While surfing the web a year ago I stumbled upon the weirdest of DIY audio cults - people sculpting flashy amplifiers, to which they ascribed magical levels of sonic performance, around a 'common garden' IC amplifier chip - the type you might find in a car stereo, portable stereo, or TV. Jumping further into the paranormal, the concept is supposedly based on some mythical "47 Laboratory Gaincard", with mysterious construction features like ultra-short signal paths and tiny power supply capacitors. These beasts go by the strange name of "Gainclones".
But this made me think. Monolithic (or single-chip) amplifiers as a rule are scoffed at by audiophiles the world over, being considered "cheap and nasty scum of solid state", so the question has to be asked - "can anyone hear the difference at all"? By this I refer to the whole untidy brawl between the audio objectivists and subjectivists, fuelled by (apparently) unmeasurable qualities of things like power cables having (apparently) obvious audible effects.
Amplifiers with vanishingly low THD specs are quite common these days, so why on earth would an IC amp with a 0.1% THD spec be treated as such a sonic holy grail? This is kind of good, but by no means the best. To put it in perspective, it is maybe 100 or even 1000 times worse than some high end amplifiers, and we're not talking about the more graceful valve (tube) distortion here, but nasty-as-it-gets class-B "transistor distortion". Note that many people still do consider 0.1% distortion to be beyond the limits of human hearing, double blind trials have failed to show it is detectable, so many (cheap) audio circuits are still designed using 0.1% as a benchmark of inaudibility - on the assumption that going any better is merely pandering to placebo effect (or plundering the bounty of placebo effect, depending on your point of view).
Over the past few days however, while developing a chip amp, I stumbled over some surprising results that have cleared up a lot of this confusion. Read on to find out.
A Gainclone isn't a mythical beast, and neither is the "47 Laboratory Gaincard" it is modelled on:
The 47 Laboratory Gaincard is 'just' a model of amplifier manufactured by 47 Lab, a Japanese audiophile brand, with the usual quirky and minimalist approach. By all accounts it is a very good little amplifier, and while I am in no position to judge (having never tried one), the technical basis looks sound.
A Gainclone, surprise surprise, is a clone of the Gaincard, so it usually has these features:
In fact, Gainclone has come to mean "any minimalist chip amp intended for high quality sound", and their popularity has continued to grow since I first stumbled onto their existence at the beginning of 2003.
So that's that mystery solved. They're not that special. People (like me) have been building chip amps for years with exactly these same features.
Hang on. This is exactly the sort of misguided reasoning which is responsible for all the confusion. Firstly, this would be assuming all amp chips are designed to that magic 0.1% distortion spec mentioned above. Secondly, 0.1% THD definitely is audible for the right signal. (Perhaps THD should stand for "transistor" harmonic distortion rather than total harmonic distortion?)
To cut a short story long: Back in my university days I bought an old Discman over the internet (yes they had the internet back then), along with Stereophile Test CD 2. I tested this on an HP dynamic signal analyser, one of the tests was to verify the 0.3% THD claim for track 24. This track has a 500Hz tone distorted through a transistor amplifier, which is very audible - the result was indeed something like 0.31%. (I actually had remembered the figure as 0.1%. So much for memory.)
Fast forward to the present, and I'm testing a chip for that amp development I mentioned above. This particular chip (which I won’t name) seemed to have good enough distortion specs (0.05% from 50mW to 12W at 1kHz), and be ideal for my application. It was fine without a load (typical for class-AB amplifiers), but on load I was having to work hard to get distortion at low signal levels down to the noise floor of my analyser (about 0.001% - something I have come to expect from a modern high quality amplifier - irrespective of whether this level of distortion is audible or not). (More notes below on this below) So I hooked up a small speaker, hoping the signal would be too quiet for the distortion to matter, but even at a few milliwatts I could hear the distortion components I was working with - these were in the range of 0.1% to 0.01%. After hours of experimenting with distortion-reducing ideas I managed to get the chip working acceptably for the application (distortion measuring below 0.006% was inaudible for the vast majority of situations).
The Gainclone concept I had stumbled across a year earlier had left me confused, but this latest experience turned it into a grating inconsistency: Here I have the best amp chip I could find (to work in the application), measuring cleaner than many chip amps, yet it clearly remains sonically impure. So I had the brilliant idea (it was probably 4am, any idea seems brilliant then) to test my other "Gainclones" while I had the distortion setup going...
Of course these aren't actual Gainclones as I built them before I knew of the concept, but they use much same techniques - perhaps for different reasons:
The first one is an LM1875 based amplifier I built for our main MLS tester. I never expected much out of this chip, with a good THD+N spec of 0.015% but rapidly rising THD at lower output levels, typical of class-B transistor amplifiers. On specs alone, I'd put it in the same category as the chip I was using above. In actual fact I couldn't face using it in an audio application, and ended up putting it in a parts drawer to be used as servo motor driver or something one day! This circuit is made with a combination of veroboard and point to point wiring, and is very compact. Its main function is to play an MLS signal (essentially a square wave), but on occasions I have routed music through it and been surprised at how good it sounds.
The second amplifier is more conventional (I actually have a few of these). It is built around a DSE K-5612 50W+50W amplifier kit based on LM3886 chips (February 1995 issue of Silicon Chip magazine). It gets used for testing and sometimes demonstrating our speakers. I went for these kits because they were very easy to build and modify, and had one of the best amp chips available at the time with an acceptable THD spec of 0.03%, and a minimum THD of 0.002% hinting at better performance potential - plus indications on the web that they sounded good.
(There's more info on these amplifiers in the detailed tests section below.)
Testing the LM1875 Gainclone was particularly easy, being part of the LAUD system used to measure distortion. All I needed to do was hook an 8 ohm resistor and a probe up to the amp output terminals, and do a low-level 1kHz single tone FFT.
I was stunned (kinda was already anyway - being 4am): The 3rd harmonic was always better than 90dB down, typically 95dB. The 5th was always better than 100dB down, typically -105dB = 0.0006%! (the residual of the raw LAUD analyser). The 2nd harmonic was the one that led to the THD spec of 0.015%, but that's ok. So:
The LM3886 amp was much the same, but without the even harmonics (2nd and 4th), leading to the lower THD spec, while retaining much the same potential for audio performance.
Maybe the result doesn't tell the whole story, because these chips are set up as big power opamps with huge amounts of negative feedback and may behave differently on transients or high frequencies - but I'm thinking that it does tell the whole story. There's certainly no denying the linearity of these "Gainclone" chips.
So that's the Gainclone issue resolved. In case you were hanging out for an executive summary, here's one:
There you have it. There's no magic - not in my opinion anyway. Short feedback paths, small filter capacitors and the like are interesting features but really are just good power opamp circuit design (as I'll expand on later on). I know there are other brands of chip which apparently work well, one day I might test some.
That's the original question answered, but if you're interested in more details, tests and amplifier building hints, then read on...
A day later, I got out the big guns, and fired up a more serious notch-filter based distortion rig which has a distortion residual for the 5th harmonic of -126dB (0.00005%).
Summary of test conditions
Circuit details for devices under test (also see description of these above)
Some notes about my distortion testing philosophy / approach
Results summary
Notes
Test conclusion
The harmonic distortion that matters never exceeds 0.001%, which equates to a power level ten billion times lower than the fundamental (0.001%^2 = 1E-10). This is better than the performance of most CD players, including many high end models.
Chip amps have always had a good following. They may be seen as the "scum of transistor tyranny" by some, and certainly do have their limitations compared to many discrete designs, but most designers appreciate the advantages that chip amps can bring:
Remember the people at National are 'just doing their job'. They are not part-time hobbyist designers, working with dubious quality information off the web, limited experience, limited resources, and making subjective judgements about everything. They don’t have to arrive at good solutions by accident.
They have (or at least can have) full understanding of and control over how every part on the chip behaves. They have the resources to simulate everything, run prototypes, and have the responsibly to apply an objective and thorough approach to everything (especially these days). They either design or have the option to alter the design of every component they use. Project briefs like "design a superb quality high power monolithic audio amplifier for our new range" are bound to attract audio enthusiasts, creating an environment (or design team sub-culture) that is naturally motivated, sonically competent, with a healthy balance of subjective input and innovation. I've seen indications of the same thing at Tripath, Philips and Analog Devices (from their documentation, or talking to their designers), and I'm sure it occurs elsewhere, so this is not a National sales pitch. When it works, it really works, and I hope the designers at National are pleased (if slightly amused) to see their chips gain a cult following.
Disclaimer: I have never built a "Gainclone" as such, because I thought they were bizarre until I measured the ones I had already built (that will only make sense after reading the text above). These ideas are pulled from my general experience, or experience with other chip amps and may be a little bit - what's the word - "empirical". Enjoy.
Filter caps: Get supply filter caps as close to the amp chip as possible, but keep them small because you don't want ultra-high current pulses going into them if they're right by the chip. If you want to use bigger caps, put some resistance and/or inductance in series with the transformer, or use a smaller transformer. (If you had an ultra low signal level opamp on an input board in a piece of instrumentation, you wouldn't run a dirty supply to a thundering great big cap hovering over the input pins, so why do it in an audio amplifier?). Remember that highly nonlinear audio power currents will run all the way back to any caps you have in parallel, while pulses of current will run from the power transformer all the way to the cap closest to the chip (more sophisticated solutions then become obvious). But with the PSRR of the Natsemi amp chips, why would you go to all this effort when a bit of voltage ripple isn't going to hurt? Small 1000uF caps close to the chip are sounding better all the time.
Bypass caps: Only play with these if you know what you're doing (or have time to trial properly). Parasitic inductance and low ESR can cause resonances.
Rectifier: In my opinion not a lot beats big old bridge rectifiers. They're slow, so they don't generate a lot of noise to start with, and can (should) always be bypassed with fairly big capacitors and/or snubbers. I personally think fast recovery diodes here are pointless. Use soft recovery types if you can. If you get interference on your TV (as with the LM3886 amp described here actually - must fix it), you know something is wrong!
Transformer: If everything else is right, it shouldn't really matter. Just make sure it is not susceptible to mains-borne interference, doesn’t howl away audibly, and doesn't radiate a magnetic field strong enough to melt metal casework next to it. The choice is well documented.
Grounding and layout: This is "everything". If you don't understand, do what the datasheet layout notes say, and remember that power supply currents are highly nonlinear and all wires are resistors. The effects usually go in this order:
Negative feedback: These chip amps are terrible for oscillation. They're fast, so they're going to oscillate if used improperly. Keep the negative feedback input pin node as short as possible so it doesn't capacitively couple to anything else. Keep the other side of the resistor close to the chip output so it doesn't pick up inductance of the output leads (won't hurt the speaker signal, but will hurt the negative feedback). If you look at the 47 Lab pages, you'll see a description of exactly the same battle with almost no mention of fringe concepts such as "time smear".
More stability stuff: That’s not enough, they'll still oscillate if someone on the other side of the world starts thinking about it (the more Gaincloners there are, the bigger this problem becomes). Use the Zobel networks as prescribed, and the output filter thing. Put a tiny cap across the inputs of the chip if you absolutely have to. Bypass the supplies better. Don't run the amps at low gain. Even that's not enough and I must admit to not really knowing what's going on. (I didn't know what oscillation was until I tried building a chip amp from scratch, and I've designed a few amps in my time.) If your amp is hissing, getting hot, or making crackling noises on transients (more so than clipping), it is probably oscillating. Some amps actually sound better like this and 400kHz (or 22 MHz I think I measured at one point) won't blow up your tweeters, but it's unlikely to be stable.
Miniaturisation: In general it's nowhere near as important electrically as it is mentally. I miniaturise things mainly because I think it is cool, or because it uses less materials, or because clients want it. It does reduce parasitics, including nonlinear parasitics (such as dielectric absorption, microphonics) and EMC problems, but the benefits are almost always secondary to good grounding and layout as described above. If you build circuits right, size doesn't matter until you get up to UHF frequencies. In short - it doesn't hurt to miniaturise your Gainclone, but don't expect it to make much difference.
No mute: Looking at the circuit diagrams, the mute works by disabling current from the input stage. Extra parts on silicon, or unwanted signals getting in could be a problem. Really, I wouldn't worry, but if you feel the need...
Inverted mode: Use it if you want, well documented reason (common mode rejection). Again, I wouldn't worry because the gain is much higher than unity so there is little common mode swing at the inputs.
Omission of + input ground resistor (inverted mode only): Yes, only use one if needed for protection. It is a relic from days when opamps had high / unpredictable bias currents. All it does is increase noise (not that it's a problem at line levels).
Absolute polarity: Absolute polarity is very audible with the right signals, so it is important to get the speaker wires the right way around. Your woofers should move out when you apply positive voltage to the amp's input (and no, this has nothing to do with dogs for those of you chuckling along in the background).
DC offset voltage: Personally I'd leave it alone. A bit of bias current shouldn't hurt.
Class-A bias: This is something I haven't seen applied to Gainclones yet. You put a resistor (or choke) from one of the supplies to the opamp output. This may or may not improve things and generally needs an ever-expanding amount of work (including listening tests) to get right. It makes little difference to chips like the LM3886.
Temperature effects: Another approach I haven't seen applied to Gainclones yet. Transistor (BJT) gain goes up as the chip gets hotter, therefore distortion can reduce if they are kept hot. However I've never managed to find any usable effect in a chip amp.
Heatsink: Not as critical as most people make out, because music doesn't have anywhere near the average power requirement of a continuous test tone. With discrete amps you do need a good safety margin, but these chips are very good at protecting themselves.
Soldering: A note for the inexperienced - with a controlled temperature soldering iron (which is something you should try to get) there should be no need to worry about soldering heat destroying the chip. The melting point of lead based solder is not much higher than the rated operating temperature of the chip. Get plenty of heat in there to ensure you make a good joint. Capacitors (in nearly all their forms) don't like to be melted, and neither do thermal fuses in case you haven't already found this out the hard way.
Speakers: Just a note that high impedance, inefficient speakers can clean up "transistor" distortion very well. And a note about speaker efficiency - you don't need high efficiency speakers as a rule with a 50W amp. It's only 3dB less than the "industry standard" 100W. You don't necessarily need high efficiency speakers with a 5W amp. Try it and see. And don't worry about blowing up your tweeters with an "underpowered" amp, this is another myth (it's the volume control that blows tweeters, not power amp clipping).
Output DC protection capacitors: Not that anyone does it any more, but this is another relic from the past, when amps used to blow up and take out the speakers. Use fuses if you're worried, 5A fast blow protects the LM3886 (or rather the speakers) nicely. I'm always blowing fuses for silly reasons (these are test amps, remember), and I've never had one blow in use.
Chip selection: Don't trust the datasheet specs! They are correct, but not necessarily relevant or useful for comparison. I have found out the hard way!
Disclaimer: I'd prefer to avoid this section because it's just asking for a fight, but there are a growing number of people who are shocked when they find out that some common audio design techniques have little or no theoretical basis, and no objective evidence to back them up. They feel ripped off, lied to, by the audio industry - and either turn away from it altogether, or become rampant objectivists who prefer to spout their theoretical beliefs rather than risk continuing a search for the truth. I'm not saying that the concepts here are irrelevant - for example it was only a few decades ago when the "1% distortion is generally inaudible" rule of thumb held true. I'm just saying "open your eyes before you open your wallet".
Wire: Definitely not an argument I want to get into here, but you should know that the concept of sonic differences between things like copper and silver wire is totally foreign to electronics theory (which is the basis of amplifier design). I'm not saying there is no difference, just that there is a massive amount of hype around which has never been justified by measurement (either electrical, or in objective listening trials). I do agree that using silver wire is very cool, and if it were to impart a sound of its own I should hope it would appear brighter, more revealing, and a bit less "warm" than copper.
"Stored energy" and "speed" of filter capacitors: The concepts, as usually presented, are actually quite foreign to the field of electronics. They seem to have parallels with the actual theoretical concepts of "stored energy" and "speed", but not enough to justify the strange theories and effects attributed to them. The explanations seem to have much more to do with the perception of the sound than any technical reason/s behind it.
Time smear: This is a meaningful defect for transmission systems, storage systems, coding (such as MP3) and even speakers, but it has no real meaning for amplifiers, which rarely have any real means of storing (delaying) information to the extent needed to justify blanket application of the word "smear". The closest parallel with accepted theory seems to be "group delay" or "phase shift", which is easily measurable. Again, the audible effects may be perfectly real, but the explanation is probably barking up the wrong tree.
Mechanical vibrations: Another contentious subject - is an amplifier's performance affected by acoustic vibrations? Undoubtedly yes, and severe problems with microphonics will be noticeable, but vibration treatments and associated theories frequently drift off into a dreamland unsupported by any actual experience. Another word for this is "guesswork" - so be careful. A lot of mechanical treatments seem to have less to do with vibration and more to do with making something look cool. That can't hurt.
Resistors and capacitors: Yet another contentious subject. It is well known that these parts are imperfect (V does not equal I * R etc), but most resistors are near ideal for audio and this often has very little to do with how expensive they are. Imperfections in capacitors are more pronounced, but very few people know that most modern electrolytic capacitors perform as well or better in distortion tests than polyester for example. It's an area overflowing with hype, assumption, conjecture, emotion and belief. But that can be fun.
Non-ferrous construction: From wiring to resistors to casework, this has a basis in reality, because magnetic effects in steel are nonlinear and eddy current losses are therefore also nonlinear. However, it also has a basis in "fashion". DIY designers would benefit if it was possible to separate the two.
Burn in: This also has a basis in reality - dielectrics of capacitors in particular can take a while for charges to "bed down", and impurities (contamination) and stresses in semiconductors may be relieved after a period of use, leading to lower noise and possibly reduced distortion. But when people start talking about burning in of wires, bang goes any sense of technical credibility as seen by the vast majority of engineering professionals - along with the likelihood of you returning a product that may have had no real effect in the first place.
In closing, the effects described above are rarely measurable in any clear sense, so the design process often resorts to guesswork, using rough subjective evaluation for feedback. This is better than doing nothing I guess, and represents the application of a certain level of "art", "innovation" and "attention to detail" which isn't a bad thing. But you should remember that there is a huge difference between clearly measurable harmonic distortion in an amplifier chip, and conjecture over the sound of a short piece of silver versus copper wire.
My intention isn't to provide a comprehensive set of links here, just some pointers to get you started...
Q: Can you tell me more about Gainclones or help me out with my design? A: No, sorry. I can't help you with your Gainclone design, because I am not a Gainclone guru (not in the cloning the Gaincard sense). See our services page if you wish to inquire about my contract design services (audio design is my job, so I can't be doing it for free).
Q: Can I have your Gainclone schematics and assembly instructions? A: See answer above. There are plenty of designs on the web, kits are available, and National has always been good with their application notes. All this stuff is documented from a practical point of view and I suggest this is your best bet. Check out our products page for our range of audio products, some of which will be of interest to gaincloners.
Q: Tell me more about your distortion test setup so I can criticise it, or replicate it. A: No. If you haven't got the time, tools and knowledge to replicate these measurements yourself, don't criticise mine. If you have a genuine question, specific suggestion, or are convinced I am in error, please let me know.
Q: Can I have the raw test results? A: Not at this stage sorry. I want to hang onto them because they contain information which might be commercially sensitive to our company (which is in the business of designing amplifiers). If you want to copy the graphs, please link to this page or at least credit where they came from.
Q: You use a lot of polysyllabic words. What do they mean? A: Check out our glossary (when it arrives).