Why Does High Resolution Sound Better Than CD?
I took this quote from the opening of the introduction page of HighResAudio.com, the German digital music download site that features most of the tracks offered through their license deals with the big three labels. You might think that we’d be way past this very basic question by now, but in reality we’re not. Because the answer to the question is not clear.
Just today, I received a brief email from a reader. He was having some difficulty downloading the files I prepared for the research project undertaken a couple of weeks ago. After we managed to get him the files, he responded:
“Thanks for teaching me about FTP downloading. I was clueless. Now that I have had the opportunity to listen to the files I would like to add a comment.
A number of your posts have claimed that it was an unfair test to compare MP3 – CD – HiRez when the original source material was from standard resolution sources. I believe it is also an unfair test to compare MP3 – CD – HiRez when the source material is sourced from well recorded true HiRez sources. I can only say that I have never heard MP3 sound so good.
What I did hear in the first set was that the low level cymbals tended to blended into the background in the file that I believed was an MP3 file.
It appears that the true HiRez digital resolution of the source might be more important than the delivered file resolution.
Perhaps the best that can be would be audio that is recorded, mixed and processed at 192khz and then down converted to 96khz for distribution.”
The point that Mike made was that the fidelity of the source material is a major factor in the quality of the delivery files no matter the format. He’s right and he’s not the first person to point that out. I remember a phone call years ago from a guy that was over the top about the sound quality of the AIX recordings he heard…and it turned out that he was listening to the Dolby Digital versions because he hadn’t yet acquired a DVD-Audio player. The quality of sound was there in spite of the lossy format.
The fact is High-Resolution Audio doesn’t always sound better than CD spec audio. It has the potential to be better but only if the source recording contains more stuff than a CD can deliver.
The answered offered up on the HighResAudio site cites three major factors to explain why HRA sound better than CDs. I’m going to focus on the first one and then return to the others in subsequent posts. They say:
“1.) The CD offers a resolution of 16 Bit and therefore 65,536 discrete steps to digitize an analogue music signal. That limitation causes distortions especially audible as noise within quiet parts of the music. That quantization noise is always present and comparable with the noise of a tape or LP but because of its digital nature more distracting.”
And here’s the solution to this very poor description of quantization errors.
“1.) Modern electronics is able to reach a dynamic range of around 120 dB by choosing higher bitrates of 24 Bit. The quantization noise is negligible because of the huge amount of quantization steps available.”
The number of discrete levels offered by 16-bits is 65,535 (I start counting from 0 so I get one less than they did). With a properly dithered input signal, the “distortions” they claim occur are really just very low level noise spread across the entire frequency spectrum. I usually don’t call “noise” distortion because it doesn’t affect the actual signals just the background against which we experience the music. And the noise of a 16-bit system is over 90 dB below the maximum level…much lower than the noise of an analog tape or LP (which hover around 50-60 dB below the signals. And in direct contradiction to their reasoning, the noise is virtually impossible to hear unlike the distractions presented by analog tape or vinyl LPs.
As to their solution, modern engineering can capture dynamics using 24-bit words and reduce the noise floor even further. I’m not sure what “quantization steps” are but I believe they’re referring to the increase in the number of amplitude levels available using 24-bits. The quantization noise is still there but the noise floor is much, much lower than that produced by a CD.
But it’s all for nothing if the sources that you’re capturing in 24-bit PCM audio were made from analog recordings. Their sources don’t need the extended number of bits in high-resolution audio because they only use about 10-12 bits.
More coming.
+++++++++++++++++++
I’m still looking to raise the $3700 needed to fund a booth at the 2015 International CES. I’ve received some very generous contributions but still need to raise additional funds (I’ve received about $3500 so far). Please consider contributing any amount. I write these posts everyday in the hopes that readers will benefit from my network, knowledge and experience. I hope you consider them worth a few dollars. You can get additional information at my post of December 2, 2014. Thanks.
Käre Mark,
I COULD NOT AGREE MORE: A number of your posts have claimed that it was an unfair test to compare MP3 – CD – HiRez when the original source material was from standard resolution sources. I believe it is also an unfair test to compare MP3 – CD – HiRez when the source material is sourced from well recorded true HiRez sources. I can only say that I have never heard MP3 sound so good.###
It really does come down to the original source recording. If there is tremendous fidelity and great sound at this point, the delivery format is less important.
So, why the digital sound is still not indistinguishable from the natural sound ?
This is all about timing between the samples. A human unconsciously detects time periods at the very least 2 microseconds, while DXD resolves only 2.6 microseconds. Hence, 768 kHz audio is already a necessity.
The better the sampling {natural or artificial, unimportant}, the more cohesive & consequently more embossed the sound{s}.
A few days ago I took an old gratis CD {i.e. 44.1/16/2}, made several required improvements upon a track & played it through a hardware intrinsically incapable of HD & what was heard simply stunned: the singer’s voice was awfully natural sounding, practically indistinguishable from that you can hear on live performance, thereby, with all the record studio dynamics of ~120dB.
I can’t really follow your thoughts…96 kHz/24-bits is plenty.
Mark, as I’ve mentioned before, my background is EE but I’ve never worked in the audio field. Even so, after reading your daily postings for more than a year now, the terminology is becoming second nature to me.
And yet, you surprised me today by you statement that “It appears that the true HiRez digital resolution of the source might be more important than the delivered file resolution.”
It was an Aha moment and I had to agree that even if I can’t easily distinguish between the different sampling rates of your files, they all sound better than some of the mp3 files I’ve downloaded from Amazon, even though they are 320k. It has to be the source that makes the difference and that really does make sense.
Maybe you should emphasize even more how important the source is and concentrate less on trying to convince people that the container format or the sampling rate is what they should be looking at. Yes, I know that this is what you have been saying in essence, but it was only by reading between the lines, so to speak, that I really got it.
Keep up the good work.
I’ve tried to make the point that the source recording is what matters…AND what defines the potential fidelity and ultimately the resolution of the delivered file. I’m glad you’ve manage to follow things and get the essence of the message.
Thinking that quantization noise is what is depicted, intentionally or unintentionally, in pseudoscientific or marketing material as small ladder steps in the frequency diagram is a common error. This leads people who don’t understand digital signal procesing to think that more samples would mean less quantization noise which is incorrect. More samples just means a hgher frequency captured. Quantization noise is manifested exactly as its name implies, as small distortions correlated with the signal not ladder steps. Dithering removes this correlation as you well know and makes this noise random. Noise shaping moves this random noise to freqs that don’t matter.
Confusion reigns in high-end audio in favor of marketing messages or information from people that just don’t know what they’re talking about. I don’t claim to know everything…but I do know people that provide insights when I’m unclear…including from knowledgeable readers.
Just take a close look at this :
http://sharp-world.com/contents/1-bit/benefits/high-resolusion/index.html
This website page is just plain wrong on so many levels.
A prime example of totally misleading marketing material such as I was refering to in my previous comment. Engineers laugh at the stuff their companies use as promo material. This is absolutely incorrect. People should try to read some basic material on digital signal processing rather than marketing stuff. One cannot understand the issue by looking at marketing material or reading the audio press.. Many good introductory papers have been linked to in these pages. Hoping that basic digital processing theory becomes standard high school material. Then we might be saved…
There is a lot of good information out there. But the marketing people tend to spin things to their advantage…I understand their motivation. But when reviewers, editors, and bloggers get on board with the spin and misinformation, I have difficulty. The whole Synergistic Research “Atomosphere” video is nothing short of embarrassing…but it generates interest. It was a featured products at a recent audiophile meeting here in LA.
Isn’t level 0 one of 65,536 available levels going from 0 to 65,565?
Yes, exactly.
* Upsampled MP3 — incredibly sweet {embossed toy sounds which you can even touch by hand}
* Upsampled CD — literally blows your ears away {LIVE performance}
** DXD/DVD-A are nowhere near {laughable sounding}
Not true.
Mark wrote: “The number of discrete levels offered by 16-bits is 65,535 (I start counting from 0 so I get one less than they did)”.
Shouldn’t you get one more: 2^16=65536 plus the zero which can’t be expressed as an exponent?
Dave, in computer memory counting begins with the value of 0 not 1…as it does when counting with your fingers. So whenever I teach my students about powers of 2, I refer to 65535 as the maximum represented value not 65536 (which cannot be represented without going to another bit). For example, 7 in decimal is 111 in binary. But the number of available value is 8 because the first value is 000.
Mark wrote: “For example, 7 in decimal is 111 in binary”
No it’s not 7 in binary is 2^0 +2^1 +2^2, ie 011.
Dave…111 in binary equals 7 in decimal. Your calculation is wrong. 011 binary equals 3 in decimal. 2^2 + 2^1 + 2^0 is 4+2+1 which equals 7. Are you with me?
Doh, what am I talking about, 7 is 1×2^0+1×2^1+1^2^2, ie 111 (and it’s the other way round 1×2^2+1×2*1+1×2^0); I’d forgot my basic UNIX; I was looking at the exponent, not the multiplier. Nontheless, 2^16 in binary is 10000000000000000 whereas 65535 is 01111111111111111.
You know there really should be an edit button on here, once “post comment” is hit typos are set in stone. For the record 7 = 1 x 2^0 + 1 x 2^1 +1 x 2^2. Thank you.
I think we’ve got it now Dave. Thanks.
Hi Mark
I also had trouble with cyberduck and could not download the files , so could not do the test.
cheers Lance
The problem that most people have with the FTP site and Cyberduck is that they click on the ftp://buckeye.dreamhost.com link instead of entering the words “buckeye.dreamhost.com” into the FTP site field.