INTRO: My AMB Mini3 headphone amp review immediately generated questions about virtual grounds. People wanted to know if they cause so many problems why use them and what’s a better option? And what about other 3 channel headphone amps? It turns out this is an interesting topic and worthy of some mythbusting.
GRAINS OF TRUTH: Not to start off on the wrong foot, but this sets the stage for what’s likely behind 3 channel amps. Audiophiles sometimes latch onto things with at least a grain of truth behind them only to apply the original truth in ways that don’t make any sense. Vibration damping is one of the more obvious examples.
PAINT SHAKERS: Please indulge me here… Let’s take a quick detour and look at vibration damping. Vacuum tubes, and certain components in very high gain circuits like phono preamps, can be “microphonic”—vibrations can affect them in audible and measurable ways. This is the grain of truth. Audiophiles responded with various elaborate methods to reduce vibration—platforms, support cones, sand filled equipment racks, and even “sonic bricks” to set on top of your equipment. If it only applied to tube gear and phono preamps, that would be one thing, but what’s good for tubes must be good for everything, right? So vibration damping is widely marketed for, and applied to, solid state pre amps, power amps, DACs, receivers, etc. This is where the grain of truth becomes myth. The alleged benefits to mainstream gear never survive blind listening tests and defy detection by even the most sensitive measurements. The fact is, you could clamp a DAC into a running paint shaker and I doubt it would sound any different. In short, the supposed “benefits” are very likely imagined. It’s mostly a giant, but surprisingly popular, waste of money. And all this money is wasted because a “solution” is being applied where there isn’t a problem to begin with. But at least vibration damping is generally harmless. Virtual grounds generally result in worse performance.
SPLIT BIPOLAR DISORDER: No it’s not a mental diagnosis for Charlie Sheen. It’s what happens when most op amp circuits are paired with a single power supply. Op amps are a primary building block of audio circuits. And they work best with two power sources or “rails”—one positive and one negative—known as a “split” or “bipolar” supply. This allows their outputs to be referenced to zero volts making them easier to use and perform better.
WHY USE VIRTUAL GROUNDS? In a headphone amp the only arguably valid reason to use a virtual ground is when you can’t use a real one. The AMB Mini3, for example, runs from a single 9 volt battery and uses a virtual ground to divide 9 volts in half forming a 4.5 volt reference. The input and output are then referenced to 4.5 volts as if it were a zero volt ground. This electronic slight-of-hand tricks the op amp driving the headphones into thinking is has dual +/- 4.5 volt power supplies. So far so good.
A FAKE GROUND BY ANY OTHER NAME… There are many names for these slight-of-hand tricks such as:
- Virtual Ground – This is the most generic term for any circuit or configuration that creates or synthesizes a ground. It could be just 2 passive resistors as shown in the diagram to the right. Here a single battery is divided by two resistors forming a midpoint that can be used as a virtual ground. This is done in many low cost, low power op amp circuits and can work reasonably well where the maximum currents are less than a few milliamps. The more current you need, the lower the resistor values have to be, and the more power is wasted as heat. It’s “good enough” for certain applications and widely used.
- Rail Splitter – This is generally an active device designed to maintain a voltage at half the power supply voltage—splitting the “rails”. Just like the resistors, it creates a “mid point” forming a virtual ground for the rest of the circuit. It’s hopefully more power efficient and/or provides a more stable, lower impedance ground than resistors. The TI TLE2426 is an example of a simple commercial rail splitter IC designed for low power circuits. It will work up to about 20 mA which is great for most low level op amp circuits but still not enough to do a respectable job in a headphone amp—although some have tried and are promoting such designs.
- Active Ground – The triangle in the second diagram can represent a buffer or amplifier such as the OPA690 op amp used in the Mini3. This allows more current capability than from a conventional rail splitter like the TLE2426. This is required for something like a headphone amplifier which require more than 20 mA.
- 3 Channel Active Ground – When the virtual ground is similar or identical to the left and right amplifier channels it’s often called a 3 channel active ground. AMB uses this description to describe several of their amplifier designs such as the M3. Other examples are the Tangent Pimeta v2 and Triad Audio Lisa III. The newest so called 3 channel design is the FiiO E11.
- Multiple-Grounds – This is where things get more confusing. The AMB Beta22 aka ß22 (B22) amplifier “recommended” configuration uses 3 channels. The B22 already has a dual split power supply and a real ground. So why create a virtual ground with a 3rd channel? Remember the audiophile vibration reduction from above applied under the wrong circumstances? This is essentially the same thing—a “solution” without a problem. The third diagram to the right shows the dual power supply with its real ground, and the virtual ground created from the split power supply rails.
WHY TWO GROUNDS? So if there’s already a real ground, why would anyone create a virtual ground? The following claim from the AMB website for their real ground B22 amp in 3 channel form is the typical justification:
“Offers improved performance by having an active ground channel amplifier for the headphone's shared "ground return" wire. The ground channel amplifier sources or sinks the return current from the transducers, which would otherwise have been dumped into signal ground or power supply ground. This shifts responsibility for the high current reactive load of the headphones from signal ground to the tightly regulated power supply rails, thus removing the primary source of signal ground contamination. The headphone transducer "sees" symmetrical output buffers with equal impedance and transfer characteristics on both sides, rather than an amplifier on one side and a capacitor bank of the power supply ground on the other. This results in lower output impedance, greater linearity and reduced stereo crosstalk.”
MYTH BUSTING: Most everything in the quote above is relatively accurate, if a bit misleading, until you get to the last (underlined) sentence where the benefits are stated. Let’s look at each of these mythical benefits in more detail:
- MYTH: LOWER OUTPUT IMPEDANCE - Using an identical copy of the amplifier as a virtual ground doubles the output impedance. In a conventional two channel amplifier the ground terminal for the headphones (or speakers) would terminate at the central ground for the amplifier via a short length of wire or ground track on a PC board. Let’s say it’s 2 inches of 18 gauge wire. That’s 0.001 ohms or 1 milliohm (mOhm). A high quality headphone amplifier might have an output impedance of 50 mOhms—50 times higher. So the output impedance isn’t lowered, but doubled. See the diagram at the top of this page. Strike One!
- MYTH: GREATER LINEARITY - This has the same problem as the output impedance above. Any amplifier has non-linearities especially compared to a 2 inch piece of wire. The virtual ground amp has to source and sink all the current from both loads and is going to generate distortion and non-linearities in doing so. The perfect amplifier does not exist. But a piece of wire is pretty close to perfect and offers much greater linearity. More on this later. Strike Two!
- MYTH: REDUCED STEREO CROSSTALK - If we start with a properly designed amplifier, a shared virtual ground will always increase the crosstalk compared to a conventional star ground. Why? The higher output impedance of the virtual ground is shared between the channels—and it turns out this degrades the crosstalk by somewhere around 20 - 40 dB. More on this later as well. Strike Three!
- MISLEADING: SYMMETRICAL IS BETTER – If you follow the current path in any device with 2 terminals, like a driver in a headphone, it doesn’t have a concept of “symmetrical”. Your flashlight doesn’t care which bulb terminal is hooked to the switch—it’s the total loop that matters. And, believe it or not, a headphone driver can’t tell if it has one side connected to ground or not. If symmetry was so important, why do so many of the best performing systems on the planet have their their drivers connected to ground? So while “symmetry” seems nice to a human, to an electron going through a headphone driver, it’s meaningless. If you don’t believe me, read up on Kirchhoff’s Laws.
3+ STRIKES AND YOU’RE… Busted. Or at least the myths are. But, if you need more evidence, please keep reading. And if you’re an engineer or serious geek, you’ll find more details in the Tech Section.
GROUND ISOLATION (added 5/15): Isolating the input ground by using a virtual ground only for the outputs does remove the load ground currents from the input ground system. But it also creates new problems. With a conventional star ground the common mode rejection of the audio amps can compensate for many imperfect things in an amplifier. But if you isolate the grounds, you remove the ability to compensate for ground impedance. Even in much larger power amps, with 100+ times greater ground currents, the ground currents are an entirely manageable problem. And, regardless, the many other side effects of using a shared virtual ground for the load cumulatively are far worse than any arguable ground problem in a properly designed conventional amp. Two grounds also have the potential to create other problems. See the Tech Section for more.
WHAT HAPPENED TO DUAL MONO? It’s interesting when audiophile beliefs directly conflict with one another. Many audiophiles believe the closer you get to a true “dual mono” signal path, the better the sound. The less that’s shared between the stereo channels, the less interaction there will be, the greater the stereo separation and purity of the sound. Unlike some audiophile mythology, this at least has some basis in fact. This is often carried to the extreme of “mono block” designs that are isolated right down to having a left and right power cord. With that in mind, have a look at the diagram. Note the amplifier in the middle, providing a very critical ground reference, is shared by the two channels. The roughly 100 mOhm output impedance is shared between the channels seriously degrading stereo separation and creating other problems. It’s the opposite of dual mono! So which audiophile belief should have higher priority? The one that has some rational basis in fact (dual mono) or the one that doesn’t (shared virtual ground channels)?
OVERWORKED AMPS: The problem is worse than just degraded separation. The shared ground amplifier has a hard job. It literally has to deal with twice as much current as the left and right amplifiers and more than twice as much power dissipation. More current means more non-linearities and distortion. And more dissipation means more heat. The OPA690 shared ground in the AMB Mini3, for example, can get seriously hot which degrades its performance and puts it in danger of failing sending potentially damaging DC to the headphones. If all three channels are the same, as they are in most of these designs, you either have to seriously over engineer the left and right channels so the same design can handle greatly increased demands of “ground duty”, or if the amps are optimized for normal headphone requirements, the ground amplifier is going to be overworked and seriously limit the performance of the left and right channels. Neither is a very attractive solution.
SHARED DISTORTION: Even when the ground amplifier is operated within its capabilities, its noise and distortion will contribute to both channels. All amplifiers have distortion, and in this case, none of it is likely to cancel out (in true balanced amps the even harmonics cancel to some degree). And, worse, a shared third channel creates entirely new kinds of inter-channel distortion that get added to both channels.
PLEASANT HARMONICS: Anyone who’s ever played around with a piano or keyboard knows some notes go well together (and form pleasant chords) and some create a much more painful effect. It turns out that also holds true for distortion products. If you play the note “A” in the 4th octave, that’s 440 hz. If you add another “A” in the 5th octave, that’s 880 hz and they go together and harmonize rather nicely because the 2nd note is exactly one octave higher than the first. The 2nd harmonic distortion product of 440 hz is, guess what, 880 hz. That’s why 2nd harmonic distortion is considered relatively benign—it blends in nicely. Odd order harmonics are less pleasant, but they’re still harmonically related to whatever is playing making it much more likely the brain will perceive them as part of the music.
NON-HARMONIC CONFLICT: In a normal two channel amp with a real ground, nearly all the distortion products—especially any audible ones—are confined to each channel and largely harmonically related to whatever is playing. But, if you share a nonlinear ground between the two channels, all that harmonic bliss goes out the window. Now you have distortion products in the right channel that are not related to what’s playing in that channel because they’re coming from the other channel which is playing something different. And visa versa. This type of distortion, created by the virtual ground, is considered far more audible than the usual harmonic variety. And the Mini3 has these distortion problems even at modest listening levels. See the Tech Section for more.
DUAL MONO REVISITED: So, just for reference, the diagram at the right shows a conventional 2 channel grounded amp. It’s much closer to being a “dual mono” design sharing only a relatively insignificant 1 mOhm of common ground impedance. This conventional design has the potential to reduce shared everything (crosstalk, distortion, etc.) by around 40 dB which likely pushes all of it safely into inaudibility.
BRIDGED AMPS: It’s worth noting a bridged amp shares the same basic concept as active ground amps. Both route all the “return” currents to the power supply rails and you can run a bridged amp from a single supply. The bridged amp has the advantage of not sharing an amplifier between the channels so it can be a true dual mono design. But despite all these supposed advantages, bridged amps are generally regarded as inferior by audiophiles. Why? Likely because they have more non-linearities, higher distortion, perform poorly with low impedance or highly reactive loads, have generally inferior clipping behavior, and a higher output impedance. They also require nearly double the circuitry so you might pay more for less performance. Bridged amps are mostly only favored for very high power applications—like rock concerts—where ultimate sound quality is not the primary goal but their power supply benefits are substantial. This is the exact opposite of a headphone amplifier where you want very high quality but only need very low power.
BATTERY LIFE (added 5/15): In a battery powered device every milliwatt can matter. Unfortunately an active ground wastes a lot of energy as heat. All the current routed through the virtual ground generates considerable “V*I” losses. In the Mini3, for example, the virtual ground can waste a several hundred milliwatts above and beyond the quiescent power. Even with just one channel operating, the virtual ground will dissipate more heat than the output amplifier. That’s not good for battery life and helps explain why a much smaller amp like the FiiO E5, with a vastly smaller battery but similar levels of output power, can have longer battery life. And with the burden of both channels it can overheat and becomes the weak link in the entire amplifier.
COMMON SENSE: Active/virtual grounds are hardly a new idea. They’ve been used in op amp circuits for decades and Crown developed one for large power amps in the 70’s. If this 3 channel active ground topology really helped headphone amp performance, why is it so rare—especially among commercial products? High end audio manufactures are always looking for ways to differentiate and market their products. And none of the top audio amp designers (to my knowledge) including Bob Cordell, Douglas Self, John Curl, Walter Jung, and others, have done any 3 channel designs of any kind. I conducted multiple searches of the massive Audio Engineering Society publication library which includes some pretty obscure stuff and came up with nothing but crickets. Could this be another “sonic brick”?
BOTTOM LINE: I’m sure there are plenty who have 3 channel headphone amps and think they sound amazing. But I’ve heard that about Sonic Bricks sitting on DACs too. The real truth is a well designed 2 channel amp will always outperform a similar 3 channel or “virtual ground” amp when a true split power supply is available. If you’re genuinely stuck with a single power supply, then you have to pick the best compromise. In some cases, that might be a sufficiently robust virtual ground. But for products where a split supply is available or readily possible, like the AMB B22, three channels just doesn’t make any sense except possibly as a band aid if the B22 has other design problems.
TECH SECTION: For the geeks, and engineers, here’s more supporting information.
SHARED DISTORTION: A 3 channel or virtual ground design generally shares the critical active ground reference between the two stereo channels. This means any distortion or non-linearities created in the virtual ground by one channel will end up affecting the other channel. Because a virtual ground has much higher output impedance and greater non-linearity than a real ground there’s going to be a lot more undesirable interaction between the channels. That’s exactly what I observed in the AMB Mini3 . Below is the spectrum from the left channel driven at 1 Khz to 1 volt (well below clipping). In this first spectrum, the left channel has the virtual ground all to itself. Note the normal distortion products only at harmonics of 1 Khz (2 Khz, 3 Khz, etc.):
DISTORTION FOREST: Without changing the left channel, the right channel is also driven to the same 1 volt level but at 300 hz. Below is the same output from the left channel but look at the new “forest” of distortion products! The largest new spike is at 300 hz and represents crosstalk. The rest are due to the complex interaction of both channels with the non-linearities of the virtual ground. For anyone wondering, the difference between the 1 Khz tone above, and the one below, is plainly audible. AMB claims the OPA690 virtual ground can handle 190 mA of current. Despite being under that current level, the OPA690 is doing a very questionable, and audibly poor, job here:
SHARED CROSSTALK: With 100 mOhms impedance for the ground channel, 1 volt output, and a 15 ohm load as above, the left channel will induce somewhere around 9.4 mV of signal onto the virtual ground. This is only –40 dB below the output signal. With a real 1 mOhm copper ground, however, the math changes to 94 uV and –80 dB. The real ground does much better at isolating the channels—about 40 dB better. Some might be wondering why the 300 hz spike above isn’t closer to –40 dB. There are multiple answers. First, I don’t know the actual output impedance of the OPA690 in the Mini3. I used 100 mOhms as a fair and typical number for the example calculations. Second, the two sine waves were out of sync with each other and alternately cancel and reinforce each other via the shared ground amp in frankly scary looking ways. Every spectrum sweep was different. And, finally, the feedback in the AD8397’s left channel may see some of the 300 hz crosstalk through the load as an error signal and be partially correcting for it (i.e. the left channel actually produces its own 300 hz signal that’s out of phase with the distortion product—another scary thought). But, regardless, the real world distortion and crosstalk result is pretty bad and plainly audible.
AN EXPERIMENT (added 5/18): The above measurements were from the Mini3 which uses a different active ground from the output amplifiers. I wondered what would happen if I directly compared a conventional ground to an active ground using the same amplifier. This would come closer to simulating the M3 and Beta22. I don’t have any 3 channel amps, but with the floating inputs on the dScope, it’s easy to use a 2 channel amp and just ground the input to one channel. The result is more like the B22 in the 4 channel “balanced” configuration as there’s no shared ground. A 3 channel design would perform much worse as shown above for the Mini3 due to the added intermodulation distortion. It’s worth noting the amp below is operating Class A/B (the B22 is Class-A) so it’s creating all those nasty power supply currents active grounds are supposed to clean up. See the diagram below under Band Aid Solution for the configurtion.
THD vs FREQUENCY: Here’s the THD vs frequency at 2 volts output into 150 ohms. The yellow trace is one channel conventionally grounded. The blue trace is connecting the same load across the left and right outputs and grounding the input to one channel. This amp uses the TI TPA6120 output buffer which is capable of very low distortion. You can see at 1 Khz the active ground configuration has nearly 3 times the distortion and at 20 hz it’s 0.0015% vs 0.009% which is about 6 times more. Even in active ground configuration this isn’t really enough distortion to worry about but it’s worth noting the active ground does indeed make the performance significantly worse:
THD vs OUTPUT: Here’s the THD at 1 Khz plotted versus output voltage into 150 ohms. Again, you can see substantially higher distortion at all output levels with the active ground configuration, but again, it’s plenty low for both versions. The Beta22 might perform very differently but the idea was to test the generic concept:
DIGGING DEEPER UNDER THE GROUNDS: I mentioned earlier a virtual ground might help performance of amplifiers with an improper ground scheme. So here’s one of many examples of what can happen if you get sloppy with ground paths or don’t understand proper grounding:
WHEN GROUND ISN’T GROUND: The “RG” resistors in the diagram above represent impedance in a single ground trace or area of “fill” as used on the Mini3’s PC board. It’s one continuous section of copper, but in reality, there’s an impedance between any two points on the copper due to resistance and parasitic inductance (I could have more accurately used “Z” instead of “R”). In this case the relatively high ground “return” currents from the headphones are going to show up as a signal across RGa and RGb as shown by the red arrows. This is a problem because the input ground connections (blue and green) are also tied across those impedances—so you end up with multiple ground signals superimposed on each other across RGb and RGa. If you have two stereo channels sharing even part of the above ground trace it’s easy to see how one channel’s output ground currents will could show up in the other channel’s input circuitry creating crosstalk. This is just one relatively simple example. There are other places, like the power supply filter capacitors, where the output signal can create ground currents that have to be carefully considered. Even decoupling cap grounds require careful treatment.
ELEGANT SOLUTION: If you understand what’s causing the problem above, the solution is obvious. The widely accepted practice of a “star ground” (versus random shapes of copper “fill” on a PCB) provides independent paths for each of the ground signals. In reality this means 3 traces (and/or wires) instead of sharing one. They all converge at the true reference ground point of the amplifier. Now each ground signal has its own private path to ground. It’s like having a house with 3 kids and each now has their own bathroom instead of having to share a single bathroom—conflict (and crosstalk) resolved:
BAND AID SOLUTION: If the above is too simple, not enough for the marketing department to work with, or you don’t know how to route grounds, you can devise a much more elaborate solution. You can add a whole amplifier to do nothing more than act as a ground. Now the input grounds of the amps driving the headphones are blissfully free of those output ground currents, but, as we’ve seen, this expensive band aid comes with some rather severe side effects. I’d find a new doctor who can treat the root cause rather than merely covering it up. Using a proper star ground is a much cleaner, cheaper, and better performing solution.
POWER SUPPLY NOISE & RIPPLE: I’ve also read third channels are justified because they disturb the power supply less. It’s another example of a grain of truth being applied wrong. A conventionally grounded Class A/B amplifier alternates its current consumption between the positive and negative power supplies. This “back and forth” can create variations on the supplies. It’s easy to produce scary looking (sometimes highly unrealistic) simulations of this effect and some have been posted in forums. With an active ground, or for that matter a bridged amplifier, the load on the power supply is more consistent creating less variation and that’s the “sell” and justification for the third channel. But first of all, the variations are proportional to the output current of the amplifier and the value of the filter capacitors in the power supply. This means with the low currents involved in a headphone amp, and large capacitors, this effect is already close to nil. Douglas Self calls it Induction Distortion and talks about how to reduce it to below the threshold of measurement just by using proper layout, grounding, etc. And this is in amplifiers with currents perhaps 100 times greater than a headphone amp. So if it can be reduced below the threshold of measurement in conventionally grounded big power amps, it’s a very safe bet it’s a non-problem in headphone amps.
CMRR: The Common Mode Rejection Ratio of an amplifier is its ability to reject (ignore) precisely the sort of power supply noise that allegedly justifies a ground channel. The CMRR of the AD8397 op amp in the Mini3 is quite good. I don’t know about the CMRR of AMB’s M3 or Beta-22 amps. But if it’s poor, that problem should be addressed rather than using a ground channel, with all its negative side effects, as a band aid.
CLASS-A (added 5/18): A true Class-A headphone amp neatly gets rid of the power supply problems completely. If the B22 really is fully Class-A as AMB claims, it’s not subject to these problems which removes this part of the argument for a 3rd channel.
POWER SUPPLY IMPEDANCE (added 5/18): As further justification for 3 channel designs, at least one designer has said “the real ground on a conventional amplifier is the midpoint between the power supply's capacitors whose internal impedance also contribute to the overall ground impedance.” He goes on to explain the Effective Series Resistance (ESR) of the power supply capacitors can be higher than the output impedance of his amplifier channels. Therefore, he argues, an active ground can outperform a real ground. But this is not true because of how NFB works and the amp’s CMRR. Any voltage developed across the capacitors due to their ESR is almost entirely compensated for by the NFB and CMRR (common mode rejection). This essentially lowers the effective impedance of the power supply caps so low as to be relatively insignificant. See the next paragraph.
NFB’s ROLE (added 5/18): The way most amplifiers lower their output impedance, including all of AMB’s designs, is with negative feedback (NFB). Generally the more NFB, the lower the output impedance. NFB constantly compares the input of the amp to the output and corrects any differences. So if we ground the input and output signals at the star ground, the NFB in even a single channel amp will make sure the output follows the input—even if that ground point is “moving around” due to ground currents, capacitor ESR, etc. Everything is relative to that single point. Any errors related to the ESR in the power supply capacitors looks like common mode noise to the amplifier and is rejected. This is the same mechanism allowing AMB’s ground channel to maintain ground even when the supply rails are “moving around”. The power supply capacitors are only referenced to the supply rails and ground. Any signal/noise/ripple across the capacitors is a common mode signal that’s rejected even in 2 channel amplifiers with a proper star ground scheme. See CMRR above.
THE BLAMELESS AMPLIFIER (added 5/18): Douglas Self has an excellent book The Audio Power Amplifier Design Handbook. Not only does he spend several pages talking about PCB layout and the finer points of proper grounding, but he also extensively digs into just about every measurable imperfection you can accuse an amplifier of right down to plated steel output jacks causing magnetic non-linearities. And he doesn’t just talk about it in theory, he builds his designs and measures them with an Audio Precision analyzer. He measures distortion, crosstalk, output impedance, noise, and more. And he ultimately came up with “the blameless class A amplifier” which, he argues, is close enough to perfect to be entirely transparent. Its distortions of any kind are so low they’re difficult to even measure. And, guess what, it has only 2 channels and uses a conventional star ground. All the supposed evils AMB and others are trying to solve with active grounds, are missing in action in Self’s designs. And if you don’t like Self, Bob Cordell has done much the same thing with much the same result. And if you don’t like either of them, there’s Bruno Putzleys and others. It’s really clear the issues the 3 channel guys are trying to solve, have been solved by others in MUCH more elegant ways.
ANOTHER ARGUMENT: Because virtual grounds are used in certain high power pro-sound amps, some claim that validates their use in headphone amps. But the main criteria for these ultra-high power amps is efficiency—not ultimate sound quality. They need to be small (for their massive power ratings), light, and not trip circuit breakers when you have big racks filled with them. As mentioned above, an active ground smooths out the load on the power supply. This makes it possible to use a smaller and lighter power supply that works more efficiently with the AC line. Bridged amps have similar benefits but they have a hard time with low impedance loads and cannot have any of their speaker terminals grounded. Back in the 70’s Crown came up with their Grounded Bridge topology which is a type of active ground. It’s a way to increase power supply efficiency while still retaining low impedance drive capability and ground referenced outputs.
WHAT’S GOOD FOR… damping tube vibration must be good for damping transistors… and what’s good for a 2000 watt pro-sound power amp must be good for a headphone amp…. Uh, No. It’s like “Wow these giant tractor tires work great out in the corn fields so I bet they’ll give my sports car lots more traction and it will handle better!” Yes I’m being sarcastic, but when you look at some of these designs and the claims being made, the analogy fits surprisingly well. It’s just that tractor tires are more intuitive than behind-the-scenes audio engineering. Just because something works well in a 2000 watt professional power amp doesn’t mean it’s needed, or even desirable, in a 0.2 watt headphone amp. Sports cars don’t plow fields.
SINGLE SUPPLY WOES: So what if you’re stuck with a single power supply? Here’s my take:
- Consider A True Dual Supply – It’s all about priorities. If sound quality and performance is the number one priority, consider doing whatever is necessary to use a proper split supply with a real ground. This might mean two batteries in a portable device or something more than a single voltage DC wall transformer for a home device. This is the cleanest, most efficient, and very likely best performing, solution. For battery devices, however, see Dual Battery Concerns below.
- Create A True Dual Supply With a Real Ground – There are many ways to create a split bipolar supply from a single supply. You can use a DC-DC converter with a bipolar output or two DC-DC converters. You can also use just one DC-DC converter, or a charge pump, and create just the negative rail from the positive source and use the “raw” supply as the positive rail. Even the little $20 FiiO E5 amp uses a charge pump to generate a negative supply in this way and it works quite well. The downside is these converters create noise and waste around 10% of the power they produce. But, with proper design, the noise should not affect performance within the audio band and the FiiO amps easily beat the Mini3 for battery life. High current active grounds are typically even more power hungry and waste even more power as heat.
- Use a “Dual Mono” Bridged Amp – Bridged (some call them “balanced”) amps offer an interesting solution when you have only a single power supply. The two amps, with no signal, are quiescent at half the power supply voltage so the load sees zero net volts. As you apply signal they both move, in opposite directions, away from the center point. The advantage over a virtual shared ground is each channel is completely independent. Plus you get roughly double the output swing and up to four times the power. And, like a virtual ground, the “return” current from the load is handled by the supply rails rather than the amp’s ground paths. Bridged amps also create less power supply disturbance. But the negatives are significant. See Bridged Amps in the first section. And, perhaps worst of all, bridged amps require a 4 wire headphone connection.
- Use Coupling Capacitors – This is often frowned on but it does have its advantages. It’s cheap, simple, efficient, noise-free, offers the best battery life, doesn’t limit current, and provides inherent DC protection for expensive headphones. A lot of very well regarded OTL tube amps use output capacitors and have their fans. The downsides are increased low frequency roll-off, phase shift and distortion. And there might be “pops” during power up/down. To provide some perspective, the distortion of a good output cap will generally be lower than the distortion created by an active ground. So, if an output cap might be offensive to audiophiles, an active ground should be even more offensive.
- Use A Virtual Ground – The Mini3’s virtual ground does an OK job when higher impedance loads are used and the current demands are minimal. But it really falls down in some areas as discussed in this article. See the next paragraph for some additional ideas on improving virtual grounds.
A BETTER VIRTUAL GROUND: Based on what I measured with the Mini3, testing another virtual ground design, and some further thought, I can offer the following ideas:
- 3X Current Capability – A shared virtual ground should have at least at least triple the maximum combined current expected from the load. The Mini3 didn’t follow this rule and it shows clearly in the measurements. AMB rates the Mini3 at 300 mW per channel into 33 ohms. That’s 135 mA per channel of peak current or 270 mA total. The TI OPA690 virtual ground, however, is only rated for 160 mA “min” and 190 mA “typical”. Based on my measurements the OPA690 is already unhappy at 160 mA of peak current and falls on its face if you try to go higher. This is likely partly because the OPA690 isn’t even designed for audio use—it’s a video/data op amp where distortion isn’t a big issue. And op amps are typically not specified for this sort of application. So I think it’s good to be conservative in specifying a virtual ground. For the Mini3 that would be 800+ mA peak current capability. Such a device would likely have higher quiescent current and further hurt battery life. It’s an ugly compromise which is yet another reason I don’t favor virtual grounds for headphone amps.
- Dual Mono Design – If your single power supply voltage offers plenty of voltage swing then dual active grounds (versus bridging) might make sense. Essentially you have 2 mono amplifiers, each with its own virtual ground. This neatly eliminates the nasty crosstalk, “double current burden”, and highly audible shared non-harmonic distortion. It does, however, increase size, cost and power consumption. This is similar to the four channel version of the AMB B22 but the B22 already has a dual supply and real ground so it’s clearly better off with just 2 channels—not 3 or 4. This doesn’t require a 4 wire connection but might be tricky with a 3 wire output due to differences in offset voltage between the virtual grounds.
- Input Ground Isolation (updated 5/15) – The Mini3 isolates the input grounds from the output ground. This might, intuitively seem like a good solution to some, but in reality it degrades the CMRR and NFB of the audio amps because the output ground is no longer referenced to the input ground. The whole purpose of NFB is to make the output look as much like the input as possible. But, by splitting the grounds, you literally insert all the non-linearities and flaws of the virtual ground into the output of the amplifier. Ground isolation can also cause cause DC offset and/or stability problems. Any phase shift between the two grounds under load could degrade stability. And at ultrasonic or RF frequencies there very likely will be some phase shift. Grounding of power supply decoupling and filtering capacitors may also become a problem—which ground do you use?. If you have to use a virtual ground, it is a much better idea to use a conventional star ground with it and not isolate the input and output grounds in the misguided attempt to avoid “ground pollution”.
DUAL BATTERY CONCERNS: The Tangent website implies dual battery designs are a dangerous choice due to DC risks: What happens if one of the batteries dies before the other or becomes disconnected? It’s likely not pretty. Here are my thoughts:
- Single Battery Designs Are Not Immune - Even a single battery headphone amp using a virtual ground may well deliver lots of DC when the battery drops sufficiently low. Op amps can only maintain their DC operating point down to a certain power supply voltage. Below that threshold they tend to do ugly things. So this problem is far from unique to dual battery designs. It may also apply to many virtual ground designs.
- Use A Protection Circuit – If you’re worried about DC there are really only two options. You can AC couple the output (using capacitors or transformers) introducing non-linearities. Or you can use a protection circuit that’s fast enough to save the headphones if there’s a problem. Such circuits are commonplace in audio power amps and should be in more headphone amps. How much did those nice cans cost?
- Use A Battery Management IC – If you’re just worried about dying batteries, a battery management IC may solve the problem and handle charging functions. Most have a low voltage shut down feature to disconnect the load when the battery drops below a certain point.
PRIORITIES: It all comes back to what I said earlier about choosing your priorities. If the best audio performance is the number one goal, it seems like bad engineering to seriously compromise the design with a virtual ground. The charge pump generated negative supply in the $20 FiiO E5, for example, works surprisingly well and arguably much better than the virtual ground in the far more expensive Mini3. It adds some ultrasonic noise, but that’s minor compared to the virtual ground problems in the MIni3. And amps with real grounds and proper symmetrical bipolar power supplies work even better.
FINAL THOUGHTS: To me, it’s obvious a conventional 2 channel amp with a true split power supply and real star ground is the best path to headphone nirvana. For single supply applications the right virtual ground used in conjunction with a single star ground, might be a reasonable compromise—not an isolated ground using a video op-amp third channel as in the Mini3. If you’re OK with a 4-wire output, that opens the door for some additional options. I may be publishing more reviews of virtual ground devices in the future. But do check out the Cmoy amp review for an example of how even a dirt cheap, dirt simple, real ground amp with a $0.50 op amp can outperform a much more elaborate 3 channel design like the Mini3.