An Upsampling Primer or Why Make More Bits?Why upsample when 16 bits/44.1 kHz sounds just fine? Can more bits really make music sound better? That is what I wondered. I have been happily living in Redbook land for the last 10 years but was wondering if digital technology has advanced in all those years - enter the
Holo Audio May KTE DAC and HQPlayer upsampling software. The May is a non-oversampling R2R DAC that uses a new linear compensation technology that solves the accuracy errors caused by resistor tolerance, which, after compensation, reaches a variance of 0.00005% tolerance accuracy. The Holo DACs are excellent at 16/44.1 but they are also a blank slate because they are also capable of supporting DSD1024 native and PCM 1.536MHz input using upsampling software.
I didn’t know how upsampling worked so I dove in. This primer is the result of my research. I hope it is educational and entertaining. There is a lot of misconception about what oversampling is and how it works. But first, lets start at the beginning…
Paul McGowan, Founder and CEO of PS Audio wrote in his June 9, 2012 blog:
“It’s clear that if you take a redbook CD and upsample it to 192kHz 24 bit you’ve wasted your time and your bandwidth and memory.” 1He has since changed his mind about upsampling.
The current PS Audio DirectStream DAC upsamples the incoming PCM to 20 times the nominal DSD rate (20 x 64 x 44.1kHz) which is 56.488MHz (as in million). Paul’s thoughts about digital reproduction has obviously changed in the last 10 years.
----
Does Upsampling Improve Sound Quality?In the Copper Magazine June 9, 2018 issue PS Audio's Paul McGowan wrote an article:
Getting something for nothinghttps://www.psaudio.com/blogs/pauls-posts/getting-something-for-nothing“When we upsample a 44.1kHz 16-bit file to a higher rate and depth, like 96kHz 24 bits, we typically get better sound quality. And since the magic of upsampling just sort of works at the touch of a button, we seem to be getting more for nothing. After all, the file size is considerably bigger. There must be more there. Right?
"So, how does that work? How can a program know what went missing from the original recording so it can add it back in?
"There are actually two things going on. The first, and least important, is interpolation. Interpolation is a mathematical process that adds more data points through intelligent guesswork and statistical analysis. Simply put, if our steps are moving in a predictable pattern: 1, 3, 5, 7 then it’s likely we can add the missing steps: 2, 4, and 6, as additional data points so we wind up with 1,2,3,4,5,6,7.
Perhaps more important is the choice of filters. With standard CD rates of 44.1kHz we need to have a fairly steep filter so we don’t run into trouble above 22kHz. By increasing to 96kHz we can apply a much gentler and better sounding filter to our digital data and this, in my experience, is responsible for the majority of what we might consider better sound from upsampling.”
2The companion video explains what he is talking about in plain language.
How does upsampling increase information?https://www.psaudio.com/blogs/ask-paul/how-does-upsampling-increase-information/----
All DACs OversampleI have read this statement all over, from Paul McGowan's blogs to posts by Jussi Laako, HQPlayer's developer, and on the Signalyst website.
The World is DSDPaul McGowan – March 9, 2016
“All modern DACs and ADCs (Analog to Digital Converter, the opposite of a DAC) are Sigma Delta based converters. PCM, all 32 bits of it, is first converted to a format that’s close to 1-bit DSD and all subsequent processing inside your DAC, from the lowest cost Audioquest Dragonfly to the over $100,000 whacko daco, is no longer a many-bit PCM process. And this has been true for many, many years.
“Fact is, when 24 bit DACs started appearing in the marketplace, chip designers were forced to move from straightforward PCM ladder-DACs to DSD-like Sigma Delta converters. Why? Because resolving 24 to 32 bit accuracy is next to impossible with part tolerances that have to be accurate beyond the abilities of manufacturers to produce.”
3----
Actually that statement is not quite true, the Audio Note, BorderPatrol, Denafrips, MHDT Labs, TotalDAC and Holo Audio DAC’s do only 1x sampling. Other DAC’s do internal signal manipulation, they just don't tell you and only the developer knows what is happening inside that chip.
The following post by Jud also talks about filtering, I’m starting to see a trend here.
----
The History of Internal UpsamplingSo why, instead of just saying what goes on inside their (by all indications very fine) DAC, did W4S give Kilroy a reply so laden with marketing-speak that he wound up thinking they were saying the exact opposite of what they actually meant?
It all goes back to the first CD players. (I know, why do we need to look at something that happened 30 years ago? Just hang on, the context is actually helpful to understanding, and I'll try as much as possible to put it all in plain English.)
Many people thought the first CD players sounded pretty bad. Many engineers thought the reason for this was the filters used in those early players. Filters?
OK, digital audio is based on the Sampling Theorem (the name Nyquist gets mentioned a lot, though there were other guys too, like Whittaker and Shannon). Basically what the Sampling Theorem says is, in the world of mathematics, you can take samples of any signal (like music) and reconstruct it to mathematical perfection, as long as you take samples more than twice as fast as the signal changes (more than twice the highest "frequency of interest," which with music meant for us humans is the upper frequency limit of our hearing, ~20,000Hz). Thus the CD sample rate of 44,100Hz, a little more than double 20,000Hz. Once you've got the samples, you filter out the higher frequency half (the higher frequencies are stopped, the lower frequencies pass, so it's called a "low pass" filter), and you're left with the lower frequency stuff - music! Easy peasy, right? So why did folks think the filters might be at fault?
The Sampling Theorem requires a perfect filter, and no such animal exists outside of a mathematical proof. The perfect filter would allow all the frequencies we wanted to hear to pass, and perfectly block anything above that like a brick wall. For that reason and possibly due to the image of the response such a filter would show on a scope (a perfectly straight dropoff to silence, looking like a wall), these are known as "brick wall" filters. That's what the designers of the early CD players tried to get as close as possible to, the perfect "brick wall" filter. But because these filters were working in the real world and weren't perfect, they had a couple of problems.
One problem is called "aliasing." Any high frequency that's supposed to be blocked but gets through because the filter isn't perfect is mirrored or "aliased" around the filter cutoff frequency to create distortion in the audible frequency range. So let's say you've got an imperfect brickwall filter with a cutoff at 22,050Hz (half the 44,100Hz sample rate) and some noise gets through at 30,050Hz. No problem, you can't hear 30,050Hz. But that noise mirrors or flips around the 22,050Hz cutoff point to create an "alias" at 14,050Hz (30,050 is 8000Hz higher than the cutoff, so the alias is 8000Hz lower than the cutoff), which you very well might be able to hear. And of course there will be leaks through the imperfect brickwall filter at other frequencies, and these frequencies interact with each other to create yet more distortion/noise at audible frequencies.
The second problem is something called "ringing," which I won't go into too much except to say two things: (1) many people think ringing "smears" transients in the audible range; and (2) ringing is caused by a sharp filter transition - the sharper the cutoff the worse the ringing, so a brickwall filter is the poster child for causing ringing.
The CD and earliest DAC designers figured out a way around this, called "oversampling." (This was happening right around the time the earliest separate DACs were being created.) If you could first raise the sample rate way above 44,100Hz, and use a filter with a gradual cut starting at a higher frequency, that would accomplish two things: (1) by the time you got down around audible frequencies you wouldn't get much leakage, so it would help stop aliasing; and (2) the filter wouldn't need a sharp cutoff, so it wouldn't ring. The industry pretty rapidly settled on "8x oversampling" as a standard, meaning 44.1kHz input rates (CD) would be raised to 352.8kHz in the DAC, and 48kHz rates (DVD) would be raised to 384kHz in the DAC.
This oversampling is more properly called "interpolation." The thing is, when you do interpolation, you need filters for that, too. So it was kind of good news, bad news: no more bad sounding brick wall filters, but now the quality of the sound you got from the DAC's output depended on how good the interpolation filters were.
But wait a second, I thought the input to my DAC got turned into DSD along the way.
Yep, that's a little further along in DAC history. The 8x oversampling DACs mentioned above all used something called "Pulse Code Modulation," or PCM. Engineers found out you could do digital to analog conversion a lot cheaper using something called "Sigma Delta Modulation," or SDM. (DSD is a type of sigma-delta modulated signal.) Pretty soon the chips built into just about all DACs were *first* using interpolation filters for 8x oversampling to 352.8/384kHz PCM, *then* running the result (still in the chip) through a sigma-delta modulator to get a DSD-type signal, before finally converting that to the analog (music) output of the DAC.
This is still the standard way just about all DACs work internally today - interpolation, sigma-delta modulation, conversion to analog.
But hey, what about all this "upsampling" foofaraw W4S was telling me they just won't do, like it was being cruel to animals or something?
This involves another DAC problem you have likely read references to, called "jitter." Again I'm not going to go into any depth; just to note that jitter can cause distortion and DAC designers take steps to get rid of it. One thing DAC designers found out they could do to help minimize jitter is called "asynchronous sample rate conversion" - ASRC. This involves interpolation, which as you remember virtually all DACs these days do as standard processing anyway. But instead of interpolating to an integer (8x) multiple of the input rate, it's thought to be better for jitter minimization to interpolate to a non-integer multiple - for example, interpolating 44.1kHz input to 384kHz. This became known as "upsampling." Same exact type of mathematical operation - interpolation - just to a non-integer rather than an integer multiple. So why do W4S and a lot of other DAC manufacturers make the sign of the cross when you say "upsampling" instead of "oversampling"?
A few years after the advent of ASRC, a different jitter minimization technique with a similar sounding name was developed, called asynchronous USB input. Many people thought/think DACs with asynchronous USB input sound better than DACs using ASRC, and in any case async USB input was then (and is even to a fair extent these days) considered the "latest and greatest" DAC technology. So when the manufacturer of a DAC with async USB input tells you "Heavens no, we don't upsample!", they are most definitely *not* saying they don't interpolate in the DAC to high PCM sample rates and sigma-delta modulate the result before converting to analog. What they're saying is "We don't use none o' that stinkin' old-fashioned ASRC technology, nosirree Bob, nuthin' but the very latest async USB here!"
All right, long journey, but here it is in a nutshell: Yes, anything you feed to your DAC below 352.8/384kHz rates does get internally raised to 352.8/384kHz rates, then modulated into a DSD-type format before being converted to analog and sent to the DAC output. W4S wasn't denying that, they were telling you your DAC has what they consider the latest and greatest technology for jitter reduction and good sound.” [4]
----
We don’t need no stinkin' upsampling...
Time Waits For No-One...
The Saga of the Audio Note DAC 5 D/A ConverterThe Audio Note DACs are an exception, they don’t use any digital filters and only use 1x sampling. Enjoy the Music has an article by Peter Qvortrup, owner of Audio Note UK that explains his thought on digital reproduction.
https://www.enjoythemusic.com/magazine/manufacture/peterandac5converter..htm“After much trial and error we found a way of removing the digital filter and incorporating a carefully designed analogue filter after the D to A conversion. Needless to say, this goes so completely against the grain of all current opinion because a lot of the spurious signals above 20kHz are still present in the output of the converter after the analogue filtering. The filter is a 3rd order design with a silver wired inductor and silver capacitors and is so designed to slowly attenuate the higher harmonics in a natural way, preserving as much of the musical waveform as possible, each filter is dynamically matched to within 0.5dB of its partner in the other channel across the full frequency spectrum, to achieve best possible channel balance.”
A friend built the ANK 5.1 Signature – Level 5 Ultimate DAC and it sounds excellent, he also said Peter is developing a R2R Ladder Resistor NOS DAC, it will be interesting to see if the new R2R DAC will do high sampling rates like the Holo DACs."
----
R2R DACs vs. chip DACs vs. FPGA DACsTed Smith, PS Audio DirectStream’s lead developer, has a wonderful (long) thread on AudiophileStyle. I learn a lot from reading his posts.
https://forum.psaudio.com/t/r2r-dacs-vs-chip-dacs-vs-fpga-dacs/11518“In particular, if the DS [DirectStream] is upsampling by 2, then every other sample is identical to the input. The other every other sample contains the exact same information, the samples are filled in assuming the ADC was band limited, which all are.
"The neat thing is that if you throw away the original samples in the upsampled stream and then upsample it again you get the original samples as the new fillins. It’s hard to claim that the signal is being manipulated or distorted by proper upsampling. Nothing is lost…
"(The DS never upsamples by just a factor of two, but the same thing happens: upsample by 147, pick a sample, skip 146, pick the next, skip 146… and you have the original information. If you then upsample that by 147, you’ll find your exact original samples every 147 samples if you start at the correct point.)
"Not using an upsampling filter is what grossly changes the result. Especially in the high frequencies.”
----
In August 2020, Ted Smith talked about the NOS Philips TDA 1543 chip in the
BorderPatrol NOS DAC (and still used in 2022). His comment reminded me of my first DAC, a Scott Nixon Tube DAC I bought used in 2006 that also used the Philips TDA 1543 chip. The SN Tube DAC was a NOS design with no digital filter. The sound was pleasant but eventually I became bored because the music just laid there, there was no PRaT, and now I know why. The Van Alstine Insight I bought in 2008 was livelier but still rough around the edges compared to modern DACs.
6Moons has a review of the Scott Nixon Tube DAC
https://6moons.com/audioreviews/nixon/tubedac.htmlTed writes:
“It’s [Philips TDA 1543 DAC] a very simple chip. For people who want NOS or R-2R style DACs it’s great. Also most people don’t seem to read the datasheet, the chip requires an opamp to meet its specs (which aren’t great), but many simply use resistors on its outputs which adds a lot of 2nd harmonic distortion and restricts the dynamic range.
"If your favorite music doesn’t have a lot of high frequencies and it doesn’t get too loud, the chip works well.
"Without an output filter the timing and waveshape of impulses is lost and PRaT will suffer, but if the original recording is already rolled off this won’t be a problem.
I suspect most people that really enjoy orchestral or, say, grunge [or rock] would be disappointed, but jazz trios, women’s solo voice, etc. should sound fine.”
----
Ted Explains DSDTed Smith talks about PCM, DSD, noise shaping and how upsampling can make the music quieter. A fun rant about those little PCM/DSD chips in almost every DAC/CD player.
https://youtu.be/xafLYw6EuZ4----
Developing and Listening to SunlightTed Smith talks about the development of the Sunlight firmware for the DirectStream DAC: Quad rate DSD and getting the correct clock timing
https://vimeo.com/537470248----
HQPlayer Explanation by Jussi Laako - HQPlayer Developerhttps://www.audiocircle.com/index.php?topic=182724.msg1919512#msg1919512"First a bit explanation on time and frequency domain, please excuse me for some technical jargon. Frequency is signal change as function of time. Thus a signal has presentation in both frequency and time domains. "Linear phase filter" is a filter where all frequencies pass with same time delay. "Minimum phase filter" is a filter where all frequencies pass through as fast as possible, higher frequencies faster than lower ones. Longer/steeper filters change faster from passing frequencies to not passing frequencies as function of frequency. Shorter/gentler filters transition more slowly or "gently" from pass to stop as function of frequency. More accurately the filter wants to detect frequencies and transition pass/stop faster, longer time the filter has to "look" at the signal. This has side effect called "ringing" or rather "time blur". On the other hand, extremely short filter like a one that looks only at single moment cannot filter anything at all, because it sees only single point of time at once without any history or future (so it cannot detect any frequencies as those are a change over time). Linear phase filter takes equal amount of history and future into account during calculation. The problem in this is that it is kind of unnatural for something that is going to happen in future to affect already the present. Minimum phase filter on the other hand considers only from present to past, so it doesn't reflect things that are coming in future. This "ringing" is already in most RedBook recordings, since in most cases the ADC has gone through down-conversion and possibly another round at mastering from 24/96 or similar to RedBook. "Apodizing" filter is one that replaces or modifies this original ringing with it's own - that can be less than the original. All the filters explained below are more or less "apodizing" unless otherwise noted.
"Why is "filtering" needed? Because otherwise upsampling/oversampling produces alias (distortion) components in frequencies above the original one. In down-conversion case it is even worse, because those components are produces below the original ones. D-A conversion also produces these components above half of the sampling rate frequency, and those are then removed by the analog reconstruction filters.
Higher the sampling rate seen by the D-A conversion stage, simpler the following analog filter can be. Digital filters can easily outperform analog ones. Removing those spurious frequencies by filtering is called signal "reconstruction".
----
And now you understand upsampling, how it works, and terms like ringing, interpolation, steep filters and noise shaping make sense. Just don’t try to explain it to someone at a party.
----
ConclusionAll DAC’s, CD players and software music players use some sort of digital or analog filters and possibly some behind the scenes upsampling. How well they sound is up to you. I think one reason for equipment churn is the stock filters chosen by a manufacturer doesn’t work well with your electronics and over time a deficiency is heard (or boredom sets in). A software player/upsampler like HQPlayer that allows a user to fine tune a digital system is one solution. But buying a new DAC is a lot more fun; all you need is luck and money.
[1]
https://www.psaudio.com/pauls-posts/here-we-go-again/[2]
https://www.psaudio.com/pauls-posts/getting-something-for-nothing/[3]
https://www.psaudio.com/pauls-posts/the-world-is-dsd/[4]
https://audiophilestyle.com/forums/topic/24623-dsd-upsampling-over-original-format/#comment-446682
