RAISING THE BAR: Last week I introduced my own headphone amp design--the Objective2 (“O2” for short). The goal was to see how much objective performance and audio accuracy I could achieve with about $30 worth of parts. And to raise the bar further, the O2 is a “no excuses” headphone amp suitable for most any headphone and adaptable to most any source—at home or on the go. The first O2 article covered the premise, detailed performance measurements, and comparisons with a few other headphone amps including the Benchmark DAC1 Pre’s headphone output and it’s closest competitor, the AMB Mini3. If you haven’t seen the first O2 article, you might want to at least skim through it before reading this one. The final article covers all the other O2 Details, options, and a detailed circuit description.
I HAD SOME HELP: Some really smart and well respected guys, like Douglas Self, Bob Cordell, Bruno Putzeys, Jan Didden, Walt Jung, Cyril Bateman, Samuel Groner, Siegfried Linkwitz, and others, have done extensive audio hardware research and published their findings. These guys have solid numbers, math, measurements and science on their side. Their published results often have an “Ap” for Audio Precision watermark in the corner indicating they use professional instrumentation. Many have published multiple books, papers, technical articles, etc. Their work has been extensively peer reviewed and has stood the test of time. They’ve found what works best from input circuits to capacitors to grounding schemes. They helped perfect the “wheels” of high quality audio. So, rather than go off and try to re-invent the wheel as many DIY and audiophile designers seem bent on doing, I liberally took advantage of their well proven research. Very few can match their expertise in their respected fields and I’m certainly not going to pretend I can do better. So to all of the guys above: Thank you!
TRUSTING THE EXPERTS: An amazing number of audio designers apparently think they know more about audio performance than the component manufactures (and sometimes the guys in the paragraph above as well). For example, they try to design discrete op amps instead of using IC op amps. Samuel Groner tested two of Audio-GD’s discrete op amps and the results were extremely poor. In his comments he said he couldn’t understand why anyone would want to use them. You can find his impressive PDF op amp distortion paper via Google and read the results for yourself. The “roll your own” approach is something like a guy trying to build his own car from scratch in a shed thinking he can do a better job than the companies who specialize in designing and producing cars. Some audio designers take off the shelf parts, that were carefully optimized by skilled engineers using hundreds of thousands of dollars worth of equipment, and think can make them work better by using them in ways the designers never intended—like forcing certain op amps into Class A operation. It’s like buying a new Porsche, bolting on monster truck tires for more “traction”, and claiming the result is somehow better than what all those clueless Porsche engineers thought was best. The sad thing is, audio performance isn’t as obvious as car handling. And most of these designers lack the equipment to properly measure the results of their often crude efforts. If they did, they would probably realize the overall performance is very likely worse as Samuel Groner has demonstrated.
No Worry Audio
OSFA: A primary mantra for the O2 was “One Size Fits All” (OSFA). Basically I wanted an amp suitable for most any application. The first article has the details. (drawing:DBaldinger)
ACCURACY: A primary goal was maximum accuracy. The amp should get out of the way and deliver music as the recording engineering intended. That might sound like hype, but it’s mostly an objective engineering exercise to make that happen—not audiophile voodoo magic. Again, the first article goes into this further.
HIGH-END MEETS OBJECTIVITY: A lot of audiophile beliefs have some objective truth behind them. For example, resistors really can sound different. Some are relatively noisy and others don’t follow Ohm’s law very well—their resistance varies with the voltage applied creating distortion. There’s some truth to “burning in” electronics as some capacitors in certain applications perform better over time. Electrolytic capacitors in the signal path sometimes create measurable and audible problems. And if someone claims a coupling capacitor changes the sound, why not measure and analyze the differential signal across the capacitor while the amp is operating playing real music? If there’s nothing much to measure, by Ohm’s law, it can’t change the sound. If there is something there, you’ve found an area to possibly improve. I’ve done that and much more with the O2. Objective measurements help an audio designer focus on the things that matter most.
Overall Design Principal
- Cost (almost) No Object – Some go overboard with ultra-high end parts, exotic topologies (i.e. fully balanced), etc. These designs can end up being very costly as some “boutique” audiophile parts are ridiculously expensive and some of the topologies require 2+ times as many parts. Do they work any better? It’s unfair to generalize but I know lots of the parts and principals that go into overkill designs often have no measurable benefits and fail to survive blind listening comparisons. I’m sure some designs turn out great. And even those that don’t can be impressive works of art to be admired. You can put a Mercedes AMG V8 engine into a riding lawnmower but the result probably belongs behind a rope in a museum rather than trying to mow lawns. In other words, the engineers at Lawn Boy can probably build a better lawn mower than some guy in a shed using Mercedes parts. I go into this more later and it’s one of the key reasons why many Cost no Object designs are flawed—the implementation is at least as important as the parts and cost
- Latest Ideas – Many in this hobby are used to getting a new phone/PC/iPod/etc. every year or two. An Intel Core i7 is easily faster and better than a Core 2 CPU for example. But that doesn’t translate to analog audio. Some of the best audio op amps are 10+ years old. The only thing analog I can think of that’s really changed in the last decade are “Class-D” power amplifiers. Some DIYers and boutique manufactures “invent” new topologies like 3 channel designs but those too rarely provide any real world improvements and are often a step backwards as I’ve shown. The challenges of implementing an accurate headphone amp were solved a long time ago. The “newer is better” axiom might apply to smartphones, but it doesn’t usually apply here.
- Design By Ear – A lot of DIY designers, and apparently even some small commercial ones, lack the right test equipment. Some use RMAA but it’s loaded with limitations and problems. So most of them, not wanting to spend five figures on instrumentation, are at least partly designing in the dark. But they creatively argue that’s ”OK” because they depend mostly on their ears. But that’s been proven deeply flawed in multiple ways. Most audiophiles dislike blind listening (as it could undermine a lot of their beliefs and hence is often criticized) so they use what’s known as sighted listening. And along with that comes a strong unavoidable psychological bias that’s hardwired into all our brains. Multiple studies have shown we humans easily hear things that don’t exist. So the “design by ear” crowd are genuinely fooling themselves. NuForce admits to designing this way and their products have had some embarrassing problems as a result. There’s a lot more information and many references on this topic in my Subjective vs Objective article.
- As Cheap As Possible (but make it look nice) – This is the mantra for a lot of commercial designs being sold. And “cheap” often extends to the R&D time that went into the design or lack thereof. It took Benchmark a few years to develop the first DAC1. But some companies, like Audio-GD, FiiO, etc. crank out new designs every few months. Obviously there’s not a lot of R&D going into each Audio-GD product unless they have an army of design engineers but one guy claims to design all of it. A lot of companies are trying to profit from the “headphone craze” however they can and it shows. Cheaper designs, faster to market, means more profit. This assumes, of course, enough people buy them. And if you’re an advertiser on Head-Fi the odds seem good you’ll have a loyal fan club no matter how bad the product is. Samuel Groner tested a few Audio-GD products, a Head-Fi sponsor, and the measured performance was awful. But they’re still immensely popular among many Head-Fi members who drink the subjective Kool Aid. So while this approach might work, I honestly don’t know how some of the purveyors sleep at night.
- Purely Objective – This is how most of the mainstream big guys do it—like say Behringer or Sandisk. They survey the competition, come up with detailed specs and requirements to be a bit better than Brand X, and figure out how to make a product that meets all the goals yet can sell for less. It’s not sexy but it’s a formula that works. Unlike in the audiophile world, their products usually do meet the claimed specs and consumers at least get what they pay for.
- Harmless Excess – This is my philosophy. When it can’t hurt, and the cost is low, I’ll throw in an audiophile tweak here, a better part there, etc. Douglas Self and others have shown in many applications polypropylene capacitors don’t measure or sound different than similarly constructed polyester caps. But if the the two types are close in price, I’ll go with with the audiophile preferred polypropylene. And I do the same with performance goals. A lot of studies claim you can get away with 0.05% distortion at least everywhere but the midrange. But if I can keep everything to 0.009% or better across the board, without other compromises, it provides extra peace of mind. Those are just two examples.
ISN’T LISTENING MORE IMPORTANT THAN MEASUREMENTS? I think both are important. And that’s why, for the first O2 article, I conducted blind listening tests comparing the O2 to the well reviewed Benchmark DAC1’s headphone amp. It’s best to make all the right measurements and do lots of listening. If you want the listening to be unbiased, you have to do it blind. This has been demonstrated dozens of times.
CAN DIY BE BETTER? I think it can but I suspect it rarely is. Someone who really knows what they’re doing can certainly beat the misguided “Design By Ear” and profit hungry “Cheap as Possible” guys. The “Purely Objective” camp is a bit more challenging and depends on the product, price, etc. Can I beat Behringer’s current headphone amps? Yes. But that’s because they’re designed for musicians monitoring a mix while playing live which isn’t a terribly critical application. If Behringer made an audiophile-grade headphone amp that would be a bigger challenge. And it’s even harder to beat a company like Benchmark, Violectric, or Grace Designs as they’re much less concerned with shaving every penny out of the design and they’re very good at audio design. It’s all they do. In general, those engineers with their knowledge and equipment can turn out a better product with inexpensive mainstream parts than nearly any DIY designer can manage in his basement with RMAA no matter what parts the DIY guy uses. I know that probably sounds harsh to lots of DIYers in their basements, but in my experience, it’s just reality. They often have no idea how poorly their designs perform in some areas.
DESIGNER COMPONENTS: Some audiophiles are “component snobs”. Someone once told me the Benchmark DAC1 isn’t worth considering because its Alps volume control only costs a few dollars. But, being objective, the DAC1 has great crosstalk performance (a weakness of some volume controls), good channel balance tracking, the volume control feels solid, turns smoothly, and doesn’t make any audible noise when turned. So what exactly is wrong with the volume control? The answer: Nothing significant. But some think you’re supposed to spend way more to get those hidden designer labels. They can go enjoy their latest issue of the Robb Report. They’re after something very different than simply getting the most accurate sound and the O2 isn’t their kind of amp.
DESIGNER PARTS GONE WRONG: I tested a commercial headphone DAC with a fashionable trendy DAC chip and op amp in it. The PC board layout looks fancy with everything arranged neatly in rows (always a bad sign—more on that later). But whoever designed it apparently couldn’t be bothered to read (or perhaps understand) the datasheet for the DAC. The oversampling digital filter in the DAC chip—a very critical aspect of a DAC—defaults to 24/192. But as a USB DAC it runs at 16/44. Because the filtering is all wrong high frequency content in music creates alias artifacts that are “mirrored” down into the audio band. It measures poorly on the dScope and I’m pretty sure you can hear all the extra high frequency garbage. A typical Head-Fi subjectivist might buy this DAC, hear the extra high frequency crud, decide the added HF content is newfound musical “detail”, and give the half baked DAC a glowing review on Head-Fi. Others run out and buy one and, courtesy of subjective bias, hear what the first reviewer described. Next thing you know the Half-Baked DAC Company becomes a Head-Fi sponsor, and well, you can see where this is going. But the real crime is whoever designed it either never properly measured it, or if they did, they didn’t care they got it wrong. But hey, it looks nice and uses all the right fashionable components. That’s what matters most, right?
DESIGNER COMPONENT CHALLENGE: Some claim specs alone don’t tell you how something like an op amp will sound. I believe if two op amps meet clear some basic measurement criteria, they will sound so similar it’s next to impossible to tell them apart. Anyone’s who’s skeptical might be interested in my Op Amp Blind Listening Challenge.
IMPLEMENTATION IS EVERYTHING: Like the DAC mentioned above, I’ve seen all sorts of products that use the right parts but got the details wrong and don’t work very well. Just routing a single ground signal wrong on the PC board can seriously harm performance. I’ve seen designs that measure great on RMAA but are simultaneously oscillating at RF frequencies. The O2 demonstrates proper implementation can yield genuinely excellent performance without using any designer or expensive parts. Some of the O2’s measurements are pushing the limits of even my dScope audio analyzer.
HEADPHONE AMP DESIGN 101: For those interested, the rest of this article discusses what goes into designing a headphone amp, some of the trade-offs, and how the steps were applied to the O2. I also bust, or at least dent, a few myths here and there. If you just want to build an O2, it’s not required reading. But if you’re interested in learning more about what makes a good headphone amp, and specifically why the O2 ended up the way it did, it’s worth checking out. And if you consider yourself a DIY or commercial audio designer it might be genuinely useful.
STEP 1 REQUIREMENTS: It’s considered good engineering practice to start with fairly detailed requirements. Without decent requirements you don’t know where you’re doing, it’s harder to keep your eye on the ball, you end up wasting a lot of time, and it’s harder to know when you’re done and if you even got it right.
1-1 BASIC REQUIREMENTS:
- As much of a “One-Size-Fits-All” (OSFA) design as reasonably possible
- Performance as accurate as possible for the best sound quality (see the first O2 article)
- Portable (rechargeable battery) and desktop (AC line) operation
- Reasonably small portable size to fit an inexpensive off-the-shelf enclosure
- 7 hours minimum battery life with a 20+ hour low power version
- Brief short circuit protection
- No risks to headphones with power up/down, low batteries, etc.
- Switchable gain (2 gain modes)
- Power LED
- DIY friendly design with no surface mount components
- Reproducible: No critical components, matched parts, critical construction techniques, etc.
- As many components available from a single vendor as possible to save shipping costs
- As low cost as possible while meeting all other requirements
1-2 WHAT HEADPHONES? Obviously a headphone amp is to drive headphones. But which headphones? In keeping with the OSFA requirement, I wanted to find close to the “worst case” headphones possible. After quite a bit of research it seems the 38 ohm planar HiFiMan HE-4/HE-5LE and 50 ohm HE-6 are especially tough cans that need lots of current. They’re the kind of headphones that make tube amps run and hide in the closet. And for needing lots of voltage, the Beyer DT880-600 needs the most of voltage of any dynamic/planar cans I could find. For wild impedance swings and ultra high sensitivity my Ultimate Ears Super Fi Pro 5s and Etymotic ER-4s help round out the assortment.
1-3 HOW LOUD? Music with a wide dynamic range is the most challenging to play loud because the peaks are much louder than the average level. Such music may not seem loud but the peaks can be seriously challenging for your audio hardware. For such music to approach live levels, you need to cleanly reproduce peaks of 110 dB SPL. If you want to know where that number comes from, see More Power?
1-4 THE CALCULATIONS: To start with a well known example, the Sennheiser HD650 is rated at 103 dB at 1 V RMS input into 300 ohms.The HD650 needs 2.2 Vrms to hit 110 dB SPL. The math can be found in the More Power article. 2.2 Vrms is about the limit of say the Mini3. And, not coincidentally, most find the Mini3 gets loud enough with the HD650s—but just loud enough So this provides some correlation that 110 dB, and this approach, seem to really work in the real world. Now to apply it to our two worst case cans:
- HiFiMan HE-4/5 – They’re rated 87 dB SPL at 1 mW and 38 ohms. To drive them to realistic peaks of 110 dB SPL they need a whopping 200 mW (enough to fry some headphones). That’s about 2.8 V RMS and a peak current of 104 mA per channel in 38 ohms.
- HiFiMan HE-6 – Rated 83.5 dB SPL at 1 mW at 50 ohms. They need 447 mW to reach 110 dB which is 4.7 V RMS and 133 mA of current. Despite their higher impedance they’re even more power and current hungry than the HE-4/5 above. There’s good reason here for tube amps, and plenty of other amps, to run and hide.
- Beyer DT880-600 – These need 43 mW at 600 ohms to hit 110 dB which is 5 V RMS. This is way beyond what most portable amps can manage. Even the new FiiO E11 can’t come close nor can the Mini3.
- Extra Headroom – In the spirit of OSFA, it’s best to have some extra headroom above and beyond the theoretical limits. That way the O2 won’t end up on the ragged edge. Generally 25% is considered enough headroom for voltage and current capability. So 133 mA * 1.25 = 166 mA and 5 V * 1.25 = 6.25 V RMS. I’m sure in the real world there are still a few cans that won’t quite get loud enough, but I’ve tried to find some of the most challenging that are currently in production (the K1000 “ear speakers” don’t count).
- Noise – Testing shows noise that’s 85 dB below the maximum listening level will usually be inaudible under nearly all conditions. The Ultimate Ears and Shure IEMs hit 110 dB SPL with only about 100 mV of input. 85 dB below that is 5.6 uV or –105 dBv (103 dBu). This would be a noise level of 25 dB SPL with these headphones. See: Noise and Dynamic Range
1-5 AUDIO SPECIFICATIONS: Here are the complete audio specs. This isn’t some watered down list of specs with the bar set conveniently lower than it should be. It’s the real deal. I believe this to be the true point of diminishing returns and amps that can pass all of the following on a real audio analyzer get my seal of approval. Most of these criteria are supported by well respected research and/or are generally accepted guidelines as to the thresholds of audibility. Some thresholds are not black and white so it’s best to error on the side of being conservative (more accurate) and that’s what I’ve done here (many of these are explained a bit more in the first O2 article):
- Output impedance less than 2 ohms
- Input impedance >= 10K
- Frequency response +/- 0.25 dB 20 hz – 20 Khz 400 mV 16-600 Ohms
- Phase response less than +/- 2 degrees error 100 hz - 20 Khz 16-600 Ohms
- Absolute phase: Preserved
- Slew Rate greater than 3 V/uS using 10 Khz square wave near full output 600 Ohms
- Distortion under 0.01% 20hz – 20 Khz into 16 – 600 ohms from 10 mV – 400 mV RMS
- Channel separation better than -40 dB @ 16 ohms and –60 dB @ 150 ohms 400 mV RMS
- Channel balance error less than 1 dB at any setting down to –45 dB below max volume
- Noise under –105 dBv (103 dBu) unweighted (5.6 uV or -97 dBr referenced to 400 mV)
- DC offset under 5 mV typical, and ideally, under 20 mV worst case
- 100% stable with any realistic reactive load from 16 – 600 ohms
- Transient ringing and overshoot tightly controlled with all realistic headphone loads and 0.01 uF
- 166 mA per channel, both channels driven, peak current capability at < 1% THD
- 6.25 volts RMS on AC power at < 1% THD into 150 ohms
- 4.5 volts RMS on DC power (nominal battery voltage) at < 1% THD into 150 ohms
2-1 DISCRETE, IC OR BOTH? Tubes and single ended designs were ruled out in the previous article as they’re notably less accurate and far more likely to let their presence be known and get in the way of the music. They’re also not battery-friendly. That leaves a push-pull solid state design with 3 main choices:
- Fully Discrete – Some designers, such as Kevin Gilmore and John Lindsay Hood, prefer fully discrete designs. And Douglas Self has published some truly excellent fully discrete power amps for driving speakers (his “blameless amplifiers”). While such designs can work very well, they require a fair amount of work to get right. And some aspects can be especially tricky—most notably finding the optimal bias level in a Class-AB design and stabilizing the bias over a wide range of device temperatures. PSRR and CMRR typically suffer without hand matching components. High frequency stability can be rather challenging as well. And discrete designs are generally more power hungry than an IC design as the transistors will require more bias for low distortion—not a good thing for battery operation. Discrete designs can also be fairly fussy about components which conflicts with the “easily reproducible” requirement. And, finally, it’s really hard to beat the big IC semiconductor companies that spend millions on R&D. Want proof? Check out Samuel Groner’s Operation Amplifier Distortion paper (Google it) and compare the discrete op amps he tested to the ICs. No contest. The discrete designs fail by a wide margin.
- Hybrid – There are many headphone amp designs floating around using an op amp to drive a pair of discrete output transistors in each channel such as the AMB M^3. There are many in databooks. But you don’t see many of them fully tested. Basically they suffer most of the same bias and stability issues as described in the paragraph above. Crossover distortion is especially difficult to fully correct with feedback. It helps a lot if they’re class A but that’s not suitable for battery operation. And some are more difficult to stabilize without using output inductors or 10+ ohm series resistors. It can be done, but it’s very difficult for such designs to match an IC with a Class-AB discrete output stage in most areas except current capability.
- ICs (myth busted) – If you want to compare a 60+ watt chip amp for speakers to say a Doug Self blameless discrete design I’ll put my money on the discrete Self amp. But at headphone levels, it’s a different ball game. You don’t see Doug Self off trying to build a better op amp or buffer IC out of discrete parts. With only a few exceptions ICs are the best way to go. For those who claim not to like the alleged “sound” of op amps they should consider the entire signal chain. Some favorite recordings widely used by audiophiles as demo material were recorded and mixed using hundreds of op amps. The mixers, equalizers, compressors, processors, etc. were all full of op amps. And there are very likely op amps in whatever digital source gear they’re using. So the majority of music is already well steeped in op amp goodness. And there have been plenty of blind tests that support the transparency of op amps. To those who insist op amps sound bad, I say let’s arrange a blind test!
2-2 CLASS-AB OR CLASS-A? Class A amps can have some significant advantages—especially if you’re unable to fully optimize a Class-AB design including the bias operating point. Given the relatively low currents in a headphone amp the main advantage of Class A is getting rid of crossover distortion. But here’s where those IC designers at the semiconductor companies and their expensive R&D labs get to show off. They can nearly always remove crossover distortion from Class-AB designs much better than a discrete designer can. This is partly because it’s far easier to manage the thermal tracking issues when everything is on a single die so the bias point can be very precisely controlled in an IC. They also have tricks at their disposal a discrete designer can only dream about. So, in a nutshell, they can get class A performance from an IC on a class B power budget. It’s the best of both worlds. This is shown in the residual THD measurement from the last article. And for those forcing op amps into Class A that weren’t designed to be operated that way, unless you can fully measure the results of your hack, you may well be making things worse. If It’s not broken it’s usually best not to try and “fix” it—especially if you can’t fully test the result.
2-3 SINGLE OR MULTI STAGE TOPOLOGY? As explained in the Cmoy With Gain and Mini3 articles there are some significant compromises trying to use one stage for both the gain and output stage. It’s like a high performance front wheel drive sports car—few exist because the front wheels have to handle all the steering and power duties which ends up compromising both. So a single stage design was ruled out and a two stage topology is the obvious choice for the following reasons:
- Lower Distortion – Adding voltage gain requires using less feedback, and less feedback means higher distortion. Most of the distortion in a headphone amp is in the output stage so you want the most feedback possible for that stage which means operating at unity, or nearly unity, gain.
- Lower Noise – The gain stage is where most of the noise comes from. If the volume control is after the gain stage reducing the volume also reduces the noise with it. In a single stage design, with the volume control at the input, you get all the noise all the time at any volume setting. With the volume control at the input you also amplify the Johnson Noise of the volume control itself which, in many headphone amps, dominates the overall noise. So there are huge noise improvements to having a second stage with the volume control between them. See the O2 noise measurements in the previous article for more details and the proof.
- Lower DC Offset – The gain stage is where significant DC offset is usually generated. Using two stages allows isolating this DC from the output stage and hence the headphones.
- Higher Stability – A 2 stage design can be inherently more stable see 2-5 Feedback below.
- Component Optimization – With two stages the components (ICs) for each stage can be optimized for their task. High current devices that can drive headphones don’t make the best gain stages and vice versa. With two stages no such compromises are required.
- Controlled Impedances – The first stage provides a known low impedance source for the second stage. This allows the components, like the volume control, coupling caps, bias resistors, etc. to be optimized for maximum performance without worrying about what’s used to drive the amp.
- Lower Power Consumption – Compared to 3 or more stages, a 2 stage design will generally use less power for longer battery life.
2-4 AC OR DC COUPLED (myth busted)? I know many high-end subjective audiophiles don’t like caps in the signal path but, in reality, their benefits can far outweigh their negatives when you use the right cap properly. Douglas Self, Cyril Bateman, and others, have conducted tests of capacitors and how they affect audio signals. And, lending some truth to the audiophile beliefs, there are circumstances where capacitors cause problems—including electrolytics in the signal path and EQ circuits where caps have a significant AC voltage across them. But a properly sized high quality film coupling capacitor isn’t one of those circumstances. For a cap to do more than simply attenuate the sound by a tiny fraction of a dB, requires a non-linear voltage across it in operation. The isolated inputs of the dScope allow analyzing the voltage across any cap while the amp is operating. And, besides a miniscule bit of linear attenuation (which is expected due to the ESR), the non-linear components are lost in the noise which is a few microvolts. It’s not even remotely close to any threshold of audibility. Proper coupling caps have also been shown transparent in blind and audio differencing tests. So the only negatives are usually cost and space but the benefits are substantial:
- DC Input Protection – A direct coupled amp like the Mini3 will amplify any DC offset at its input and send that amplified DC straight into the headphones. And, because audiophiles often dislike coupling caps, audiophile and DIY gear (including DACs) are more likely than consumer gear to have significant DC offsets. At 14X gain, just 70 mV of DC at the input would mean 1 full volt of inaudible headphone destroying DC output. Is that risk really worth the zero proven benefits from DC coupling?
- DC Offset Reduction – As mentioned in 2-3 above, the gain stage generates most of the DC offset. AC coupling keeps it away from the output stage so it never reaches the headphones.
- Silent Volume Control – DC bias currents through a pot wiper are hard to avoid in a fully DC coupled design. And with bipolar input opamps, it’s usually audible like it is with the Mini3 and FiiO E9. When you change the volume with no signal you hear an obvious “rustling” noise in the headphones. That’s just tacky. It makes it sound like the pot is worn out when it’s just a marginal design that’s to blame.
2-5 LOCAL OR GLOBAL FEEDBACK? With a multi-stage amp you have to decide between global feedback, local feedback or a combination of both. To most easily meet the goals of stability and transient response local feedback was chosen. It has the following advantages:
- Gain Stage Isolation – Reactive loads create phase shift between the output voltage and output current. Local feedback isolates the phase shift enhancing stability.
- AC Coupling – Each stage needs DC feedback to operate. So an amp using global feedback must be fully DC coupled. That prohibits using AC coupling between the stages which has other advantages (see 2-3 and 2-4 above).
- Volume Control – Global feedback generally requires the volume control be outside the feedback loop. This has significant noise consequences as discussed in 2-3 above.
2-6 DC SERVO? Does the design need a DC servo? Using a servo lets the amp be fully DC coupled without worrying about the offset voltage. But servos require power and this is a portable amp where battery life is important. Servos are generally more applicable to larger discrete power amplifiers where they can address compromises with the design of the feedback loop (different AC and DC gains). The O2 doesn’t have these problems to begin with. And the DC offset and frequency response of the O2 are genuinely excellent without a servo. A servo would just drain the batteries faster with zero benefit, take up board space, make the amp more expensive and add complexity.
2-7 POWER SUPPLY VOLTAGE? Before going too far with selecting components (like op amps) it’s good to know the power supply voltage. To achieve 6 - 7 volts RMS output you need about 20 V peak-to-peak. Accounting for the appropriate voltage drops, a +/- 12 volt (24 volt total) power supply should work depending on diode drops, and how close the stages can swing to the rails under load. This is another area where some designers get carried away. Higher supply rails limit your choice of components and create much greater power dissipation (waste heat) in the output stage. Power is a function of the square of voltage. So as the power supply voltage goes up, the thermal losses are exponentially greater and the amplifier is less efficient. You want the voltage just high enough to get the job done and no higher.
2-8 INPUT CIRCUIT? The higher the input impedance the more stray noise pickup you get if the amp is connected to an un-terminated cable or gear that is powered off. And much less than 10K can excessively load the outputs of some DACs, preamps, etc. So 10K was chosen as optimal. An RC filter provides RF protection with a cutoff around 3 Mhz. This is low enough to filter out most RF energy while not creating phase shift in the audio band with even higher impedance sources. Cell phones operate at 800+ Mhz and are the most common source of RF problems. A good input circuit should also have some series resistance to help limit current into the op amp if it’s overloaded and provide greater ESD protection. But the larger the series resistor, the worse the noise performance of the amp so it’s a trade-off.
2-9 GAIN STAGE: It’s an IC-based design so the gain stage will be an op amp. The first choice is what topology? Inverting (shunt feedback) or non-inverting? The inverting stage has some benefits for common mode rejection but it’s at a disadvantage for noise performance which is critical in a OSFA headphone amp. A non-inverting design is also absolute phase correct keeping the purists happier (assuming the output stage is also non-inverting which it is).
2-10 WHICH OP AMP? With a topology defined, which op amp out of the hundreds available is best? The semiconductor websites are the best place to start, then all the datasheets, and ultimately you have to start testing the best candidates in the desired application. A given op amp won’t be specified at the exact gain, impedances, etc. of your circuit so the datasheet can only tell you so much. And testing can reveal other surprises. Doug Self, for example, found the expensive Analog Devices OP275 performed worse than several less expensive options. Nearly two dozen op amps were tested in developing the O2 and I have published two articles on the process:
2-11 VOLUME CONTROL: This is a critical part of a headphone amp and what’s mostly responsible for meeting the channel balance requirement. So it’s worth keeping a few things in mind:
- Location - See 2-3 above for why the volume control is best positioned between the two stages rather than before the input/gain stage. The main downside to the “volume control in the middle” approach is it’s easier to overload the gain stage which always runs wide open. The O2 has a gain switch to address this problem but it’s something those altering the default gain settings should be aware of. See: Gain Stage Overload. A compromise is a “split gain” design where you divide up the gain between two sections but this requires the output stage, in a 2 stage amp, be capable of gain (many buffers are not) and usually requires additional components taking up board space.
- No DC - The same high quality film capacitor that provides DC input protection in the O2 also isolates the bias current of the next stage from the volume control rendering it completely silent with no “rustling” noise when the volume is changed. If you put even the tiny DC input bias current of most op amps through a pot wiper—especially if it’s before the gain stage—you’ll get noise in your headphones when you adjust the volume. The Mini3 and FiiO E9 suffer this problem.
- Channel Balance – All pots have some channel balance error. Due to the logarithmic nature of perceived volume, and the way voltage dividers work, the channel balance error will be greatest at the lowest volume settings. And it’s different for every pot. The dScope has a real time channel balance measurement so you can literally turn the pot and watch the number in dB change in real time. It’s not uncommon for the louder channel to swap back and forth as the pot is trying to average a 0 dB error over its range. The dScope makes it relatively easy to find the worst case imbalance and it’s nearly always in the first 5% of the range. The key is to intelligently set the gain so you avoid using the few few percent of the volume control’s range.
- Stepped Attenuators – These are great when they’re well implemented. But beware of the eBay versions. All those tiny SMT resistors on a DACT? What are they exactly? Thick film SMT resistors perform poorly for audio. And even thin film SMTs get much worse in smaller sizes. They get more noisy, and more alarming, their voltage coefficient rises dramatically. That means they don’t follow Ohm’s law as well—their resistance literally varies with the voltage applied. That creates distortion as the voltage is constantly changing in an audio amp. And the switch contacts, wipers, etc. in a cheap stepped attenuator are prone to wearing out as they get a lot of use. They’re rather expensive to make properly. So, without detailed measurements with an audio analyzer, I wouldn’t trust an eBay (or other unknown) stepped attenuator.
- Electronic Pots – Most electronic pots are not well suited for high-end audio. Many run from 3V or 5V single power supplies and the audio signal has to stay in that range. So they have to be capacitor coupled and the audio signal has to be referenced to the midpoint of their power supply. That’s a big problem in a high quality amp. They’re also, at best, limited to about 1.7 V RMS before they severely clip the signal. There are some higher voltage chips and even some with bipolar power supplies but they’re expensive (around $10 each). In most cases you need a microcontroller to run them but a few have pins for up/down buttons. Sometimes they’re built into chip amps (like the one in the FiiO E5 and E7).
- Choices – Lower values are generally better to reduce the Johnson Noise. But if it’s in the input stage, you have to consider loading on the source gear. I think so many use the Alps RK097 because it's hard to find anything better without spending a lot more. Noble, Panasonic and Bourns makes some similar pots but they’re not better that I know of. There are some interesting motorized options as that’s more typically what the high-end manufactures want. The Alps RK27 "Blue Velvet" pot is a really nice pot. But it's currently only stocked in a 100K value which is way too high for most headphone applications (except tube amps) and it wouldn't fit on the board. There are also cheap RK27 clones around made in China.
- Taper - Alps makes two slightly different audio tapers for volume control use “3B” and “15A”. The 3B taper is at 50% (-6 dB) at half volume. The 15A taper has a more gentle taper up to 70% and then it becomes more steep. If you listen mostly at lower volumes the 15A taper is likely the better choice. If you listen often close to full volume, the 3B might be better.
2-12 OUTPUT BUFFER CHIP? I considered a lot of options for the output stage. A dedicated high current buffer seems like an obvious choice and here are the main options:
- National LME49600 - This part performs very well. But it’s only available in a surface mount package, it’s relatively expensive, and worst of all, it’s too power hungry for a battery operated design. And, because it has no gain, it doesn’t work with the chosen 2-stage topology here without giving up several other things (especially noise performance and potentially stability). See the BUF634 below for more.
- Linear Tech LT1010 – This is a fairly old part and only is rated for 150 mA of current which is below the 166 mA spec. It also has relatively high quiescent current and several specs are not as good as the 49600 above. It suffers the same topology issue as the 49600.
- TI BUF634 – This is available in a TO220-5 through-hole package, and it has a lower power mode, but they’re over $12 each at Mouser and there’s still no voltage gain so you can’t use local feedback. Running it without feedback creates excess distortion. And using global feedback creates stability issues and requires a DC coupled design with the volume control at the input rather than between the stages. That creates DC offset issues—especially at higher gains—and far more noise for all the reasons already mentioned. So to properly use the BUF634 in this design, you need a third op amp stage to provide the feedback for the buffer. And if you use global feedback you may need a DC servo. And either one would help kill the battery life. Not to mention there’s not enough space on the small PCB or either option. Plus, just the required pair of BUF634s would cost more than all the other parts on the board combined.
- TI TPA6120 – This has been a popular part I suspect partly because it’s much cheaper per channel than the above 3 choices. But as I documented in my FiiO E9 and DIY QRV09 reviews, TI specifies at least a 10 ohm output impedance and it has seriously high distortion at low frequencies. Plus it’s nearly impossible to properly solder (it’s surface mount with a hidden heat sink pad). It’s also power hungry with high quiescent current. It can, however, be configured for voltage gain and local feedback.
2-13 OUTPUT STAGE CHIP AMP? I looked at several “chip amps” designed to drive headphones and I didn’t find any that met the requirements. Most are designed for low voltage operation from around 3 – 5 volts. I also looked at “chip amps” made to drive speakers as they have higher voltage capability. But the only ones I found with suitably low quiescent current for battery operation have relatively poor performance specifications as they’re designed for lo-fi portable audio gear. So much for chip amps.
2-14 OUTPUT STAGE OP AMP? By process of elimination (discrete designs were ruled out earlier) the choice is down to an op amp. But which one? I scoured all the semiconductor websites looking for high current output, low distortion, through hole packaged, 24+ volt power, low distortion, op amps, and came up with only a few parts that even came close to meeting most of the criteria. Here’s the breakdown:
- Analog Devices AD8397 – The 8397 is is used in some “fashionable” headphone amps like the new FiiO E11 and the AMB Mini3. The good news is it can manage around 300 mA of peak current which is impressive. It will also (barely) run from a +/- 12 volt supply to meet the voltage swing requirement. The bad news is it’s surface mount only which makes it impossible to directly socket and opens up lots of DIY issues. The AD8397 can be pre-mounted to a SOIC-to-DIP8 adapter and used that way but that makes an already expensive ($6.50) op amp more expensive and decreases its already marginal stability. Because this op amp is much faster than it needs to be for audio use it’s far harder to properly stabilize. AMB documented having problems with it and being unable to get it to work at supply voltages higher than the +/- 4.5 volts in the Mini3. There are quite a few stern warnings in the datasheet. It’s also not short circuit protected and known for blowing up if it’s even slightly abused. So it needs some form of protection—usually series output resistance which compromises the performance. It’s also rather dissipation challenged in this application. Both channels in a single SOIC8 package with peak currents of 166 mA and a 24 volt power supply will far exceed the dissipation limits. So it’s off the list.
- TI Burr Brown OPA551/2 – These single op amps were my first choice as they look good on paper, are rated for 200 mA of current, have high open loop gain, and even some cool features like thermal shutdown. In testing, however, they proved to be something like the AD8397—not quite as bad but still high strung. Admittedly my prototype board wasn’t optimal but the OPA551 didn’t like reactive loads without using output inductors or substantial series resistance. You can get clever with the feedback to presumably compensate for capacitive loads but it kind of works against the OSFA mantra as it’s hard to optimize it for all possible loads without using an output inductor. The very high current limit of 380 mA is also less than optimal. Finally, they’re relatively expensive at $5 each and you need two of them.
- TI TLE2062 – This is an interesting part that’s specified into loads as low as 100 ohms—headphone territory. The quiescent current is less than 0.5 mA per amplifier section which is great for battery operation. Amazingly, despite the lower power, the slew rate is more than fast enough. It’s also rated at 80 mA max which is far higher than most op amps and it’s short circuit protected. But the unity gain bandwidth is less than 2 Mhz which is less than optimal. Likewise the CMRR and open loop gain could be better. The open loop output impedance is also rather high which, in real world use, limits the voltage swing into low impedance loads. Still, this part has some promise as I’ll discuss later.
- JRC NJM4556 – JRC is an interesting company with its roots in audio and analog. Unlike the big US companies like National, TI, etc. they tend to either offer lower cost versions of existing designs, or they design application-specific analog parts. I suspect the NJM4556 is one of the latter as it’s unique and almost made to order as a headphone op amp. It’s rated at 70 mA of current specified into 150 ohm loads. It was designed for audio use. In the eBay Cmoy it managed around 100 mA peak and overall rather impressive performance. In the same prototype setup as the OPA551 the NJM4556 performed significantly better in several tests and was much more stable. The NJM4556 is the optimal speed for audio use and no faster. This makes it very stable without any fussy special requirements, output inductors, or series resistance required. That’s worth a lot in this application. The one remaining problem is the output current. The 4556’s 70 – 100 mA obviously falls short of the 166 mA requirement. But, the 4556 is a dual op amp with two well matched amps on a single silicon substrate (die). What will it do with both amps in parallel to double the current capability? The answer is, with appropriate measures, it does good things! It produces over 200 mA which approaches the 250 mA of even the LME49066 and BUF634 above. It turns out using one 4556 with paralleled sections for each channel meets all the requirements. I’ve also tried to blow one up with brief short circuits playing music at clipping and so far so good. And you can buy 7 of them for the price of one OPA551. To my knowledge, this is the first time the 4556 has been paralleled for headphone duty and it’s a big reason the O2 can deliver great performance at such a low price.
2-15 OUTPUT STAGE DESIGN – With the 4556 specified, the rest of the output stage needs to be optimized. First, you usually can’t simply connect two op amp outputs in parallel. They may not share the current equally and any difference in offset voltage will significantly increase the quiescent current. Measurements with the dScope demonstrated just 1 ohm of series resistance works nicely with half a dozen different 4556 samples (some from different production lots). These are effectively in parallel so the output impedance is approximately 0.5 ohms which is well under the 2 ohm goal.
2-16 CURRENT LIMITING – It’s important to protect the output stage from at least brief short circuits and current limiting can help a lot. But intelligent current limiting can also help protect the headphones in a OSFA design like the O2. The FiiO E9, for example, can put out over 1 watt into 16 ohms. That’s enough to send many headphones up in smoke and much more than any 16 ohm headphone I’m aware of needs. FiiO’s solution was to toss extra resistors in series with the 3.5mm jack resulting in serious frequency response and damping problems with balanced armature IEMs. A much better solution is active current limiting. This is discussed more in the first O2 article. The end result is the O2 still has more more power than any headphone I know of in the 16 – 32 ohm range should ever need. But it’s only about 1/3 as much as the E9 puts out at 16 ohms which should help save expensive headphones if someone gets careless with the volume, plugs something in with it cranked up, etc. And it does it without 43 ohms (FiiO E9) or even 10 ohms (QRV09) of output resistance. Active current limiting is the solution that makes the most sense. Why don’t more amps use it?
2-17 STABILITY (one myth confirmed) – Audiophiles claim at least some output inductors sound bad and there just might be something to that. I tried at least half a dozen different inductors in the output of the QRV09 trying to get rid of the datasheet mandated 10 ohm resistor for the TPA6120 and it didn’t go well. All the ferrite inductors significantly increased distortion. And trying to use an air core inductor, on that very fast amp, caused instability. Output inductors are common on power amps driving speakers but seem a bit more problematic in this application. Some of them also are relatively large in terms of PC board real estate. For these reasons I wanted the O2 stable into any reasonable load without inductors. The two stage design with local feedback goes a long way towards achieving that by isolating reactive loads from the rest of the amp. Combine that with an output stage that’s not’s better suited as a video amp, the right PC board layout, proper power supply decoupling, what’s described in the next paragraph, and you have a nicely stable amp.
2-18 COMPENSATION - Op amps get complex, literally, when you near the end of their useful bandwidth. There’s complex math involved, complete with imaginary numbers, to calculate their behavior, including things like phase margin in a given circuit. To do it right, you have to take into account stray parasitic capacitance, stray inductance, and more. Those things are difficult to model in calculations or simulations. Despite the fact the O2 uses internally compensated op amps you still have to verify their stability and transient response. In a low gain application, an op amp might still show ringing on square waves due to the phase margin becoming small. The generally preferred solution is to apply additional compensation in the feedback loop. This, in addition to the dominant pole compensation, is used to optimize the transient response and stability. Amplifier circuits that exhibit significant ringing often have dangerously low phase margin. In my experience this can have a negative impact on their sound quality although some have perceived the “different” sound of a ringing amplifier as somehow better. See Op Amp Myths and Op Amp Measurements for more details. The compensation in the O2 was optimized using a very fast ( < 20 nS) rise time square wave and a fast (> 50 Mhz) scope. The resulting –3 dB point of the O2 is around 250 Khz and the phase shift at 10 Khz is less than 1 degree. Tthe slew rate is still in excess of 3 V/uS. Why does anyone need more bandwidth and “speed” for an audio amp? Would you rather it starts oscillating at 500 Khz when your friend plugs his $1500 HD800s into it? Output stages don’t like being RF transmitters and get really hot trying. The heat ultimately makes them fail and the resulting DC could easily destroy the HD800s. Your friend will hate you and your marginally stable amp. Half-baked stability is never good.
2-19 GROUND TOPOLOGY: Anyone who’s read my Mini3 review, Cmoy review and/or my Virtual Ground article knows I strongly prefer a proper bipolar power supply with a real ground. And it needs to be a star ground as shown at the right where each functional area of the amplifier has its own private path to a single central ground reference. A true zero volt referenced bipolar dual power supply almost always works best in a headphone amp.
2-20 BATTERY POWER SUPPLY: The O2 only needs less than 200 mA DC clipping a sine wave in both channels into 15 ohms. In real world use with music the current is under 60 mA DC total under even difficult conditions. Just considering battery operation for the moment there are several choices:
- Bipolar DC-DC Converter – This allows using a relatively low voltage battery like a single 3.7 V li-Ion cell, pair of AAs, etc. A DC-DC converter generates dual bipolar supply rails at the desired voltage (+/- 12 volts for the O2). If the battery becomes too low the converter shuts down and the amp shuts off gracefully. But DC-DC converters are expensive and are typically only 50% - 80% efficient when not fully loaded so 20+% of the battery capacity is wasted. They also create substantial electrical and magnetic noise (EMI) that will find its way into the audio circuitry no matter how hard you try to keep it out. You can see an example of this noise on the blue square wave at the end of the FiiO E7 review.
- Single DC-DC Converter Or Charge Pump – This is a lopsided version of the above solution. Only the negative rail is generated and the battery is used “raw” for the positive rail. For the O2 that would mean at least a 9 volt battery. The FiiO E5 and E7 use a single charge pump built into the output chip amp. The supply has asymmetrical impedances which can degrade performance. So you would be lucky to get 3 hours instead of 7 – 9 hours. Battery life would be rather poor with only 50% of the total battery capacity (watt/hours) plus the losses in the converter. And it may have unpredictable behavior when the battery gets low. The plus is the amp could be smaller.
- Dual Batteries – A battery is quieter than any power supply. Which helps explain megabuck battery powered phono preamps like the Nova Phonomena. Two batteries are also 100% efficient—all the battery power goes to the amplifier rather having some wasted in a power converter or even more wasted in a virtual ground/third channel. The main downside is the cost and space of two batteries and, more important, possible headphone damage if one battery becomes disconnected or dies first. This issue is addressed in 2-23 below.
2-21 AC POWER SUPPLY: There are several options for the AC power as well:
- DC Trickle Charge - A cheap solution is to always run the amp from the batteries and simply trickle charge them with a 24 volt DC wall adapter. The batteries would equally divide the DC charge current between them neatly solving the problem of needing two DC supplies from an AC source. The downside is the batteries are required even for AC operation making this a poor solution for a full time desktop amp. And, worse, the power drain exceeds the maximum continuous “float” charge rate for 9 volt Ni-MH batteries so the AC adapter would extend the battery life but the amp would still eventually die playing even while plugged in. Or the current would have to be set high enough to eventually “fry” the batteries if the AC was left connected with the amp turned off. Neither is acceptable.
- Center Tapped or Dual Transformer – A conventional bipolar power supply is made using either a center tapped (3 wire), dual secondary (4 wire), or two independent AC transformers (like the QRV09). So far so good. The problem is wall transformers with a 3 or 4 wire AC output are extremely rare. So you have to build the power supply from scratch which involves working with hazardous AC line voltages. The supply would also need to be built into a separate enclosure (expensive) or the O2 itself greatly increasing the size for a portable amp.
- Dual DC Power Adapter – I’m not aware of any wall adapters with suitable dual DC outputs, but there are table top “brick” power supplies (like most laptops use) with dual outputs. The problem is the decent ones are relatively expensive (around $40+) and are overkill for this application. They’re also switching power supplies intended for digital devices and typically have some noise in their outputs which probably will find its way into the audio circuitry.
- Single AC/AC Wall Adapter - An inexpensive AC/AC two wire wall transformer can provide a true split supply using half-wave rectification. While half-wave might seem less than ideal, the performance is in the implementation and the advantages are numerous. Wall transformers are already safety agency approved and available for local power/plugs in various countries. They come with standard barrel connectors allowing the use of a small inexpensive power jack on the amp. The amp can operate perfectly without the batteries installed. There’s no high frequency switching “hash” to worry about. All that makes an AC wall transformer the best choice for the O2. Anyone doubting the performance of a half-wave power supply should check out the performance results in the first O2 article including the incredible noise performance. All those measurements (marked “AC”) were using this power supply. It works great!
2-22 BATTERY CHARGING: Again, there are more options:
- Charge Controller IC – These typically use multi-stage charging with a faster rate until the battery is close to full then switch to a trickle charge. They’re great but the only options for a dual 9 volt bipolar battery set up are expensive and not DIY friendly (surface mount).
- Constant Current IC – This solution is used in the Mini3. But, it turns out, it’s actually a disadvantage for 9 volt Ni-MH batteries. See the next option.
- Resistive Taper Charging – The battery specs say you can charge a NiMH battery at 1/20 the battery capacity indefinitely. Higher than that and you’ll cook the battery, and much lower will take forever to charge. That limit is about 10 mA for the O2’s batteries. A resistor can be sized setting the charge current well under that (about 6 - 7 mA) when they’re fully charged which helps prolong the life of the batteries when the amp is left plugged in. But, even better, the charge current in the O2 is proportional to the battery voltage. This speeds charging compared to a constant current source. The maximum current is around 50 mA if the battery is completely dead which is still a safe value. So no constant current limiting IC is required. It’s also dirt cheap and takes up very little board space.
2-23 POWER MANAGEMENT: As mentioned under 2-20, dual batteries present a significant risk. The problem of potential headphone damage has to be addressed. I developed a unique circuit that manages the DC power rails of the amplifier and offers some additional benefits besides DC protection. I’m not aware of anything similar in other headphone amp designs. It serves five purposes:
- Headphone Low Battery Protection – The circuit shuts the amp down long before the amp can become unstable due to one battery dying before the other. So there’s never any DC at the output.
- Battery Disconnect Protection – If one battery becomes dislodged the amp will immediately shut down preventing a large DC output. In my development adventures this is a more likely scenario than low batteries. There’s also the chance one battery is not fully connected when you turn on the amp. The circuit prevents the amp from turning on unless it’s safe to do so.
- Power On Transient Suppression – Controlling the timing of the power rails greatly reduces turn on transients. The power circuit reduces the turn on “click” by approximately a factor of ten.
- Cell Reversal Protection - 9 volt batteries, with 7 cells, are easily damaged by what’s known as cell reversal if they are discharged much below 1 volt per cell. Without protection the O2 (or a Cmoy) will operate down to battery voltages of under 3 volts. The listener may have no idea the batteries were being over-discharged and permanently harmed. The power management circuit shuts the amp down when the batteries drop much below 1 volt per cell (7 volts each). So users are free to just listen until the amp shuts off. No harm. No foul.
- Hysteresis – When the amp shuts off due to low batteries the battery voltage will rise because most of the load is removed. A simple circuit would just turn the amp right back on again. And you end up with the O2 doing DJ rap special effects with your tunes. So the circuit is designed to keep the amp off when the batteries are low.
2-24 LOW POWER VERSION: A lot of portable music players these days will play 20 – 40 hours on a battery charge. So one of the O2’s requirements is 20+ hour battery life in a low power version so you can fly from LA to Sydney with ease on a single charge even with a multi-hour layover along the way. Those who want to use the amp on the go, or where there’s no AC power handy, will probably be happier with the low power version. See: Low Power Option
2-25 ENCLOSURE: With most of the amplifier design roughed out, how small can the enclosure be? There are not that many high quality low cost enclosures on the market—especially if you want a metal one. The best I could find was the “BEX Series” by Box Enclosures and the B2-080 was the smallest one that would conceivably hold two 9 volt batteries and the 80-ish components on the O2’s circuit board. The B2-080 is a high quality rigid extruded aluminum case that you could probably drive a car over and it would survive. Plus it’s under $11 with front and rear panels and hardware. It has slots to slip the PC board into so no mounting hardware or modifications are required. There’s also the slightly taller B3-080 that can accommodate the same PC board but has more height to allow extra panel mounted jacks such as a 1/4” headphone jack, RCA input jacks, etc.
2-26 FRONT AND REAR PANELS: While it’s temping to put external components at both the front and back of the amplifier that requires using two of the most expensive single component on the entire Bill of Materials—the custom machined panel. That makes a big difference in the total cost. The B2-080 and B3-080 come with both front and rear panels so if the rear panel is blank, it’s free. So all the controls and jacks are arranged along one side of the PC board. That way only one machined panel is needed (it’s under $17 from Front Panel Express) instead of two. It also allows using deeper custom enclosures without having to worry about anything matching up with the back of the enclosure.
2-27 PC BOARD: The enclosure mandates the PC board be no larger than 100mm x 80mm. This is where things get really, um, “fun”. Figuring out if everything will fit can take many hours or sometimes even days of trying with the PCB layout software. And if it doesn’t fit, you’re forced to either start compromising your design and toss out parts, or you give in and move to a bigger enclosure, a bigger board, and more or less start over. The more cramped a PC board is the harder and more time consuming it is to place and route. For those who have never routed a board, nothing can cross over anything else that’s not supposed to be connected. So it’s like a 2D Rubik’s Cube puzzle to get all the signals where they need to go, especially when you take 2-29 into account.
2-28 CHANGES & ADDITIONS: Just like in software development, hardware is often a moving target. You might get the first prototype built and finally working right and then you or someone else discover it needs to be changed. For the O2 the suggestion was made to add a gain switch. But there wasn’t space. The solution was to get creative and switch to 1/8 watt sized resistors. There are 20-ish resistors in the O2 so that made room for the gain switch but required re-routing the entire board. It also created a new problem I would discover later in 2-31.
2-29 PC BOARD ROUTING: The O2 has nearly 80 components. Where they’re placed and how they’re interconnected is a seriously critical aspect of performance. While you can find lots of info on how to route a tricky HDMI digital video signal, or make a strip-line antenna feed for a WiFi module, there’s surprisingly little info on routing analog audio boards. Even Doug Self hasn’t published much on the topic. It’s much less black and white than most other aspects of audio design and, to be honest, it’s something of an art. It’s very much an acquired skill you can’t master overnight. There’s even an underground blotchy photocopied “manuscript” paper dating back to the 80’s that’s passed on from senior engineer to junior engineer like some kind of secret manual and right of passage. It’s all about the “black magic” of PCB design. I’m seriously not making this up. Here are just some of things that can make a big difference in measured audio performance:
- Ground Routing – As mentioned earlier a star ground is essential. The diagram to the right is what often happens but it’s not how to do it. That oval bit at the bottom is a typical “ground fill” island on a PC board. And within it critical ground currents overlap and interact with each other. That creates distortion and other problems. With a 2 layer board, and a design of any complexity, you will quickly run out of routing options if you try to return every single grounded component to a single point. So you have to know which things can share a ground return and that’s design specific. And you have to plan for anything “off board” that can destroy your best efforts. For example if someone uses non-isolated metal connectors, and panel mounts them, those grounds you so carefully routed end up connected together elsewhere (via the panel) and you end up with ground loops and/or other serious problems. The Mini3, as another example, has exposed ground “strips” on both sides of the PCB in an attempt to ground the PC board to the enclosure but this creates two giant ground loops through the upper and lower halves of the metal case. An enclosure should always be grounded at a single point and never have ground currents flowing through it.
- EMI Loops – Think about where the higher currents flow. You want to keep the “loop area” of those currents as small as possible. For example, I’ve seen several boards that more or less route a power supply rail up each edge of the board and the ground up the center. It’s very logical, neat and tidy. But it’s very poor practice for analog audio. All the currents flowing from the rails, through the load, and back to ground are spread wide apart creating a single turn coil or inductor with your entire circuit nestled right in the middle of each “loop” where all the resulting EMI fields are the worst. And even a 3 channel or bridged (balanced) amp doesn’t solve the problem because the return current is to the opposite rail which is still on the other side of the board. It takes some serious thought to imagine all the current paths and the loops they form. Then it’s even more of challenge, especially on a cramped board, to minimize those loops while still being able to route the entire PCB in 2 layers.
- Inductive Currents – Don’t run high gain input signals parallel to anything with much current flowing. There will be inductive coupling that can (depending on the signals) significantly increase distortion, degrade crosstalk, and worst of all, create instability if it’s out of phase with whatever it’s next to or generates positive feedback to an amp input. Doug Self has examples of 100 times greater distortion and I’ve seen similar results from various half-baked designs.
- Watch Parasitics – If you dig deep into datasheets and/or application notes, especially for faster parts, you’ll often find some advice on keeping the part stable. One problem can be parasitic capacitance and inductance. Amplifiers hate positive feedback as it turns them into oscillators. Yet I see all sorts of questionable routing where outputs are coupled to positive inputs in ways they never should have been. You may not always want ground fills around the input pins of op amps, for example, as that creates parasitic capacitance that can degrade stability.
- Use The Reference PCB Design – I know some engineers who work for semiconductor companies. It’s not uncommon for one or more engineers, more or less full time on the project, to spend months getting their reference designs correct. They might spin a half dozen iterations of the PC board and they’re making measurements with hundreds of thousands of dollars worth of instrumentation. Their reference design is essentially the company’s business card for that part. It’s the first thing many potential customers will lay their hands on and evaluate. It’s important it be as correct as reasonably possible so the part makes a good first impression and they get more design wins. Why then do so many DIYers and small commercial designers just ignore these carefully engineered reference designs and screw everything up thinking they somehow know better? This is also blatantly a problem with DIY DACs. Use the reference design as much as possible. If you really think you know better than the guys at National, Analog Devices, TI, Linear Tech, etc. you’re very likely wrong. I’d love to be in the same room when you try to make your case to the guys (or gals) who sweated all the tough details of their reference boards.
- Forget Aesthetics – It’s apparently hard for some to resist but the optimal layout rarely has everything arranged in neat tidy rows and columns like a spreadsheet. Electrons have no concept of aesthetics but they don’t generally like going further than they need to because someone thought an IC or resistor looked better over there. See the next item.
- Think Small – It’s tempting to think of copper PCB tracks as nearly “perfect” but they’re not. A four inch trace at a typical width of 15 mils from one side of the O2 to the other is around 0.13 ohms. That may not sound like much, but it can be a lot in some parts of the circuit. Traces also have inductance and capacitance and couple to both components and each other. A little math helps put this in perspective. For the distortion goal of 0.01% at a realistic listening level of 400 mV (my standard reference level) that distortion can be reached with just a single unwanted anything of only 40 microVolts (0.00004 volts or 40 uV). It doesn’t take a lot of ignorance to end up with an extra 40+ uV here and there from a sloppy layout. And before you know it, even using the best op amps and parts on the planet, your design won’t produce anything resembling really low distortion. Moving a single track can change the distortion from 0.005% (-86 dB) to 0.1% (a lousy –60 dB). Really! PCB design is way more important than many realize.
- Ground Planes vs Floods/Fills – A ground plane, by definition, is uninterrupted. The idea is ground currents will follow the shortest path back to their source. And that generally keeps loop areas very “tight” which is what you want. But if the ground area is interrupted, it’s no longer a plane, and the ground currents likely will have to take long paths around obstacles that interrupt the “plane”. It’s like walking anywhere you want on a football field, versus breaking the same field up into a bunch of little islands surrounded by water with only bridges here and there to get across. It’s entirely different. And it rarely works very well. The random divided areas are more accurately called “floods” or “fills” and they’re what you get when you fill unused areas on the PC board with copper after you have routed all your traces. If you want a real ground plane use a 4 layer board (which are much more expensive—especially in small quantities) and don’t route any signals on the ground plane. Otherwise you have to carefully plan for every ground current and use star grounding practices—floods or not.
- Never Use Auto-Routers – A low speed digital design will typically at least work if you simply route the power and ground tracks (or set up suitable design rules for them) and turn the software auto-router loose on everything else. But it’s suicide for high quality analog audio work. The auto-router has no concept of what makes an op amp oscillate as just one example.
2-30 COST ENGINEERING: This step is sort of sprinkled throughout the process. But in choosing everything from the topology to the components if you’re designing to a budget you have to keep an eye on the pricing. Nothing is really expensive in this design but with 80 parts it does add up. The most expensive items, especially in low quantities, are the PC board, enclosure, front panel, rechargeable batteries, and the AC wall adapter.
2-31 SPECIFYING COMPONENTS: This gets surprisingly complex if you’re trying to keep the costs down and not compromise the design. Some things to consider:
- Try To One Stop Shop – Shipping costs can really add up. If you’re forced to buy parts from 5 distributors that’s probably $50+ in total shipping costs right there. The goal here was to get as much as possible from one source and I managed to do get everything that’s on the PC board (which is a complete working amplifier) and the AC wall adapter from Mouser. You can even get the batteries, solder, and tools from them if you want.
- Try To Combine Values – I won’t name any names but I’ve seen DIY designs with a 1K resistor in series with the LED and a 1.1K resistor in the feedback loop of an op amp and no other 1K resistors in the entire design. Why not use 1.1K for the LED? It will be ever so slightly dimmer. Big deal. There are lots of examples like this where I see similar, but different, non-critical values being specified. It just makes the design more complicated, more likely something won’t be in stock, etc. In the commercial world it also usually costs more money.
- Look Up Similar Values – As explained earlier, there was a need to switch to mostly 1/8 watt sized resistors due to a lack of space. A few resistors in the O2 are especially critical for noise. But when I tried to find low noise resistors in that size I discovered Mouser only had limited values. And the prices for the best mil-spec ones were $0.30 or $1.19 (same resistor, slightly different resistance value). Often what matters in audio is the two channels be well matched (hence the use of 1% resistors) but the absolute value isn’t that critical. Nobody cares if the input impedance is 10K or 10.1K for example. Mouser might have a big customer that uses huge quantities of the 10.1K value so they can offer a much better price on that value. So there are a few weird looking resistor values in the O2 and that’s why. Same part, way cheaper, no difference in performance. This also happens with many other passive parts like capacitors.
- Stick To The Big Companies – Distributors like Mouser don’t sell junk like you’ll find on eBay and at smaller outfits like Jameco, etc. They sell what are known as franchised lines which means they’re a top tier distributor for those parts with full factory support from the component manufactures. Most of their customers are making hundreds or thousands of units at a time. And the last thing a distributor wants is for a customer to have to re-work 1000 boards because of some substandard part. Avoid no-name components, surplus, small dealers, and eBay parts if at all possible. Always remember eBay is often used to liquidate reject merchandise.
- Passive Components Can Matter – The issue here isn’t so much the brand, but the details. Thick film SMT resistors for example are awful for audio use. Always use thin film. Metal film through hole resistors work much better than carbon film. The ESR of power supply capacitors can vary by a factor of 10X for the same value capacitor but you have to download the datasheets and look it up. Other capacitors can have very significant differences. Doug Self and others have written extensively on this topic. You can’t just use any parts and expect similar performance.
- Avoid eBay Components – A lot of the components being sold on eBay are rejects that can’t be sold elsewhere. Big companies do incoming inspection on parts, and if they fail, they’re sent back. In China many of the parts that weren’t good enough to even use in kids toys end up on eBay. And many of the audiophile components are fake knock offs. That Alps Blue Velvet pot from a Chinese or Hong Kong vendor? It’s probably not made by Alps.
- Try To Avoid Single Source Items – Lead times on many parts are horrific right now. Some parts have 52 week lead times. So, more than ever, single source parts may become an impossible roadblock. Sometimes you can’t avoid it but at least try to look for two versions of a component that can work. Like the B2-080 or B3-080 from Box Enclosures both work for the O2. Plus each comes in several colors. So it’s unlikely they’ll all be out of stock even if they are made by only one manufacture.
Prototypes and Testing
3-1: SIMULATION: I have multiple simulation tools including some with four figure price tags. All of them are ultimately based on SPICE which dates back to at least the early 80’s if not the 70’s. It’s basically from an era when a computer the size of a living room struggled to do in a day what an iPhone can do in a fraction of a second. Despite its very crude origins, computer simulation is still useful for some things. For example I used it in designing the power management circuit for the O2. But when you’re worried about things like –90 dB vs –70 dB of crosstalk, or 0.007% THD vs 0.07% THD, or 105 dB S/N or –95 dB S/N, simulation falls flat on its face. The reason is implementation. A lot of high-end performance is more about getting the power supply, grounding, PC board layout, decoupling, parasitics, etc. correct. And simulation glosses over most of those or it gets seriously complex trying to model all of them. It’s also really easy to leave things out or otherwise have flaws in the underlying models you’re not even aware of. Simulation can be a useful, but crude, tool for getting in the ballpark. But don’t pretend it represents real world performance.
3-2: DIGITAL VS ANALOG: If your design is say a PIC microprocessor that measures the room temp you just hack together a protoboard and test it out. Digital signals under 5 Mhz or so are fairly tolerant of sloppy implementation. But if it’s a piece of high-end audio gear things are different. You can’t build a headphone amp like the O2 on a piece of perfboard, protoboard, etc. and expect it to work anything like what it’s capable of. Unfortunately, the only way to properly test it is with a properly routed PC board. And that takes a lot of time to design. Getting the board made is usually at least $70+ with shipping and takes at least a week or more. If you want it faster, plan on more like $200.
3-3: ITERATIONS: All that work you put into the first PC board above is usually thrown out the window rather quickly after you build the first prototype. There are often enough changes after you test it you’ll have to significantly modify the design which usually makes all those painstakingly hand routed traces unusable. Sometimes, just for the prototype, you can get out the X-acto knife, cut traces, tack wires on, hang parts off the bottom of the board, etc. to patch things together temporarily. But, ultimately, you have to spin a whole new PCB.
3-4 MECHANICAL ENGINEERING: We electrical engineers sometimes have to put our mechanical engineering hat on. Everything that’s supposed to interface with the front panel has to be mounted on the PC board the correct distance from the edge of the board. If something sticks out too far, or not far enough, it won’t work in the enclosure. You also have to worry about parts interfering with each other, hitting protrusions inside the enclosure, or being too tall. And things like 3.5mm connectors, the power jack, Alps volume pot, power switch, gain switch, battery connections, right angle LED, etc.are unlikely to be in your PCB CAD software library. So their mechanical “footprints” must be defined from scratch. And if you get even one pin or hole wrong by a fraction of a millimeter, the component may not even fit in the board. All these things can also require multiple spins of the board to get correct.
STEP 4 TESTING: The test results for the O2 are shown in the first O2 article. Before those measurements were made, however, there were many similar measurements conducted unveiling problems that needed correcting. Lots of DIY designers, and even some commercial ones, get a design to where they think it sounds good and they “blindly” release it based mainly on their ears. But that’s often a serious mistake. Just ask Schiit Audio and the guy who designed the Half Baked DAC mentioned earlier. If designers run the sort of measurements you’ll find in the first O2 article, they will discover if their design has serious problems. NuForce has been caught with their pants around their ankles multiple times because they failed to make the right measurements. And the AMB MIni3 doesn’t come close to some of its performance claims. Many don’t seem to bother and/or lack the capability to make all the right measurements. Everyone should remember you can’t hear ultrasonic oscillation, headphone damaging DC, and other things that are serious problems and/or red flags. So just because you think something sounds fine doesn’t mean it is fine. And RMAA still leaves a lot of stones unturned.
BOTTOM LINE: I’ve tried to document what goes into designing and testing something like the O2. There are many more steps involved than have been covered here. But, hopefully, I’ve at least touched on the highlights. In the bigger picture a headphone amp like the O2 is relatively simple. A headphone DAC, for example, is considerably more involved and a remote controlled pre-amp/USB DAC like is much more involved still. Unless you’re in a position to know you can somehow do better than the experts, including the semiconductor companies and published audio designers (i.e. Self, Cordell, et al.) I would suggest following their proven solutions as closely as possible.