Something I have noticed over the years commenting on the audiophile hobby has been how incessant and persistent various themes tend to be. Just like the apparently never-ending arguments of "digital vs. analogue/vinyl", or "CD vs. hi-res", or "subjective vs. objective", there has been this mostly friendly banter between those who feel that essentially "bits are bits" vs. those who think there is significantly more to digital transmission than bit-accuracy.
Seeing recently this article "Why the 'Bits is Bits' Argument Utterly Misses the Point" from Upscale Audio published compelled me to write this post to explore the topic further with a review of measurements and some demo tracks for readers to listen to themselves. I don't know how long the Upscale article has been on the site since there's no date or author listed, and was only made aware of it through the Darko.Audio Facebook page (it seems Mr. Darko felt the article was accurate, really?).
While the article claims that some people have "missed the point", let us examine their points and see if perhaps it might be the author(s) that are a bit too aggressive in making these arguments. After all, it is 2019 with decades of development in digital technology that impact our lives in more sophisticated ways than just audio reproduction. It's hard to imagine there are huge lacunae in our knowledge of digital communications and digital-to-analogue conversion of audio frequencies.
0. A preamble about digital...First, remember why digital is "good". Digital data consists of discontinuous representations of information quantized as either "0" or "1" which we call the "bit". With higher speeds over the years and more "0" and "1"s available, we can represent evermore complex information from ultra-high resolution audio, video to bewildering amounts of "big data".
Since each bit is either "0" or "1", mechanisms are in place to ensure integrity and protocols allow excellent accuracy ("perfect" even) for data transmission. The electrical signal for "1" is significantly different from "0", thus making it comparatively easy to transmit the data without guesswork when the bits are transported. For example, USB uses a differential voltage pair of wires with "1" being when the D+ line is ≥200mV higher than D- and the converse, digital "0" being when D- is ≥200mV higher than D+ on the receiver end. That differential voltage between D+ and D- of ~400mV is not a trivial amount and provides good signal-to-noise ratio such that we can quite easily these days enjoy high integrity, multi-megabyte-per-second data transfers with commodity, consumer devices even at very low cost.
[There is more complexity to the USB situation around modes of operation and speed identification, etc. especially with newer iterations. No point getting bogged down with this stuff; you can read some more here if interested.]
Remember that years ago there were some tests of USB cables using eye diagrams in Hi-Fi News & Record Review. These "eye diagrams" can tell us all kinds of information like signal-to-noise, jitter, rise time, bit period, etc. and whether all of these parameters over thousands if not millions of captured samples measure up to the specifications to ensure that bit errors are within the acceptable tolerances. Not just USB, but all digital interfaces (SATA, ethernet, S/PDIF, PCIe...) can go through these kinds of checks.
|Sample USB eye diagram showing the ~400mV differential, 2-level Pulse Amplitude Modulate (PAM2).|
Although USB data transfer consists of just 2 differential voltage levels, the article makes reference to more complex systems like Gigabit Ethernet with 5 levels (PAM5), and 10GbE which I wrote about last year with 16 levels (PAM16, referred to specifically in the Upscale article). They have even funkier eye diagrams than USB:
Of course this is complex, but the nature of technology is such that over the years, complexity becomes commonplace and easier to achieve with commodity components. As the signals get more complicated and speeds go faster, yes, things like cable specifications do need to be better. As long as compliance testing has been performed and the cable achieves the specified parameters, it's fine. Even if expensive to begin with, the "magic" of market forces including economy of scale eventually makes technology affordable.
Note that not all digital communications implement error correction. For example, if you're playing a video game online and transmitting using the UDP protocol over ethernet, you just want fast "realtime" data transfer with minimal latency to the server and the occasional data error likely isn't going to be critical for gameplay. Back in 2016, I explored this with the ODROID-C2 machine looking at ethernet error using UDP which was still very low without error correction across a home network. By the way, these days, other custom protocols can be implemented over UDP so that the server/client can be smart enough to selectively control what data is important and which packets with errors can be ignored.
For audiophiles, other than ethernet with TCP transfers, most digital audio transmission actually do not have error correction (even if detected). The old S/PDIF interface (coaxial, TosLink) did not have error correction since it was a unidirectonal flow from source to receiver, and USB data transfers to your DAC using today's asynchronous protocols do not implement error correction (even when detected, your DAC will not ask the computer/streamer to resend an erroneous data block). Like playing the video game online and not noticing the occasional error, the vast majority of the time you won't hear an issue either with the audio stream. However, when data errors are sufficiently numerous, as I demonstrated here with a very poor cable, you will notice the problem from non-bit-perfect data transmission.
I hope everyone is in agreement that digital therefore provides an exceptional level of error-free storage and data transfer; a level of "perfection" if you will. "Bit-perfect" copies of music and "bit-perfect" transmission of this data to one's DAC is simply to be expected and normal. These days, "non-bit-perfect" transmission to a DAC likely suggests malfunctioning hardware/drivers, unintended software settings (like running audio through an OS software mixer), or intentional data manipulation (eg. EQ or DSP).
With this preamble, let's get back to that "Bits Are Not Bits" Upscale article then and consider why there are major issues with what they're saying / implying. "Bits are not bits" arguments generally agree that the signal remains "bit-perfect" but feel other anomalies significantly still affect the sound.
I. Noise: "Digital is actually analogue" exaggeration...Yes, the signal that represents the digital data does electrically manifests as an "analogue" waveform. Of course it does, and indeed the rising and falling edges of the signal are not perfect square waves; we can easily see this in those eye diagrams above. However, the reality is that transfer techniques can tolerate imperfections with significant margin such that ultimately the data remains error-free; the digital data remains "perfect".
This point about "there's no such thing as digital" was somehow made into a big deal as I recall by John Swenson back in 2013 in this interview. To this day, I still see the occasional reference to the idea as if this was some remarkable revelation. He said in that article (bold emphasis mine):
At a first glance let’s look at a voltage on a wire. It can have many different voltage levels on that wire, 0, 10, 300, -2.75, 13765.45 etc. This is the infamous “analog” realm, the voltage on that wire can be pretty much anything.
What digital does is quantize those values and say “anything below a certain value (the threshold) is low and anything above the threshold is high”. This is the fiction part.While most of what he says is true, it's that little twist at the end suggesting "something's not right" that is problematic. Notice this is often how people start conspiracy theories ;-). So what exactly is fictitious?
Notice how in the article, from that questionable statement on, Mr. Swenson wanders into speculative territory suggesting that "noise" now can somehow magnify and cause problems, yet still below thresholds so the digital data remain error-free. He then proceeds with dragging in concerns about the ground plane and yet more noise that might or might not be problematic, resistance and inductance that might or might not be issues, then he throws in capacitance effects, and the "return current", and so on and so forth... Sure, maybe there are some really terrible and noisy USB DACs out there. But all of this piling on of speculations without actual reference to specific devices or magnitude of the issues he's talking about does nothing but overwhelm the reader, thus directly sowing the seeds of fear and uncertainty! What a mess of an article with no context provided yet some writers in the audiophile community seem to blindly hold these speculations up as the words of some kind of enlightened "brilliant" designer?
By 2015, much of these speculations made their way into the "one-port USB hub" known as the UpTone USB Regen, and more speculations can be found as "Swenson Explains". Despite lots of words, that page still provides essentially no details as to the magnitude of the problem he's trying to "fix" nor does he show evidence that the device even did anything to the sound from a USB DAC. Others examining the device and later variations apparently could show no improvement, or perhaps even slightly higher noise (yet still garnering positive subjective testimony).
You see... The problem here is not necessarily that the "bits are bits" people (like myself) deny the existence of noise in digital circuitry, or claim that the electrical waveforms are perfectly square. Rather, there's no evidence of what he says should be of concern to audiophiles using reasonable digital gear for a long time now!
For example, over the years of listening and testing, I have never been able to show that noise level was high with simple "USB-stick" DACs like the AudioEngine D3, GeekOut V2, SMSL iDEA, or even the AudioQuest Dragonflys (v1.2, Red, disappointing performing Cobalt) from a few weeks back despite their proximity and use of a computer USB port for power and data. Furthermore, as I showed in late 2018, using even a USB TEAC UD-501 DAC from 2013, there was no severe worsening in noise from the DAC output whether I fed it with a low power, "quieter", battery-powered Raspberry Pi 3 B+ or a mains-powered Intel i7 computer with nVidia GTX 1080 GPU from audible frequencies up to 192kHz. At most, when the power-hungry CPU + GPU were running at 100% (who does this while listening to music?), only a few tiny noise anomalies could be seen! Here are a couple of previously published graphs to demonstrate this point:
As far as I can tell, there is no need for esoteric "audiophile" equipment or the USB Regen as filter / signal "regenerator" since noise levels are already excellent. Can John Swenson show us where it is he sees problematic noise in his digital systems (USB or otherwise)? Perhaps give us an example after all these years where the USB Regen actually improved the performance of even an inexpensive DAC?
So what about "noisy" network switches and the like, also mentioned in that article? As shown years ago, on a relatively complex home network like mine (you can see the network configuration with various devices here), ethernet noise with multiple switches and computers on the network remains minuscule - here's an example using a battery-powered Raspberry Pi 3 streamer with DAC:
|(Don't worry about that 37kHz noise... It's a limitation of the Focusrite Forte, not to do with ethernet noise.)|
Again, if there really is a problem, why doesn't the Upscale Audio article writer(s) demonstrate examples where digital cables show the problematic "antenna effect" that they claim? While the idea sounds fine, where is there evidence that more "shielding against gigahertz transmissions" will improve a typical DAC or streamer's output? Could it be that the engineers who produce modern DACs already know about these issues and have figured this out years ago?
BTW, speaking of blurred "digital as analog" ideas, remember that others in the audio media have made similar claims/comments such as UHF Magazine stating that "the CD can be said to be an analog disc" back in 2014. Yikes. Thankfully, I think audiophile magazine writings have improved somewhat since then and I haven't seen such types of comments in awhile.
II. Timing: The jitter exaggeration...Let's not spend too much time here since measurement after measurement of jitter with modern asynchronous USB DACs have demonstrated excellent performance these days. Ditto with ethernet-based DACs. Yet, jitter remains some kind of perennial beast that must be tamed in the eyes of various audiophile writers, magazines, and of course certain companies!
While jitter was worse back in the days of synchronous USB and with older S/PDIF equipment (much improved these days), there is no evidence that a cable is able to change jitter performance - it cannot. Jitter is a function of the sending and receiving devices themselves when using reasonable-length cables adhering to standard specifications. For synchronous USB, S/PDIF, HDMI, the sending device's clock accuracy will affect jitter performance while with asynchronous USB and ethernet, the accuracy of the clock is based off the DAC itself.
Again, have a listen to the simulated jitter samples I posted last year. As you can hear for yourself, jitter has to be at massive levels before it can be heard. Unless you truly own terrible (or faulty) audio equipment, I doubt jitter will ever be audible. I would love to see results of a blind test with typical picosecond jitter showing otherwise.
III. Timing: The drift exaggeration...This is mentioned in the Upscale Audio article a few times briefly. Drift refers to the idea that clocks do not keep perfect pace. True, clock oscillators are not perfect and will function at slightly different speeds. Again, like jitter, let's not get too concerned - why? Because we're still talking about tiny parts-per-million (ppm) levels of inaccuracy!
Have a look at this page "Clock accuracy in ppm" for an interesting overview on clock precision between typical crystals, TCXO ("Temperature Compensated"), OCXO ("Oven Controlled"), rubidium, and cesium atomic clocks. In the table describing accuracy of the various types, notice that a typical crystal may have accuracy as poor as +/-100ppm. That sucks, right? But then again, that's an inaccuracy of only 8.6 seconds out of 24 hours! Do you think this temporal inaccuracy which will lead to a very slight change in pitch is going to be noticeable?!
We can actually measure the drift between two devices as hobbyists... Check out the new software DeltaWave Audio Null Comparator currently in beta (1.0.37b). Years ago, I had used Audio DiffMaker which worked well in certain situations but had its limitations. DeltaWave goes above and beyond DiffMaker's abilities by a big margin!
Remember earlier this year, I ran the "Do digital audio players sound different playing 16/44.1 music?" blind test? Well, we can load up the same track recorded from different DACs, and compare the temporal drift between the DACs (same highly accurate RME ADC used of course). Loading up the Stephen Layton "Chorus: For Unto Us A Child Is Born" track played back from the ASRock motherboard and comparing this to the Oppo UDP-205, we see this very linear graph of clock drift in DeltaWave:
|Note: Y-Axis is the # of samples of offset. Above 0, the "Compare" track is faster, and number of samples "ahead" vs. the "Reference". For this comparison, I had set DeltaWave to upsample to 192kHz,|
If I just examined the drift of the Oppo UDP-205 itself compared to the "ideal" bit-perfect data, there's barely any drift as recorded off the RME ADC - <6ppm!
Are 30 and 6ppm variations like this audible? To help answer that question, let me create some demo samples for you to download and listen for yourself.
As audiophiles, we love Diana Krall :-). So to start, we have a 90 second clip from her song "Let's Fall In Love" off the album When I Look In Your Eyes. (As usual, samples used under "fair use" for the purpose of education... Please delete the files when done listening. If you like the music, please purchase it.)
Track 1 is the original 90 second segment of Ms. Krall with a 2-second fade out. Track 2 was manipulated with drift going back and forth between 100ppm and 200ppm (remember, typical low quality crystals may drift +/-100ppm).
If we compare the original with the "drifty" version, the Clock Drift graph looks like this in DeltaWave:
|DeltaWave settings: upsampled to 192kHz.|
0 - 10 seconds - no changeThis is significantly worse than the difference between a good DAC like the Oppo UDP-205 and the built-in audio from a computer motherboard!
10 - 40 seconds - slowed clock by 100ppm (0.01%)
40 - 70 seconds - sped up the clock by 200ppm (0.02%)
70 - 90 seconds - slowed clock by 200ppm (0.02%)
Focus particularly at the transition points at 40 seconds and 70 seconds. At 40 seconds, we're going from 100ppm slower to 200ppm faster - a change of 300ppm. At 70 seconds, the sound will go from 200ppm "fast" to 200ppm "slow", a sudden 400ppm change in speed/pitch! These kinds of "speed shifts" are beyond even the worst DACs out there.
Can you hear the timing difference? Can you hear the change during those transition points? How bad is it? If you're wondering what to listen for, drift will result in very minor pitch variation depending on amount of drift. A 200ppm error is something like 0.3% of a semitone shift in pitch. Would even the most "golden eared" audiophile detect this difference with the best gear and multi-thousand-dollar cables?
But wait! There's more :-).
Let's add some "nasty" jitter! Like I did with the simulated jitter samples linked in the previous section, let's use Yamamoto2002-san's WWAudioFilter (188.8.131.52) and throw in a few jitter sidebands - in total 5.5ns worth of cumulative jitter resulting in sideband distortion up to +/-5kHz from the primary signal. This is what the 16-bit J-Test would look like with this amount of sinusoidal jitter added:
You will not find this high amount of jitter in any half-decent DACs out there. Such a level of jitter should automatically disqualify this DAC from being called "high-fidelity" and objective reviewers would rightly criticize the performance (even if subjective-only writers don't notice!). Track 3 is the sound of Ms. Krall with this amount of jitter applied to the data.
Finally, let's put the duo of fluctuating drift and jitter together. We have Track 4. Have a listen...
Not good enough, you say? Don't like Ms. Krall's music? Okay then, how about a hi-res big band style male vocal track instead?
Thanks to Dr. Mark Waldrep, I've also included similar manipulations with an excellent sounding sample from AIX Records; Steve March-Tormé's "On The Street Where You Live" from Tormé Sings Tormé. This is a true high-resolution 24/96 recording with excellent dynamic range as usual with these AIX recordings. Have a listen to Tracks 5-8. Like with the Diana Krall sample, I've included processing but with even more intense +/-200ppm speed fluctuation every 10 seconds through the whole clip! Flipping back and forth between +200 and -200ppm timing variation results in the equivalent of 400ppm sudden change each time. For Track 8, we have both the 400ppm fluctuation every 10 seconds along with 5.5ns jitter.
Here is the graph from DeltaWave showing the temporal drift pattern described above for "On The Street Where You Live" Tracks 6 and 8 (Track 7 is 5.5ns jitter added with no drift over the seconds):
|DeltaWave settings: upsampled to 192kHz.|
So, dear readers, based on what you hear with these temporal anomalies "baked in" compared to the original music, do you think you should be worried about ppm variation even worse than typical standard crystals plus jitter from maybe the worst of the worst commodity DACs?
These days we have companies selling products with very high quality clock stability featuring "femtosecond" accuracy for low jitter and "master clocks" with very low drift. Remember that clock drift can be an issue in production studios when multiple devices have to run in synchrony, especially with audio-visual projects. In that situation, having a master clock might make sense (here's a good article). However, when listening to music at home with a single DAC, would something like the dCS Vivaldi Master Clock (US$14-15k) which promises +/-1ppm accuracy make sense? And obviously this begs the question of just how inaccurate is the Vivaldi DAC's (~US$36k) internal clock to begin with such that adding a Master Clock supposedly improves sound!?
Furthermore, what then do you make of this John Quick of dCS interview on AudioStream recently? Is there anything in the content of that interview that provides a cogent argument for owning a master clock (other than maybe bragging rights or the box looks cool)? Did the contents of that article clarify anything for you or did it just make things more obscure?
[For the record, I have listened to dCS DACs before. The Ring DAC sounds good, and workmanship is top notch. But this doesn't mean that the claims around the Master Clock should not be questioned or one should not consider the cost-benefit.]
IV. Concluding thoughts...If a non-audiophile reads the Upscale Audio "bits are not just bits" article were to come and ask you:
"Wow, audiophile friend! When did you realize that there were so many issues like 'noise' from digital cables, timing problems such as 'jitter' or 'drift'?"
What would you say?
Did you hear noise in the digital transmission between your CD transport and DAC over the years? Did you personally hear jitter? Have you noticed significant pitch differences between CD players and DACs, thus concerned about the accuracy of the crystal oscillators and drift?
Perhaps not surprisingly, when I speak to audiophiles over the years, they generally don't seem too concerned about these issues firsthand. Most couldn't really describe what jitter sounded like (again, have a listen here). I've never heard of an audiophile complaining of terrible drift. Rather, most would point at claims made by manufacturers or refer to online press and magazine articles as being the source of their discontentment.
As you can see in the subheadings above, I used the word "exaggeration" to describe the noise and temporal anomalies being suggested by the Upscale Audio article. These are not the only exaggerations - the article also makes claims about SMPS noise which we have already addressed recently, and also points at digital filters which of course can be different depending on how they were programmed and has nothing to do with digital data or transmission (a topic well discussed over the years). Like with other claims, the article just "hits and runs"; mentioning these issues quickly with no actual examples or depth to the discussion. I find this style of article-writing rather unfortunate, irresponsible even, and endemic in much of the Industry-sponsored audiophile literature (here's another example from iFi/Abbingdon).
Overall, remember that digital is not like analogue audio where noise is commonly a factor in the storage medium (eg. dirty, poorly pressed, damaged or warped vinyl LPs), or have significant timing abnormalities in the playback system (eg. wow and flutter of turntables are orders of magnitudes worse than digital jitter or drift!).
When it comes to digital audio, if you hear a big difference between different devices, IMO, don't freak out about inherent "noise" in digital technology (any poorly designed equipment can have noise) or temporal error (jitter or drift) as the
As audiophiles, I agree that we should aspire to achieving the best fidelity we can, including to the point of insuring that the resolution is beyond human perception. As such, I'm certainly not suggesting that we should be satisfied with 1ns cumulative jitter, or be happy with low-accuracy 50ppm clock oscillators (much less >5ns jitter and 100-400ppm/0.01-0.04% variance as in the demo tracks)! As I said in my SUMMER MUSINGS a month back, it's good to not become neurotic slaves of a hobby either as extreme "subjectivists" blown away by all manners of testimonies or extreme "objectivists" chasing after femtoseconds and down to the smallest of decibels. For me, it's about that balance of achieving a level of science and engineering for the quality of reproduction we desire that can be verified objectively, validated for ourselves with enjoyable subjective listening and getting it done with reasonable value for the money. The wise audiophile also recognizes that there are commercial influences out there desiring to distort and misinform, most likely for financial reasons.
I therefore must agree with the aspirational "Most Interesting Man in the World" pictured at the start of this post. Based on what I have heard and found with today's gear, subjectively and objectively, indeed "Bits Are Bits" as far as the consumer/audiophile is concerned. If you don't like the sound, the issue has to do with the device itself, not some esoteric "bits are not bits" rationale. Only poor quality equipment would need any special attention to "noise" filters for digital data communication, exotic clocks for accuracy, or requires anything more than normal digital cables.
"Stay thirsty my friends" for good music, and justifiably skeptical of vague audiophile articles.
Hope you're enjoying the music as we head into the latter half of August...
Addendum: August 18, 2019
With all this talk in the comments of HDMI and "snake oil", readers might be wondering what a "non-snake-oil" company and the types of technical information they provide can look like. Here's Exhibit A:
While it is unfortunate that Oppo no longer makes audio gear, that post is a beautiful example from 2017 of what honesty and technical competence looks like when a company wants to engage and educate the customer.
Notice the technical background information and rationale provided. Notice the fact that they demonstrated with objective results how the goal was achieved (eye diagram, J-Test). While they understandably did not disclose the comparison Blu-Ray player they measured from another company, the fact that they compared the UDP-205 to another device is a good sign that they examined the competitors' products.
Notice also that there's a humility to the tone of the presentation. The HDMI jitter was reduced from 53.82 to 50.67ps only (6% reduction). No dramatic claims about improved "air", veils lifted, soundstage "opened up" or any such flowery talk. As Oppo said, already jitter performance is excellent and while they were able to improve it even further, I highly doubt the engineers would have proclaimed these improvements are being massively audible! Such is the advanced level of digital performance these days. (Objectively, I was able to verify indeed that the HDMI "Audio" output performed with lower jitter when audio sent to my Yamaha receiver.)
Would we ever see forthright discussion articles like this from John Swenson/UpTone, Mapleshade, AudioQuest, Nordost, Synergistic, Shunyata and countless others for their "digital" gear and cables? Would we ever get articles of this technical magnitude from the likes of specialist stores like Upscale Audio? I certainly would not hold my breath...
Remember, the Oppo UDP-205 when it was available up to earlier 2018 only had an MSRP of US$1299. How many pieces of cable by some of the companies listed above cost significantly more than that!? And with what justification?
Addendum: July 2021
For further reading on a specific device that makes certain claims, consider the AudioQuest Jitterbug FMJ and how useless it is.