DIY? There’s a significant group of audiophiles into Do-It-Yourself (DIY) projects. And DIY seems to be on an upswing and gaining popularity. Headphone amps, being relatively simple, are especially DIY-friendly. There’s a strong emotional element to DIY that can put a huge smile on your face. These designs are not subject to commercial constraints allowing higher-end components to be used. Some believe this gives DIY a big advantage. But, as I’ve said before, a good chef can do more with everyday ingredients than an amateur can with expensive exotic foods. Implementation is everything. So how does the Mini3 measure up? Is it a good design worthy of a DIY’ers time and money?
THE MINI3: As I see it, the Mini3 is an attempt to improve on the basic Cmoy while still minimizing the cost, size and complexity. It comes in “high performance” and “extended run time” editions which only differ in the op amps used. I tested the high performance version which is touted as driving low impedance loads with an impressive 300 mW into 33 ohms. You can buy a Mini3 fully assembled for around $180 which is rather expensive for essentially a Cmoy amp or you can buy the circuit board, parts, case, front/back panels, etc. for around $100. AMB makes several strong claims such as:
“High-end sonics from a tiny package -- the Mini³ is carefully designed for excellent performance, rivaling amplifiers many times its size and cost.”
AMB LABS (revised 5/14): There used to be a paragraph here saying many favorable things about AMB Labs “hi-fi par excellence”. But, since this review was published, and I attempted to discuss several of the erroneous specifications and claims for the Mini3 on the AMB forum, AMB has banned me from their forums. So it’s a bit more difficult to praise a website that not only makes a lot of misleading and erroneous claims but censors even their factual discussion. AMB is not doing the DIY community any favors with either of these things.
MINI HEADPHONE AMPS: Some might be familiar with the Cmoy designed by Chu Moy and various similar “mint” amplifiers built into Altoids tins. A lot of portable audio gear runs from a battery voltage around 3 volts and has a tough time driving some high impedance headphones. Amps like the Cmoy typically run from 9 or 18 volts allowing them, in theory, to do a better job with such headphones. You can find several variations on eBay but its important to understand their limitations. A Cmoy can be worse than no amp at all in many applications.
SMALL & SIMPLE: It doesn’t get much more basic: Input, Output, Volume (with on/off switch), and a DC jack for charging. It’s small and easy to toss in your backpack. You plug in your source and headphones, turn it on, and listen.
AUDIBLE HISS & NOISE: I listened to the Mini3 with some full size headphones and didn’t notice any obvious problems. It was fairly quiet even with my SuperFi 5’s and dead quiet with full size cans.
VOLUME & GAIN: The Mini3 gets plenty loud with low impedance or high efficiency headphones. But if you listen to music with a wide dynamic range on low efficiency high impedance headphones it might clip off the peaks or not play loud enough. It has about 14.5 dB of gain which is on the high side for the more efficient headphones it’s best suited for. I prefer amps with a selectable low/high gain option. As a DIY project, if you build it yourself, you can lower the gain by reducing the value of two resistors (R4L and R4R) as needed.
BATTERY LIFE: Even with a brand new fully charged battery, my tests show you can only expect about 4 – 5 hours per charge from a typical rechargeable battery. This is substantially less than most portable players. As the battery ages, that number will only get worse. I’m not sure how AMB claims 10+ hours for this version as the highest capacity NiMh batteries can only bump the runtime to about 7 hours. The 3 channel design, high performance op amps, and rail splitter, combine to require a lot of power. It’s possible the “extended run” version gets closer to its 20+ hour rating but, for reasons I explain later, I don’t recommend it. See the Tech Section for more on all of these topics.
MEASUREMENT SUMMARY (revised 5/13): The Mini3 did a few things well but failed to meet several published specifications (see the start of the tech section where I specifically tried to recreate several of AMB’s test conditions). The good news included flat frequency response and it’s fairly quiet. But, mainly due to using an “active ground” that’s shared between the channels (a three channel design) it doesn’t like low impedance loads, has poor channel separation, excessive high frequency distortion and it’s prone to a particularly objectionable kind of intermodulation distortion between the channels. See the Tech Section and my article on Virtual Grounds for more details. The following table compares my measurements to the AMB provided specs, the $99 FiiO E7, and the $20 FiiO E5:
|Measurement||Mini3 Result||AMB Spec||FiiO E7||FiiO E5|
|Frequency Response||+/- 0.1 dB Excellent||+0 -3dB||+/- 0.1 dB Excellent||+/- 0.1 dB Excellent|
|THD 1 Khz 150 Ohms||0.002% Excellent||0.001% 330 Ohms||0.003% Excellent||0.005% Excellent|
|THD 1 Khz 15 Ohms||0.017% Good||0.001% 33 Ohms||0.03% Good||0.012% Good|
|THD 20 hz 15 Ohms||0.01% Very Good||Not Specified||0.09% Good||0.6% Poor|
|THD 20 Khz 15 Ohms||0.45% Poor||Not Specified||0.06% Excellent||0.05% Excellent|
|IMD CCIF||0.043% Fair||Not Specified||0.03% Good||Not Measured|
|IMD SMPTE||0.009% Very Good||0.006%||0.008% Excellent||0.006% Excellent|
|Noise (ref 400 mV)||-94 dB Excellent||-93 dB||-91 dB Good||-86 dB Fair|
|Max Output 15 Ohms||104 mW Excellent||300 mW 33 Ohms||113 mW Excellent||108 mW Excellent|
|Max Output 150 Ohms||38 mW Fair||Not Specified||23 mW Fair||22 mW Fair|
|Output Impedance||0.9 Ohms Very Good||< 0.1 Ohms||0.13 Ohms Excellent||0.7 Ohms Very Good|
|Crosstalk 15 Ohms||40 dB Poor||88 dB||63 dB Very Good||46 dB Fair|
|Channel Balance Error||1.14 dB Fair||Not Specified||0.2 dB Excellent||Sample Problem|
DIY PRIDE: It’s easy to think something you soldered together yourself sounds amazing. I’ve been there. DIY efforts are often labors-of-love. And, because of that, the subjective impressions are even more biased than usual (see: Dishonesty of Listening). It’s really easy to fall in love with something you created.
DIY DESIGNS: Some DIY’ers tinker with their own “one-of-a-kind” designs while others prefer to let someone else do the design work. The Mini3 design I tested has been around since 2007. It’s supposedly a mature design that’s been extensively discussed on forums and suitably optimized based on the iterations that came before it. It’s well documented including fairly complete specifications on the AMB website. Yet, even after all that refinement, it still has some significant problems. If a design that’s been around for several years has these sorts of problems, what does that say about a design someone throws together in their basement? Unless they have a way to properly measure how it really performs, they really don’t know what they’re getting. RMAA, as used by AMB and most DIYers, has lots of problems and issues I discuss here. Obviously, RMAA didn’t cut it with the Mini3. Several of the published RMAA measurements are wrong by a wide margin.
DIY MEASUREMENTS: The Mini3’s performance suggests RMAA and typical basic instruments are often not enough to verify audio gear is working as intended. At the beginning of the Tech Section there are several examples where the numbers claimed by AMB are literally impossible. First of all, a good audio designer should know what range of performance is realistic. If numbers are impossible, even in theory, they should suspect something is wrong with their measurements. And, if you can’t properly verify the performance, how do you know you got it right? And it’s been well proven sighted listening can be very misleading.
DIY vs COMMERCIAL: Commercial manufactures have far more resources to design and test their products. They also generally make sure what they’re shipping actually matches the advertised performance. So it’s not surprising the $20 FiiO E5 can beat or rival the $180 (assembled) DIY Mini3 in several areas. Being mass produced in China, of course, doesn’t hurt either. It’s much the same story with the Behringer UCA202 $29 DAC. But such products are neither unique nor offer much in the sense of accomplishment—like a DIY project can. But some guy in his garage or basement using a soundcard and RMAA will have a hard time, no matter what designer parts he uses, matching experienced engineers with $100,000+ R&D labs at commercial companies. It’s like some guy trying to build a new car in his shed. Does he really think he’s going to do a better job than Mercedes or even Volkswagen?
DIY VERDICT: It’s not fair to judge all DIY designs by the Mini3. DIY offers a level of satisfaction and customization that’s not possible with commercial products. For some, these things are the main priority. DIY is also more cost effective than it used to be with more options today for PC boards, enclosures, parts from Asia, etc.
IMPROVING DIY? In my opinion, the DIY community could benefit, and save a lot of money, with more solid engineering and less audiophile inspired “tweaks” including questionable design topologies. My 3 Channel article discusses how audiophiles sometimes end up going down the wrong path leading to products that measure, and likely even sound, worse. Those looking for an off-the-shelf design as a starting point for their project should ideally have reliable published specs to help them decide if a given project will meet their needs, is worth the time, expense, etc. You can’t just return a DIY product to the store if you end up not liking it. So DIYers should have as complete and accurate of information as possible before choosing a design to build.
ALTERNATIVES (revised 5/28): Being purely objective, the $20 FiiO E5 outperforms the Mini3 in several areas (see table above). It’s much smaller, has better battery life, is happier with low impedance loads, and you can buy two of them for the $40 cost of just the Mini3’s front and back panels. The FiiO E7 at $99 is a decent USB DAC and headphone amp along with 80 hours of battery life, bass EQ, and more. An excellent DIY sites worth checking out is:
- diyAudio.com – This the best DIY forum I know of
- Small size
- Flat Frequency Response
- Doesn’t come close to several published specs
- “Ground Channel” creates more problems than it solves
- Weak overall performance into low impedance loads
- Not enough power for inefficient cans
- Poor battery life
- Poor channel separation (crosstalk)
- Prone to some stray noise pickup (poor ground design)
- Potential high frequency instability
- DC coupling/marginal design could damage headphones
DIY DECISIONS: I believe those choosing among various DIY designs deserve honest information before they invest lots of time and their own money. Several tube DIY headphone amps, for example likely offer much worse performance than the Mini3, but at least I’m not aware of any making the sorts of misleading objective claims that AMB makes for the Mini3. I strongly believe objective claims for things like power output, battery life, crosstalk, etc. should be reasonably accurate.
MINI3 vs CMOY (added 5/27): I just finished testing a dual battery classic Cmoy design based on a much cheaper op amp than the Mini3. It outperformed the Mini3, often by a significant margin, in nearly every test. See my Cmoy eBay Amp Review for more.
BOTTOM LINE: Does the Mini3 “rival amplifiers many times its size and cost” as claimed? In my opinion no. At $180 in pre-assembled form it’s not exactly cheap. And even as a DIY project if you buy the PCB, case, front/back panels, etc. you could easily get close to the price of several other options. But DIY isn’t always about being cheaper. The Mini3 stakes out an awkward middle ground. It doesn’t have enough output to be an ideal match with really power hungry cans. And it has significant distortion into low impedance loads rendering it a poor choice for many portable headphones. It might work reasonably well with some high impedance headphones if you’re OK with the limitations.
AMB’s MEASUREMENTS: The “Specifications” section of the AMB website lists an impressive set of specs and measurements. Unfortunately, a lot of them are wrong and might mislead prospective DIY’ers considering the Mini3. Here are some examples where I tried to recreate the AMB test conditions:
- CLAIMED POWER: 300 mW @ 33 ohms vs REALITY: 98 mW @ 33 ohms. AMB’s 300 mW requires 3.15 V RMS (V = SQRT(P*R) = SQRT(0.3*33) = 3.15). That’s 8.9 volts peak-to-peak. In reality, the output can only get within about 0.25 volts of each rail into 33 ohms per the AD8397 datasheet. There’s a 0.7 volt diode drop off the battery plus a 6.2 ohm output resistor which drops about another 1.6+ volts. So even with a freshly charged battery at 9 volts, you end up with only about 6.2 volts of swing at the output. That would be good for 145 mW into 33 ohms best case. But, especially with both channels working, the OPA690 virtual ground has a very hard time supporting the required current and its distortion rises rapidly. So, in reality, the Mini3 only measured 1.8 V RMS into 33 ohms or 98 mW at 1% THD. That’s only 1/3 of what AMB claims but represents 154 mA of total peak current which is very close to the 160 mA limit specified by TI for the OPA690.
- 300 mW RMS TAKE TWO: The AMB specs page says “These tests were run with the amplifier operating on battery power.” But, just to be fair, I also tried an external 15 volt DC power source. The Mini3’s 7812 regulator and 1N4001 diode limit the internal supply voltage to 11.3 volts under load regardless of the DC supply voltage connected. In theory, that’s good for about 3.2 V RMS of output. But into 33 ohms, the Mini3 still has a hard time because current, not voltage, is the problem. It hits 1% THD at 2.0 V RMS or 121 mW RMS. That’s still well under half what AMB claims. The problem here is 121 mW is 86 mA of peak current per channel or 172 mA total out of the OPA690.
- 300 mW RMS TAKE THREE (added 5/28): If you only drive one channel, and use the AC adapter, the Mini3 can in theory manage 3.2 V RMS into 33 ohms. This is 310 mW. I didn’t try to verify this, but I suspect it will hit 1% THD before it reaches 300 mW because of the relatively high peak current required. But, if you ignore the usual 1% THD limit, it might actually make 300 mW into 33 ohms with just one channel operating running into clipping on AC power.
- CLAIMED POWER: 30 mW @ 330 ohms vs REALITY: 18 mW Working backwards the claimed 30 mW is 3.15 V RMS. Where have we seen that number before? Perhaps someone at AMB only calculated the power from the battery voltage and neglected all the other losses in the circuit and never made any real measurements? The reality is 2.45 volts into 330 Ohms at 1% THD = 18 mW—about half the claimed power. This is a somewhat marginal number for many headphones.
- CLAIMED BATTERY LIFE (revised 5/10): 10+ Hours vs REALITY: 4 or 5 hours. Typical NiMh 8.4 volt batteries are 180 – 200 mAH. These will yield about 4 – 5 hours run time in typical use. There are some 300 mAH “max” or 270 mAH “typical” 8.4 volt batteries and you might get 7 or so hours with one of these. If you look at a typical battery datasheet, you’ll see the 35 mA typical consumption of the Mini3 works out to a battery life of 4 – 5 hours.
- CLAIMED THD: 0.001% @ 33 ohms vs REALITY: 0.01% This one’s a bit tricky because, like so many RMAA results, there’s no mention of what level was used for the test. But at my standard listening level of 400 mV the THD+N was 0.01% into 33 ohms. The Mini3 just doesn’t like low impedance loads. The dScope graphs for 15 ohms are shown later and show distortion is relatively high at every output level and much worse at high frequencies.
- CLAIMED STEREO CROSSTALK –88 dB vs REALITY: -46 dB: AMB quotes RMAA crosstalk into 33 ohms as 88 dB. The “shared ground channel” design of the Mini3 does horrible things to the crosstalk but here’s some very revealing math assuming 1 volt of output into AMB’s 33 ohms: AntiLog(-88/20) = 0.000004 volts (40 uV) of signal in the muted channel. Because of the isolated input and output grounds, all ground impedance in the Mini3 contributes to the crosstalk. So in the working channel the ground impedance forms a voltage divider with the 33 ohm load. To get the output in the muted channel below the required 40 uV (-88 dB) you have: 33/(1/0.000004 - 1) = 0.0013 ohms (1.3 mOhm)! This is an impossible number. Even just the output jack in the Mini3 has a 50 mOhm spec—38 times higher! Add in the copper PCB traces, ferrite bead, and most of all, the output impedance of the OPA690 ground amp, and you end up with about 150 mOhms total which gives somewhere around –46 dB. I suspect the –88 dB AMB measurement is with no load making it an entirely useless measurement. And because RMAA measures everything at once, that would also mean that all the other measurements are also with no load. The dScope graph for 15 and 150 ohms is shown later.
HIGH PERFORMANCE? Lots of claims are made for the Mini3’s performance. The High Performance Edition uses a relatively expensive ($6 – $7) Analog Devices AD8397 op amp in a modified Cmoy-like configuration. This op amp is notable for its unusually high peak current capability. AMB claims it can “drive low and high impedance headphones with authority” and has “very high slew rate, wide bandwidth, low output impedance and low distortion.”
EXTENDED RUN TIME VERSION: The longer battery life version uses the National dual LMH6643 and single LMH6642 op amps which draw less quiescent current. But they’re only rated at 75 mA of peak output current, and as you’ll see in the measurements, that just makes a serious problem much worse partly because of the 3 channel design. They’re also not designed for audio use, have relatively high distortion, higher noise, and way more bandwidth than is needed. I personally think they’re a poor choice for any audiophile application. I did not test the extended run version.
3 CHANNELS ARE BETTER? Several AMB designs use 3 channels. AMB makes all sorts of claims for the “3-channel active ground topology” of the Mini3 and some of their other designs. Here’s the impressive sounding description for the Mini3:
“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 power supply rails, thus removing the primary source of signal ground contamination. The headphone transducers "see" active amplifiers 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.”
- Greater Linearity? Nope! Active amplifiers have distortion and non-linearities. To have the headphone “see” two amplifiers instead of one increases distortion and decreases linearity with two active devices adding distortion instead of one. You essentially double the potential non-linearities.
- Reduced Stereo Crosstalk? Nope! When you share a fake ground between the two channels, you dramatically increase the stereo crosstalk as shown in the measurements and the math is shown a few paragraphs above.
- Lower Output Impedance? Nope! This would require the virtual ground channel to have negative output impedance. The reality is the virtual ground raises the output impedance.
FAKE GROUNDS (updated 5/13): The Mini3, in part, prompted me to write an entire article on virtual grounds. AMB chose a single battery design using a virtual ground and, in this case, it causes several problems—some likely audible.
UNSAFE SOUND: Some purists want as little as possible between them and their music even if it means no protection. The Mini3 is fully DC coupled and has full gain at DC with no capacitors to come between you and the music. That might sound great but it’s risky for your headphones—especially for a DIY design. If your source has DC offset or leakage (many do) the Mini3 will faithfully amplify the DC and send it onto the headphones. You can’t hear DC but even small amounts can damage headphones. And, if anything goes wrong with the Mini3, including even a bad solder connection, it’s likely your headphones could see LOTS of DC and be damaged. The OPA690 “virtual ground” is especially stressed in this design. It gets seriously hot when the amp is driven hard into low impedance headphones. If it fails, your headphones could easily be damaged. With op amps, if pretty much anything goes wrong, their outputs usually slam hard into a supply rail. And because of the virtual ground design, there’s more to go wrong with the Mini3 compared to a dual battery Cmoy or the self-contained and self-protected chip amp in the FiiO E7 and E5.
MY USUAL TESTS: It’s clear much of the Mini3’s published data at 33 and 330 ohms is, at best, optimistic. To more thoroughly evaluate the performance I made my usual measurements at 15 and 150 ohms consistent with the other headphone gear I’ve tested. For more on how I test, please see: Testing Methods
FREQUENCY RESPONSE: The Mini3 was impressively flat. The graph below shows both channels into a 150 ohm load (pink/red) and also a 15 ohm load (yellow/blue). You can see about 0.25 dB channel balance error at this volume setting (50%) and about a 0.5 dB drop into the 15 ohm load. The Mini3 also has a bit more HF roll off into a low impedance load perhaps due to the ferrite bead on the output. It’s about 1 dB down at 80 Khz. The insignificant LF roll off is likely the DC blocking in the dScope:
THD+N 150 OHMS: Here’s the wide spectrum plot into a 150 ohm load, both channels shown, at my standard headphone reference level of ~400 mV RMS. The distortion is very low:
THD 15 OHMS: Using a more challenging load, there’s about 8 times more distortion and the highly audible 3rd harmonic is rather high at –75 dB and cause for some concern. Anything above -80 dB is considered potentially audible by many experts. Because the current required is well within the limits of the op amps, I suspect the 6 ohm series output resistor in the feedback loop might be partly to blame here:
THD RESIDUAL: The residual shows the harmonic components and some minor crossover distortion. The waveform is somewhat oddly shaped likely from the combined distortion products of the OPA690 trying to maintain the virtual ground:
THD 20 KHZ: One disadvantage of a “single stage” design like the Mini3 is the hard working output stage also has to provide gain. This is sort of analogous to why the best sports cars are not front wheel drive. It’s a compromise when the driving wheels also have to handle steering duty. With the Mini3 there’s less feedback available at high frequencies to correct the output stage distortion. And, perhaps combined with other problems (like that 6 ohm resistor previously mentioned and possible borderline instability), the result is a lot of high frequency distortion. It went from a reasonable 0.016% at 1 Khz to 0.45% at 20 Khz—nearly 30 times higher:
THD 20 HZ: At 20 hz the Mini3 is much happier but the nasty 3rd harmonic is still flirting with 80 dB:
IMD CCIF: Again, the Mini3 is not very happy at high frequencies into a 15 ohm load. The huge “mountain” reaching up to almost –50 dB around the 19 & 20 Khz tones indicates potentially objectionable odd-order IMD distortion products and is almost certainly audible under some conditions. And remember this is only at about 400 mV output, it gets much worse at higher levels:
SMPTE IMD: The Mini3 does much better here with nearly all distortion products below –90 dB:
OUTPUT IMPEDANCE: From the reference level of 400 mV into 150 ohms, the Mini3 produced 404 mV into 100K and 381 mV into 15 Ohms. This works out to an output impedance of 0.9 ohms. This is acceptably low but the 6 ohm series output resistor may stress the negative feedback loop under some conditions which is likely responsible for some of the high distortion seen above.
MAXIMUM OUTPUT & THD+N vs OUTPUT LEVEL: This graph kind of tells it all. Into 15 ohms the Mini3 is constrained by distortion that quickly hits 1% at only 1.25 V RMS of output and barely dips below 0.01% at all. 1.25 volts is 118 mA of peak current per channel and both channels are driven. So the OPA690 “ground channel” is having to source and sink 236 mA of peak current but is only rated for 160 - 190 mA. The ugly result is obvious below. Into 150 ohms the Mini3 is much happier with distortion under 0.002% over nearly all the output range up to well over 2 volts. Compared to the $99 FiiO E7 the Mini3 does worse at 15 ohms and better at 150 ohms with 104 mW and 38 mW respectively:
GROUND CHANNEL BLUES: Given the poor performance into 15 ohms I wondered how many of the problems noted above were due to AMB’s “virtual ground channel”. If the OPA690 is the limiting factor, it’s going to create an even bigger problem with real stereo music. In the above tests the right and left channels are reproducing exactly the same signal. With stereo music, however, the two channels each place different demands on the OPA690 but the OPA690’s distortion will be shared between the channels. So, put simply, each channel’s distortion adds to the other. This is especially bad because some of this “shared” distortion will not be harmonically related to the music in that channel and thus far more likely to be audible.
THD 15 OHMS ONE CHANNEL ONLY: Here’s just one channel driven to 1 volt of output into 15 ohms. The distortion is a moderate 0.03%:
THD 15 OHMS BOTH CHANNELS: Here’s the exact same measurement as above, but the other channel is now also operating. Notice the distortion goes from 0.03% to 0.32%—ten times higher. This makes it obvious the problem is the shared OPA690 “ground channel”. It’s just not up to properly supporting both channels into a low impedance load—even at only 1 volt of output you have the highly audible third harmonic hitting almost -40 dB. Listening to this signal, versus the one above, the added distortion is plainly audible using the dScope’s monitor feature:
INTERCHANNEL IMD: The above points to a serious IMD problem between the stereo channels due to the shared virtual ground. To test that theory here’s one channel driven by 1 Khz and the other channel (not shown) driven by 300 hz—both to 1 volt RMS. Note the OPA690 is operating “within spec” here as the total peak current is under 160 mA. The –54 dB spike at 300 hz is crosstalk from the other channel. The relative “forest” of other spikes are the combined IMD distortion mostly due to the shared OPA690. You can’t see it in a single still image, but the “forest” was constantly changing due to the way the 300 hz and 1000 hz signals overlapped their peak current requirements from the OPA690. This is a significant problem. A residual of –53 dB is 0.22% and much of it is not harmonically related making this sort of distortion much more likely to stand out and be audible. Indeed, using the dScope monitor, the 1 Khz signal sounded plainly cleaner when the other channel wasn’t in use. The difference wasn’t subtle:
CHANNEL SEPARATION: Given the above, you might expect poor crosstalk performance. Indeed into 15 ohms it was only about 40 dB which is fairly lo-fi. Into 150 ohms it was a respectable 61 dB. By comparison, the FiiO E7 is a huge 20 dB better with 60 dB into 15 ohms:
NOISE: The Mini3 can be impressively quiet--significantly better than the FiiO E7’s analog input for example. The 60 hz hum spike is from the Mini3’s PC board picking up stray AC fields. It’s nothing to worry about at only –110 dB but it does show the Mini3 is a bit more prone to EMI problems than some designs including even the lowly FiiO E5:
CHANNEL BALANCE: With the volume control near 9 O’clock I measured the worst-case channel balance error at 1.1 dB. This is just over the audible threshold of 1 dB and probably not a big deal but the FiiO E7 does considerably better at < 0.3 dB across the entire range:
GAIN: The Mini3 has about 14.4 dB of gain (unity gain is about 11 O’clock on the volume control). The op amp itself runs (assuming accurate resistor values) at 14.9 dB but there’s some slight attenuation in the input circuit. Into 15 ohms 210 mV RMS of input was enough to produce output clipping with the volume set to maximum:
PHASE: The Mini3, as expected, does very well on this test—perhaps too well. It’s considered good practice to restrict bandwidth for a variety of reasons but the Mini3 runs “wide open” to whatever bandwidth the AD8397 can manage:
SQUARE WAVE PERFORMANCE: The blue trace below is the output of the Mini3 into Sennheiser CX300 headphones representing a real-world reactive load. The red trace is the input signal from a waveform generator. The “overshoot” on the output could be a sign of marginal high frequency stability and the “squiggle” that follows is an oscillation with a period of about 250 nS or 4 Mhz. This, combined with the dramatically higher distortion seen earlier at 20 Khz, is a good indication the Mini3 may need more work to assure stability under all conditions:
SLEW RATE: If you look at the table in the graph below you’ll see the Mini3 managed about 0.64 volts of “slew” in 60 nS between points “1” and “2” on the blue curve. That’s about 10.7 volts/uS and is nearly 20 times faster than the Mini3 needs to be. The Mini3 only needs about 0.6 V/uS to reproduce any musical signal you’re ever likely to feed it. It’s been well demonstrated by Douglas Self and countless others 0.2 V/uS slew rate per volt RMS of output is more than enough. So even into a high impedance load on AC power, the Mini3 can only manage about 3 volts RMS. Hence it only needs a slew rate of 0.6 V/uS. Anything more is meaningless and may require design trade-offs that degrade the performance in other ways. You can also see the period of the ringing is about 250 nS. Here’s the result:
- Don’t Use A Ground Channel: The OPA690 ‘fake ground’ is the biggest problem with the Mini3. It adds a lot of distortion that could be audible. It also destroys the channel separation, is the main bottle neck limiting output power into lower impedances, and seriously decreases battery life. The OPA690 is nearly a $5 part plus another $3 for the TLE2426 and related parts. The $8 could buy another NiMh battery and a real bipolar power supply. A simple trickle charger could be used from a 24 volt DC wall transformer to charge it. There would be a huge increase in performance and longer battery life. And it removes a likely point of failure that could easily destroy a pair of headphones with unwanted DC. Dual batteries would greatly increase the output swing into high impedance loads making the Mini3 much more suitable for power hungry full size cans. This change would, however, require reviewing/updating some other aspects of the design.
- Remove The 6 Ohm Output Resistor: This crude form of short circuit protection is significantly compromising the performance. It dramatically increases the feedback error signal and reduces maximum power when driving low impedance loads. I suspect it’s partly responsible for some of the high distortion levels. And, worse, with a reactive load (like headphones) the voltage drop across the resistor will be out of phase with the output of the AD8397. This creates a phase shift in the feedback loop potentially reducing stability (phase margin) and leading to possible oscillations—like the hint of 4 Mhz ringing I observed. If the AD8397 cannot survive a short circuit without the series resistor, there are other op amps that are short circuit protected and might offer better overall performance. Another option would be a more elaborate protection scheme. A final (low cost) compromise might to move the resistor outside the feedback loop—especially if you only plan to use higher impedance conventional headphones (not BA IEMs).
- Improve High Frequency Stability: There’s enough evidence the Mini3 might be marginally stable to consider some improvements or at least investigate further. The AD8397 is a fast op amp and it demands very careful implementation. The AMB Mini3 History page describes having oscillation problems during development—I’m not surprised. The Mini3 is roughly 20 times faster than it needs to be. I would add a compensation capacitor to the feedback loop creating a dominant pole to greatly improve stability. AMB runs this AD8397 “wide open” for bandwidth when there’s no arguable benefit, and only risk, in doing so. I’d also look hard at the PCB layout, power supply decoupling, etc. It’s also highly possible the isolated input and output grounds are degrading stability.
- Update the PCB Layout: I saw indications the Mini3 is more susceptible to external fields than it should be. This is almost always because of poor PCB layout and/or ground problems. There are also signs of marginal high frequency stability and this may also be a PCB layout problem (see above). Ground fills, as used on the Mini3 are cut up into “islands”. Ground currents are forced to take a much longer path around all the interruption compared to a real ground plane. So ground fills tend to work nothing like a real ground plane unless very careful attention is paid to all the ground current paths.
- Remove PCB “Ground Strips”: Enclosures should never be used to carry signal ground currents. They should be grounded only at a single point using a nut and washer for a consistent low resistance connection. The Mini3, however, has long ground strips the length of both sides of the PCB that make contact with the “rails” inside the aluminum case (see photo above right). This creates multiple problems. Anodized aluminum is an excellent insulator, and even raw aluminum quickly forms an insulating layer of aluminum oxide when exposed to air. So any connection between the case and the ground tracks that slip into the “rails” is likely to be intermittent and rather random. That’s a recipe for intermittent noise and/or having ground paths change over the life of the amp. Having ground signal currents flowing through parts of the case may have contributed to the hum/EMI problem I observed. It’s really poor design practice—especially here where it forms “loops” from one side of the board to the other through the top and bottom halves of the case.
- Add an Input Coupling Capacitor For DC Protection: A high quality input capacitor would not change the sound but it would provide much needed DC protection and help to prevent expensive headphone damage. Such capacitors have been proven transparent with measurements, blind listening tests and audio differencing.
- Add RFI Protection: It’s considered good design practice to have an RFI filter on the input stage of audio op amp devices. RF (say from a cell phone) picked up by external cables can easily be demodulated at the AD8397’s input creating dc offset, noise and/or distortion. A simple capacitor to ground in each channel and a ferrite bead would be a huge improvement. Again DC can damage headphones.
- Revise The Specs & Claims: Especially with a DIY project people expect honest information to base their project decision on. AMB should not claim an active ground channel “results in lower output impedance, greater linearity and reduced stereo crosstalk” when the opposite is true of all three. And they should make proper measurements (or find someone who can) and avoid publishing misleading specifications.
FINE PRINT: The Mini3 PCB was version 2.01 and built to the standard AMB design minus D4 which has no effect when the battery is installed. It should be consistent with those available from MisterX (Marshall Wyant), YBM Audio, or Wilmer Workshop. Except as noted, the Mini 3 was run from a freshly charged 8.4 volt NiMh battery for all the tests (consistent with AMB’s Specifications page). My instrumentation and methods are detailed here: Testing Methods
TECH SECTION SUMMARY: I wasn’t expecting perfection from a simple single op amp Cmoy design, but I was still relatively disappointed in the AMB Mini3. It didn’t come close to meeting many of its published specifications even when I tried to re-create AMB’s test conditions. And it has some potentially significant design problems. AMB claims it can “drive low and high impedance headphones with authority” but, in reality, it doesn’t much like low impedance loads even at typical levels of only 400 mV. With 80 ohm or greater headphones, it will likely work reasonably well as long as you’re OK with the short battery life, DC risk, potential stability issues, and it has enough voltage swing for your cans. But I strongly suspect the AD8397 op amp would perform better in more of a classic “Cmoy” design running from 2 batteries. That eliminates the need for the third “virtual ground” channel that causes so many performance issues with the Mini3. Dual batteries, however, create other potential issues—see my Virtual Ground article.
REAL GROUND FOLLOW UP (added 5/27): See my test of a classic 2 battery Cmoy which, despite using a much cheaper op amp, performed better than the Mini3 in nearly every test. It’s further evidence virtual grounds don’t work as advertised.
O2 DIY AMP FOLLOW UP (added 8/7): I have since designed a similar amp using 2 batteries and a conventional ground. Despite using much less expensive op amps, the O2 DIY Amp outperforms the Mini3 in every measurement. It can also suitably drive most any headphones, has longer battery life, battery protection, RF protection, and costs less. The O2 is open source hardware and the design is available to anyone.