Back to the classroom
Join Our Community Subscribe to Paul's PostsThanks for indulging me my story of the early struggles with an outdated education system. I hope we’ve come a long way since those days of industrialized factory schools churning out something; although I am not sure what.
The saving grace of my education through public schools was the few standout teachers I had (and I suspect you had as well). Hats off to those brave souls that really loved to teach and share and gave their all towards that endeavor, despite an antiquated school system that hadn’t changed much since it started, back in the Henry Ford days.
Let’s finish up with DoP and move on to other subjects. If you’ll recall, DoP is a format that permits single bit audio (DSD) to “trick” a computer or network into thinking its Multi-bit Audio (PCM). Because computers don’t recognize Single-bit Audio, they don’t know what to do with it. And that brings me back to what I asked you to memorize: that Bits are Bits. To a computer a video bit, math bit, audio bit, or mouse bit all look the same. Bits is bits. Remember?
Now, don’t get turned around just yet. Many of you wrote in to say “if bits are bits, why do cables matter, sample rates, etc.?” Ok, that means you’re thinking too much! I’ll likely call Mr. Shannon out of retirement to give you a good whack for doing so. 🙂 When I write “bits are bits” I am not referring to how those bits are delivered, their speed, their timing and all the little details critical to how those bits are going to sound. No, I am referring to the bits themselves: the ON and OFF pulses are all pretty much the same.
So if bits are bits, how do Single-bits differ from Multi-bits and how does a computer know the difference? Computers use something called a header, which is a small group of bits that represent an instruction set. The header always comes first and tells the computer “I have PCM or I have DSD or I am USB” and with these instructions, the computer (or DAC) knows what to expect next and what to do about it.
The problem we run into with Single-bit Audio is two fold: it’s a stream and doesn’t have a consistent “reminder” header telling the computer what’s going on and most modern computers were never instructed to deal with Single-bit Audio even if they did have the instruction set. They only know Multi-bit Audio. Think of this as getting a set of written instructions in a language you can’t understand. That could change, over time if operating systems of Microsoft and Apple wanted to include Single-bit audio, but for now they don’t.
So here’s what some clever guys did to get around this problem: they put sheep’s clothing on the pack of wolves. If you had a flock of sheep and wanted to get a pack of wolves into that group without the sheep noticing, you’d wrap the wolves in sheepskin so the sheep would be fooled. Now let’s relate that to Single-bit Audio (the wolf) and Multi-bit Audio (the sheep).
Single-bit Audio is a stream of on/off bits. The more ON bits the more energy we generate to make a loudspeaker move. Multi-bit Audio is not a constant stream, but rather a stream of coded packets (their bits also make more or less energy to the loudspeaker, but they must first be decoded, unlike the simpler Single-bit stream). Think of Single-bit Audio as a flowing stream of water and Multi-bit Audio as a long unbroken freight train. Each of the cars in our imaginary freight train has an identifying marker that lets an inspector know what’s inside each of the cars of the train. This is the header I spoke of.
If we want to take our continuous stream (Single-bit Audio) to a new location, one thing we could do is divert a section of the stream’s water and fill up one of the freight train cars. If we repeated this process, using all the freight train cars, we could transport an entire stream intact to anywhere we wanted. How did we manage this?
Picture our stream running in an overhead pipe with a gate on the pipe that opens up at identical intervals, releasing water. Below the pipe is our train. With each water release we fill up one of the cars of the train; the train then moves forward and the process repeats. Using this method, the train can travel anywhere it needs to go. What we’ve done is expertly divided up the water in the stream into containers (the train cars) which allows us to transport the water intact to a new destination. Once we arrive all we need to do is reverse the process and we again have a flowing stream. The water is identical, it’s in the same order it started with and it was never converted to something else for transport.
Now you start to understand how DoP does not convert Single-bit Audio to Multi-bit Audio, it simply breaks the continuous stream of bits in Single-bit Audio into identical groups, wraps an additional piece of info around it and fools the computer into thinking it’s actually something it is not. The computer’s happy, the data is intact.
Remember that the bits are bits and the computer can’t tell one bit from another? Once we add this header that is lying to the computer, falsely identifying the attached group of bits as PCM, the computer lets them through because it simply cannot tell the difference and doesn’t care (they pass through unmolested but the computer still doesn’t know what to do with the DSD bits and thus can’t play them without a program to help it understand them).
For the more technically minded here’s the details. The Single-bit Audio stream is running at 2.822mHz (2,822,000 bits per second). The DoP “converter” simply culls out 16 bits from the Single-bit stream and attaches an 8 bit header to it, then repeats that process over and over again. The bits it is grabbing have no timing information associated with them; they are simply a group of exact ON/OFF bits in the exact order they started with. 16 audio bits, plus 8 header (information) bits equals 24 bits. So a DoP packet is 24 bits long (8 info bits and 16 data bits). Each of these newly minted packets collectively run at 176.4kHz. Sound familiar? Sure, you’ve heard of high resolution audio running at 176.4/24 bits? Right. That’s single rate DSD in DoP clothing. You can do the math yourself: 16 x 176,400 equals 2,822,400 which is the exact sample rate of single rate DSD. Simple eh? And the other added 8 bits? They account for the larger file size of DoP vs. native DSD (bigger by 1/3).
Here’s a picture of what that looks like.
The orange squares to the left represent the 8 information bits attached to the 16 bits of DSD to the right. This is our freight car filled with Single-bit water (the wolf we’re trying to hide), and the 8 header bits provide the sheep’s clothing to trick the computer into believing the next set of bits are PCM, not DSD.
Lastly, one of the 8 information bits lets any interested party know the data is really DSD. Thus, if the DoP data is put into a DAC that understands that information and can process it, all it needs to do is throw away the 8 bits of information data, connect all the 16 bit wide groups of DSD together and play the DSD stream perfectly. No conversion process ever took place.
Now, let me go grab a cold glass of perfect water from the stream, uncontaminated from any dreadful impurities, despite the fact it arrived in a freight train with a wolf as engineer.
If the 8 header bits are thrown away, why do we need 8 if them? Why not 15 audio bits and one header bit to signify its DSD? Less size, easier load.
Because the header is doing much more than just saying it’s DSD or not DSD. Remember, the header has to look like PCM, has to tell the computer the information that follows is PCM. There are standards of information in those headers, sample rate, bit depth, channel orientation, file types – all required for the computer to verify and authenticate it.
Remember, the standards of PCM based audio have been around for many decades. These standards are what they are. If we were to start over and want to make DSD work, we wouldn’t do DoP – because it’s a hack – but one that works without any damage to the signal.
Hi
Q-
Does your DSD in the latest unit work the same as theDSD in the phono pre amp you are selling?
If not is the preamp upgradable?
Larry
Very clear Paul ….
Going back to 2012, I always thought the bridge with the PSII sounds better than USB with a good cable (at least for me) because the ethernet cable conveys only signal with no jitter issue … only bits whatever how they are packed + I2S connection…
So can we make the following analogy ?
Ethernet /bridge connection may have the same effect as DSD for transportation purpose : bits with no timing issue on the route to the computer, which makes the quality insensitive to the cable.
Even not being single bit, can we see it as a kind of Multi-bit direct stream Audio ? : it looks like it is bits-packs conveyed and deciphered with clock treatment only within the dac / i2s ; another “X”oP thing 🙂 ?
So if the analogy is merely right, i have 2 questions:
– do you experience a significant improvement with the DSD vs PSII/bridge ?
– and DSD/USB vs DSD/Bridge ? (I speak of SQ only, not of the bridge features for network management purpose, which remains of course a good reason to use it)
Thank you
Thanks. Fortunately, the new DirectStream doesn’t pay much attention to how things are delivered to it, whether their full of jitter or not at all. We’ve managed to get DirectStream to sound pretty close on all manners of input.
Having said that, there are, of course, differences. The Bridge and USB inputs all deliver data without timing info attached – DoP or otherwise. There is also a new means of getting DSD directly over USB that we’ll look at updating DirectStream to handle as well.
Thanks very much for the explanation Paul. Based on what you said then 16-bit PCM does not contain 16-bits of music information. Does that mean it’s only 8?
No, unfortunately it’s not that simple. I probably made it too simple. PCM doesn’t use a header and uses just the 16 bits, up to 24 bits (for S/PDIF).
There is a bit (embedded in the channel status frame) that identifies the data as either PCM or not, but this is not part of the regular data frame. DoP expects it to be set for PCM to make sure that no one throws it away along the road.
So devices expect the data coming over S/PDIF to be PCM, and if it really is then there is no problem. Because PCM doesn’t have a header, a PCM device still thinks that DoP data is PCM. That’s what makes it work, the fact that most equipment isn’t smart enough to recognize it as being anything other than PCM.
Thanks for your time and the explanations. Very informative!
A pulse is a pulse. A bit is not a bit. The significance of each pulse depends on what the system is that it’s being used in. In a strictly amplitude versus time system a pulse is what will make your speaker pop, the larger the pulse, the louder the pop. In DSD, a pulse at a given time tells the decoder to kick the output voltage up by a fixed amount. The absence of a pulse tells it to let the output fall by a same amount. A pulse does not become a bit until it is used as one method of digital encryption. As with any system whether it’s the only system a circuit is designed to encrypt or decrypt or not, the pulse only means something in the context of that particular system of encryption. The system RBCD chose is a 14 bit binary encoded word system. Each pulse or bit has a value depending on where in the agreed sequence it occurs in the word. Having utilized the system that evolved from computers this base 2 system reads the sequence of bits in each word from right to left. It could have been designed to read from left to right or in any other sequence. This just happens to be the system chosen. These types of encryptions don’t have to be restricted to a binary code either. For example, a third bit represented by a negative going pulse could have signified the digit 2 in a ternary or trinary code. 0, +, – for no pulse, positive going pulse, and negative going pulses could have achieved 64,000 levels with only a ten bit word. In hexadecimal where 0 and fifteen different levels of voltage discrimination could be encoded and decrypted the same number of levels could have been achieved with a 4 bit word.
http://en.wikipedia.org/wiki/Ternary_computer
The binary system for audio and video data was undoubtedly chosen because it is relatively easy and there was a lot of hardware and software already around that could be made use of. It was much easier to understand than learning new coding systems. You haven’t lived until you’ve tried to perform simple arithmetic calculations in hexadecimal (base 16.)
I think we can all agree that noise is the most pressing problem in digital as well as analog systems, and the elimination of noise is the engineer’s first task. Failing to eliminate noise is failing to solve the fundamental problem in digital systems.
From what I gather from your explanation Paul, and the math that confirms the explanation, there is no practical difference between DS and 176/24.
They are one and the same at the end, with one major exception. And this is the end.
If DS is not converted, just held in a 16 bit word length in the same box car as PCM, it is being converted indeed, it is PCM, no? The box car is the word length. The math appears to be the same, just approached one from the top down, the other from the bottom up. Or perhaps better described, one at the input (door to the box car) the other to the door on the other side.
This reasoning may be why a number of top designers disagree with DS advantages as being hype and nothing more. the only “more” is that DS adds noise that was eliminated with PCM.
http://craigmandigital.com/education/PCM_vs_DSD.aspx
My analysis is that DS may sound “smoother” to some listeners alike analog not because DS mimics analog – it does not – but rather DS has similar noise characteristics as analog: more of it the higher the frequency one is dealing with.
We can see how such noise is tolerated by LP lovers, for the noise can mask other issues – acting like dither figuratively, smoothing the staircase so to speak.
http://www.audiostream.com/content/ayres-pcm-dsd-comparison
http://homerecording.com/bbs/equipment-forums/other-equipment-reviews/dsd-vs-pcm-head-engineer-phillips-179930/
Now you need to test both to see if one can reliably tell one apart from the other.
It should be an interesting test indeed.
I don’t like people speaking for me by putting words in my mouth.
“I think we can all agree that noise is the most pressing problem in digital as well as analog systems…”
I entirely disagree. Noise is not a problem in RBCD systems or better. Where there is noise on RBCD recordings it was inherent in the source signal whether as noise recorded on an analog tape recorder or in the analog system, microphones picking up room rumble or other spurious sounds, or other equipment malfunction. The background of DDD recordings made in a quiet environment even at high gain assuming the amplifiers etc. are not introducing noise is dead silence. Whatever problems high quality digital formats have inherent in them, noise is not one of them.
RE Soundminded’s assertion: “I entirely disagree. Noise is not a problem…” referring to my comment “in digital systems.”
To his statement, let me recite a personal, but appropriate story. I have a relative, fairly close, not blood, who was a systems engineer who graduated at the top of his class from Cooper Union, regarded by the industry as the top of it’s class. He ran a major facility that employed tens of thousands. He was an angry and bitter man who didn’t get along with anyone in the family and at work too, which seriously impacted his prospects. He disagreed with everyone – that disagreeable personality was his real profession, not his slide rule.
Everyone however thought that he was bright. When I met him at the age of 17 or so, I didn’t think he was really that bright and maintained my impressions through decades, after which a few others in the family eventually saw it my way. The man wasn’t really bright, but he knew how to use a slide rule better than I did. He also had a lot of textbooks about engineering that I didn’t possess. He was also taller than I and had a better head of hair.
He could figure out why everyone around him was wrong about everything and he was right about everything, but not why his entire family and all his friends were alienated. Interestingly, everyone in the family disagreed with him, but he was right and everyone else was wrong.
But he had his numbers, opinions, his textbooks and his very own anus.
In other words, he was the center of the universe and the universe was centralized at the Forum in Rome. Never mind that the Forum started crumbling 2000 years ago.
Back to the noise that Soundminded says doesn’t exist or are the least of digital’s problems:
[PDF]
EE273 Lecture 5 Noise in Digital Systems Today’s Assignment
cva.stanford.edu/books/dig_sys_engr/lectures/l5.pdf‎
Oct 7, 1998 – Noise in Digital Systems. October 7, 1998. William J. Dally.
[PDF]
Noise and Spurious in Digital Systems and Digitized Signals
http://www.highfrequencyelectronics.com/Archives/…/HFE0906_Tutorial.pdf‎
6.915 Digital Systems Engineering
http://www.ai.mit.edu/…/6…/6.915.html‎
Massachusetts Institute of Technology
Sep 4, 1996 – Noise in Digital Systems: Noise is the dominant concern in the design of both signalling and timing conventions.
Digital Systems Engineering Home Page – Stanford University
http://cva.stanford.edu/books/dig_sys_engr/
QUOTE in the intro: “As technology continues to advance, issues of signaling, timing, power, and noise become increasingly important.”…..”Band-aid fixes rarely exist for these types of problems; however, they could have been easily avoided by proper design if the engineers involved had been knowledgeable about noise, signaling, timing and power. By writing this book we hope to help eradicate the widespread ignorance, and often misinformation, in these areas, and in doing so, help avoid disasters of this kind in the future.”
Effects of Clock Noise on High Speed DAC … – Texas Instruments
http://www.ti.com/lit/an/slaa566/slaa566.pdf‎
How DACs Work
http://www.msbtech.com/support/How_DACs_Work.php‎
At loud signal levels quantization errors manifests themselves as noise.
Reducing noise on DAC output | Microchip Technologies
http://www.microchip.com › … › dsPIC33F Topics‎
I could go on an on about THE NOISE IN ONE’S HEAD that doesn’t exist either, but I trust these will suffice for the time being. However if Soundminded still thinks that noise in digital systems are hardly a concern, he is welcome to his opinions that no one in the engineering field would agree with.
I might have put it another way: Noise is NO LONGER the largest concern for audio digital systems because of the recent advancements made, but there is still noise that can and does exist as stray voltages that play a role in jitter…and noises that can get induced in a circuit by other means, magnetic, RF and so on. But I have no idea why I, a non-engineer, has to explain it to one who is.
Did you ever know someone who knew so little about something he didn’t even know how little he knew? Well that’s the digital noise I get from your postings.
Ever hear of “dither?” That’s the NOISE that is DELIBERATELY ADDED to mask quantization at the very lowest sound levels recorded in digital formats. It is so low many experts say you really can’t tell any difference with it or without it. Now why would anyone think of deliberately adding noise even at such a low level? BECAUSE OTHERWISE WITHOUT IT THERE ISN’T ANY.
Soundminded, you need to refer to the references I posted. Dither has little or nothing to do with it. Nice try.
Not only don’t I read your links, I don’t read most of what you post except when I scan them to see if you have topped your prior remarks with something I find even more ridiculous making less sense than what you’ve posted in the past.
Of course, we already know that All Roads Lead to Rome and that you’re closed-minded and immune to evidence from MIT, Stanford, Texas Instruments and Microchip Technoligies – as well as I could post dozens of other authorities who know much more about electronic engineering than you do – and you should continue in your arrogant ways, because you know that the laws of physics can be violated at will, especially by you, and you know that a microphone or a battery of microphones hear exactly the same way as a pair of human ears hear because the microphones are made of the same organic material as the pair of human ears, with the same geometry and resonance characteristics.
You are, Soundminded, so much smarter than all the world’s audio designers and engineers COMBINED that you had figured out all the problems in audio reproduction than they could not, and that is why your only proof of life are your words in lieu of achievements.
If there were a MOUNT RUSHMORE for engineers (and I’m certain you’re not reading this because you’re so smart) on which the world’s top engineer’s heads were carved, for example like Gustave Eiffel, yours would top them all.
Mostly because you’re not reading this and because you’re not full of yourself. But again, nice try.
Flattery will get you nowhere, at least not with me. Maybe someone can lend you a chisel and hammer and you can start sculpting. Forget Rushmore. You want to impress me? Carve the top of Everest.
I now understand why Paul enlisted the two of you to the same Blog; you are a laugh a minute, and provide the lighthearted respite needed, while we’re struggling to understand the basics.
This reminds me of some of what I learned as a molecular biology major at University of Colorado in (yes!) Boulder. The 5′ untranslated region upstream of the mRNA that codes for proteins. Art imitates life. I believe there is a depth of information we could generate from the described linear information that we have not pursued. For example, tertiary structures. I am sure this is possible, as we do this every day in our brain from a nervous system that is in essence linear via action potential quanta. I have some ideas about this but not enough knowledge to know if it is possible to reconstruct…
After reading the aforementioned the bilateral tinnitus I have is starting to sound good!
BTW will dsd remove the “bits “tinnitus?
Larry