An FPGA is a pure digital device with millions of transistor gates. How does it then convert digital audio to analog? Have a question you want to ask Paul? Go to www.psaudio.com/ask-paul/
I love the fact that Paul references the Intel 8080 that was introduced in 1974 and discontinued 30 years ago. Paul has been in the business a long time and has respect in the industry as well as from me. IMO PS audio has the best all around audio channel. Thanks for the information Paul
Yeah I found that reference awesome aswell. Building a 8080 system myself today with old parts and I started wondering, did they really stop? Well, yeah, Intel for sure. the 8086 followed and 8085, while intels 8051 can be found everywhere. lansdale.com/ is apparently still making them :P wow. probably for the military and obscure industrial machines. I doubt they will ship me a sample of some of their IC (I could try, but got most parts already second hand). Now samples for a diy audio device running on old tech would be cool tho.
Another way to visualize single-bit DAC is of two forces (negative and positive) that drive the speaker magnet inwards and outwards. The DAC has control of "where" the cone is positioned between full off and full on due to how long a 0 or 1 is "held".
I started with Cakewalk 1.0 in a 486 SX 😂. For where we are today, and after hearing that description, I can see where the beta testing and driver building for 40 years forced the hardware and software to converge. We aren’t far from having those chips networked like the cerebrum “on call” and using an atom based nano computing system that will end the debate on differences between digital and analog sound.
There are several companies that use off the shelf Burr Brown or ESS DAC chips and use a processor before it to convert everything to DSD. The advantage of FPGA is, if you are good enough, you can program it to your hearts desire and customise it to better suit the topology and parts used. When you use processors like XMOS / XCORE and off the shelf DAC chips you are limited by the fixed instructions set/filters. Technics SU-G30 does something very interesting, converts everything to DSD and sends the bits with little filtering directly to the GAN-FET output transistor without the use of a DAC (1 transistor per channel!!!). GAN-FET is the only transistor quick enough to cope with the MHz oscillations of a pure DSD signal. Any other FET or BJT transistor would require an analog signal.
Wow Paul, That was a brilliant and such a simple explanation that even a simple person like me could understand. I now want a direct stream DAC . it can never get outdated. Just be updated- I get it. Thanks.
Yes. Paul's DAC is futureproofed and worth spending on. His amps are lousy in comparison and get beat to death by high end Yamaha (A-S2100 and above), Marantz, etc.
W A Hi WA, I have my end game Power Amp from Joe Rasmussen in Australia. However Dac’s keep evolving and getting better, so if and when I can afford Paul’s FPGA. This would be my end game DAC - only hope I live long enough for that to happen.
@@paulpavlou9294 I thought i bought my end game amp 20 years ago. I still drool over stuff 20 years later. The only end game is the end of your wallet... haha
@@paulstubbs7678 Yeah, MQA comea to mind though i am not hopping onnthat bandwagon. Paul's business model may be that he wants you to buy more dacs from him in the future. So, updates on this one may get killed. I primarily got this DAC because i have a handle on FPGA from past work experience). What i listen to these days is my "mod" of Paul's dac. I made it better :). I can revert back if needed to what Ted Smith thinks i should listen to.
Nikolas I’m new to the concepts of DSD only have read a little bit about it, but he provided a base intuition for me to read more about I feel. Would you mind pointing out what you found inaccurate?
Wonderful explanation! I've heard you say more than once that you can actually hear the output from a DSD-signal. I'm curious: How good or bad would it sound (without any post-processing), just pure DSD-signal to amp?
Great explanation. I 1st became aware of this approach way back in 1986 when Technics released a line of compact disc players using 1-Bit D2A converters. I have a background in engineering & saw the advantage immediately. Rather than try to engineer the tight tolerances of a resistive ladder for 16-bit PCM converters for mass production; - A single pulse stream could be modulated to give much more reliable conversion across a filtering node - especially in mass-production. I'm sure your approach leverages faster clock speeds today & tighter pulse formation. What is the equivalent PCM bit-depth your process is capable of mimicking (If I may ask)?
Can this be combined with class D power amplifier in a single unit? Would that not be the most pure way to build the sound system? For example, each monoblock or dual mono amp has it’s own dac for each channel, the control unit/preamp is digital input and digital output to the power amps after splitting the digital channels and controlling volume/tone etc
Paul is partly right, it is possible to do analog signal processing within a FPGA, I have done it, and have the patent for it. I am using it in optical and radio communications, now looking at ways of using this technology for magnetic tape record and play back system. It is very true, there is a lot flexibility built in to FPGA's, and I am always finding new applications for this technology.
If you've been following Paul on youtube you know this! FPGA is a processor you can program acurst as you like. You can program them with a program in a way that the chip behaves exactly as you want. Regular DAC chips are pre-programmed to do a specific thing and you can't change it! The closest you can compare it to is hardware video decoding vs software video decodin on your PC or phone. Don't know if it'll be right to compare with this?
How I see it. Sound is basically variable amplitude over time. So in order to control the amplitude, the FPGA is configured to output pulses at variable rates. To increase the average amplitude, it will ramp up the pulse density. To lower the average amplitude, it will decrease the pulse density. This happens fast, at MHz rates. So the digital source provides amplitude data which is then used to control the density of the pulses at the output. The signal is filtered at the output to even out the edges of the signal, typically with a capacitor, to get a nice analog wave. To answer the waveform question. A sine wave, is basically a signal with varying amplitude, shaped by a soft curve, and if that variation repeats at rates (hz) within our hearing range, we will hear it in the form of sound. So for an FPGA to reproduse that, the output will produce varying pulse densities based on the amplitude of the wave at any given time, and that's it. Hope it helps.
The question: Can we need resistors from out of the FPGA to build this DAC ? or, we can acces to discreet resistors from into the integrated cicuit FPGA ? I say resistors to build the ladder of voltage divider. Thank you in advance.
FPGA is nothing new, Sony have been using that since the early 90s. But some company markets the heck out of it and the millions of taps to make you feel it’s magical jeez. I’m glad PS Audio doesn’t overblown their use of FPGA.
Mike the best part is that the company that makes a big deal about the tap length in their dacs do not even use the fpa as the converter. They still have a delta sigma chip to do the conversion as far as I know. There are many companies that use fpga tech in their dacs. Metrum, denfrips, dCS are just a few. For the company that is advertising tap length it is mostly for marketing. Their designer is also quick to put down other dac designs say that his is superior and eve the cheapest out performs other companies dacs.
It is really an I2S to class D Converter with a great interpolation filter in the middle so you can have an almost infinitely precise signal at the end.
Another reason why the FPGA with DSD method is soo "analogue" is because it pulses at 2.8 MegaHertz or much more. That is orders of magnitude more detailed than Sound Waves, relative to Sound time frame of Khz it adjusting much faster. All being done with 1 bit swithching.
So it must be that there is little or no switching noise to comb out as well? Or is that what gets filtered? It sounds like a moving coil setup where you get lots more detail but in order to bring the gain up there's all that juice to keep quiet. You might like my friend's preamp design for that stage. He uses a photocell arrangement that eliminates filter caps, or any resistive components. The signal to noise ratio is not measurable, it simply doesn't seem to exist. In his words: "It has an input impedance too high to measure, output impedance is a couple ohms, bandwidth is to 60Mhz, slew rate is a couple hundred V/microsecond. There are no capacitors or resistors (other than the photocell) in the signal path. There is no potentiometer in the signal path. It uses photo cells to control the volume. It does not use optocouplers."
There is actually a ton of noise generated by the DSD output, so they have to use noise shaping, which is a feedback loop that pushes that noise way above the audio band, and analogue filters to get rid of it.
The next generation will have nearly twice as much to fit our needs. The upsampling, converting the data to a 50-bit number, filtering, all take enormous resources. Which is why it's prohibitive in a fixed chip that needs to be sold for only a few bucks. The FPGA unleashes the power to do what we need.
@@Paulmcgowanpsaudio Thanks Paul, I was surprised that better upsampling imposes such a requirement. Not so long ago, an university with a campus was built around such a computational performance. At all, why do we need better upsampling? Can we hope you’ll try to shed light on this for laymen like me in a future episode? I wish the best from Hungary: Csaba Kelemen
@@Paulmcgowanpsaudio Well as I'm sure you know, the economics here are a bit more complex. There are ASICs (custom designed chips) in many consumer products that do cost a few bucks and are far more complex that many FPGAs. Your product's FPGA could be reduced to a custom chip that just does cost a few dollars if not even pennies in some cases. Consider the image scaler in a $80 LCD monitor. It comes down to basic Henry Ford mass production. It can cost upwards of a million to have a custom chip designed and debugged. In fact the initial designs are most often built with off the shelf FPGA chips. Then once perfected a custom chip is designed which can be produced for as little as a few dollars or less. But you need a massive product market, like smartphones and TVs, to recover the cost of that custom chip design. Your use of FPGAs is quite common in professional products that also have a limited market where the cost savings of a custom chip cannot be recovered. The other large advantage that is also not common on commodity consumer products is the ability to reprogram the FPGA in the field using a USB port. You have taken good advantage of this yourself with the DirectStream firmware upgrade program.
@@andydelle4509 Thanks, Andy, and you're right. However the cost of an ASIIC is astronomical and makes no sense for the numbers of DACs we produce. Secondly, unless they too were programmable, it would mean we'd have to stay with what we have rather than taking advantage of the freedoms of updates the FPGA's afford.
@@Paulmcgowanpsaudio 50 bit, wow your not mucking about, kind of leaves 24bit audio in the dust. How about a 'simpler' one with only 24bit math for those of us with shallower pockets, but want more than off the shelf chip DAC's offer.
It makes sense to convert PCM to DSD before feeding it to most modern DAC's and you can do a pretty decent job with a laptop and music software. I wish Paul would really push the point of how different PCM and PDM (DSD) are even if he disagrees with my heretical extremest view that DSD is actually an analogue capture system. Stay well everyone.
Here's another thought: analog is actually digital. Think about the way a vinyl record is made. It is, obviously, made of vinyl, which consists of molecules, big ones as molecules go. It isn't a glassy smooth surface under a microscope, even when it should be. It's bumpy, and the needle isn't sharp enough to follow all the contours of vinyl, so it rides across the high points from bump to bump on a microscopic level. You also can't have any instantaneous sharp transients or the needle would get stuck or more likely gouge out the sharp corners, something that digital has no problem with. But anyway, if you begin with a smooth analog signal and pour lumpy stuff into a mold of that signal, you end up with a quantized approximation of the original signal, something much like digital, but bad digital. I got that from a study done at the University of Glasgow, if I remember correctly, but it has been several years since I read it and I can't find the link.
@@angelwars3176 it makes perfect sense if you read the study, which, unfortunately, I can't find right now. What they were saying was that records don't follow the analog signal, but rather a bumpy approximation of it due to the nature of vinyl molecules, much like digital is itself a bumpy approximation of the original signal, not physically like a record, but electronically. At any rate, it's just a thought.
@@brianmoore581 Yeah but 'vinyl' is bumps in a composite material not 1's and 0's binary words so in what sense is a vinyl record digital? Paul called me out on his daily blog site last year citing "certain people making comments that DSD was not digital" Bottom line for him is DSD is digital because it uses 1's and 0's bits - perfectly correct but the key understanding is how it uses them in comparison to PCM.
@@angelwars3176 well, it's not digital as in zeros and ones, obviously. What they were saying was that it is quantized, as in coming in in individual pieces of information as it rides across the high points of each molecule, then skips to the next. I suppose that is analogous to the quantization error of digital sample rates. I'm sure you've seen those stair step diagrams of different digital formats showing how closely or not they follow the original analog signal. They were saying that the same thing applies to vinyl due to its construction. In other words, vinyl doesn't really follow a smooth analog signal, but comes in "samples" corresponding to the high points of the vinyl molecules. In other words, the signal is quantized, represented by discreet steps as opposed to a continuous uninterrupted signal, which makes it digital, at least from that point of view. Digital as we know it is binary, as in ones and zeros. But all digital isn't binary. There are, apparently, from what I have read, other forms of digital like trinary, or base 3, even a base 10 form used in some early computers. Binary, base 2, is obviously the simplest, most reliable, and the standard of our time. But those are computer languages. Does digital exist outside of a computer? The information encoded in your DNA is a form of digital, base 4. Are the discreet firings of nerve impulses from the hair cells in your cochlea considered digital? Are the firings of your neurons in your brain considered digital? They either fire or they don't. There is no continuous signal. So in that sense they are digital. I suppose if you consider that we live in a quantum universe, everything is ultimately digital. There are only so many rods and cones in your eyes, kind of like the pixels on your TV screen, and there are only so many discreet photons of light that illuminate our view. The air we breathe and hear through is made up of discreet atoms, not a continuous substance. We experience the world as if it were analog, but deep down it's all quanta, which would be digital. Well, that's getting into another topic, I suppose. Kind of splitting hairs on definitions, perhaps, but if we see things as either analog or digital in a broad sense, then it's all ultimately digital. Back to audiophile reality, digital audio obviously produces an analog signal, so the only thing digital about it is the way the information is stored. The vinyl record stores its own information in a molded series of rather big molecules which only approximate the original analog signal, a quantized version of the original analog signal, assuming it was originally analog, or perhaps a further quantized version of the original quantized digital recording. Anyway, definitions are arbitrary, depending on how closely you look at things. And it's just a thought from somewhere out in left field. In purely audiophile definitions, records are still analog and DSD and PCM are still digital.
Thank ytou misters Puals! That was interstings. Maybe you can use that to make sound processores such as what Lexicons or Eventide or TC ELectronics makes. I remmebr some jargon from the late 1980s about 1 bit processing for CD players, but it was probable just jargon to make them look better than they reallywere. THANKES YOUES AND THUMBBSS UPS!
PC Audio is quite unique with this FPGA approach and in a high-end system, it makes sense. Most modern DAC chips are single-bit or Sigma Delta but always with fixed silicon. Even what is in your smartphone or in a US$10 iPhone lightning to 3.5mm adapter cable is with such DAC chip (often part of a larger chip). A mobile phone DAC is obviously not of the highest performance, but when your iPhone doesn't sound great, it's mostly due to the music compression rather than the almost invisible DAC itself. The specs of your smart phone DAC is typically much superior than audiophile DACS of a few years ago, e.g. www.qualcomm.com/products/aqt1000
Thanke you MIne Frends! You have loits of things other don tknow ro wont say. I gacve up in 1978 when I figure out I could neer afford the Pioneer and Kenwood and other stuff in ROlling STone. THAMKES YOU ABD THUMBS UPOS!
Lloyd Stout: I am a teapots short and stouts. her i smy Handles here is my spouts. Is this the article you wanteds.... This standard provides an overview to the family of IEEE 802® standards. It describes the reference models for the IEEE 802 standards and explains the relationship of these standards to the higher layer protocols; it provides a standard for the structure of IEEE 802 MAC addresses; it provides a standard for identification of public, private, prototype, and standard protocols; it specifies an object identifier hierarchy used within IEEE 802 for uniform allocation of object identifiers used in IEEE 802 standards; and it specifies a method for higher layer protocol identification. Ethernet runs on the MAC address NOT on IP numbers, Its in teh ARP tabels thankes you and thumbbsss UPS!
The higher you make the DSD 'sample frequency' the easier it is to filter it down to a pure analogue signal, without any high frequency digital pulses that 'may' make the following equipment unstable.
