tom Petty's DTTP HDCD decoded using that plug in and it is brick walled but at -6db. So, I guess I get some benefit from HDCD but too bad they didn't extend the peaks in the encoding.
     
  18. benintune

    benintune New Member

    Location:
    Plano, TX
    I'm not sure how WMP 11 handles it but in WMP 9 it only decodes when playing the cd. If you rip it in wav or a lossless codec you can burn a cd and also play that cd and it will decode that. I haven't been able to play an HDCD file on my computer with the HDCD decoding.
     
  19. tps

    tps Forum Resident

    Location:
    Philadelphia, PA
    Another HDCD thread... Well, if you read the Euphonix manual, you find that:

    1. The peak extension is done by reversing the action of a peak limiter in the HDCD encoder. The characteristics of the peak limiter are fixed. That is, they are embedded in the encoder and decoder rather than transmitted by information on the disc. For peak extension, the information encoded on the disc is a simple On/Off signal. The HDCD peak limiting/extension compresses/expands the topmost 9 dB into 3 dB. Presumably it works best when the audio signal does not spend much time in that region. Discs which have already been "brick-walled" probably should not use peak extension.

    2. Low level enhancement is done by a companding algorithm that kicks in below a fairly low level, my memory is that it's in the -40 dBFS dBFS region. This is probably useful mainly for classical and certain jazz recordings. Most rock recordings don't have many stretches at that low a level. Once again, the companding characteristics are embedded in the encode/decode algorithm. All that is transmitted on the disc is an On/Off signal.

    3. The HDCD allows the encoding engineer to choose on-the-fly among a small number of digital interpolation filter coefficients that will be applied by the decoder. Once again, the coefficients are embedded into the decoder. The information on the disc is just a "switch" which selects.

    The HDCD patent and white papers explain that the HDCD switching signals are encoded using psuedo-random sequences into bursts of data which replace the bottom audio bit for brief periods. IIRC, they specify the use of maximal length pseudo-random sequences, modified in a manner that makes it easier for the decoder to achieve a "lock" with the sequence. But they are very careful not to specify the exact sequence anywhere. I've burned a few brain cells trying to make an educated guess at the parameters of the HDCD pseudo-random code, but with success so far. The first goal of my investigations was just to write an HDCD "detector" program which would scan my hard drive for HDCD encoded audio.

    BTW -- in my investigations I found some specifications for a device that, to my knowledge, has never been built: an HDCD decoder with DIGITAL output. Pacific Microsonics specifies that the digital output format of the decoder should be at least 88.2 KHz, 20-bit resolution to preserve the benefits of the decoding process.
     
  20. PMC7027

    PMC7027 Forum Hall Of Fame

    What kind of digital output do you mean? In my audio system I have a Theta Digital DSPro Gen III D/A processor with HDCD decoding. I can take the digital output of the D/A, burn it to CD or record it to DAT and the HDCD encoding is preserved.
     
  21. tps

    tps Forum Resident

    Location:
    Philadelphia, PA
    No, this would be an HDCD decoder with digital output, so the HDCD encoding would be decoded, not preserved, at the digital output.
     
  22. PMC7027

    PMC7027 Forum Hall Of Fame

    You are correct, I have never heard of an HDCD implementation that gives access to the digital data that comes out of the HDCD decoder before it goes into the D/A chip.

    That doesn't mean that someone familiar with digital audio, such as the many people who professionally modifiy CD and SACD players, couldn't tap into the digital data stream between the output of the HDCD chip and the input of the D/A chip.
     
  23. SiriusB

    SiriusB New Member Thread Starter

    Location:
    New York
    well...there's nothing about the 'expanded' waveforms that couldn't have been 'encoded' in a plain CD in the first place! The dynamic range is still well within CD spec (96dB) after all. So the actual 'need' for HDCD remains dubious, in my opinion, at least in regard to dynamic range.

    .
     
  24. SiriusB

    SiriusB New Member Thread Starter

    Location:
    New York
    Yes, it is. As long as your flac file is made from a bit-perfect rip, the HDCD flags are preserved and read by an outboard HDCD DAC. As best I can tell, though, Windows Media Player won't recognize them unless they're read from a CD drive-- so you'd have to decode the flac to wav and burn it to a disc, to use WMP for decoding. I'm still investigating this.
     
  25. SiriusB

    SiriusB New Member Thread Starter

    Location:
    New York
    Yes, I mention this in the first post of the thread. Admittedly a rather longwinded post, though.

    :D
     
Thread Status:
Not open for further replies.

Share This Page

molar-endocrine