Schiit Happened: The Story of the World's Most Improbable Start-Up
Jul 14, 2015 at 10:13 PM Post #6,994 of 75,537


New Head-Fier
Jun 11, 2015
Not kidding, just a serious question since a lot of companies tune to a particular signature. Ultrasone, Beyerdynamic, McIntosh are a few examples from the top of my head.

It's easier to tune a guitar amplifier (Marshall, Fender) to a particular sound signature (isn't supposed to be transparent) and a headphone too (supposed to be perfect, but unfortunately modern headphones are not there yet). Amps and DACs on the other hand are very close to being transparent and to maintain a distinct sound signature you would have to introduce significant deviations from this "transparent" state. It's probably also hard to make solid state and OTL amps sound similar.
Jul 14, 2015 at 10:29 PM Post #6,995 of 75,537
Jun 9, 2010
Newhall, CA and Corpus Christi, TX
We don't "tune" or "voice" our products to any particular signature, since "signature" implies "coloration." I believe Mike (baldr) discussed this in a post, but I can't find it at the moment.
Schiit Audio Stay updated on Schiit Audio at their sponsor profile on Head-Fi.
Jul 14, 2015 at 11:30 PM Post #6,997 of 75,537


Aka: epoon2, realfacer
Nov 23, 2004
  We don't "tune" or "voice" our products to any particular signature, since "signature" implies "coloration." I believe Mike (baldr) discussed this in a post, but I can't find it at the moment.
Jul 15, 2015 at 10:26 AM Post #6,998 of 75,537
Jun 9, 2010
Newhall, CA and Corpus Christi, TX
2015 Chapter 12:
On Measurements (With a Side Order of Sanity)

Okay. Big subject. Big chapter. But it’s a great subject: what do we measure, and why do we measure it?

When Schiit first launched, we downplayed the instrumentation aspect of our designs—providing only a handful of measurements, without attribution or explanation of how we arrived at them. This led to early speculation (in some corners of the web) that we were doddering mouthbreathers, rubbing magic stones and praying to the audio pantheon for help with more sound-good magick.

So, why didn’t we provide complete measurements from the start? Well, plenty of reasons, none particularly intelligent in hindsight. From the top:
  • Our experience has been that many measurements, at least the commonly given ones, don’t correlate highly with perceived sound quality, even when those measurements differ by several orders of magnitude. Current example: compare a Magni 2 to a Vali, and tell me you can hear that the THD is 100x higher on Vali than Magni 2 at 1V.
  • Our experience has also been that measurements vary from product to product, and cherry-picking one product with stellar measurements and using it as a reference is disingenuous at best, and once fudge-factors are applied to actual measurement ranges, measurements look a whole lot less impressive.
  • And, finally, our experience has always been that measurements can be spun as well as any fancy marketing-speak; how something is measured, at what point, with what filters and weighting, with what equipment…if you work at it hard enough, you can produce impressive numbers from virtually any product, numbers which then encourage people to argue about the relative superiority of a -103dB A-weighted SNR vs a 98dB unweighted SNR, without knowing the technical reality behind it.
“Okay, so now you said that measurements are meaningless,” you might be asking. “Why bother at all?”

Because measurements are absolutely meaningful—it’s just that what measurements are meaningful changes at which part of the product development and production process you’re in. Without measurements, it would essentially be impossible to develop our products…and without deep, high-quality measurement capability, the products wouldn’t be as good as they are.

And, with that, let’s start this introduction to what we measure and why in an unconventional place: by introducing you to the equipment we use, and why we use it. Relax, there won’t be any test on this at the end.

The Metrology Players

“Metrology” is just a fancy word for the science of measurement. It’s sounds more appropriate (and more costly) when used to describe a $3.5MM atomic force microscopy system, but it applies just as well to the gear and techniques we use.

So let’s take a look at what’s on my desk, starting with the measuring stuff.

Stanford Research SR1 Audio Analyzer. This is the big boy that does most of the heavy lifting. The SR1 outputs both analog and digital signals, and inputs both analog and digital signals, so you can look at distortion, noise, gain, max output, frequency response, jitter, and a whole lot more…no matter if you’re measuring a preamp (analog in, analog out), a power amp (analog in, BIG analog out), DAC (digital in, analog out), or ADC (analog in, digital out.) Mike has his own, kitted out with all of the high-precision options in the digital domain. Mine is a more basic model, but it still does some very good digital analysis.
  1. Why is this is on my desk? I use it all the time. It’s a lot easier to hook up a design, even in breadboard form, and know exactly what it’s doing, then to squint and guess (well, at least after the basic power-up stuff happens…but I’ll get to that.) It takes much of the mystery out of how a design is performing (at least when there isn’t a cable/grounding/instrument glitch, but I’ll get to that, too.)
  2. How much will this set me back if I want one? Ouch. Starts at $10K, can get $13K or so with options. More info at:
  3. What alternatives are there? Audio Precision is the leader in this space. They cost more (the one model they have that can beat Stanford in all aspects is about 2.5x the price). There’s also the Prizm DScope, which is about the same price as the Stanford, but with not-as-good performance in the digital domain especially. There’s also Rhode and Schwarz, which cost a mint and we have no experience with them, so I can’t comment.
  4. No, really, what if I want to do measurements like this for cheap? Get a QuantAsylum QA400 for $199. It won’t do everything the Stanford does, and it will absolutely blow up if you overload the inputs (the Stanford will not), but it’s a great way to get started. We use one in production. QuantAsylum also has a much more interesting product coming, the QA405, that looks like it might provide a significant portion of the capabilities of the Stanford/AP/dScope universe for a lot less...but at the moment, it's vapor. Beyond that, it’s soundcards and open-source software.
  5. Hey, can I use an old Sound Tech or HP distortion analyzer to do the same thing? Uh, no. Get a QA400 and invest in resistor padding/protection box and get much more comprehensive measurements.
  6. Hey, my scope does FFTs, can I use that? Uh, no. If that scope goes below about -80dB on its FFT scale, we’ll eat a hat…er, I mean, be very surprised. The Stanford can resolve down about -150dB.
Agilent (now Keysight) MSO-X 2024A Mixed Signal Oscilloscope. This is the device that they play with in the intro to the old Outer Limits series. In general terms, it takes an analog waveform input and displays it on a screen. Like most scopes these days, the Agilent is a digital scope, which also means it can easily look at one-shot and transient events. It also has a ton of other capabilities like built-in function generators, 4 channels of analog, and 8 channels of digital, so you can do mixed-signal stuff and logic analysis. Mine is one with all the options, so I can do all of that. And, best of all, what it displays is very high-res and analog-looking, which is a welcome change from early digital scopes.
  1. Why is it on my desk? For quick “Does it work? Is it oscillating? How’s the clipping look?” type measurements, a scope and function generator will keep you a lot more sane than being stapled to an audio analyzer all the time. Plus, it lets you look at noise on the grounds, in power supplies, etc in a more natural manner.
  2. How much will it set me back? New, about $3500. Check eBay for refurbs. Mine was an apparently unused refurb with all cables, etc for $2K.
  3. What alternatives are there? Tons. Agilent/Keysight is just what we use. If you’re a Tek aficionado, don’t get up in arms. There are also a ton of good Chinese oscilloscopes out there…if you’re willing to put up with less analog-looking waveforms and funky UIs, there are some amazing values. The Siglent SDS1102 can be had for $350-ish and does a good job, if you’re just starting out.
  4. Danger Will Robinson! Do NOT get a USB oscilloscope or pocket oscilloscope for your only scope, unless you have endless patience with software UIs or can decipher and control what is happening on a scope with cryptic numbers and four buttons.
Tektronix 7603 Oscilloscope with 7A22 and 7A18 Modules. In addition to the fancy new digital Agilent scope, my desk also supports a giant old 1970s-era Tektronix mainframe oscilloscope. Yes, a fully analog beast. Yes, we’re talking hand-assembled modules made from discrete components here. While this may seem like an anachronism, it really isn’t. See below.
  1. Why is this heavy, ugly, giant-ass thing on my desk? Because it offers some capabilities that modern digital scopes don’t. First and foremost, the 7A22 plug-in goes down to 10uV per vertical scope division. Compare that to the new Agilent, which does 1mV. When you’re looking at ground noise or power supply noise, that’s 100X more resolution on the old scope.
  2. How much will it set me back? Get one on eBay for $150-300. Modules are $50-100. Prepare to reinforce your desk, especially if it’s from Ikea.
  3. What alternatives are there? Want the same capability on a new digital scope? Then you’re buying a high-$ Tek and a blindingly expensive differential probe amplifier.
Fluke 8842 5.5 Digit Multimeter. This is a good desk multimeter that allows you to measure things like AC and DC voltage, 2-wire and 4-wire resistance. It has 5.5 digits, which means it has enough accuracy to chase down things like shorts on a board. It ain’t very modern, though…good modern stuff goes to 6.5 digits easy. Look at the Fluke 8846A or something like that.
  1. Why is it on my desk? First reason: for checking if your power supplies are working, or if you put the wrong resistor in somewhere, or if you have DC offset, or a board short…you need a good DMM, period. Second reason: battery-powered multimeters are always out of battery when you need them most…and you usually forgot to stock up on the 9V batteries.
  2. What will it set me back? Used, $150-250. A new fancy 6.5 digit Fluke will be about $1000-1200.
  3. Alternatives? Come on, get a Fluke. You know you want one. If for the name alone. (A test and measurement company being called “Fluke” must rank up there with Schiit.)
Fluke 179 Multimeter. At the same time, I also have a battery-powered Fluke.
  1. Why do I have it? Because sometimes you need two multimeters. At least if the batteries are good.
  2. What will it set me back? $300 new, $150-200 used.
  3. Alternatives? Come on, get a Fluke. Seriously.
Der EE LCR Meter. One thing that most multimeters don’t do is measure inductance and capacitance. Hence, I have a cheap LCR meter. Cheap because I don’t use it much.
  1. Why? This is invaluable for checking transformer designs, especially picky quadfilar stuff for circlotrons.
  2. What will it set me back? $100-ish.
  3. Alternatives? Tons. LCR meters can get crazy expensive. If you need it, cool. We don't.
Seek Thermal Camera. This is a widget that plugs into an Android phone (or iPhone, if you get a different model) and shows graphically how hot something is getting. It’s similar to the Flir One, and all of the standalone Flir units, but cheaper. Mike has a real Flir. Dave has a Flir One. All are good options.
  1. Why is it on my desk? Because it beats the hell out of losing thumbprints on hot components. Or, more seriously, because it shows you exactly what’s getting hot on a big PC board, and how hot it’s getting. If you have a couple of SOT-23s running at 150 degrees C, they won’t be running for long.
  2. Why Android? Because my Schiit phone is Android. My Centric phone is iOS. I also use a Nexus 9 and an iPad 3. And a MacBook Pro and an Ultrabook. Haters, hate equally.
Siglent SDG1025 Arbitrary Function Generator. Technically not a piece of test equipment, since it only outputs various waveforms, but I’m including it because I usually use it with the Agilent scope.
  1. Why is it on my desk? Because if you want to look at square wave performance, or open-loop frequency response out to 10MHz, you won’t be doing it on a Stanford. Plus, it has knobs rather than a keyboard, so it works a lot more “analog-y” than punching in a number on the Stanford, or running an automated test routine.
  2. How much will it set me back? About $350 on eBay.
  3. Alternatives? Wow, cheap much? Get one. You’ll thank me.
Of course, there’s a ton of other stuff on and around my desk that is used as ancillary to measurement, like:

Powerstat Variac. This is an ancient device that allows you to bring the power up on a new design slowly (variable-AC, get it?) Powerstats last forever, so you only need to buy one. Per tech, that is. Very useful if you don’t want things to instantly smoke when you botched something on the board. Hook up with an AC ammeter in series, slowly turn up…if it spikes, well…you messed something up.

Mastech HY3003F-3 Power Supply. This is a dual-channel 30V power supply. If I want to put something together fast on a breadboard to see if it works, I usually don’t go through the trouble of building a power supply. That comes later. For tube stuff, I usually use the supply out of an old Lyr or something like that.

Weller WES51 Soldering Station, Yihua 898D Rework Station. The Weller is a fancy soldering iron with good control over temperature and interchangeable tips. Necessary for working with PC boards, duh. The Yihua is an inexpensive hot air blowing device that allows you to work with surface-mount more easily. Nothing fancy.

Whew. That’s a hell of a lot of stuff. But now that you’re introduced, let’s talk about actual measurements.

The (Expected) Measurements (That We Publish)

“Huh? What? Caveats already?” you ask. “Why not publish both the expected and unexpected stuff?”

Hold on. We’ll get to that. What I want to do first, though, is break down the typical measurements you expect to see on a piece of gear, discuss what meaning they have for us, what we expect to see, and if there are correlations to audible differences.

So, here we go:

Frequency Response. This is the range of frequencies that the device can reproduce, usually expressed with a variance. For example, "20Hz-20kHz, +/- 0.1dB." A device with this specification can reproduce all frequencies from 20Hz to 20kHz with no more than a 0.2dB variance. Yes, 0.2dB, not 0.1dB. You saw the +/- there, right? Kinda makes speakers with 45-18,000Hz +/- 3.5dB not look so hot, hmm? That means they could easily be down 7dB at 45 and 18kHz.
  1. How hard is it to get great numbers? For electronic equipment, not hard. A flat frequency response from 20-20kHz is cake. For example, most of our equipment is -6dB at 0.16 Hz and 300-500kHz (without input filtering at 150kHz or so, so it doesn’t reproduce AM radio). That makes 20-20kHz at 0.1dB nooooo problem. It’s also why we sometimes have two numbers, one at +/- 0.1dB and one at +/- 3dB. The only struggle would be with capacitor-coupled tube amps like Vali and Valhalla 2. In those cases, the bass -6dB frequency is defined by the RC filter of the coupling cap and the load, or about 7-8 Hz for a 32 ohm load and 0.7-0.8 Hz for a 300 ohm load.
  2. What do we shoot for? Flat. Duh. It’s actually very hard to make something that’s not flat, unless you are deliberately including tone controls. Otherwise, something is monumentally amiss.
  3. Does frequency response correlate to audible differences? Sure it does, if it is grossly off. But “grossly off,” these days, should be the exclusive domain of transducers—that is, headphones and speakers. It is very easy to get ruler-flat response from electronics.
THD. This is the total harmonic distortion, usually expressed in a percentage at a certain output and frequency, or over a certain frequency range for a specified output. For example,” <0.003% at 1V RMS, 20-20kHz,” or “<0.01% at 150W RMS into 8 ohms.” There are a ton of gotchas with this measurement, including the “total” word. Total harmonic distortion includes all harmonic components—2nd, 3rd, 5th-order stuff, and more. That makes it a less illuminating measurement than looking at an FFT and saying, “Well, it’s all 2nd order at 80dB down, you can’t even see the 3rd in the noise.”
  1. How hard is it to get great numbers? It depends on the topology. It is dead-easy to get great numbers for a super-high-feedback, IC-based design. At that point, THD will be dominated by layout quality (especially in power amplifiers) and power supply design (ditto for power amplifiers.) For a tube run at low plate voltages with no overall feedback, well, you may be looking at several percent distortion—and at that point, it could be audible. Numbers can be made to look better with weighting, especially if you’re talking THD+N, but these designs will usually be exposed by a high noise floor.
  2. What do we shoot for? It depends on the topology and the design. In general, we shoot for -6odB minimum, or 0.1%, at typical outputs. The only amp we make that doesn’t make this number is Vali. Most everything is much, much better, though.
  3. Does THD correlate to audible differences? Like frequency response, only if it’s grossly off. While 0.1% THD is high for an electronic design, it is VERY good for a transducer. Again, the transducer dominates.
IMD. This is intermodulation distortion, an important measurement because it can give us an idea of how a design performs with respect to non-harmonic distortion, which is usually more objectionable than harmonic distortion. This measurement is usually expressed as a percentage at a certain output, and type of IMD measurement (CCIF, SMPTE.) For example," <0.006% at 2V RMS full-scale output, CCIF." The type of measurement is important, because CCIF usually measures 19kHz and 20kHz tones (or another pair of tones 1k apart—we use 19 and 20k) and looks for a resulting 1k tone. SMPTE uses 60 Hz and 7kHz (or two other nonharmonically related tones.) We’ve found both of these tests to better illuminate what’s going on with a marginal design, which has led to some specific tests that we do that go a bit farther…but I’ll get to that later.
  1. How hard is it to get great numbers? Like THD, it depends on the topology, and it’s pretty easy to get good numbrs from a high-feedback design. For poorly designed tube amps, it can be high. Numbers aren’t usually going to look better with weighting, though, so IMD is less easy to game than THD or THD+N.
  2. What do we shoot for? Same as THD. As with THD, only Vali doesn’t meet the -60dB number.
  3. Does IMD correlate to audible differences? Again, only if it’s grossly bad…but since it’s not harmonically related, it may be more audible than harmonic distortion…and transducers usually swamp the electronics numbers, again.*
*It’s not like we hate transducers or anything, they’re just harder to get right from a measurement standpoint. That’s why it’s always best to start with a headphone or speaker you love, before going crazy on amps and DACs.

SNR. The Signal to Noise Ratio of a product is easily the most highly correlated to actual audible differences…simply because a noisy product is, well, noisy. Plug sensitive headphones into it and listen to it hiss. Or hum. Or both. Typically expressed as a dB number below a reference, with weighting, like: "-102dB, referenced to 2V RMS, A-weighted." Pay attention to the reference and the weighting, because that’s where the first number can be gamed. Big time. For example, if we measured Ragnarok’s low gain output in the same way you normally rate a power amplifier—that is, referenced to maximum output and A weighted—it would have a staggering spec like -135dB. However, this is because it’s referenced to 29V RMS and A weighting rolls off where most of the power supply noise will be.
  1. How hard is it to get great numbers? Again (you’ll get tired of hearing this), it depends. If you’re gaming the numbers, rating A-weighted referenced to max output, it’ll be a lot easier. If you’re using a simple, single-ended design with excellent power supply rejection and/or a low-noise power se easy as long as you understand thermal noise…that is, the fact that high-value resistors just sitting there in the signal path (or used in gain stages) create thermal noise. Pick a 30k/3k pair for a diff amp drain load and source degeneration respectively, and your noise levels will be a lot higher than 3k/47 ohms, for example. Tubes are more variable from device to device and manufacturer to manufacturer, so you may get noisy tubes and quiet tubes—hence bigger fudge factors being applied to tube amps. However, all in all, it’s not tough to get good numbers. Just remember to watch the reference level and weighting when comparing products.
  2. What do we shoot for? At least -100dB at low gain, or at full scale output, unweighted. We get there most of the time. One big exception: Mani. You try delivering 1000x gain and see how it works out, noise-wise. That’s just the reality of phono preamps. Now, -100dB may not sound particularly impressive, but -100dB unweighted really translates to inaudibility in practical applications. And, as I mentioned before, we’re notoriously conservative on our numbers. We usually add a 6dB fudge factor to the published specs, based on worst-case measurements. And, it really matters where the noise is. Noise at 60Hz with no harmonic components (like from tube heaters run with AC) is surprisingly hard to hear. Noise at 120Hz from glitch rectifiers and unsnubbed transformers with tons of harmonic components is buzzy and irritating.
  3. Does SNR correlate to audible differences? Absolutely. As stated above, a noisy component will be noisy…you’ll hear it.
Crosstalk. This is how much of one channel bleeds over into another. You’ll usually see these numbers expressed as “-XXdB over Y-Z range,” like: "-68dB from 20 Hz to 20 kHz." In the old days, crosstalk was an important spec, because phono cartridges were so limited that they could only deliver about 20-30dB of channel separation. Today, you’d think it would be easy to have infinite channel separation, but the reality is that everything inside the box influences everything else…PCB traces “talk” to each other electromagnetically, poor power supplies can bleed channel content across from one to each other, even output jack resistance and load impedance comes into play.
  • How hard is it to get great numbers? Harder than you think, especially when you can see some numbers based on insufficient information or complete fantasy. Crosstalk numbers of -80dB, -90dB, and -100dB and more should be looked askance at. Is that only at 1kHz? Then maybe. It’s a lot easier to isolate PCB traces at 1k than 10k or 20k. Is it into no load? Again, then maybe. But real numbers based on an actual 20-20kHz measurement and with actual physical output jacks that have actual physical resistance are usually a lot lower.
  • What do we shoot for? At least -60dB from 20 Hz to 20 kHz, and at a 32 ohm load if the product is a headphone amp, or 8 ohm load for a speaker amp.
  • Does crosstalk correlate to audible differences? Even in this day and age of potentially infinite channel separation from a digital source, probably not…even at -60dB and higher.
Output Impedance. This is the impedance of the output stage of a product. It will be more important in power amps, where a lower output impedance is better (unless you’re of the “current output” frame of mind, which I won’t discuss—but if it works for you, have fun!). This is usually expressed in terms of ohms, or, in the old days, in terms of “damping factor” into a specified load. For example: "0.05 ohms, or a damping factor of 160 into an 8 ohm load." These are the same measurements. For preamps, you may be looking at 75 ohms or 600 ohms (or even higher for some tube pres). It’s important that power amps be low output impedance so they do not affect the frequency response of the headphones or speakers they are powering (transducer impedance varies with frequency, whereas for amps and preamps it should not—or vary predictably due to a Zobel network…like I said, this gets deep really fast…) It’s important for preamps and source components to have a stated output impedance that is not too high, or long cable runs may be problematic.
  1. How hard is it to get great numbers? One more time: it depends. For high-feedback designs, it’s dead-easy to get low numbers. If you throw 100dB of feedback at something, you can have essentially a 0 ohm output impedance. (Or, if you wanna be tricky, use an error-correction output stage for true 0 ohm output impedance, or even negative impedance…and watch out—we were messing around with an error-correction stage that welded a 1/4” headphone plug to a Neutrik jack when I pulled it out while the amp was playing.) For tube amps, low output impedance will be much harder, since the intrinsic output impedance is much higher and there is less gain to play with. Hence the 14 ohm output impedance of Valhalla 2 in high gain mode.
  2. What do we shoot for? As low as possible on amps, and 75 ohms for SE output sources and 600 ohms for balanced output sources. The SE and balanced output impedance is just set with resistors in sources.
  3. Does output impedance correlate to audible differences? Yes, it can. If you use a high-output-impedance headphone amp with multi-driver IEMs, you’re going to run into gross frequency response problems (because they use crossovers). The same thing happens to a lesser extent with dynamic headphones that have impedance variations at certain frequencies. And the same thing happens with speakers. In addition, source components with too high of an output impedance and long, highly capacitive cables can roll off high frequencies.
Maximum Power. This is how much power an amplifier can put out, usually expressed in watts RMS into a specified load, or multiple specified loads. For example, “1.5W into 32 ohms,” or “60W into 8 ohms and 100W into 4 ohms.” These numbers are very important for understanding how an amp will work with your specific transducers. If you’re trying to use an amp rated for 110mW into 32 ohms for HiFiMan HE-6 headphones, you aren’t going to be very happy. But also, if you’re using an amp rated for 1000W into 99dB efficient speakers, you also may not be happy if the amp is noisy.
  1. How hard is it to get great numbers? Heh. We should turn this around and say, “Which numbers?” At one point, the Federal Trade Commission regulated how power output was to be measured, since the amp companies were gaming the numbers so much. They don’t pay very much attention to that today. But here’s what you’re looking for: power in RMS watts per channel, all channels driven, into a specified load. Not RMS? That doubles your stated power. Not per channel? Hey, you can add the numbers! Now Ragnarok is a 400W amp (both channels into 4 ohms, not RMS.) Pretty tricky, right?
  2. What do we shoot for? Depends on the application and the amp. Typically 1W or higher into a typical load for headphone amps, and 50W or higher into 8 ohms for speaker amps.
  3. Does maximum power correlate to audible differences? Absolutely. If your product can’t drive your transducer without clipping (distortion), you’ll hear it. To be fair, though, most headphones don’t need much power to run.
Full-Scale Output. This is a simple measurement for source components, typically DACs, specifying how much voltage it puts out for a full-scale signal (0dB). You’ll usually see this in terms of RMS voltage, like “2.0V RMS.”
  • How hard is it to get great numbers? This really ain’t about great numbers, it’s about knowing what kind of output you have.
  • What do we shoot for? The consumer standards: 2.0V RMS for single-ended sources, and 4.0V RMS from balanced sources.
  • Does full-scale output correlate to audible differences? Believe it or not, it can. If you’re trying to compare a DAC that has 1.5V RMS output to one that has 2.5V RMS output without level matching, well, the higher output one is usually going to sound better.
Gains. This is a simple measurement of how much a product amplifies a signal, usually expressed as a simple ratio or in dB. For example, “Gain = 20 (26dB).” This is a much more important measurement for complex signal chains, such as vinyl…where you have to start with 100 to 1000x gain, then factor in the 400mV output spec into the gain of your amp, and decide if you still need an active preamp. For simpler systems, it’s important to know that the higher the gain, the noisier the product.
  1. How hard is it to get great numbers? Again, this isn’t really about good or bad numbers, it’s about appropriate numbers.
  2. What do we shoot for? The ability to drive an amplifier to its full output with a standard consumer-level source, and gain switching for transducers that need lower noise.
  3. Does gain correlate to audible differences? No, except for the level-matching caveat above…stuff that’s louder tends to sound better.
Power Consumption. This is a simple measurement of how much power a product uses. Well, usually simple. In the case of speaker power amps, the FTC mandates measurement at full output into the lowest specified load. So that leads to big numbers for Ragnarok that don’t really reflect its idle power. It's rated in watts, like "400W."
  1. How hard is it to get great numbers? Depends on what you mean by great. Low power consumption either means low-standing-current design or “standby” modes with keep-alive transformers and high complexity.
  2. What do we shoot for? We don’t. Every product uses the power it needs to.
  3. Does power consumption correlate to audible differences? LOL. No.
Size and Weight. Yep, these are specs too. We take them. We provide them. No, they don’t correlate with sound.

Whew! This is getting deep. But hold on…there’s more. I still haven’t talked about what we don’t publish…and how we use measurements throughout the process…and one of our not-so-standard measurements…

The Measurements (We Don’t Publish)

Wait. Stuff we don’t publish? Why not? Are we embarrassed? Are we hiding something?

Nope. No embarrassment, no hidden agendas. The handful of measurements we don’t publish fall into really only two categories:
  • Too easy to misunderstand.
  • Too technical to matter to anyone other than us.
Let’s have a look at some of these measurements.

Jitter. Oh gawd, people love jitter. They just love, love, love it. They love to throw numbers like 0.2pS at the screen and talk about “femto” clocks and how their jitter is just the lowest possible number it can be. But there are several problems with this:
  1. The number of individuals who can actually measure sub-pS jitter is very, very, very low—it requires a $30K instrument that does exactly one thing: measure jitter
  2. “Femto” clocks—clocks with femtosecond level jitter—frequently only have femto performance at very high frequencies…and are dependent on layout, logic, power supplies, PCB noise, etc, etc…so they may not really mean anything
  3. Most don’t measure jitter where it matters—at the word clock to the DAC—because the number will be higher than those sub-pS readings.
We measure jitter on all our designs at the word clock with an interval analyzer. On the best designs, we can get down around single-digit pS numbers. On other designs, it may be 50+ pS. In either case, they’re not impressive when you’re comparing to 0.2pS. So we don’t publish those measurements. Nor do we publish our eye diagrams, jitter impairment tests, and jitter spectrums from the Stanfords…but a quick look at Yggy’s jitter specs, tested here independently, confirms exceptional performance.
  1. So, how hard is it to get great jitter numbers? I could be snarky and say, “Not hard if you make them up.” But, bottom line, very hard, especially when measured where it matters.
  2. What do we shoot for? 2-digit pS in non-Adapticlock products, single-digit pS in Adapticlock products, both measured at the word clock with an interval analyzer, and correlated with a benign jitter frequency distribution as measured on the Stanford.
  3. Does jitter correlate to audible differences? It shouldn’t. Modern numbers, even on some fairly terrible interfaces, really should be below the limit of audibility. We'll leave it at that.
Open Loop Gain, THD, Frequency Response, Slew Rate. These are the same as the gain, THD, and frequency response measurements outlined in the “what we publish” section above, but applied to the gain stage without feedback, or “open loop.” Since we do primarily discrete designs, and primarily designs that have open-loop bandwidth greater than 20-20kHz, it’s important for us to characterize the designs in an open-loop state. This helps us optimize them for their particular usage. Note that these are early-stage measurements.
  1. So, how hard is it to get great open-loop numbers? Not too hard with inherently linear stages—from complex multistage solid-state amps to simple tube designs that are run with proper voltages.
  2. What do we shoot for? Great numbers for both linearity and bandwidth. And by “great,” we mean less than 0.1% THD, greater than audio bandwidth, etc.
  3. Do open-loop numbers correlate to audible differences? According to some of the Pundits That Be, unless the slew rate is insanely low, no. However, we have noted sonic correlations between a constant -6dB per octave falloff outside the flat passband to infinity (no lumps, humps, bumps, or other weirdness going on.) Of course, pure objectivists will say we’re fooling ourselves on that one. But hey, maybe if we can convince ourselves that there are differences when there aren’t, maybe they can convince themselves that there aren’t differences when there are. Neener.
Gain/Phase Margin. Ah. If you’re not familiar with Bode plots, this one will be relatively boring. Briefly, these are measurements that determine how stable an amplifier design will be, if you’re going to use feedback. There’s always a time delay (phase difference) between the output and input signals of an amplifier. This time delay varies by frequency. And if the time delay means that the output signal is in-phase with the input with an open-loop gain greater than one, boom! You have an oscillator. An amplifier doesn’t do anything until you give it an input, and then it amplifies it by a specified amount. An oscillator just starts singing by itself. Which you absolutely don’t want in audio…especially when some circuits can oscillate at frequencies that take out FM radio!
  1. So, how hard is it to get good gain/phase margin numbers? It depends on the design…more complexity and greater bandwidth (as with current feedback) make it more challenging. However, the main thing to note is that measurement of, and compensation for, an amp design should be done at the PC board level, not in simulation or on a breadboard, because parasitic coupling comes into play here.
  2. What do we shoot for? An amplifier that isn’t an oscillator into any sane load.
  3. Do gain and phase margin numbers correlate to audible differences? Again, they shouldn’t, at least if the amp is actually stable into all loads.

Measurements, Beginning to End

Measurement isn’t something you do once to a single golden sample and call it done. It’s an ongoing process, from the first breadboards to “naked” prototype PC boards to production qualifiers in chassis to production itself. Here’s a run-down of what we do at various stages in the game. It may not be entirely complete, but hey, this is a looooonnggg chapter, and I’m doing this from memory.

Early Stage Measurements. We’re talking breadboards and naked PC board prototype stage here. This is when we’re either just playing around (on breadboards) or trying something that may or may not end up being a product (on a prototype PC board, usually without a chassis, usually green rather than red…we play with quite a few things that never become products.

At this point, what do we look for?
  1. Gross instability and oscillation. This is the biggest problem in both analog and digital designs. This measurement usually starts with a scope and a waveform generator, because it’s easy to just take a look and see what’s going on…as well as see the fundamental of the oscillation if it is oscillating. Is the sine wave “fat” on the scope? Then it’s oscillating. Is a square wave reproduced with a ton of overshoot and ringing? Then it’s only conditionally stable. If it’s apparently stable on the scope, we move to the Stanford…and that’s where it becomes interesting, because sometimes stuff that looks stable might have a noise floor of only -100dB, when they should be -130dB to -140dB. This means it’s oscillating at a low level or only conditionally stable…which means more compensation is needed.
  2. THD up to maximum output. While we’re on the Stanford, we’ll look at THD performance, but usually in terms of what specific harmonic components are present. Typical notes in my engineering docs might be: “-95dB 2nd, -115dB 3rd, 4th and up in noise at 1V RMS into 32 ohms, -75dB 2nd, -90dB 3rd, -115 4th, rest in noise at 15V RMS into 32 ohms. Of course, these numbers will change based on PCB layout, so they have to be re-measured every time.
  3. Maximum power output. Where does it clip? We look at this visually on the scope and at 5% THD on the Stanford.
  4. Clipping behavior. What does it look like when it clips? Does it have sticking problems (nasty waveform irregularities that sound bad and can damage transducers)? If so, that needs fixed before moving further.
  5. Noise. This is another important one, and is 100% the domain of the Stanford. I noted that a giveaway to unstable behavior is a high noise floor, but that’s not the beginning and end of it. Is the noise floor where you’d expect it to be, given the overall circuit design? Where are the power-supply-induced peaks at 60, 120, 180Hz? How big are they? Running down where the noise is coming from is usually the domain of the antique Tek scope—with that, you can see exactly what grounds are being contaminated, and by what.
  6. Heat. What’s getting warm on the board? This is where the thermal camera comes in. Calculation only gets you so far…if you have a couple of 5W resistors each dissipating 2W…next to a hot transformer…better measure it. Though, in our case, it’s usually miscalculating the power dissipation of an SOT-23 transistor or something like that. If it doesn’t fly off the board before the thermal camera comes out, that is.
  7. Input performance (digital). In addition to the above tests, digital devices get verified for different input frequencies and levels, as well as different levels of degradation of the input signal, to make sure they lock to the signals we specify.
  8. Jitter performance (digital). Also on the digital side, we’ll typically take a look at jitter performance once everything is on a PC board.

Qualifying Measurements

Once we get past the naked board measurements, we get into stuff that’s intended for production. So that means we’re now measuring boards in chassis…hopefully final boards, but the measurements let us know just how final they are.

This is where we repeat everything above, and pay more attention to the actual numbers across the board (including IMD, output impedance, and other stuff I didn’t mention above.) We’ll measure 4-6 prototypes and see if they are all similar, or if they’re all over the map. If they are similar, and the numbers are what we expect, then it’s then on the path to production.

Well, except for a few more measurements…
  1. Multiple qualification. Yeah, we like our Stanfords, but Dave has an Audio Precision SYS-2722 as well. And yeah, it’s not the latest Audio Precision, but if the AP numbers and the Stanford numbers agree, that’s a good sign that everything is right with the world.*
*Getting consistent measurements isn’t as easy as you might think. A bad cable, a power cable running over an input cable, noisy AC (our building is very bad), RF interference, ground loops (like from forgetting the scope is still connected when you’re running the analyzer) and even broken equipment all affects your results. We actually blew up the analog generator section of one of the Stanfords…not enough to get it to quit working, but enough to make it behave oddly (wrong output at some ranges) and poorly (high noise and distortion.) Once the problem was confirmed with a loopback test, the instrument had to go back to NorCal for repair. And this is on professional-grade measurement gear. On a QA400 or sound-card based system, it can be much worse.
  1. Stress tests. This is also where we get to short the outputs, stress the protection system, and see if the products survive. It is also where we see if static will cause problems with digital and analog inputs, and where we’ll run tube designs for a few weeks (or months) and re-measure tube characteristics to make sure there’s nothing unexpected going on.
  2. Custom tests. And sometimes we come up with custom tests. For Ragnarok, we were having problems optimizing the algorithm that controlled its bias and DC offset (read, they were blowing up unexpectedly), so Dave wrote a special version of the Ragnarok firmware to push out the numbers applied to the DACs that set bias level about 1X a second, together with the instantaneous readings for bias current and offset. This allowed us to see in real time what Ragnarok was doing, and make changes to the algorithm that eliminated the problem.
Production Tests. And here’s where we do it all again—on most of the first run of products, as well as first articles of each run and spot checks throughout. Plus additional testing that I’ll get to.

Now, some of you are saying, “Wait a sec! Does that mean that you don’t run everything through the Stanfords?”

In short, no. Why? Two reasons:
  • Because the variation in production (and failure rate) is exceptionally low. In a typical run of 2000 Magni 2s, for example, 2-4 of them will fail on first power-up.
  • These failures are usually gross and easily seen on a scope.
  • While the Stanfords are excellent instruments, they are cumbersome to use in production. Yes, we could spend the time setting up automated tests, but automated tests can be easily as fallible (false positives, false negatives) as other instrumented tests, and they provide no benefit for gross failures.
That said, should we be measuring everything in production with automated test equipment? It’s getting to that point, yes. Testing all the input sample rates on a multiple-input DAC and all the I/O on a Ragnarok is pretty time-consuming. So we may be looking into adding some Audio Precision gear specifically for that (they have some stuff specifically targeted at pass/fail testing…and those instruments would live at our PCB assembly house, so they have to be simple to use.

Beyond that, production has its own set of measurements we need to do…measurements that cross over into the actual production process.
  1. Programming and verification. Lots of our stuff needs to have firmware installed, and the firmware needs to be verified. No testing can be done before the firmware is installed and verified.
  2. Biasing. Some of our products need manual biasing of the output stage, like Lyr and Mjolnir. This means we actually have to adjust potentiometers and measure voltages across an output resistor to set the bias…and then verify the rest of the measurements.
  3. Tube Matching. Some products use tubes that need to be matched. We do this in-circuit for the specific product the tubes are used for. And, for even more detailed work (such as looking at a new design or qualifying a new batch of tubes), we have a computerized curve tracer as well.
And then, of course, after all of this, there’s a final listening test. Yes, our final test is subjective. You’d be amazed at what kind of things it can pick up—including tons of stuff that would sail right through a scope or a Stanford. Stuff like scratchy volume pots, operational noises or glitches, or interference only at specific output levels is something that automated or instrumented testing isn’t going to find easily.

Measuring the Unexpected

Okay. Is that enough? No. Let’s go deeper, and talk about one of the measurements we do that is off the beaten path. This measurement appears to correlate at least loosely to subjective impressions, and it unearths some surprising problems in gear that otherwise measures very well.

“So why not release it for the world?” you ask. “If this is such a breakthrough, everyone should be using it!”

Well, we’re not sure it’s a breakthrough. Our sample size is very small. And any correlation it has with sonics is loose at best. And it won’t matter for the hard-core objectivists who have decided that there are no sonic differences between competently designed components, no way, no how, nuh-uh.

And it’s not a breakthrough test. It’s a simple extension of the old IMD idea, but this time with three, four, or five sines—a multitone distortion test. The theory is the same as IMD—can we reveal non-harmonically related stuff with a more complex signal? Usually we use four tones, ranging from 50 Hz to 15,000 Hz, but we’ve run more and less. We’ve used different tones. We’re still playing with this, so don’t take it as gospel.

Aside: it is dead-easy to set up a multitone test on the Stanfords, but I’m not sure how easy it is to do on other products.

So what do we see when we do multitone tests? More non-harmonically related stuff in designs that sound not so hot. Sometimes in very surprising places…not even related to the beat frequencies themselves. Like the example of the Perfect DAC.

The Perfect DAC was not one of ours. It was sent to us by a friend who wanted to get some measurements for it. This was a delta-sigma DAC, manufacturer and chipset redacted, with a very fancy power supply and all the buzzword-compliant stuff people like to hear about these days. We said, “Sure, why not.” And ran it through its paces.

And…in terms of standard measurements, this DAC blew everything we’ve ever measured away. I mean, vanishingly low noise floor, virtually undetectable power supply harmonics, insanely low THD, flat frequency response…

…until you looked at the IMD, which gave numbers a bit higher than you’d expect, given the THD results. And the numbers weren’t related to the 1K spike…they appeared down low, below 100Hz.

What? We ran through our multitone test (it’s easy to do digital multitones on a Stanford as well, not sure about other analyzers) and the low-frequency numbers went bonkers. As in, there was a broad range of non-harmonically related distortion components from 10-90 Hz, at a fairly high level (-50dB or so). -50dB is potentially audible. And it was up nearly 90dB from the baseline measurement.

So what happened? I don’t know. With digital, there are more variables, and noise-shaping and decimation are math-intensive, algorithmically based operations. Perhaps there’s a glitch in their algorithm. I don’t know. It’s not our DAC, and it’s not something we were going to spend the time to dive into.

So…while we putter around confidently with all of the accepted measurements, maybe there are still realms out there where “here be there monsters.”

That’s why we still listen. And measure. And come up with new measurements. And listen again.

And I’ll leave it at that.
Last edited:
Schiit Audio Stay updated on Schiit Audio at their sponsor profile on Head-Fi.
Jul 15, 2015 at 1:02 PM Post #7,003 of 75,537
Jun 9, 2010
Newhall, CA and Corpus Christi, TX
Was the "Perfect DAC" ever available on Indiegogo?

I actually don't even know if it was ever commercialized. I didn't recognize the product. Though it did look like a commercial product.
Schiit Audio Stay updated on Schiit Audio at their sponsor profile on Head-Fi.
Jul 15, 2015 at 1:08 PM Post #7,004 of 75,537


Headphoneus Supremus
Dec 12, 2003
Shawnee, KS
Indeed, I think this chapter is worth sticking at Sound Science or something.

Users who are viewing this thread