by Allen Farmelo
with Matthew Agoglia
“If it sounds good, it is good.”
– Duke Ellington
“The commonly held intuitive notion that internal clocking must be better than external clocking is simply not always true.”
– Grimm Audio, “Clocking” White Paper
“I get on my knees and pray / that we don’t get fooled again.”
– The Who
This report is part technical paper, part cultural history, part investigative journalism, and part gear review. The cultural history section of this paper considers how audio clocks went from being thought of as synchronization devices to aesthetic devices (affecting sound). The emergence of entrenched debates over external vs. internal clocking is also considered in depth, and an attempt to draw those debates to a close is made. The investigative journalism centers around a first-in-print disclosure of the Digidesign 192 converter’s inherent jitter, a fact that helps to clarify the rise of aesthetic clocking, as well as the confusion and debates around it. The technical paper content puts forth an accessible technical paradigm with which the end-user of digital audio systems, who we presume lacks a deeper technical understanding, can begin to more effectively evaluate those systems. Some deeper technical concepts are explored in an accessible manner, and recommendations for listening are made. Finally, the gear review section presents the first comparative round-up of 10 MHz audio clocks, a relatively new breed of clocks for the pro audio market designed solely for their sonic impact, and we compare them to standard crystal-based generators, as well.
A Note on Reading
Because this is a long report that explores many ways of looking at the complex topic of audio clocks, it may be challenging to read. Much of what we have to say is divergent, and therefore it is tucked into footnotes – which for this online version you can conveniently hover over and read in a separate window that will pop up (you can also click the footnote to read it at the bottom). We have also included a table of contents that you can use to leap to different sections of interest. We do, however, encourage a top-to-bottom close reading, as we believe this report (when taken as a whole) presents a healthy intermingling of multiple perspectives on the topic.
Table of Contents
- Word Clock – a signal used to synchronize digital audio devices
- Internal Clock – an oscillator inside a digital audio device that drives the conversion process
- External Clock – a free standing clock signal generator that is synchronizing another digital audio device
- Jitter – deviation of a periodic signal (such as Word Clock) from its presumed ideal timing
A Menacing Topic
Digital audio clocks have become a menacing topic for the pro audio field. The technology is exceedingly complicated, knowledge of it esoteric, and the average end-user’s grasp of it is typically flawed and misleading. Even more menacing is that, despite this general lack of understanding, the pro audio field has witnessed a decade-plus of entrenched disagreement between those who are certain that external clocking can never improve an interface’s performance and those who assert that it can and usually does. The argument really heats up when the anti-external camp (typically technicians) accuses the pro-external camp (typically practitioners) of having fallen prey to placebo effect and marketing hype. No one likes being told their ears are elluding them, especially those whose professional reputations rest on the acuity of their hearing.
The core problem in the argument, however, is actually the tendency for both sides to form generalizations based on their individual experiences. Here’s why: because any combination of clock, converter and/or digital interface forms a unique digital audio system with its own highly specific performance profile, there really is no practical way to predict how any combination of clock, converter and/or interface is going to perform together. Forming broad conclusions about internal vs. external clocking – or about the efficacy of one specific product, or even about any type of product – is truly a flawed enterprise. Because manufacturers pretty much never disclose key information about their products’ abilities to perform with other types of devices (and that’s presuming that they even know), and because measuring digital devices is a prohibitively costly and esoteric task, it is difficult to make much meaningful headway beyond listening to specific systems. However, because performance is so system specific, any generalizations based on listening also fail us when it comes to clocking audio systems.
Interestingly, generalizations about internal vs. external clocking fail whether they are formed via subjective listening or via objective measurement. We will deal with this point thoroughly below, but for now suffice it to say that readers who strongly side with one or the other method of evaluating audio equipment will find their interests addressed below.
If the behavior of digital clocks and converters is so system specific, how is it that such entrenched generalizations about internal and external clocking emerged? One explanation is that popular magazines failed to publish the accurate suggestion that the efficacy of external clocking is essentially a crap-shoot, but these magazines did publish the inaccurate notion that internal clocking always trumps external.1 Certainly gear reviews (including our own) and marketing campaigns (often elegantly synchronized with gear reviews) have played a major role in promoting the idea that external clocks were a must-have. But there were also technical papers that made sweeping (and therefore flawed) generalizations that internal clocking was by definition always better. Interestingly, there is also a specific product that has contributed to the widely held belief that external clocking improves the sound of a system: that product is Digidesign’s 192 model of converter and digital interface. Considering the 192’s strange legacy is a good inroad into this menacing topic.
Digidesign 192 & The Rise of Aesthetic Clocking
In 1994 Digidesign (now Avid) impressively asserted its stronghold on the professional DAW and converter market with their 16-bit Pro Tools TDM systems. These early Pro Tools systems required the use of Digidesign’s 8XX series converters, thus establishing the company’s long-standing practice of making third-party hardware incompatibile with Pro Tools. In 1997 Digidesign released 24-bit TDM systems, and then in 2002 Digidesign asked the industry to make a major shift to their new Pro Tools HD systems. Pro Tools HD required Digidesign’s new computer-mounted processing cards as well as their “blueface” 192 converters and digital interfaces that interest us here. Eventually there were a few third-party interfaces that could trick Pro Tools into working with them (Apogee and Lynx were the most popular), but these alternatives to the 192 were relatively uncommon. In essence, during the 2000s the 192 became ubiquitous in tracking and mixing studios, and thousands of those systems remain in use to this day.
While certainly an improvement over Digidesign’s earlier converters, the sound of the 192s disappointed many engineers and producers. Some of these people noticed that clocking to another brand of external master clock improved the sound of the 192 converters and interfaces, and by 2003 Apogee began advertising their Big Ben external clock as capable of “bringing clarity, depth and accuracy to recording and mixing that will be heard immediately.”2 The Big Ben was a big hit, especially among those using Pro Tools HD systems, to whom Apogee and their dealers marketed directly and aggressively. It won a TEC award in 2004.3
Until the Big Ben marketing campaign, the idea of using an external clock for aesthetic reasons was relatively new to the broader pro audio field. Previously, master clocks were boring boxes that synchronized various digital devices, but by the middle of the decade the pro audio field had become obsessed with the idea that external master clocks were key ingredients in a digital studio’s sound. Between the mounting distaste for the sound of the Digidesign 192 and the powerful marketing behind Apogee’s Big Ben, a new market for “aesthetic clocking” was born right at the peak of one of the biggest economic bubble’s in modern history. It didn’t take long for other companies to start marketing their clocks as sonic enhancers. Most notably, in 2008 Antelope Audio released the first 10 MHz clock aimed at the pro audio market for a total package price of around $8000.4 Antelope’s marketing department was savvy and well connected. Quite quickly, an elite echelon of mixing and mastering engineers including Howie Weinberg5 and Michael Brauer began to endorse the Antelope “10M” clock. The era of aesthetic external clocking was in full swing, and all the while Digidesign continued to produce and sell thousands of 192s straight through the end of the decade.
Was The Digidesign 192 Actually That Bad?
If you had concluded that internal clocking was always better than external clocking, you’d have to conclude that those who spent $8k on an external clock had grown delusional. Accusations that devotees to external clocking had caved in to placebo effect and/or marketing hype were not uncommon. However, for those who worked on Pro Tools systems all day every day, it was obvious that external clocks were improving the sound of these rigs. With nothing but subjective impressions to go on, some of the most reputable recordmakers grew convinced that external clocks – and particularly that $8k Antelope 10M, if you could afford it – were a must-have. As it turns out – in fact – these practicing subjectivists were right.
Though the evidence has existed for years, we will now expose for the first time that the 192s actually did measurably exhibit a marked reduction in jitter when clocked externally. What’s more, when synchronized to different models of clock (e.g., an Apogee Big Ben vs. an Antelope 10M), the sonic character of the 192’s output does indeed change. This was why, despite the unavailability of hard evidence to support the notion, so many people began to talk about the sound of clocks.
As a professional field, we could have reached consensus around the jitter profile of the 192 much earlier – if only we had connected one single measurement to the mounting subjective preference for external clocking. That missing measurement has been hiding out in a highly readable and publicly available Grimm Audio white paper6 since 2008, the same year Antelope released their 10 Mhz clock.
In their paper, Grimm Audio explains very clearly how digital audio clocks work, then – in Section 7 – we learn of “a popular DAW system” that is consistently shown to be improved by an external clock. What Grimm Audio fails to mention is that this “popular DAW system” was Pro Tools HD running a 192 interface7 – the most widely used digital audio system at the time, responsible for tracking and mixing a likely majority of the world’s recordings for well over a decade.
We understand why Grimm Audio – an equipment manufacturer whose reputation has been built on the quality of its products rather than on competitive marketing – chose not to name the Digidesign 192 in their paper. At the same time – as consumers and end users – we can express our disappointment that the authors had to make that choice. Not specifying the 192 is indicative of a kind of professional courtesy that, in the end, fails to connect technical research to the end-user base in direct ways (more on this below). Understandably, applying journalistic pressure on a relatively large corporation is not the goal of a technical white paper, but at what point does a discovery that could expose a significant shortcoming in a ubiquitous professional product – and possibly even improve the sound of recorded music world-wide – go from an anonymous example of a theory to something that ought be named specifically and publicly?
When a company like Avid virtually corners a market by intentionally making it difficult for third-party hardware developers to build compatible products, it becomes especially important for the user base to remain wary of the technology and to become vocal when it falls short. We professionals spent over a decade arguing in the abstract about internal vs. external clocking – and even more abstractly about subjectivism vs. objectivism – rather than openly and collaboratively investigating and revealing the true nature of the specific tools we were using. This should serve as a lesson going forward. We need to build bridges – not trenches – between objective and subjective evaluations if we are going to hold the noses of tool makers’ like Avid to the grindstone of excellence.8
Feb 4, 2016 – Upon reading Pink Paper #002, Eelco Grimm reached out and sent us the following information regarding the Digidesign 192. It’s a fascinating look at why, and how, this machine was designed with so much inherent jitter. Grimm’s comments can be read in this footnote –>9
It is one of our aims here to push our career field to demand better digital recording devices. This is not to suggest that as a field we are not concerned with the quality of our tools, but being good at being concerned – that is, being an educated and sharp consumer base vis-a-vis the manufacturers who build our tools – has not been our strong suit for quite some time. Let’s consider why that might have been.
How Confusion and Disagreement Emerged
How is it that we as a career field we spent over a decade missing the opportunity to connect readily available (and aligned) objective evidence and subjective opinion of something as important as the fact that the majority of our professional audio systems had inferior internal clocks and were thus forever embedding less-than-ideal amounts of jitter into countless recordings? Why, instead, did we start bickering so aimlessly on the topic?
The short answer is that, as a profession, pro audio has lost some of its knack for building meaningful consensus around both technology and aesthetics, and especially around the subtle interplay between them. This loss has a lot to do with the rise of the DAW and the amateur recordmaker, but also with how knowledge, information and communication have transformed in the digital age.
We can call the 2000s the decade of prosumer audio, a time during which the pro audio industry (its manufacturers, its marketing departments, and its publications) rapidly shifted from serving the commercial concerns of professional studios to serving the artistic concerns of individuals. Thousands of untrained people (mostly musicians) were jumping into recording, mixing and mastering, and given the mid-decade economic bubble, these eager novices also had a lot of money to spend. Suddenly a new wave of prosumer oriented publications, as well as ad-supported online discussion forums (notably Gearslutz), were blasting ads and how-to tutorials that sent the message that recordmaking at the pro level really only required some cash for gear and a bit of boiled-down knowledge. “Buy this, do that, realize your vision” became an industry refrain.10 Simultaneously, large studios were closing up shop, thus cutting off the humbling, multi-year apprenticeships that had served as the transmission lines of deep knowledge for nearly a century.11 By the end of the decade, Gearslutz had swelled to unprecedented size and – rather frighteningly – had become the primary means by which information and knowledge about recording was being shared.12
Not surprisingly, many experienced pros fortressed themselves in invitation-only discussion groups, coming out mostly only when hired by Gearslutz to do a few days of public Q&A. All the while the AES struggled to build a younger membership beyond those interested in specifically technical careers, though the lack of appeal for the artistically minded amateur and the price of admission were significant barriers.13
The point here is not to bash amateur recordmakers, or to gripe about the inescapable downsides of populist technology, or to wax nostalgic for a lost era. This point is to show how – during this highly volatile decade – the field was losing its grip on how to build meaningful consensus around the delicate interplay between recording technology and practiced aesthetics. This was definitely not a professional climate in which nuanced white papers or peer-refereed journal articles were garnering much conversational bandwidth (let alone comprehension), and a topic as complex and subtle as clocking digital audio systems was likely to have been grossly mishandled, as it was.
Another explanation as to why debates over clocking grew so heated and unfounded has to do with the human tendency to form general principles from specific observations. This is generally known as inductive reasoning.14 It has helped our species survive (“Big orange cats are dangerous!”), but inductive reasoning also sits at the center of our proclivity to form inaccurate generalizations (“Big orange cats attack on Thursdays!”). As we’ve noted, because clock performance is so system specific, generalizing from specific instances serves us very poorly when it comes to making sense of digital audio clocks.
That all-too-human tendency to form faulty generalizations in a field of largely untrained practitioners gathering in online forums and gobbling up a new wave of marketing targeted straight at them added up to a decade-plus of aimless bickering.15
However, not all the arguing was going on among amateurs, and more than a few experienced professionals became embroiled in the debates over clocking. Pro-level rifts often emerged between mastering engineers (MEs) and mixing/tracking engineers (M/TEs). MEs typically only require two channels of conversion and rarely rely on Pro Tools, so they are free to choose from an array of audiophile-grade stereo converters. Meticulously designed and built with top-shelf components, mastering converters typically do not sound better (or all that different, and often worse) when clocked externally. In fact, some of them do not even have ports for externally clocking the D-A.16
Knowing what we know about the converters MEs and MT/Es tend to use, it’s fairly easy to see what happened: MEs rarely heard an improvement when clocking externally and M/TEs usually did. The rift between these two points of view arose when members of these two groups began to assume that their systems were indicative of all systems. Even the pros are not immune to the failings of inductive reasoning. To this day it seems that MEs are still more likely to believe that external clocks are snake oil, and M/TEs are still more likely to prefer them.17
We could go on about the various reasons for the emergence of the rifts over clocking, but we feel the brief overview above paints a picture clear enough for any of us to see a little of ourselves in it. We now turn to the role of objective jitter tests in furthering these rifts.
How Objective Jitter Measurement Can Fail
The rifts over external clocking reached enough of a fervor to inspire a few pioneering pro audio writers to set about objectively measuring the effects of external clocking. Intuitively, we’d think that these objective jitter measurements would have told us what we wanted to know and put an end to the debates, but it turns out to be a little more complicated than that. Let’s look at one such study and figure out how even a properly performed set of objective measurements can still fail to be helpful.
One widely read jitter test report is Hugh Robjohns’ piece entitled “Does Your Studio Need a Master Clock?” (published in Sound on Sound in June of 2010).18 Even though his equipment and measurements were accurate, in this particular report Robjohns makes a few mistakes that lead him to some rather inaccurate conclusions.
Upon testing the effect of a handful of commercially available external clocks on a number of commercially available digital interfaces, Robjohns concluded that external clocks can only either increase jitter, or have almost no effect on jitter. The problems with his report are: (1) he only tested four converters,19 (2) he didn’t test Digidesign’s 192 converter (an odd omission as this would have been the big concern when he published in 2010), (3) he used the audio output measurement technique,20 and (4) based on his measurements of the four converters he tested, Robjohns wrongfully generalized that external clocks are incapable of improving the performance of any digital interface.
Robjohns also takes another interesting leap in this article. Based on his assumption that external clocks can never improve a converter, he concluded that those who prefer external clocking must prefer the sound of more jitter. It is possible that some people might prefer the sound of more jitter. However, Robjohns remained blind to the likelihood that adherents of external clocking were running jittery Pro Tools systems and, therefore, were actually preferring less jitter when clocking externally. Nonetheless, Robjohns’ assertion about people liking jitter caught on as a plausible – and polite – explanation as to why intelligent and reputable engineers so often preferred their systems to be clocked externally.21 After a while it wasn’t uncommon to hear engineers toss up their hands and concede that perhaps they did prefer a more jittery system – though that concession never quite felt right, nor was it ever meaningfully substantiated.
Regarding Robjohns’ measuring technique, if he had tested the Digidesign 192 he might not have detected the 192s inherent problems. There are many questions that hover around tests which look at the audio output only, rather than directly measuring jitter on the clocks or time-domain performance of converters (both of which require expensive equipment and esoteric know-how). Though crucial to the topic at hand, a look at these testing methods is well beyond the scope of this paper. We plan to take testing methods up in a future Pink Paper.22
Robjohns’ article and the erroneous conclusions in it are a strong reminder that, even when objectively measuring jitter in a digital audio system, it’s truly a pointless exercise to use the performance of one system – or even a few – as the basis for predicting the performance of any other systems. Further, the testing methods themselves need to be reevaluated. However – speaking hypothetically – had Robjohns in any way been able to expose the inherent problems with the 192, the his paper would have gone off in an entirely different direction, as would have the general understanding. Unfortunately his report was widely read, and widely believed without much meaningful scrutiny.
The Aesthetic Perspective: Sighted Listening
For over a decade, the only consistently correct voice on external clocking was that of working engineers who performed neither objective measurements nor double blind tests. Subjective evaluations will never satisfy the requirements of empirical science, but based on the clocking conundrum of the past decade, perhaps it would be wise when considering digital audio systems to put a bit more stock in the subjective impressions of our colleagues.23 As Duke Ellington famously said, “If it sounds good, it is good” – and perhaps a playful reprise for the digital age could be “If it sounds good, it’s probably less jittery.”
While some studies on human perception of things like jitter, sampling rate, filter design, cabling and more are starting to show up in the professional literature,24 the likelihood that substantial resources are going to be spent on cutting-edge objective research into pro audio’s interests is slim-to-none, because audio is simply not a research priority. In the meantime, the ears of experts may be the best, and often the only, “test equipment” we have on hand.
Yes, we do mean sighted listening (not double blind tests), and long term immersive listening (rather than short term comparative listening). In other words, recordmakers at work. We recognize that this recommendation cuts straight to the core of an unresolvable debate between objectivists and subjectivists, a debate so heated among audiophiles that one forum has banned discussions on the topic.25 We would never claim to be able to suddenly resolve this timeless debate; instead, we accept that – because it is inherently unresolvable – our energies would be best spent building meaningful connections between objective and subjective evaluations, rather than arguing about their relative validity.
As Bob Katz wrote to Allen prior to publication:
“The issues with jitter are so difficult and expectation bias so prevalent that, I am sorry, your “immersive” sighted tests will be treated with skepticism. There’s no comfortable answer here in the question of blind testing. You’re damned if you do and damned if you don’t! I’ve been on both sides of the issue and at this point when I reach a sighted conclusion I simply have to label it as anecdotal. That’s as scientific as you or I can ever get if you don’t do it blind. Live with it!”
Live with it we must. However, when we find enough experts forming subjective consensus around the same feature of a digital audio system, it is time for the those capable of objectively measuring those systems to come on board and help illuminate the situation. If the strange case of external clocking teaches us nothing more than to pursue meaningful consensus between what both our ears and our oscilloscopes are telling us, then as a professional field we can say that we are maturing (or returning to our previously more mature ways). We hope this maturation is the trend for this decade and the rest to follow.26
Further, because we have revealed a decade’s worth of accurate sighted listening evaluations of jittery Digidesign 192s (see above), we hope that our more die-hard objectivist colleagues will value sighted evaluations as a valid starting point for judging the effects of clocks on digital audio systems, rather than dismissing such evaluations (incorrectly) as “placebo effect” or “buying up snake oil.”
The Technical Perspective: An Introduction to the PLL
Just as subjective impressions can lend objectivist technicians direction and insight for their investigations, even a cursory tehnical understanding of the tools end users and consumers use can help us hone our subjective evaluations. If handled thoughtfully, aesthetic and technical perspectives make wonderful partners in the quest for deeper understanding.
Slowly, the general end-user is starting to hear more about the phase lock loop, or PLL, in the marketing literature for digital converters. The PLL manages the flow of digital data in a converter and, therefore, also manages (and hopefully minimizes) the timing inconsistencies of that data delivery which we call jitter. As a rule, systems with less jitter are generally understood to sound better, so growing familiar with how a PLL manages jitter is key to becoming a more nuanced and savvy evaluator (and consumer) of digital audio systems.
When you send a digital audio signal into a converter (and most digital-only interfaces), the converter needs to synchronize with the incoming audio’s time base. To accomplish this, most converters use a PLL that tries to lock onto the incoming clock signal. To do this, the PLL’s aptly named comparator “looks at” the phase relationship between its clock signal and the incoming clock signal. If there are any changes in that phase relationship, the PLL adjusts itself so that it remains as locked as possible. This process is called establishing phase lock. The reason we call it a loop is because the PLL sends its clock signal backward to be compared to the incoming clock signal. Thus, phase lock loop, or PLL.
Just as with any electronic device, there are many ways to design and implement a PLL, but what distinguishes one PLL design from another another is beyond the scope of this article.27 For our purposes, we will stick to the essentials that end-users can use in thinking about their digital systems.
If the incoming data’s embedded clock signal is jittery, then that presents problems. A well-designed PLL will both lock to the incoming clock signal and simultaneously ignore its jitter as much as possible.28 If you’ve ever seen a snake-charmer find the sweet spot where a cobra will neither attack nor slither away, you’ve got some vague sense of how a well designed PLL does its thing. Another apt analogy might be a car’s cruise-control, which needs to make constant adjustments while still providing a smooth ride. If a car’s cruise control is too responsive the car would be constantly accelerating and decelerating, and if it’s not responsive enough the car will lose its lock on the desired speed.
A PLL that reacts too quickly to the incoming jitter fails to ignore it and will, effectively, recreate it. This is often called a “fast” or “fast tracking” PLL. Some would casually say that converters with fast PLLs are “susceptible to input jitter.” This sounds like bad news, but a well-designed fast PLL can be thought of as transparent in that it presents a relatively accurate copy of the input signal.
A well-designed slow PLL will stay locked to the input signal’s time base, but it will not recreate the jitter of the input signal. Slow PLLs can also be said to be transparent because they present the original data without the jitter of the upstream device.
Essentially converters can have one of two different types of transparency: (a) transparency that reveals the jitter of the upstream device, and (b) transparency that ignores the jitter of the upstream device. Which is better depends on what we want to hear.29
Because we generally want to minimize jitter, most of the time we will be happier with a slow PLL that ignores input jitter, but for certain D-A monitoring conversions a fast PLL that reproduces the jitter of the upstream device might be useful. If we had our druthers, we might all prefer converters with adjustable PLLs that allowed us to choose between both types of transparency. Product documentation that tells us about the nature of the PLL in any given device would also be desirable.
Another important concept to grasp is phase noise. In simplest terms, jitter is inconsistencies in time and phase noise is inconsistencies in frequency that result from jitter. For our cursory purposes, jitter and phase noise are pretty much interchangeable, but we will attempt to use the contextually correct term as we make our way through this report. We may also refer to both simultaneously as jitter/phase noise.
As a general rule, anything a designer of a PLL can do to reduce jitter/phase noise is a plus for audio systems, and the PLL is always a central concern for designers when it comes to reducing jitter. Knowing more about the nature of the PLLs in our digital audio systems will help us listen to them with greater accuity, and it will help us grasp important concepts when making a purchase.
LONG- & SHORT-TERM CLOCK STABILITY
It would be hard to overstate the extent to which we are about to reduce the complexity of audio clocking in this explanation. However, as with the PLL, even a cursory grasp of clocks can help us become more informed users and consumers.
Depending on the application of the clock, technicians are concerned with long- and/or short-term clock stability. Long-term stability is how much the clock will drift over very long periods of time (often measured in 1000 year incriments). Short-term stability indicates how much the clock will vary from pulse to pulse over extremely small fractions of a second. For audio purposes the concern is for short-term stability, as we want our zeros and ones to be delivered as evenly in time as possible, while the accuracy of the signal in 1000 years is of no concern.
As it turns out, crystals oscillators have very good short-term stablity but relatively lousy long-term stability, so for audio purposes most designers use crystal oscillators. These crystal oscillators are complex things, but ultimately they put out a square wave at the sampling-rate that the PLL and other devices in the converter use.
In the telecommunications industry, however, both long and short-term stability are important, so a breed of clocks emerged that use rubidium osciallators to discipline the crystals into having good long-term stability. The audio world has latched onto the phrase “atomic clocks,” and – despite the fact that long-term accuracy doesn’t matter for audio – atomic clocks have made quite a buzz in both the audiophile and pro audio markets, as we’ve seen.
So-called atomic clocks puts out a high-frequency 10 MHz sine wave which can then be used by a compatible audio clock. In most configurations, 10 MHz generators are freestanding units that send their signal into either (a) a 10 MHz-compatible Word Clock device (there are only a few on the pro-audio market) or (b) into a digital converter’s internal clock (presently only Antelope makes 10 MHz-compatible pro-audio converters). The new Antelope 10MX houses both a 10 MHz generator and a Word Clock device in the same 1RU unit – a first for the pro-audio market.
The way a rubidium oscillator works is rather complex, but the basics are interesting to consider as they help us understand how and why any two 10 MHz clocks might perform so differently. Inside the clock there is a crystal oscillator (some use 10 MHz crystals, others are multiples of 10MHz). Inside a highly controlled heated chamber (or oven) is some rubidium, which will put out an exceptionally stable high frequency signal that disciplines the crystal’s long-term stability. The crystal oscillator is then able to put out a very accurate 10 MHz sine wave.
As you can imagine, there’s a lot that goes into these rubidium oscillators, and designing the core rubidium clock unit is a delicate craft. There are multiple PLLs and other complicated devices used to control everything inside the 10 MHz oscillator – imagine a group of snake charmers holding a multi-headed snake in place, and you get a very vague sense of what goes on inside a 10 MHz rubidium oscillator.30 The rest of the design goes into powering and porting the rubidium oscillator in order to put out a clean 10 MHz sine wave.
One definitive thing that we can say about 10MHz clocks is that they are far more elaborate devices than standard crystal-based audio clocks, and most of that elaborateness has been designed in order to achieve the long-term stability that – as far as we know – does not improve performance for audio use. We will eventually arrive at the conclusion that 10 MHz clocks for audio purposes have been a ruse. However, in order to substantiate that conclusion, it is important that we carry you through an elaborate sighted listening process as well as some refined logic that deals with how 10 MHz technology was able to have been taken seriously by the audio community despite it being a ruse.
How To Listen To Clocks
As much as we appreciate the need for objective measurement using measurement tools – and to a lesser degree double blind testing – sighted listening remains the method by which the vast majority of people will evaluate audio equipment, including clocks and converters. While sighted listening evaluations are always anecdotal, if we can help the end user come into a better understanding of how to think about and communicate their own subjective impressions, and if we can help build bridges between subjective and objective camps, then we’ve helped move the larger conversation about digital audio forward.
With something as subtle (yet pervasive) as clocking, pointing out the dimensions of sound that clocks impact will help listeners of all skill levels develop their ears, as well as providing us with a useful vocabulary with which to describe what we hear. We believe that aesthetic sensitivity can best be developed with proper guidance and that directing our focus to specific aspects of sensory stimulus is, ultimately, what becoming an expert is all about.
Below is a list of the sonic qualities that one can listen for when swapping out clocks on a digital system. We generally believe that improvements in any of these dimensions likely indicate a reduction of jitter.31
Clarity – Many systems “open up,” exhibiting more sonic detail, especially in the high-end where finer sonic definition resides. Specific sounds such as ride cymbals and sibilance on a vocal can be more detailed and interesting, and the tweeters seem to reveal more. But the entire frequency spectrum will also exhibit varrying levels of clarity.
Changes to Soundstage Shape – Some clocks will strengthen the center image and others will appear to widen the stereo image, bringing more impact to the sound on the far left and right of the mix. One might think of this as the soundstage having a “concave” or “convex” shape.
Front-to-Back Depth – The front-to-back depth of the sonic image can change when swapping clocks, and typically the favored clock will present more depth. Often, along with this increase in depth, is an increase in the detail of reverberated sound within the recording.
Three-Dimensionality – Highly related to soundstage shape and front-to-back depth, three-dimensionality can also be thought of as the relationship of the phantom image in the center of to the rest of the sound stage. Some people call this ‘holographic’ sound, or even ‘holosonics’ becasue the sound appears (note the visual language) to be three-dimensional.
Localization & Individuation – The ability to differentiate and locate individual sounds within the sonic image can often improve when jitter is reduced. Panoramic placement can appear more precise, and individual instruments can seem to hold a more defined location.
Low-End Focus – The tightness and focus of low end can be affected by an external clock, as well as the note differentiation of melodic low end instruments like bass guitar.
Low-End Extension – The favored clock will sometimes cause a digital system to extend its low-end performance (see below for a drastic case of this).
General Ease of Listening – For many listeners, decreased jitter can create a more relaxed feeling when listening, especially over long periods. Some will notice their bodies relaxing more when listening to one system vs. another.
Richness – A highly subjective quality, many experienced listeners will report that the quality of the sound is more “rich” when jitter is reduced. Or we might say that listening is a richer experience. We suggest that listeners make personal meaning for this term and use it as subjectively as you wish. We suspect that richness might be the sum total of improvements in many (if not all) of the above characteristics.
We have veered into the kind of descriptive language one normally hears from, or associates with, audiophiles. We suggest that recordmakers further embrace, rather than shy from, descriptive language, and even look to the vocabularies of audiophiles for linguistic innovations. Developing – and even enjoying – the language we use to describe the subtler aspects of what we hear can only help us when sharing our impressions with each other, as well as with product designers who might use our descriptions to guide their creations. Language will always be the material from which we build bridges between the subjective and objective perspectives.
But we will offer up a word of discretion: if you do not hear a difference when listening to different clocks, then that is your subjective experience. Not hearing a difference is as valid and important as hearing one, and a key aspect to becoming an expert listener is to not let the opinions of others sway you into hearing things that aren’t there. The core problem here is that that guided listening can cause confirmation bias (the tendency to seek confirmation of one’s expectations) – a colossal problem in the research sciences. We are not out to convince anyone to take on our sighted listening paradigm as scientific. Instead, we aim to clarify what that paradigm is, and how one might practice and use it.
We Compare Externally Clocked Converters
Matthew Agoglia was the listening partner for these evaluations. Matthew’s mastering studio, The Ranch (www.theranchmastering.com), provided us with the necessary equipment and listening environment to try out different systems, all of which showed varying results when clocked externally, either to crystal or 10 MHz units. As we will continue to stress, each system reacted very differently to different clocking schemes. Let’s dig into some examples to see how that played out, as well as how we arrive at our final testing rig for the 10 MHz comparison.
Externally clocking the Dangerous Music CONVERT-232 – The CONVERT-2 is a stereo DAC that showed no improvement when clocked externally, and in fact sounded a bit worse when it was. This is an indication that the CONVERT-2 is a solidly designed and implemented converter whose internal clock and PLL are expertly tuned to each other. The CONVERT-2 DAC is, therefore, a good candidate as a studio’s master clock and has shown itself to be when clocking converters that do respond to external clocking.
Externally clocking the Forssell MADA-2’s A-D – The D-A of the MADA-2 is not ported for external Word Clock, but the A-D is. When clocked externally – whether to 10 MHz or standard crystal units – the MADA-2 showed no improvement, and slight degradation in sound in some cases. As with the CONVERT-2 above, we can assume that the MADA-2 is similarly well designed, with a steady internal clock and well implemented PLL that produce low jitter. The MADA-2 is a also good candidate as a studio’s master clock, and has shown itself to be.
Externally clocking the Burl Mothership – Though we were unable to listen ourselves, our friend and colleague Joel Hamilton told us that his Burl Mothership converters showed no audible improvement when clocked externally to his Antelope 10M system. Thus, he runs his Mothership on the internal clock.
Externally clocking the Cranesong HEDD 192 – Another converter that has shown no imprvement when clocked externally. The HEDD 192 has operated as Allen’s master clock for over a decade, and it turned out to be an excellent performer as master clock on our test system.
Externally clocking the Lynx Aurora – The Lynx Aurora converters regularly exhibit a sonic change when clocked externally, especially in high-end clarity and openness.33 Allen has been clocking his Pro Tools HD Lynx system to his Cranesong HEDD 192 converter for a decade with excellent results. Matthew has clocked his Pro Tools/Aurora rig from an Antelope OCX (and more recently with the 10M clock attached) for nearly as long. Joel Hamilton’s second Pro Tools rig uses Lynx Aurora converters clocked to the Antelope 10M system for a marked improvement.
4-19-16 addition: We have been made aware of the possibility that disabling Synchro-Lock on the Lynx may impact the way it responds to external clocking. Unfortunately, this requires a USB-MIDI command bridge, so it’ll be more than flipping a switch. We will experiment with that and report back here.
This small sampling of digital systems should be enough to show how system-specific the role of external clocks is, and why generalizing about them would be fruitless. With a basic understanding of the PLL, we can conclude that these units, as a group, exhibited a variation of stablility in their internal clocks. We hope that it is clear how – when coupled with only a curosry technical understanding – a sighted listening evaluation can reveal much of what an end-user needs to know when assembling a digital system.
Externally Clocking the Digidesign 192 Digital & Dangerous Music Monitor – As we already know, the Digidesign 192 has been proven to exhibit less jitter when clocked externally. Externally clocking a 192 Digital, and then sending the 192’s AES output to the Dangerous Music Monitor’s DAC revealed the greatest sonic changes from the clocks we tested. There are some relatively easy-to-grasp reasons why this system was so revealing, and why we chose it as our final test system.
As we know, the 192 will show a marked decrease in low-frequency jitter when clocked externally. The Monitor was thought of as one of the most transparent DACs of its era, but we have to be clear about what kind of transparency. The Dangerous Music Monitor uses a Cirrus Logic 8416 192 kHz receiver chip to manage the incoming digital data, and this chip does not attenuate low-frequency jitter, thus offering up a highly accurate replica of the low-frequency jitter coming into it from the 192. Furthermore, The Monitor does not contain an internal clock, and instead uses the clock encoded in the incoming AES3 signal to convert data to analog, thus effectively giving us a very transparent translation of the 192’s jitter profile. More simply put, the impact of external clocks was very obvious on this system.
We Compare Four 10 MHz Clocks
As far as we know, no one has done a comparative evaluation of commercially available 10 MHz generators, and certainly not with the pro audio field in mind. Before recently, there was hardly ever a chance to compare 10 MHz generators. Who had, or could (or would) afford, two different 10 MHz clocks? Just using the Antelope 10M system was an $8000 enterprise, and – shockingly – a quick perusal of the field brings up the Abendrot 10M clock for $36,000, the Esoteric G-01 for $23,000 and so on.
Before long, some people started opening up these clocks, including the Antelope 10M, and asking why these few and relatively inexpensive components should cost so much? There was no clear answer. One has to accept that audiophile companies (including Antelope, which operates in both the pro-audio and hi-fi markets) have been marketing 10 MHz clocks as high-end luxury goods. The business strategy is to sell just a handful of units (which conveniently ups the elitism) at enormous markups.
Perhaps that trend is starting to change, as both Antelope and Stanford Research Systems have recently offered up 10 MHz clocks aimed at the audio markets at lower prices. For $3495 you can get the SRS PERF10, an 10 MHz generator designed for audio use. You need a 10MHz compatable clock or converter to run the PERF1034, so, for example, a system that includes the Antelope OCX ($1295 street) comes in around $4500. For $5999 you can now get the Antelope 10MX, which gives you their newly designed 10 MHz generator plus an upgraded version of their Trinity clock in one box. Not cheap stuff, but at least the price is coming down.
As far as we know, Stanford Research Systems is the only company that designs and manufacturers the rubidium-based oscillator inside their own audio-specific clock. That oscillators is the SRS PRS10, which costs $1495. Most companies buy the rubidium oscillator “off the shelf,” and some companies use the SRS PRS10 oscillator in their clocks, including the previously mentioned $23,000 Esoteric G-01.35 Antelope Audio’s 10M is based around the Spectratime LCR-900 rubidium oscillator (currently a $900 item), and new Antelope 10MX appears to be built around the Microsemi SA.3Xm series of small, affordable rubidium oscillators (they cost around $900 as well, though prices vary with quantity purchased).
Once an audio company has decided on a rubidium core, their job is to implement it into a device that will output the best 10 MHz signal for audio use. Essentially what will define a good 10 MHz design is low phase-noise in the 10MHz sine wave output, which (theoretically) should translate into less jitter in the Word Clock signal being sent to the converter’s PLL. The specs for phase-noise on the rubidium oscillators used in the three clocks that we tried vary (somewhat significantly), and of course the performance of the device receiving the 10 MHz signal is essential, too.
Firstly, all three 10 MHz clocks improved the sound of our test system. The change was obvious to the ear, so we know that the OCX’s Word Clock output was changing, and, as mentioned earlier, this sonic improvement indicates that the Antelope OCX is not an optimal crystal clock on its own (further comparisons confirm this – see below).
As we got situated, we were somewhat surprised to find that each 10 MHz generator had such a distinctly different impact on the sound of this system. Further – and most importantly – the individual sonic profile of each clock held true across a vast selection of program material. Each of these 10 MHz clocks helped to reveal more detail and improved imaging from this system, but in rather different ways. This discovery runs completely against the grain of our expectations, because, intuitively, one would think that clocks based on such reputedly ultra-accurate technology would not vary much from one design to the next. But they did.
Impressions of Antelope 10M/OCX – The Antelope 10M/OCX combination brought out so much reverb and widened the soundstage so much that I (Allen) actually found myself seriously doubting a mix I’d just done for thereminist Carolina Eyck with American Contemporary Music Ensemble (ACME). This is a stripped down production of theremin and string quartet mixed in analog on my API console down to the TASCAM DA-3000 (at 96k, 192k and DSD). Reverb was via a Roland R-880 returned via analog to the board. Building the illusion that the DI’d theremin occupies the same acoustic space as the strings is all about meticulously sculpting the reverb, finding precise dry-to-wet ratios and delicately manipulating pre-delay. Clocked to the Antelope 10M/OCX, my mix simply had too much reverb, and the center image that held the theremin receded rather than came forward in the phantom center as I’d intended. Localization felt disorganized and uncertain. Because these mixes are so highly dependent on precise localization and reverb settings, this recording was particularly sensitive to the kinds of sonic changes that the Antelope 10M/OCX combination exhibited on this system.
Other music we checked out seemed similarly drenched in reverb with the center channel seemingly “sunk in” and the soundstage cinematically wide. Sigur Ros’ ultra-slow “Untitled 5 (Álafoss)” from the brackets album ( ) was made even more cinematic and enveloping by the 10M/OCX combination. Specifically, the depth of reverb on this record was even more vast when clocked to the 10M/OCX, which is saying a lot. We thought the 10M/OCX combo flattered the production in a way, but details were somewhat obscured when compared to other clocks. Emmylou Harris’ “Deeper Well” from the album Wrecking Ball – a classic Daniel Lanois production filled with rhythmically delayed details and deeply embedded looping samples – was detailed with the 10M/OCX, yet our attention was similarly pulled toward the sides and away from the center image.
We also noticed that music sounded louder with the 10M/OCX, though not necessarily nuanced or elegant – aggressive might be a fair descriptor. We believe that the Antelope 10M played a minor role in the “loudness wars” since its release, and we know that more than a few producers and labels were seeking out mastering studios with the 10M in order to gain a loudness edge. Certainly those endorsing the 10M are known for big, loud mixes that make their way into major radio rotation.36
Impressions of Antelope 10MX – The 10MX was revealed at the Audio Engineering Society (AES) convention in September of 2015. The 10MX is the first to house the 10 MHz generator and a Word Clock generator in a single unit for the pro audio market. This design points the way toward further integration of 10 MHz clocking technology into Antelope’s digital audio devices, all of which are ported for 10 MHz clock signals.
The Antelope 10MX provided a wholly different sound than the 10M/OCX combo. The individuation of elements was crystal clear, details firmly localized, and reverbs were far less enveloping and washy. The soundstage was wide, yet the center image was strong and present, so the sense of width was not quite as enhanced as with the 10M/OCX.
The 10MX was, however, slightly strident in the treble range compared to the other 10 MHz clocks we tried. The low end was very similar to the 10M/OCX combo – tight and full simultaneously, but not particularly deeper than when the rig was clocked from the crystal-based Word Clock in the OCX alone.
When the 10MX was clocking this system, the theremin and string quartet recording mentioned above was closer to the sound coming off the console. The individual quartet instruments were highly localized (the cello in particular seemed to “find its place”). However, there was a slightly-too-far-forward attack of the bow on the strings that seemed to divorce the instruments a bit from the reverb. The reverb seemed to sit behind the strings, rather than in and around them as intended. We both agreed that on this record the 10MX presented the most crystalline image, excellent localization and focused bass, though it lacked a bit of richness and depth as a trade-off. Very forward sounding.
Like it’s predecessor, on this rig, the 10MX is the clock to use if you want to present details, precision panning, and an up-front, forward sound. Relaxed elegance, however, was not this clocks forte. We imagine the 10MX could be helpful for those trying to achieve loud, forward pop mixes competing for attention on the radio or a playlist (which could explain some of the admiration it has received from its more renowned early adopters).
Impressions of the Stanford Research Systems PERF10 – The SRS PERF10/OCX combo offered up yet again another very different sound. First of all, you get what feels like a whole extra low octave out of this system when clocked with the PERF10/OCX combo. Seriously – it was as if someone had set up an auxilary sub in the room. On the Sigur Ros track mentioned above, it was as if Jupiter had finally joined the solar system. On the Lanois production of Emmylou Harris’ “Deeper Well,” Larry Mullen Jr.’s kick drum carried the kind of huge, unapologetic lowend Lanois claims to adore. And so on, across all the material we tried, the PERF10 consistently brought the bass like no other 10 MHz combination we tried.
The problem we had with the low-end, however, was that note differentiation was not spectactular. What was gained in low-end content seemed to have been lost in low-end definition.
The ratio of dry to wet from the PERF10 was very balanced – we could hear deep into the space of the mixes, but nothing was washing out either. Furthermore, the reverb stayed in and around the source instruments in a convincing way. Where the Antelope 10M/OCX was a bit too wet, and the 10MX perhaps a bit dry and forward, the PERF10 sounded “just right” in its presentation of front-to-back detail.
The soundstage was also consistently concave, and the sounds seemed to hover between and behind the speakers. The phantom image never came forward into the center of the room as it did with the better crystal clocks we tried (we wonder if this has to do with poor left/right time accuracy, which could add up to phase cancellation in the center).
I’m going to risk being ultra subjective here: I felt better listening to the PERF10 than with the Antelope 10M or 10MX. My shoulders relaxed and I could just ease into the sound of Matt’s rig running the PERF10. I was actually moaning over the sound of certain musical moments, and we both agreed that the PERF-10/OCX combination gave us the richest listening experience of the 10 MHz clocks we tried.
Impressions of the Stanford Research System’s FS725 – This clock is not designed for audio purposes, and it is terminated at 50 ohms, rather than the standard audio termination impedance of 75 ohms.37 But, because SRS was curious about our tests, and because we were curious about their clocks, we tried it out and were pleasantly surprised to find that it sounded pretty good. At $2500, we were hoping that the FS725 would come in as the strong underdog, but based on our listening we’d recommend that people avoid this clock for audio use and seek out the PERF-10, which was designed for audio applications.
Interestingly, the FS725 presented a similarly concave soundstage to that of the PERF-10, and a its relatively relaxed sonic quality was neither pushy nor loud like the Antelope units were. However, the low-end octave was not there, and overall it was a bit “papery” and unexciting with this system. The value of listening to the FS725, however, is that we were learning what another version of jitter did to the sound of this system.
Conclusion About These 10 MHz Clocks – As we will show in the next section, we can not grasp why one would invest in a 10 MHz clock for audio, but if for some reason someone felt they had to, we would point them toward the Stanford Research System PERF-10 as the less expensive and more enjoyable option.
We compare 10 MHz & Crystal Clocks
Using the same jitter-revealing test system, we were curious to learn whether there was any obvious advantage to these 10 MHz clocks over crystal clocks, so we took what we had on hand and did a comparison of crystal clocks to the SRS PERF-10/OCX combination (our preferred 10 MHz rig). As with any and all clock comparisons, we can not be certain that other systems would benefit so obviously as our test system did, but this was a very revealing exercise nonetheless, and we can make some general conclusions below that should help end users and consumers make better sonic evaluations and purchasing decisions of their own.
Antelope OCX as Master Clock – As we mentioned earlier, the OCX on its own did not reveal the kind of high-end detail, nor the width and depth as it did when driven by a 10 MHz generator (especially the SRS PERF-10). However, the low end on the OCX was more powerful than with the Antelope 10M hooked up, and the center image was more pronounced. Generally, however, the overall sonic image was not as engaging, and the center image strength was at the expense of detail and interest in the left-right sides of the stereo image. The sound was disorganized.
TASCAM DA-3000 as Master Clock – For a $999 dollar free-standing recorder and A-D-A converter, the DA-3000 was impressive as a master clock on our system. The image was wide, high-end detail robust and fairly clear, low-end solid and punchy. Localization was much better than the Antelope OCX. However, overall, the sound was not elegant or professional enough for our tastes. We would recommend, however, that anyone who owns a DA-3000 might try clocking their DAW rig to it and seeing for themselves if there is an improvement.
Cranesong HEDD 192 as Master Clock – Wow. Dealys and reverbs that were nearly inaudible with the Antelope and the TASCAM were obvious and lush with the HEDD as master clock. The depth of the soundstage was vast. All around the HEDD’s clock provided a much more professional sound with a strong center image. Localization was excellent. Low end balanced and punchy. A much more 3-dimensional listening experience all around. The HEDD was our favorite master clock on this rig. As noted earlier, the HEDD has also served excellently when clocking Lynx Aurora converters.
Forssell MADA-2 as Master Clock – A second “wow” for the Forssell. This clock was similar to the Cranesong HEDD in its revelation of reverbs and delays. Localization and individuation was stable and precise. Width was pronounced, perhaps slightly at the expense of the center image. Low-end frequencies, while full-bodied, were not as clear as with the HEDD.
Lynx Aurora as Master Clock – Delays and reverbs that were obvious with the Cranesong and the Forssell diesappeared with the Lynx as the master clock, bringing us back toward the sound of the TASCAM DA-3000 and Antelope OCX. The low-end was lacking powerful sub frequencies, hig-end detail was two-dimensional, individuation and localization were unimpressive. In Matthew’s words, “It’s a bit messy.” I felt it put me on edge, much the way digidesign 192 converters always had. Both of us being Aurora users, we are glad that these converters respond to external clocks.38
Was the Cranesong HEDD Better Than the 10 MHz Clocks? – Yes. From the crystal oscillators we had on hand for our initial listening tests, we honed in on the Cranesong HEDD as our favored crystal clock for this system (with the Forssell as as very close rival). We then returned to the 10 MHz clocks and compared. We quickly zeroed back in on the Stanford Research Systems PERF-10 as our preferred 10 MHz clock and did a comparison with the HEDD.
We’ve described the sound of these clocks above – however, the biggest differences between the HEDD and the PERF-10/OCX combination was that the HEDD delivered articulate and dynamic front-to-back depth compared to the PERF-10/OCX combination. The sonic image when using the HEDD reached into the room while still presenting reverb and delay details that reached far back into the space behind the speakers. It was a very three-dimensional and engaging soundstage. Low-end on the HEDD wasn’t quite as big as the PERF-10, though it was more clear and dynamic in general, with better note differentiation and detail.
For our test system, which revealed the upstream jitter profile, we could find no “10 MHz advantage” and we far preferred the sound of two of the crystal clocks we had on hand.
4-19-16 Addition – Grimm Audio CC1 and Dangerous Music CONVERT-2 as master clocks.
Grimm Audio CC1 Master Clock – Quick summary: Allen bought one immediately! This clock is incredible. The detail and imaging from it are superior to anything we heard, and the overall “relaxed” sound of the playback was as close to analog tape (read: jitter-free) as anything we’d tried. The Grimm CC1 didn’t make the music sound as aggressive and forward as the Antelope clocks, and that was a huge plus, because what you get instead is unparalleled depth and detail, as well as a feeling of “calm” that none of the other clocks provided. Of course, the CC1 is simply a clock with no conversion available (like the Cranesong HEDD), but it seems there is something to be said for a dedicated clock, as whatever Grimm has done inside the CC1 is topping an elite class of options as a price less than half of some of the other offerings. Hands down, the CC1 was our preference.
Dangerous Music CONVERT-2 as Master Clock – The CONVERT-2, a dedicated D/A converter, provided great sound from the system when used as the master clock. The level of detail wasn’t quite as crystal clear as the HEDD, but there was an “analog quality” to the sound that was appealing. Reverb depth was particularly good. Like all the other converters-running-as master-clocks that we tried, the CONVERT-2 could provide a great master clock in a studio that needs a high-end D/A as well and doesn’t have the budget for a dedicated external clock like the Grimm.
Listening and Logic Agree: 10 MHz Clocking for Audio is a Ruse
In the absence of jitter measurements (for now), we are left with listening and logic. In the case of 10 MHz clocks, logic and listening are aligned.
We’ve shown above that on our test system the 10 MHz clocks, while capable, were not able to outperform the crystal oscillators found in two high-end converters. Perhaps there are other systems, or 10 MHz receiver devices, which might reveal something we were unable to hear. However, our test system was excellent at passing a subpar converter’s jitter onto the speakers for us to hear. If any of these clocks was going to shine, this system provided the opportunity.
As we explained above, for audio we are interested only in short-term stability in a clock. The rubidium oscillator is an elaborate and relatively expensive device that improves long-term stability. So, the logical conclusion – and one that lines up with our listening evaluations – is that nothing is gained from adding a rubidium stabilization device to a crystal oscillator for audio use.
Certainly the claims that atomic clocks are 100,000x more accurate than crystal clocks is a deliberately misleading brag-point because long-term stability does nothing to improve audio clocking. From Antelope’s website regarding the 10MX Rubidium Clock: “A Rubidium atomic reference generator 100,000 times more accurate than the crystal oscillators, bringing vastly improved staging. transparency and imaging.” This kind of marketing is not only misleading technically, but we feel that the description of the device itself has not been born true. This is a bit like an automobile manufacturer advertising that a special car engine upgrade is 100x more efficient, but failing to mention that this improvement only applies when that engine is used in outer space.
We can go even deeper with our suspicions about 10 MHz clocks: wouldn’t any improvement to the performance of the audio system when switching over to the 10MHz generator indicate that the crystal oscillator was not performing optimally on its own? Because there are so few crystal clocks and converters with 10MHz ports for pro audio, we are left to wonder if the so-called “10MHz advantage” is not simply a case of a less-than-accurate crystal clock getting a little help via the 10MHz signal. Add in the subpar jitter profile of some of the most widely used “professional” converters, and we can see how the external clock with the 10 MHz generator attached could seem like mana from heaven. If that is the case, then the entire 10MHz clocking paradigm can be summed up as a ruse (however intentional or unintentional it might have been).
Another way to put this is that if a crystal clock is implemented excellently one would not hear an improvement with 10 MHz technology attached – and we assume that one wouldn’t bother with porting for 10 MHz anyways.
Astute readers will likely be wondering if our test system is simply a recreation of the 10 MHz ruse, and, indeed, it is. We are using a jittery converter (the 192) with a crystal oscillator (the Antelope OCX) that changes when using a 10 MHz input – that’s the ruse. However – given the equipment at our dispsal – this system was also the best way we could manage to isolate and reveal the differences in external clocks and 10 MHz generators. The readily exposed jitter in our test system is analogous to piles of contrasty colored dirt tossed on carpets during vacuum cleaner tests. Hopefully you wont have major spills like these, or jitter like the 192’s, but making the impact of these devices obvious is the goal here.
The other thing to remember is that unless you’re using our test system, you’re going to have to go and do these tests yourself on your system. The specific products that we tried – as well as products that we hope to try in the future – are all going to work differently together. We will not generalize on your behalf.
From a consumer’s point of view, if you run the cost on all of the clocking options we’ve presented so far, you can see readily that an excellent two-channel converter can also make an excellent master clock. Dollar for dollar, using a high-end converter as your master clock is an excellent solution for mixing and tracking engineers who need great sound from dozens of channels of AD and DA systems that show reduced jitter when externally clocked. Along with great clocking, you also get two channels of excellent conversion. Once you add up the expense of a 10 MHz clock rig, it’s difficult to see why anyone would spend that kind of money on a system that has mostly won favor by way of a ruse (however intentional or unintentional that ruse may have been). Typical drawbacks to using a converter as your master clock have to do with the number of clock ports available to you and the ability to run multiple sampling rates. For many studios, a more elaborate clocking solution will be necessary.
Rebuilding The Bridge Between The Technical and The Aesthetic
We no longer exist in the glory days of analog when equipment designers like Rupert Neve or Jim Rogers would drop units off at the studio for feedback from the world’s best recording engineers. Today’s “engineers” rarely know the people who design their equipment, and they rarely know much about the engineering behind the equipment they use. The digital era is one of quick product consumption, planned obsolescence, software updates, questionable hardware quality, all built on the assumption that the end user will neither reach beyond the knobs and buttons on the front panel nor meaningfully question the marketing used to sell these products. In many ways, the connection between the technical and the aesthetic that marked the analog era has been severed in the digital age. As cynical as that may sound, however, it does not mean that this new end user base can not become a demanding and intelligent one, or that we can’t find ways to rebuild the bridge between the technical and the aesthetic.
Re-building that bridge is one of our long-term goals at Pink Noise, and specifically of our Pink Paper series. What gets revealed when jitter measurements and listening evaluations are brough to bear on each other will, we believe, prove to be very enlightening. We are in the process of assembling a team who can provide such measurements. To keep up to date on that, sign up for our mailing list or follow Pink Noise on Facebook.
Until we have those measurements, however, we must refrain from making overarching conclusions based on our subjective evaluations. However, we would point back to the strange case of the Digidesign 192, the flawed generalizations of Robjohns’ piece on external clocks, and in the meantime offer up the following which Bob Katz wrote to Allen during the research for this paper: “I go by this rule: the combination of clocking that sounds warmer, wider, and deeper is probably the most accurate and probably has the least audible jitter artifacts.” When it comes to educated guesses, Bob’s “rule” is probably the one to follow….for now.
Eventually it will be much more enlightening and interesting when we know what it is exactly that we’re preferring. Bridging those perspectives will help the designers of digital systems become more in tune with how the humans at the other end were reacting to the actual functioning of their products, and it will help end users specify exactly what it is they prefer from digital systems. We don’t think it’s too much to ask the artistically inclined professionals in our field to become sharp, smart, aesthetically demanding and somehwat tech savvy end users and consumers. Even a cursory technical grasp of the tools we use can cut through miles of marketing hype to the bone of what’s actually being offered.
Dropping The Digital Ideal – Toward Multiple Transparencies
Given the variability of sound that we heard from all the different clocks and converters we’ve tried, we believe it’s time to move away from the tired notion digital audio is evolving toward some ideally transparent system and to embrace that a diverse array of beautifully executed converters – all of which will have “a sound” of some kind – is emerging. Transparency and musicality (this means ‘beauty’ we believe) will always be intertwined in an elaborate and confusing dance between the speakers.
Given the idea of multiple transparencies, perhaps the future of audio clocks is to unlock them from a fixed design and build in end-user variability. Think of it like the adjustable suspensions available on so many vehicles these days, which allow the driver to choose how the car will handle based on their preferences for different terrain. Certainly the ability to reveal upstream jitter with a faster PLL or to erase jitter with a slow PLL would be a nice option on a converter.
This suggests the notion of building other variable controls into converters. Perhaps we will one day be able to tweak the jitter, filters, chip headroom and other features that can change the sound of the digital audio systems we use. The Bricasti M1 DAC alredy offers six different user-selectable filter types. Some models of DAC from dCS also offer two filters that change the sound of the DAC. And Lavery Engineering offers the Gold 122-96 MX converter, designed specifically to be pushed hard into clipping by the end user, blatantly pursuing “a sound” from this otherwise transparent converter.
As Matthew and I talked about these ideas for building variability into converters, we both agreed that perhaps the dimensions of sound that we hear changing when we change clocks are better managed with the tools that are intended for sonic manipulation (standard fare, like EQ, compression, mid/side processing, harmonic distortion, reverb, etc….). The more we talked about it, the more we realized that our preference would be for digital conversion to continue toward a multiplicity of beautiful sounding digital systems, and for the work of sound manipulation to be left to the tools designed to do so. Others may disagree, but our happiest moments as recordmakers and as listeners has been when a digital system simply sounds beautiful.
It Is Time
As a field of professionals, it is time to become more astute users and consumers of digital audio systems. We know that most of the other devices we use as recordmakers – mics, preamps, compressors, etc – are going to change any individual sound more than your converter or your clock, but there is no single device in your digital studio that will always be in use, every moment that you work, imparting its sound on all that you do across all channels of your creations. If your conversion and clocking are not optimal, you’ve got a systemic sonic problem.
As we more deliberately abandon the notion of perfect transparency and accept that the best designers of our digital audio devices will be those who can artfully bring the technology toward beautiful sound – of which opinions will be as varied as the personalities that make up this professional field – it is our responsibility to open the dialogue between the technicians and the practitioners and to resist the temptation of believing that there is some final, ideally transparent, destination for digital audio. And we must always resist the temptations set before us by marketing hype – and especially ruses.
Certainly the language we use to describe the sound of digital audio systems can improve as a means toward these ends, and we hope that the cursory set of sonic qualities we listed above (high-end clarity, shape of soundstage, localization, etc…) as well as our general paradigm for thinking about audio clocking can begin a more fruitful dialogue for all of us as we begin to explore what is certain to be an era of expansive progress for digital audio in the coming years. The broader goal is, of course, to always move in the direction of increasingly desirable sound, and to never again allow bad design, bewilderment, and bickering to become the norm.
with Matthew Agoglia
Special thanks to: Andy Hong, Eelco Grimm, Dave Ames, Andy Peters, Dave Collins and Bob Katz
First publication: Pink Noise, January 28, 2016.
- The best, and most readable, explanation about the variability of performance from external clocks is Grimm Audio’s very clear piece “Slave to the Rhythm” Unfortunately, this tight little paper never reached a wider audience via a more popular pro audio publication. http://www.grimmaudio.com/site/assets/files/1088/pll_and_clock_basics.pdf
- From Apogee’s product copy.
- For Pink Noises’s critique of the TEC Awards program, see Pink Paper #001: http://pinknoisemag.com/pink-papers/pink-paper-001
- You need a 10 MHz compatible clock to use the Antelop 10M. At the time of release, Antelope was offering their OCX and Trinity clocks, which brought the package price up to about $8000US with tax.
- Matthew Agoglia worked under Howie at Masterdisc in NYC, and the playback rig there was Pro Tools running a Digidesign 192 with the Antelope 10M / Trinity clock system
- Read the Grimm Audio “Clocking” White Paper at http://www.grimmaudio.com/site/assets/files/1088/pll_and_clocking.pdf
- This has been confirmed for us by Eelco Grimm. The connection emerged in conversations with others in the field while researching this paper.
- Though we speculate here, it probably wasn’t just the Digi 192 converters that were benefitting from external clocks. It was also likely the ubiquitous pro-sumer Digi 002s and 003s, mBoxes, the so-called “legacy” Digidesign Mix Plus systems, as well as the popular Pro Tools HD-compatible Lynx Aurora 8 & 16 converters – all of which we’ve personally heard sound different – and better – when clocked externally. Smart engineers with well-trained ears regularly clocked these units externally – and many of them still do. Though we lack the measurements to back it up, based on the case of the 192 and a mountain of subjective opinion, we’d bet good money that these other units also exhibit a reduction in jitter when clocked externally. Indeed, it is hard to imagine that so many professionals prefer more jitter, not less, from these units.
- “When we interpret our measurements and peek at the 192’s functions, we come to the conclusion that the 192 clock is a non-crystal oscillator that can be pulled many ppm’s. This is necessary for Protools ‘varispeed’ mode, a relic from the days of analog tape machines. We believe that Hollywood filmmakers insisted that the varispeed function would not be dropped on the 192. As you can imagine, a non-crystal oscillator with large pulling range intrinsically exhibits much higher jitter than a crystal oscillator.
To improve things a bit, the 192 also has a second, crystal, oscillator that is used in master mode. However, probably to avoid clock glitches when switching from varispeed to crystal lock, the non-crystal oscillator in the 192 is simply slaved to the crystal oscillator when put in master mode. That means a 192 is always in slave mode, except when in varispeed mode. Above the corner frequency of the non-crystal oscillators’ PLL, the jitter of the non-crystal oscillator is dominant; below the PLL corner frequency the external oscillator’s jitter is dominant. Still you don’t see the jitter level drop below the PLL corner frequency because the jitter of the non-crystal oscillator rises (downwards in frequency) at the same rate as the PLL attenuates its jitter – presuming the external clock or internal crystal oscillator has much lower jitter. Result is a flat phase noise floor (jitter).
With this knowledge, let’s look at the 192 jitter graph again. From 20 kHz downwards you can see the jitter rising, until about 4 kHz where it starts to flatten out. From then on it is flat down to 200 Hz. When slaved to the CC1 master clock, it stays flat down to 20 Hz. When slaved to the internal master clock, it starts to rise again below 200 Hz. What we learn from this is that the corner of the non-crystal oscillator’s PLL lies at approximately 4 kHz (a common frequency for many off-the-shelf PLL’s). In varispeed mode (the only non-slave mode), the jitter would continue to rise below 4 kHz, going off the scale at 500 Hz or so. We could not measure this since our jitter analyser was not able to lock to the 192 in that mode. In ‘master’ mode the internal crystal oscillator keeps it level from 4 kHz downwards, until 200 Hz where the jitter of the internal crystal oscillator is apparently rising to above the flat phase noise floor again. When the 192 is slave to an external clock, the results vary. If you have a jittery external clock, the jitter could start rising at some frequency below 4 kHz. With a low jitter clock, it could stay flat down to 20 Hz. Your milage may vary, depending on the external clocks’ quality. Many will be worse than the 192’s internal crystal oscillator that is used in master mode. But most devices have a crystal clock that at least is a lot better than the 192’s non-crystal oscillator, so usually your result will be much better than the varispeed mode…
It is sad to know that so many recordings have been made with this device that are in much lower quality than possible, just to offer a varispeed mode that’s only used in a handful of recordings – that will sound horrible anyway.”
- We assume that a majority of the people in the marketing departments of the pro audio manufacturers and software companies were trained marketeers who knew how to tap into the end-users insecurities and not trained audio engineers who might have understood that those insecurities could be the seeds of humility needed to become an excellent engineer. Interestingly, it is worth mentioning that there are fields in which the promise of excellent performance from equipment is not coupled with false promises of that gear making the end-user any better. We point to the marketing of athletic gear in general as an area in which the long, arduous struggle for excellence (largely a struggle against the self, thus highly analogous here) is heralded as the reason for acquiring better tools. This attitude sets up a wholly different paradigm within athletics than we have seen with recordmaking. In athletic gear marketing the coupling of discipline with humility is potent, and it would serve as a great lesson to any pro audio marketers out there who might be concerned with advertising toward the hard reality of professional recordmaking, rather than selling products based on false promises of instant gratification.
- Music production departments at some universities were getting started during the 2000s, and many of the professionals who had previously shared their knowledge in the big studios were taking faculty positions. These programs were quite new and had not yet established themselves as solid transmission lines. Furthermore, the curricular demands of a university setting will never quite replicate the experience of making records in a big studio under an experienced professional.
- Blogging bears a quick mention here, as well. The rise of the self-edited “publication” was one of the big contributors to the degradation of the quality of information that was filling up the internet. Recording blogs were rarely run by highly experienced people, and more typically were the projects of younger people caught in a dark gully between their lack of deeper knowledge and the demands for click-thru.
- The AES made broad outreach efforts during this first decade of the 2000s. Allen also recalls being approached by the head of The Audio Engineer Society (AES) in 2007 and being asked how the AES could build more connections with “the Tape Op people.” This was barely coded language. The real question was “How do we get all these untrained people interested in the actual technical basis of what they’re doing?” Allen’s response: “I have no idea.”
- Inductive reasoning is actually more nuanced than this, involving logical inflections based on references to an existing set of truth assumptions, but for our purposes this more common definition of the phrase is more than sufficient and should not mislead us. For students of philosophy, please forgive this adaptation of what may be a less than exact definition, as well as our basing the following discussion on it.
- If you’re morbidly curious, search “external clock” at Gearslutz for a window onto the downward spiral of this bickering.
- A well implemented D-A converter should be able to match or outperform the the incoming data stream in terms of jitter, thus rendering the need for external clocking pointless. Also, because D-A conversion does not record jitter permanently into the waveform, D-A clock accuracy is, arguably, less critical. A-D converters, on the other hand, are often used in arrays, so it’s important that they are all synced to the same master clock source. For this reason, A-D converters almost always have a Word Clock input and can be slaved to an external master clock.
- Quite wisely, many M/TEs began using high-end two-channel mastering-grade converters for overdubbing, printing mixes and as their external clocks. It was an effective way to reduce cumulative jitter in a Pro Tools HD workflow. Accordingly, the marketing literature for high-end stereo converters began to push the onboard master clock as a key selling point, and for many customers – especially M/TEs using Pro Tools HD systems – it was a valid point.
- Read Robjohn’s article here: https://www.soundonsound.com/sos/jun10/articles/masterclocks.htm
- The converters Robjohns tested were a Behringer UltraCurve Pro DEQ2496, an Apogee PSX100, a Focusrite ISA428 with digi‑card, and a Prism Sound Orpheus. He also could have told us whether any of the different clocks he hooked up to the converters had a different impact on their performance or whether they were all largely the same, but unfortunately he either didn’t bother to gather that information, or perhaps he just didn’t share it.
- Please read on for how we plan to deal with this complex and highly technical topic in the future.
- For an example of how the various erronious conclusions in Robjohns’ piece have been picked up and reguritated in the blogosphere, see Justin Colletti’s revamp of these concepts here: http://www.trustmeimascientist.com/2013/02/04/the-science-of-sample-rates-when-higher-is-better-and-when-it-isnt/
- Feb 6, 2016 – This section has been edited slightly to reflect issues with objective testing methods. If you compare the Grimm jitter graph and Robjohns’ audio output graph, you readily see that they’re not measuring the same things. This difference is the crux of ongoing debates in the audio community. How to properly test clocks and converters is a subtle and complex technical topic that challenges our intuitive notion that the analog audio output would be all that matters. We are working with a team to begin to unravel some of the questions that hover around the best methods for measuring jitter and time-domain performance.
- We may be embroiled in a decade-long argument all over again with sampling rates. There have been long-standing rifts between camps of people in favor of going past 44.1k and those who feel that doing so is pointless. There are technical papers and other studies, online chatter – all the telltale signs are there that that something is amiss. Perhaps the lessons of the clocking debate can help us see the value in cooly coming together to form common understanding around sampling rate, and possibly allowing subjective assessments to lead the way.
- E.g.: “Sampling Rate Discrimination: 44.1 kHz vs. 88.2 kHz” – Amandine Pras, Catherine Guastavino, McGill University – Montreal, Quebec, Canada, AES Convention 128 (May 2010) Paper Number 8101, and “The Audibility of Typical Digital Audio Filters in a High-Fidelity Playback System” – Jackson, Helen M.; Capp, Michael D.; Stuart, J. Robert, Meridian Audio – UK, AES Convention 137 (October 2014) Paper Number 9174.
- see http://www.audioasylum.com/audio/dbt.html
- With a softer attitude between objectivists and subjectivists, perhaps the shortcomings of double blind testing (DBT) – in particular AB and ABX testing – can be more graciously investigated by our field. Such investigation should be done in order to refine those tests, not to denounce or eradicate them. In particular, our field has a lot to learn about the impact of exposure intervals (how long we listen to each example, as well as the gaps between them) on our short-term memory’s ability to execute accurate auditory discernments. Despite its reputation as the gold-standard of objectivity in pro audio, blind testing is rarely controlled carefully enough in pro audio to deliver meaningful results. Add to this the expense and rarity of objective technical measurements of digital systems and it becomes quite clear that our most accessible, most abundant, and possibly our most accurate source of evaluation for the time-being might be subjective impressions built up over sighted, long-term exposure to specific systems.
- See Grimm Audio’s “Clocking” White Paper for more on the design of PLLs.
- As well-known digital audio designer Dan Lavry puts it, a PLL’s phase detector “is designed to respond fast enough for tracking the variations in the input clock rate, but slow enough to filter out as much noise as possible.” Lavry’s White Paper can be found here: http://lavryengineering.com/pdfs/lavry-on-jitter.pdf.
- Andy Hong has this to contribute: “For A/D and D/A conversion, ideally, we would have absolute periodicity in our clocks. But imagine if we’re getting a burst of communication from New Horizons as it transmits images of Pluto. In this case, we want our receiving PLL to track the communication as “transparently” as possible, without regard for absolute periodicity. In this case, the jitter that’s in the signal should be tracked as tightly as possible, so it’s the PLL’s job to follow the embedded clock — jitter and all — and any deviation from that tracking could be called phase noise. The time domain equivalent for this deviation would be called tracking jitter.”
- For a more in-depth look at 10MHz rubidium oscillators, see Grimm Audio’s write up at: http://www.grimmaudio.com/site/assets/files/1088/picoseconds_or_ppm.pdf
- After writing this section, we learned that Dave Hill of Cranesong Audio had already prescribed a very similar set of listening instructions along with a set of audio files with varying amounts of jitter in them. This is an excellent exercise for those interested in training their ears to detect the presence and absence of jitter in digital audio. http://www.cranesong.com/jitter_1.html
- Full disclosure: Allen has worked as an independent consultant for Dangerous Music.
- Colleagues Bob Katz and Garrett Haines have reported that some people find the changes in the Aurora’s top end to be harsh when clocked to the Grimm Audio CC1 external clock. Read their report here: http://tapeop.com/reviews/gear/75/cc1-master-clock/
- There are only a few Word Clock generators with 10 MHz ports on them available for the pro audio market, notably the MUTEC MC-3+. In our tests the MC-3+ did not seem to perform well when clocked to 10 MHz generators and hooked up to the Digidesign 192, so we set it aside in favor of the OCX as our main Word Clock generator. We can’t conclude that the MC-3 is inadequate for all 10 MHz generators, but our quick impression was that it might not be.
- A quick google search will reveal that the PRS10 is used in the Esoteric G-01. E.g., see: http://www.audioaficionado.org/mbl-dcs-goldmund-gryphon-etc/27643-clock-experiments-7.html and http://www.audioaficionado.org/esoteric/28251-using-g-01-master-clock-other-brand-2.html and http://www.audioshark.org/archive/t-6298.html
- The width and loudness of the 10M may explain why it was favored during mastering of the densely produced LP Beauty Pill Describes Things As They Are, released in 2015 from Allen’s label Butterscotch Records. Allen oversaw the mastering of the LP, which was done by Emily Lazar using Antelope Audio converters running with the 10M Rubidium Clock using 192k mix prints off the console. Though others had produced beautiful masters of this music, the sonic qualities of the 10M in Emily’s final masters really suited this music – minute details in the high-end were pushed to the sides, and the center image pulled back just enough to bring a cinematic, concave quality to the soundstage. With multiple drummers, unrelenting overlays of electronics, and seemingly endless production details, the 10M provided an excellent sound for this music which also wanted to be loud and forward. Beauty Pill Describes Things As They Are was released at 44.1k and on CD with a louder, more limited master. The high-resolution masters and the vinyl source files were a less-limited master. Almost unanimously, those who hear the high-res version prefer it over the more limited versions.
- 50-to-75 ohm converters exist with BNC jacks, but we didn’t have one on hand. The challenges of successfully changing the impedance have been explored to some degree by audiophiles here and there, but we were not interested in pursuing this route ourselves.
- Lynx has always said that their Aurora converters were “Mastering Grade,” and we hope that readers will understand the meaninglessness of such claims, not just by Lynx, but by any company.