Sample rates
Join Our Community Subscribe to Paul's PostsThere sure is a lot of confusion over sample rates. We hear about CD-quality sample rates at 44.1kHz (and its multiples), or another common sample rate, 48kHz (and its multiples), and then there are multiple higher sample rates (176kHz, 192kHz as examples) and of course DSD.
Lots of numbers. All very confusing.
Perhaps a short primer would help.
First, what is a sample rate? Simply put, it’s a snapshot of the audio signal. A slice of time where we capture the voltage level of the music signal. The number of times per second we take that snapshot determines the sample rate. (Bit depth determines the loudness range we can capture within each sample)
First, what’s the difference between 44.1kHz and 48kHz and why do the two exist? The former is what Sony/Philips set as a standard for the Compact disc. When we do higher sample rate versions of this standard we get 88.2kHz, 176kHz and so forth. The latter, 48kHz, is the standard the “pros” use (because, well, they can’t use something as conventional as consumers, now can they?). 48kHz gives us multiples we’re familiar with like 96kHz, and 192kHz.
What’s painful about the above two standards is the difficulty moving between them. When recording studios record at “pro” sample rates of 48kHz they then have to interpolate down a few Hz to 44.1kHz to make something we poor consumers can listen to.
Silliness.
When we nerds talk about sample rates we use different terminology. We base our discussion on how many multiples of the base frequency (44.1kHz) are in play. So, for instance, the CD sample rate is referred to as 1fs. Its multiples are 2fs, 3fs, etc.
The sampling frequency or sampling rate, fs, is the average number of samples obtained in one second (samples per second). Think of 1fs as the minimum baseline to capture 20Hz to 20kHz.
While we might be familiar with all the differing PCM sample rates, DSD brings in a whole other dimension with its far higher sample rates. For example, standard DSD is 64fs while double rate DSD is twice that at 128fs. So what’s that mean? Well, 1fs is running at 44,000 times per second, while 64fs is running at 64 times that frequency, or 2,822,400 times per second! That’s fast, man.
And, while DSD is so much higher of a sample rate as to raise a few eyebrows, it’s instructive to remember it’s a 1-bit system compared to a basic 16-bit system like PCM (remember that the number of bits is needed to measure amplitude). This boils down to something less hair raising if we do a bit of math. 64fs (1xDSD) runs at a very high clock rate of 2,822,400 Hz (2.8mHz). Now, simply divide that by 16 (the number of bits in a PCM word) and guess what you get? A sample rate of 176kHz. Sound familiar? 176kHz is the same as 4fs PCM. So, while PCM requires 16 bits to adequately measure amplitude, and DSD needs 16 single bits to do the same, it all kind of works out in the end. (Don’t take what I just wrote about DSD and 16 bits as literal. I use it only as a means of helping form a picture. DSD is far more complicated, using a Sigma-Delta Modulator, noise shaping, etc.)
Without getting too much more in the weeds, that’ll give you a brief simplistic overview of sampling rates.
The biggest silliness can be seen in the marketing campaigns for digital audio when the most clever marketing guys explained the sampling (digitization) process for a pure sine waves – instead using the steepest transients found in music (cymbals, triangles etc). This would have shown that it is impossible to reconstruct (the peaks of) the original music signal from a 44.1/48 kHz recording.
This is interesting Paul, & thank you for taking the time to break it down into ‘simple’ (basic maths) terms (American ‘math’; English ‘maths’)
For the Nerds I’m sure that it’s great stuff & I suspect that, in theory, the higher the sampling rate the better the reproduction…in theory.
However, as a listener who is well aware of the ‘law of diminishing returns’ I will have to defer to my tympanic membranes as to the difference in sampling rates.
I, like some others here, am quite content with 44.1/16, especially when the recording itself is very well engineered.
You just hit the jackpot!
“When the recording is good”.
That’s so well said, Both American English or British English.
I have some records- very old vinyl ,that probably, by mistake ,
That sounds terrific ! We know vinyl has a dynamic range of probably 60dB unlike it’s 16/24 bit cousins.
Sample rate is easy to explain and understand as long as it is
Defined and explained correctly!
All it is about is what occurs in a
Second of time!
Sounds again like the school problem discussed a few days ago. If it’s not explained with precision confusion reigns.
Now the question is can the same piece of music played by the same peeps in the same
Place at the same time sound different is the sample rate changes as a function of changing the bit value?
Not a complex design to test. However if done it might burst a bubble.
Champagne anyone?
Larry
Don’t forget Shannon’s1 theorem which says that you need to have a sampling rate of , at least, two times the bandwidth of the signal source. One reason for choosing higher sampling rates was to simplify the design of the filters. I try to remember the discussion regarding sampling rates in ITU CCIR SG 10 and 11 but it is a too long time ago.
So this is the simplistic overview…?
I guess I’m too simple 🙁
So….”if it all works out in the end” sort of boils down to “No matter where you go, there you are”. Now that make sense?
You were right Paul. There sure is a lot of confusion about sample rates.
Sampling rate has always been an interesting topic for me. I used to teach math. Digital recording requires an established standard for encoding and decoding from analog to digital and then back to analog. I we look at analog recording on a magnetic tape the signal is encoded onto the molecules of metal. How many molecules get magnetized each second at a recording speed of 7 1/2 inches per second? Would that be a sampling rate?
Is there an accepted or predicted error rate for these different digital recording rates?
Re. 1/4″ analog tape at 7.5 ips, it’s a little more complex than that, since it’s not just how many particles are magnetized, but how much each particle is magnetized, since it is an analog signal, not a digital one.
Paul, it would be helpful if you could carry the analysis one step further. Whatever sample rates were used to make the recording, presumably analog, being continuous, would theoretically encode a higher number of samples. However, playback is a different story. Digital playback preserves the recorded samples. Analog playback entails a stylus bouncing around in the grooves of the record, so my guess is that a fair number of samples never make it to playback. Am I on the right track?
hmmmm – yesterday’s topic was “teachers” and today is “math”
I like this home-schooling 😉
Aren’t there similar mathematics in the realm of video?
The higher the film speed or shutter speed, the clearer and more detailed the final image?
Paul – maybe a post relating the two?
Great stuff!!!
One of the reasons for choosing a 48K for “pro” users is that 48K is neatly divisible by the 24 and 30 frame per second rates that are commonly used in video and film production. Thus, digital video recorders could record an even 1600 or 2000 samples per frame.
Forgot to mention 25 FPS, which is common in European PAL video systems. 48K/25 FPS yields 1920 samples per frame.
I’m going to glue a nickle to the top of my tonearm.
and tin foil on your rabbit-ear antennae?
Yes! By the way, do you know how to make my VCR stop flashing 12:00 all the time?
Thanks Paul! I’m sending my readers over to you! 🙂 Do you want to tell the story of why 44.1/16 was used in the first place? or should I 😉 The world could have been fine with 48 but then “someone” stepped in. LOL.
By the way, if anyone is interested here… on January 15 Blue Coast Records is launching 29 titles on the major streaming services… yes, Tidal, Spotify, Apple, Amazon, Qobuz, etc…. I’m still a fan of downloads, but for some the lower rez technology is easier to manage.
Enjoy your music, everyone!
Cookie Marenco
Blue Coast Records and Music
https://bluecoastmusic.com/
Happy New Year Cookie!
Same to you, Fat Rat!
Cookie
What I am concerned more is the sample. Does the sample covers every bit of the musical information, leaving no gaps. Analogue copies leave no gaps. I think human ears are more sensitive than their eyes who cannot detect the gaps in movies whose films do not cover each and every bits of the movement. I suspect digital recordings are similar, having gaps in the information they record. Repeating the samples more frequently will not provide you with information they have lost. I don’t know if I am correct but suspect I am.
So here’s this consumer’s viewpoint. In the late 1990s Sony adopted a digitisation technique known for decades (DSD) to store and preserve analogue tapes. It was a simple 1-bit stream of data of very high frequency. Nice idea. If only it had stayed in its box.
Meanwhile, since the mid 1970s music had been recorded digitally by PCM, a sort of slicing and dicing, which was explained to the music listening public. One example is a Solti/Bruckner recording, where the entire inside sleeve was devoted to explaining how PCM works. https://www.sekaimon.com/itemdetail/174026983768 (if anyone wants to give me $50 for my copy, fine).
So DSD was like a single length of string with very compact data recorded on it and PCM is lots of parallel strings (a multiple of 8 strings) with the data stored less compactly.
When the music industry decided to make digital discs, a committee spent years deciding before arriving at the final CD format, which is enshrined as an international standard of 16 bits (strings) at a frequency of 44.1khz. It was called Red Book CD. And you could put it in your pocket.
Then the s**t hit the fan. Sony decided DSD was a good way of getting more data on a disc, particularly for 5-channel audio, and worst of all decided to make it a closed system that had to use their decoder. Then people decided to use it for higher density 2-channel music. Audiophiles loved it, some still do, the music buying public didn’t. So in a few years it largely died a death.
Had downloads and streaming come along a little earlier I suspect SACD would never have happened as a delivery format, or subsequently DSD. A bit like how MQA was designed to solve a streaming speed problem that was largely solved by network providers by the time MQA came to market.
It seems a sad quirk of fate that stored and streamed music came along a few years later as it solved the problem of the physical limitations of a CD. More data is easily provided by increasing bits and frequency used, and the hardware is really cheap. A $2 printer usb cable and a $100 DAC will do the job.
For reasons I cannot comprehend, DSD persisted to some degree as a delivery format, even though it was not intended for that and has a fundamental limitation in that the encoded music data cannot be edited. It’s great for listening to archived analogue tapes. So to use DSD, most of the music has been edited in a high density version of PCM (called DXD) just to confuse people further.
So PCM was easily understood by the serious music listening public who bought digitally mastered records in the 1970s and early 1980s, even before CD and streaming. All you needed was to like Georg Solti (which I did, both conducting and walking his dog).
The fact that Paul now has to explain PCM and DSD simply indicates that the whole format thing has got out of hand.
Sony were looking for a replacement for their CD cash cow soon to be out of copyright. they had decided on DSD to archive their vast catalogue of analogue recordings and decided to use this format as a new disc to market with the added benefit of 5.1 channels and a ‘higher quality’ with as you say a copyright control built in. All of this was vs the DVD Audio rival PCM based format – it was just VHS and Betamax all over again but in this instance nobody won because in the meantime the average consumer had bought an iPod
DSD has not died out because it actually makes a great deal sense in terms of a robust and accurate way of capturing an audio signal that can be stored and delivered using a digital network. It cannot be mixed and mastered digitally because it is not actually digital.
I was at an AES seminar in 1983 presented by engineers from Zenith Corporation, the same company that created the method the FCC adopted for stereophonic FM broadcasting. The reason that 44 khz was chosen for CDs and 48 khz was chosen for digital tape was the fear of direct digital pirating. In the analog domain every generation of copies is degraded from the previous one. In principle and in fact in the digital domain the thousandth generation is identical to the first. So to prevent first generation quality pirated copying a different sampling rate had to be chosen. I guess they never anticipated that it wouldn’t be long before recordable CD technology would become available to the public where they could burn a CD from another CD in the digital domain.
Ignorance of some presents opportunity for others. It takes two years of calculus before you can understand the Fourier transform mathematics and why sine waves can be used to understand very complex waveforms with exact precision. It takes another two years of mathematics and electrical engineering to understand information theory and how and why the Shannon-Nyquist analysis is correct. So oversampling on playback doesn’t add any information to a recording, it merely repeats the same information several times as though they were recorded at a much higher sampling rate to allow the analog anti-aliasing filter to be pushed out to a point far beyond human hearing where it has no audible effect. At 22 khz it creates major phase shift and FR problems in the audible frequency range. At 8 times 22 khz it doesn’t.
The only people who challenge what is now bedrock science, engineering, and mathematics in these areas are audiophiles. Dr. Mark Waldrep conducted a valid series of tests comparing RBCD to his own true HD recordings by downconverting HD files to RBCD files. This was a valid test because the same D/A converters could be used. Over 600 participants were used in the test. And what he learned to his chagrin was that I was right and he had been wrong all along. As a PhD in recording engineering he is forced to accept facts he doesn’t like. Whatever differences there are between HD and RBCD recordings are not attributable to the more capable technology of HD. They are attributable to other factors that have nothing to do with HD.
Waldrep has accepted he was wrong and has been very decent in doing so – I don’t like your tone here saying it was to “his chagrin”. It takes a strong man to accept that what he has spent a great deal of time and money in (building and running a multi -channel 96/24 bit studio) has not ended in the result he hoped for. To be clear he promotes the recording of audio at 96/24 to allow for engineering headroom but accepts the playback would seem to not require this resolution and that dropping down to 44.1/16bit appears (from his survey) to not make any significant difference to the average end user.
Paul, I wish you had not done this post. Giving people your description of PCM versus DSD sampling rates and then referring them to that Wikipedia page on sigma-delta modulators is not useful. Unless someone has a bachelor’s degree in math, physics or electrical engineering that page is not very useful. Below are a couple of links that I think can help people understand what is going on without and heavy duty math required. Sampling rates are the single most important thing in digital audio. The higher the sampling rate the more information you capture from the analog signal. ( The next most important thing is jitter. ) DSD is really just a version of pulse density modulation ( PDM ). I have never found a better way of describing the difference between PCM and PDM in a non-mathematical way than shown in charts 11, 12 and 13 in the following link:
http://www.dspj.co.jp/~manuals/mergingtechnologies/WorkFlow/SACD_FormatOverview.pdf
As to how important sampling rates are please scroll down about 2/3 of the way in the link below and look at the actual oscilloscope images of impulse signal that has been digitized using various PCM sampling rates and a DSD sampling rate. I think this picture makes it clear why higher sampling rates are better.
https://www.magicbus.biz/what-is-hi-res-audio-.html
That link was more of an ad than anything of technical merit. The RBCD standards can go from any one of over 64,000 loudness levels to any other level over a 96 db dynamic range in 1/44,100 second. It does it reliably and virtually flawlessly. This is faster by at least 10 percent and has at least 10 times the loudness resolution that even the must acute human hearing can detect. It does this over every loudness level and frequency a human can hear. The RBCD standards have so many belt and suspenders redundancies built into it that it is virtually mind boggling. It scans each track at least 6 times and if it doesn’t get it right it scans it again. There is no music that exceeds the capabilities of RBCD insofar as the two channel stereophonic format is concerned. There are no technical improvement to it that are of any use to the end user.
As usual, we will have to agree to disagree! If this pandemic ever is over and you happen to be in the NYC area let me know in advance. You are welcome to come to my house and hear the CD to SACD difference for yourself. We can also discuss in detail why high frequency ( over 20 KHz ) is so important to human hearing.
Indeed! But soundmind is a strong believer in RBCD and one cannot argue against believes. 🙂
I’m a believer in science. And what I posted comes directly from my understanding of a lifetime of studying it.
so the discussion of same rates is interesting , but leads me to wonder….
If the premise of the Direct Stream DAC is that “pure DSD” is the cream of the crop for playback regardless of the sample rate or bit depth fed to it, then is there a noticeable difference out of the Direct Stream DAC when fed a “reference recording” of different sample rates of the exact same digital master? experience using some of the Blue Coast comparisons tells me the differences are very small when using the Direct Steam Dac as compared to using something with a Burr Brown or ESS Sabre DAC
So I guess the hardware as its interfaced with software is just as important for digital playback, as sample rates and recording techniques are for getting a great recording.
Is bit depth as important as sample rate (i.e. 16 versus 24 versus 32) or is that just marketing?
Once you get to 16 bit depth matters very little.
Mike, thank you for using our files for format comparisons.
Yes, at every stage of the listening chain, the device, cables, song manager, conversion software, DAC can alter the sound. My experience is that the DAC will alter the sound more than the format used so it’s important to compare using the same listening chain (to the best of your ability) to make a judgement as to which format you prefer.
Some people don’t hear the differences that others do. Some think the difference between 44.1 and DSD256 doesn’t make enough difference to pay the price for storage, managing files and cost of the download. That’s fine. Everyone should decide for themselves what format they want to listen to.
After 40 years of being in the studio daily for recording, mixing or mastering, I can say this… Working with DSD256 is a complete pain in the ass. The size of the files, nonsensical digital recording software and a small niche audience compared to Spotify’s fans – well, some of you may wonder why we continue to record to DSD256……
We do it because we hear a difference. It’s especially grating to listen to PCM over long periods of time. Yes, even DXD (a PCM format) is impossible to listen to after working in DSD256. Patrick (my main dig operator) can tell you that I’ve walked into the room while he’s working and said “what happened to the sound?”… and after some momentary freakout on my part, he calms my nerves to say something like, “I’m in DXD doing tops n tails”. Once he goes back to DSD256, it always sounds better and I can relax.
After an album is available for sale, I can listen on almost any format… yes, even Youtube in my office on $100 computer speakers. That said, there is an interesting comparison going on momentarily on Qobuz streaming for a few more short weeks.
We uploaded 192WAV files to Qobuz of 29 albums from Blue Coast Records for our streaming audience. They are streaming in 192WAV. At the same time, we began a distribution deal that used a 3rd party delivery service uploaded the same files converted down using their methods to 9624 and the exact same album is now rolling at Qobuz. Listening to the stream in mp3 or 44.1 quality streaming, you can hear the difference in the audio of these two uploads of the same album.
If you want to check .. look up “Kiss in a Shadow” — artist is Art Lande, solo piano on Blue Coast Records. You only need to compare the first 30 seconds of the first song on each format. The two albums have slightly altered covers to help distinguish. I can’t tell you how many of our fans wrote to me to say “what happened” and were also upset. I went through the roof at the 3rd party and Qobuz is also upset with them. We’re leaving the two versions up a little while longer, but the 9624 is going to go soon… so compare while you can. So yeah, digital converters are not the same. Am I concerned about all the millions of songs in HD this company is handling? Yes. We were never told.
Again, I’m not here to offer which format anyone should listen to or prefer. Our customers buy DSD256 well above any other format. My job is to give everyone the best versions possible in all formats.
I know what I hear and prefer. If in 600 people no one could hear the difference (not sure what was compared), so be it. Our experience has been testing hundreds of people over time in our studio and 4 out of 5 could hear a difference… now if the question is “do they care?” that’s a different answer. But if even one of 100 people cares, that is our customer.
If others would like to setup a test, it’s not easy. To do it right you’ll need an extra set of hands.
Download multiple formats of one song recorded in DSD here.. it’s free
https://bluecoastmusic.com/free-downloads
Read how to do a comparative test here…
https://dsd-guide.com/how-do-comparative-listening-test
Have a wonderful weekend, all!
Cooke Marenco
Blue Coast Records and Music
https://bluecoastmusic.com/
Cookie,
Thanks for the reply and all your thoughts. I’ve been avoiding the whole subscription / streaming ‘scene’ …. instead preferring to purchase, download and burn physical back-ups, while at the same time having large external SSD drives and fully imaged back-ups.
Personally I’m sticking with DSD recordings and remasters when I find music I really enjoy…
Mike, I couldn’t agree more. 🙂
Cookie
I’ve heard every sample rate and format beat every other. My conclusion is that we are really hearing differences in filter design, dither character and analog stages.
Many many wonderful and astute comments here. Really appreciate everyone’s insight and understanding with in the world of sample rates.
I’ve always been amazed at the psychological marketing attached to sample rates, which get just about every budding audiophile hook line and sinker to explore at least for some time. When someone says in a short paragraph that a blu ray audio or SACD has 256 times more resolution at a sample rate of 24/96k than a standard cd at 16/44.1k you are gonna buy it. Hands down. Many really honestly think that 256 times more resolution is 256 times more detail in every aspect of their music because many don’t really know the science of what that 256 times really means . In any case, there are so many theories where that greater definition really comes from. Take for example Rob Watts of Chord Electronics. Mr. Watts gives an incredible presentation explaining how his Hugo M-Scaler works.
https://youtu.be/VfscfTkHgM4
Basically the upsampling in music revolves around the transients that our brains can pick up and resister. Bit depths are of not the biggest importance. It is kilohertz range.
Honestly, given so many great comments and people on here I’d love to hear some opinions for what others think of Rob Watts and his theories of increased tap length (transients) in his DACS that cause a greater scope, clarity and resolution in our music we hear. Personally I’ve never heard the M-scaler paired with my chord DAC just yet, but I really think he is onto something and I’d love to try the M-scaler. 🙂
A few comments.
CD isn’t a standard it is a specification of PCM at 44.1/16 bits, saying ‘CD standard’ is a meaningless statement.
DSD (PDM) does NOT sample. It has bit rates NOT sample rates.
Recording studios don’t have to convert down to 44.1/16bit unless it’s for a CD release or it’s all your DAC will handle so you have to stream or download at this spec’. Based on all the collated surveys this really doesn’t appear to be a big deal.
I’ve spent a considerable amount of time researching and understanding the whole ‘Hi res’, DSD vs PCM deal and listened to plenty of music in the process and at the end of the day – as has been stated time and time again – it’s the care taken in the basic recording process, mixing, mastering that makes the biggest difference to what you pump through your speakers or headphones.