Post by KeithL on Sept 28, 2014 1:46:48 GMT -5
We receive a significant number of questions about HDCD, and there seems to be a lot of confusion about what it is and what it does. I'm going to start this thread with a brief summary of the technology behind HDCD.
(Incidentally, our ERC-3, and our previous CD players, support decoding of HDCD; the XDA-1, XDA-2, and DC-1 do not.)
OK.... on to the good part.
HDCD was developed at a point in history where CDs were virtually the only form of digital audio available. As we all know, CDs store digital audio at a sample rate of 44.1k samples per second (per channel) and a bit depth of 16 bits - usually written as 16/44. The sample rate establishes a theoretical limit on frequency response - a CD cannot store any audio signal above about 20 kHz. The bit depth allows the digital audio stored on a CD to have a theoretical maximum dynamic range of about 96 dB. (Dynamic range is simply the range between the quietest sound you can record and the loudest - that's a slight oversimplification but it will do for our discussion here.)
The point here is that a Red Book Audio CD (a "standard" audio CD) is strictly limited to these values; unlike digital audio files, you can't change the sample rate or bit depth of an audio CD without making it... well... no longer a standard audio CD. So, if you were convinced that the frequency response or dynamic range of a CD wasn't quite good enough, you'd be sort of stuck... or maybe not...
There is a process called "dynamic range compression" - which is simply a fancy way of saying "automatically turn the volume up on the quiet parts and down on the loud parts". There are pretty much two broad categories of compression: you can compress the sound linearly (as the sound gets louder, you turn it down more), or you can do it based on a threshold (where you leave normal stuff alone, but turn the gain way down only on loud peaks to "crush" them). Both are used in the mastering process - although hopefully not to excess. That sort of compression is often referred to as "asymmetrical" - since it's done on the mastering side of things and you leave it that way... and it makes things so you can hear the quiet flute over your car engine, and the bass drums still don't shatter your windows or your eardrums.
However, you can also use dynamic range compression for something different. Let's say you actually had music that had more dynamic range than would "fit" on your recording. You could use a dynamic range compressor to reduce that dynamic range before you made the recording, then use an exactly opposite dynamic range UN-compressor (called a dynamic range expander) to restore the music to its former dynamic range when you played it back. If a single device did both parts of this, it was often referred to as a compander. This method was used widely for cassette tapes - which had a rather poor dynamic range and really needed the help. Dolby Noise Reduction (DNR) was simply a dynamic range compressor/expander (high frequencies were compressed before recording, and then expanded on playback), and was built into many high-end cassette recorders. The dBx Noise Reduction system was similar, although it operated equally on all frequencies.
While this idea actually works pretty well, it has a few weaknesses. The whole process relies on being able to change the gain applied to the signal in real time (you turn the volume up and down while the music is playing). Unfortunately, there are two requirements if you want to do this without adversely affecting the sound quality, and they conflict with each other. The circuitry has to respond quickly - otherwise, if the music gets louder suddenly, it might overload the system before the volume control is turned down far enough to stop it. However, changing the gain of a signal while it's playing has another name - modulation - and, if you want to avoid creating all sorts of nasty intermodulation distortion, you have to change the gain slowly enough that the side effects of this modulation fall outside the audio range. The usual "solution" to this conflicting set of requirements is to compromise by having the compressor (encoder) respond quickly to large changes in level (referred to as "fast attack time") - because a little IM distortion is certainly better than a real overload, but have it then return to normal more slowly (referred to as "slow release time") - to avoid producing any more IM distortion than necessary. A properly designed system actually works pretty well, but I wanted to stress the fact that there is always some alteration of the signal.
The other weakness is that systems like DNR and dBx rely on the level of the signal itself to determine their action. This means that, if the signal is altered between when it is compressed and when it is re-expanded, that alteration may be magnified by the process. This is the reason why cassette recorders with DNR had to be carefully calibrated; because the compressor/expander combination would double any errors in high frequency response caused by the recording itself. dBX, not being frequency dependent, avoided this weakness, but traded it for other issues. Obviously, this is not an issue for digital systems, where we can ensure that the signal remains unaltered between recording and playback.
OK, so what has any of this got to do with HDCD?
Simple, HDCD is a compander process (like DNR or dBx). In response to the suggestion that the 96 dB of dynamic range on a CD isn't quite good enough, the HDCD encoder applies dynamic range compression to the incoming audio, and the HDCD decoder applies the appropriate amount of range expansion to reverse the process and give you back the original signal.
There is, however, one important difference. With both DNR and dBx, the processor "follows" the level of the signal, and uses the level of the signal itself to decide what to do. At the encoder, when the signal gets louder, the encoder turns the gain down; at the decoder, when the signal gets louder, the decoder turns the gain up. That makes two points where the system has to make that compromise we talked about before, between fast response and intermodulation distortion, and increases the possibility that the encoder and decoder might not perfectly match each other - which means that the signal that comes out isn't exactly what we started out with. Wouldn't it be nice if, when the encoder did its thing, we could record how much it turned the gain up or down - then use that recorded control information to tell the decoder how to exactly undo what the encoder had done? That's exactly what the HDCD process does. Along with the music, the encoder records a "control track" which can then be used to control the decoder.
But... one of our requirements is that this thing will play on a regular CD player. This means that we have to somehow squeeze our control track onto our 16/44 disc along with the audio.... which would seem to be impossible. Well, the guys who invented HDCD found a way around that. The control track actually doesn't carry much information - just the occasional "gain up" or "gain down" command - so it doesn't need much bandwidth. What they did is to format that control track so it fits into a single bit (it is a low rate serial signal). They then applied some cool math and dithered that signal in such a way that, while it still carries the control information the decoder needs, if you listen to it, it just sounds like white noise (hiss). They then take that one bit control track and REPLACE the least significant bit in the audio signal with it. It's actually pretty ingenious.
In the end, if you play that HDCD encoded CD on a machine that can't decode it, you get 15 bits of audio (they stole one for the control signal), with about 6 dB more noise than you'd get on an ordinary CD, and the audio is also range compressed a little bit - by whatever the encoder decided to do. (Note that, contrary to some claims, the encoding process actually DEGRADES the signal a tiny bit if you play it back on an ordinary non-HDCD player; if it sounds better, it can only be because the encoder otherwise has very good sound quality. Any claims otherwise are... err... untrue.) However, if you play an encoded recording back on a machine that CAN decode HDCD, it "picks off" the control signal, decodes it, then uses the information in the control signal to decode the audio. In this case that means that the 15 bit audio contained on the disc is expanded by up to about 20 dB (whatever the control track says is necessary to reverse the encoding process). In the final analysis, you lose about 6 dB, but gain back about 20 dB, for a net increase in dynamic range of about 14 dB. (Again, note that we're talking about the ABILITY to record a signal with 14 dB more dynamic range; that's only a benefit if your source material has enough dynamic range for it to matter.)
To put that in context, a well recorded HDCD is about equivalent to a digital audio file recorded at 20/44.1 (it has more dynamic range than a CD, but less than a 24 bit file). I'll leave it to you to decide whether the upside of an approximately 14 dB improvement in dynamic range is more important than the downside of any artifacts or distortion that might arise from changing the gain up and down, but, even taking the most optimistic view, an HDCD is slightly better than a CD, but slightly WORSE than a 24/44 digital audio file. (In other words, as a way to "stretch" the capabilities of the CD audio format, it makes sense but, with current technology, it is pretty much rendered obsolete by 24 bit audio files.)
Now, there are a few more interesting bits of information you should know about HDCD....
The process actually includes two different types of range compression - although both are "handled" by the control track. The original spec also included an option for the recording engineer to specify several different playback filters (which would alter the way the decoded audio sounded slightly); however, from all information I have been able to find, NO equipment ever utilized that option.
Architecturally, the decoding process has always traditionally been performed in the digital filter of the DAC (the company that originally created the technology designed digital filters). Since HDCD is only specifically used on CDs, the DACs in many CD players are able to decode HDCD, but that functionality is almost never available in a stand-alone DAC. BECAUSE OF WHERE THIS DECODING TAKES PLACE, THE ANALOG AUDIO COMING FROM THE CD PLAYER IS DECODED; AND THE DIGITAL AUDIO COMING FROM THE DIGITAL OUTPUTS IS *NOT* DECODED.
(Back when many DACs used external digital filters, the digital filters used in many DACs did support HDCD - because the most popular vendor of digital filters was also the company that invented HDCD - Pacific Microsonics - and their filters supported it. Since most DACs now include internal digital filters, and those original parts are long since discontinued, the functionality is now limited to DACs that are specifically designed to support HDCD - the ones in CD players and many pre/pros.)
Some CD audio player software can decode HDCD, including Windows Media Player (Microsoft currently owns HDCD). For those who RIP their CDs, some current CD rippers have the ability to decode the HDCD encoding in software; they then store the "pre-decoded" HDCD audio as a 24 bit audio file (24/44), which can then be played back "properly" on any standard DAC. At that point it's just a "normal" 24/44 digital audio file. My favorite ripping software is dBPowerAmp, which can do this (you have to install a free plugin); there are others.
As a final note, the original studio unit used to encode HDCDs was a combination analog-to-digital converter and encoder. Along with supporting HDCD, this unit has a long standing reputation for sounding very good, and was (and still is) quite popular for its audio quality. Perhaps because of this, HDCD encoded discs are still being released - some labelled as such and some not. (Since some discs that indicate as being HDCD don't seem to sound very different when you decode them, some folks conjecture that, on at least some of them, no encoding has actually been applied "but the switch was left on" and so they still carry the "encoded" flag. Note that it is also possible for a given CD to have a mix of HDCD and non-HDCD tracks.)
In summary, HDCD was a cool technology, which has really been rendered obsolete by the availability of 24 bit audio files - and DACs that play them. However, if you have any HDCD encoded CDs, and you really want them to be decoded properly, you'll need either a CD player or other device that decodes HDCD - or you'll have to use a ripper that does the decoding for you.
(Incidentally, our ERC-3, and our previous CD players, support decoding of HDCD; the XDA-1, XDA-2, and DC-1 do not.)
OK.... on to the good part.
HDCD was developed at a point in history where CDs were virtually the only form of digital audio available. As we all know, CDs store digital audio at a sample rate of 44.1k samples per second (per channel) and a bit depth of 16 bits - usually written as 16/44. The sample rate establishes a theoretical limit on frequency response - a CD cannot store any audio signal above about 20 kHz. The bit depth allows the digital audio stored on a CD to have a theoretical maximum dynamic range of about 96 dB. (Dynamic range is simply the range between the quietest sound you can record and the loudest - that's a slight oversimplification but it will do for our discussion here.)
The point here is that a Red Book Audio CD (a "standard" audio CD) is strictly limited to these values; unlike digital audio files, you can't change the sample rate or bit depth of an audio CD without making it... well... no longer a standard audio CD. So, if you were convinced that the frequency response or dynamic range of a CD wasn't quite good enough, you'd be sort of stuck... or maybe not...
There is a process called "dynamic range compression" - which is simply a fancy way of saying "automatically turn the volume up on the quiet parts and down on the loud parts". There are pretty much two broad categories of compression: you can compress the sound linearly (as the sound gets louder, you turn it down more), or you can do it based on a threshold (where you leave normal stuff alone, but turn the gain way down only on loud peaks to "crush" them). Both are used in the mastering process - although hopefully not to excess. That sort of compression is often referred to as "asymmetrical" - since it's done on the mastering side of things and you leave it that way... and it makes things so you can hear the quiet flute over your car engine, and the bass drums still don't shatter your windows or your eardrums.
However, you can also use dynamic range compression for something different. Let's say you actually had music that had more dynamic range than would "fit" on your recording. You could use a dynamic range compressor to reduce that dynamic range before you made the recording, then use an exactly opposite dynamic range UN-compressor (called a dynamic range expander) to restore the music to its former dynamic range when you played it back. If a single device did both parts of this, it was often referred to as a compander. This method was used widely for cassette tapes - which had a rather poor dynamic range and really needed the help. Dolby Noise Reduction (DNR) was simply a dynamic range compressor/expander (high frequencies were compressed before recording, and then expanded on playback), and was built into many high-end cassette recorders. The dBx Noise Reduction system was similar, although it operated equally on all frequencies.
While this idea actually works pretty well, it has a few weaknesses. The whole process relies on being able to change the gain applied to the signal in real time (you turn the volume up and down while the music is playing). Unfortunately, there are two requirements if you want to do this without adversely affecting the sound quality, and they conflict with each other. The circuitry has to respond quickly - otherwise, if the music gets louder suddenly, it might overload the system before the volume control is turned down far enough to stop it. However, changing the gain of a signal while it's playing has another name - modulation - and, if you want to avoid creating all sorts of nasty intermodulation distortion, you have to change the gain slowly enough that the side effects of this modulation fall outside the audio range. The usual "solution" to this conflicting set of requirements is to compromise by having the compressor (encoder) respond quickly to large changes in level (referred to as "fast attack time") - because a little IM distortion is certainly better than a real overload, but have it then return to normal more slowly (referred to as "slow release time") - to avoid producing any more IM distortion than necessary. A properly designed system actually works pretty well, but I wanted to stress the fact that there is always some alteration of the signal.
The other weakness is that systems like DNR and dBx rely on the level of the signal itself to determine their action. This means that, if the signal is altered between when it is compressed and when it is re-expanded, that alteration may be magnified by the process. This is the reason why cassette recorders with DNR had to be carefully calibrated; because the compressor/expander combination would double any errors in high frequency response caused by the recording itself. dBX, not being frequency dependent, avoided this weakness, but traded it for other issues. Obviously, this is not an issue for digital systems, where we can ensure that the signal remains unaltered between recording and playback.
OK, so what has any of this got to do with HDCD?
Simple, HDCD is a compander process (like DNR or dBx). In response to the suggestion that the 96 dB of dynamic range on a CD isn't quite good enough, the HDCD encoder applies dynamic range compression to the incoming audio, and the HDCD decoder applies the appropriate amount of range expansion to reverse the process and give you back the original signal.
There is, however, one important difference. With both DNR and dBx, the processor "follows" the level of the signal, and uses the level of the signal itself to decide what to do. At the encoder, when the signal gets louder, the encoder turns the gain down; at the decoder, when the signal gets louder, the decoder turns the gain up. That makes two points where the system has to make that compromise we talked about before, between fast response and intermodulation distortion, and increases the possibility that the encoder and decoder might not perfectly match each other - which means that the signal that comes out isn't exactly what we started out with. Wouldn't it be nice if, when the encoder did its thing, we could record how much it turned the gain up or down - then use that recorded control information to tell the decoder how to exactly undo what the encoder had done? That's exactly what the HDCD process does. Along with the music, the encoder records a "control track" which can then be used to control the decoder.
But... one of our requirements is that this thing will play on a regular CD player. This means that we have to somehow squeeze our control track onto our 16/44 disc along with the audio.... which would seem to be impossible. Well, the guys who invented HDCD found a way around that. The control track actually doesn't carry much information - just the occasional "gain up" or "gain down" command - so it doesn't need much bandwidth. What they did is to format that control track so it fits into a single bit (it is a low rate serial signal). They then applied some cool math and dithered that signal in such a way that, while it still carries the control information the decoder needs, if you listen to it, it just sounds like white noise (hiss). They then take that one bit control track and REPLACE the least significant bit in the audio signal with it. It's actually pretty ingenious.
In the end, if you play that HDCD encoded CD on a machine that can't decode it, you get 15 bits of audio (they stole one for the control signal), with about 6 dB more noise than you'd get on an ordinary CD, and the audio is also range compressed a little bit - by whatever the encoder decided to do. (Note that, contrary to some claims, the encoding process actually DEGRADES the signal a tiny bit if you play it back on an ordinary non-HDCD player; if it sounds better, it can only be because the encoder otherwise has very good sound quality. Any claims otherwise are... err... untrue.) However, if you play an encoded recording back on a machine that CAN decode HDCD, it "picks off" the control signal, decodes it, then uses the information in the control signal to decode the audio. In this case that means that the 15 bit audio contained on the disc is expanded by up to about 20 dB (whatever the control track says is necessary to reverse the encoding process). In the final analysis, you lose about 6 dB, but gain back about 20 dB, for a net increase in dynamic range of about 14 dB. (Again, note that we're talking about the ABILITY to record a signal with 14 dB more dynamic range; that's only a benefit if your source material has enough dynamic range for it to matter.)
To put that in context, a well recorded HDCD is about equivalent to a digital audio file recorded at 20/44.1 (it has more dynamic range than a CD, but less than a 24 bit file). I'll leave it to you to decide whether the upside of an approximately 14 dB improvement in dynamic range is more important than the downside of any artifacts or distortion that might arise from changing the gain up and down, but, even taking the most optimistic view, an HDCD is slightly better than a CD, but slightly WORSE than a 24/44 digital audio file. (In other words, as a way to "stretch" the capabilities of the CD audio format, it makes sense but, with current technology, it is pretty much rendered obsolete by 24 bit audio files.)
Now, there are a few more interesting bits of information you should know about HDCD....
The process actually includes two different types of range compression - although both are "handled" by the control track. The original spec also included an option for the recording engineer to specify several different playback filters (which would alter the way the decoded audio sounded slightly); however, from all information I have been able to find, NO equipment ever utilized that option.
Architecturally, the decoding process has always traditionally been performed in the digital filter of the DAC (the company that originally created the technology designed digital filters). Since HDCD is only specifically used on CDs, the DACs in many CD players are able to decode HDCD, but that functionality is almost never available in a stand-alone DAC. BECAUSE OF WHERE THIS DECODING TAKES PLACE, THE ANALOG AUDIO COMING FROM THE CD PLAYER IS DECODED; AND THE DIGITAL AUDIO COMING FROM THE DIGITAL OUTPUTS IS *NOT* DECODED.
(Back when many DACs used external digital filters, the digital filters used in many DACs did support HDCD - because the most popular vendor of digital filters was also the company that invented HDCD - Pacific Microsonics - and their filters supported it. Since most DACs now include internal digital filters, and those original parts are long since discontinued, the functionality is now limited to DACs that are specifically designed to support HDCD - the ones in CD players and many pre/pros.)
Some CD audio player software can decode HDCD, including Windows Media Player (Microsoft currently owns HDCD). For those who RIP their CDs, some current CD rippers have the ability to decode the HDCD encoding in software; they then store the "pre-decoded" HDCD audio as a 24 bit audio file (24/44), which can then be played back "properly" on any standard DAC. At that point it's just a "normal" 24/44 digital audio file. My favorite ripping software is dBPowerAmp, which can do this (you have to install a free plugin); there are others.
As a final note, the original studio unit used to encode HDCDs was a combination analog-to-digital converter and encoder. Along with supporting HDCD, this unit has a long standing reputation for sounding very good, and was (and still is) quite popular for its audio quality. Perhaps because of this, HDCD encoded discs are still being released - some labelled as such and some not. (Since some discs that indicate as being HDCD don't seem to sound very different when you decode them, some folks conjecture that, on at least some of them, no encoding has actually been applied "but the switch was left on" and so they still carry the "encoded" flag. Note that it is also possible for a given CD to have a mix of HDCD and non-HDCD tracks.)
In summary, HDCD was a cool technology, which has really been rendered obsolete by the availability of 24 bit audio files - and DACs that play them. However, if you have any HDCD encoded CDs, and you really want them to be decoded properly, you'll need either a CD player or other device that decodes HDCD - or you'll have to use a ripper that does the decoding for you.
