Back in 2003, everyone started building "Gainclones", including me! For the uninitiated, read the introduction to Gainclones article. Meanwhile, this page records my experiments with Gainclones, and covers a lot of technical background. There are some updates pending, but I need to unearth my old notes after moving house - don't hold your breath!
Building the first prototype
As you'll notice from reading this
thread on DIYAudio.com, most people use point-to-point construction
techniques. This is fine if you are using the physically small
1000µF PSU capacitors that are popular, but I couldn't bring myself
to use such low-value components... It just so happens that I'd rescued 8
decent quality 2800µF 63V capacitors - enough for 4
monoblocks...
So I wanted to prototype with a PCB from the outset, but didn't want to go the trouble of getting a board etched at this stage. No problem - an afternoon with some graph paper, my PCB drill and a scalpel produced the boards shown here.
Physical layout is an extremely important issue. Star-earthing is recommended, along with separate bridge rectifiers for the positive and negative rails to ease the managment of charging currents. This is basic common-sense stuff, and you can see how I incorporated this in the design, arranging the power tracks to avoid charging currents finding their way onto the signal earth...
Inverting or Non-Inverting?
There is space for rather more components than would be required - this is because I wanted to try both inverting and non-inverting layouts. Many people claim that the inverting topology sounds substantially better. OK, some differences are possible as inverting mode avoids potential common-mode distortion problems in the input stage of the IC, but it's really hard to believe that any differences are going to be anything but subtle in the extreme. There is a trap here: when operating in inverting mode, it is important to reverse the loudspeaker connections to maintain the correct absolute phase - this means that the red output terminal should end up being connected to ground. Absolute phase can be audible given the right source material, and you want to make sure that any differences you might hear are caused by the amplifier configuration, and not just the change in absolute phase.
This diagram shows the differences between the two modes of operation - as you can see, this is just standard op-amp practice. The power supply connections have been omitted for clarity. Please note that if you are unclear during the following technical discussion, you might like to read through my article op-amps for beginners first.
Taking the non-inverting version first, the signal is presented to the non-inverting input via C1. The input impedance of this circuit is determined by Rin. The gain is 1+(RF/RI)
The capacitor in series with RI is required to minimise any DC offsets that might appear on the output because of input bias currents. For this to work, RF should equal Rin so that the bias currents set up equal offset voltages at the input pins (hence zero DC offset because the op-amp should respond to the difference between the input pins). Again, this is standard op-amp practice. Many people choose to omit this capacitor and live with the DC offset that results - this might be OK if the DC offset is less than 50mV or so - check the carefully!
For the inverting version, the signal is connected to the inverting input via C1 and RI. This time, RI determines the input impedance because the inverting input of the op-amp forms a "virtual earth". The gain is simply RF/RI
There is a compromise with this version - input impedance vs noise. This is determined by the choice of RI. When designing high quality audio equipment, you should make every effort to minimise the impedance 'seen' by the gain stages - a good "rule of thumb" value here is less than around 2k. So if we make RI=2k, remembering that the source-impedance should be nice and low (perhaps in the order of 50-600Ω), we should have an acceptable noise performance.
But, this would mean that the input impedance is 2k - and a lot of preamps would be unhappy with this. Also, an undesirable side-effect is that the value of C1 would have to rise in order to maintain a suitable LF -3dB point. As a compromised, I used 4k7 along with a 10µF unpolarised electrolytic. My preamp is happy to drive low-impedance loads, and is also AC-coupled, so I could possibly reduce RI and omit C1...
Note that a lot of people decide to omit R2 (and C2), but this is a bad idea. In theory R2 should equal RF to minimise the DC offset (as discussed above). But as this resistor would become a source of noise, it is traditionally bypassed with C2 so that the AC impedance is low. In practice, the DC offset caused by omitting R2 isn't high enough to worry most speakers, leading people to think that they can omit it completely. However, the bipolar input stage of the op-amp becomes vulnerable when connected straight to ground - large currents would be able to flow into the input during unusual circumstances - such as what might occur during power-up or power-down. Indeed, the datasheet makes this point. Initially I decided to try making R2 1K, and omitting C2.
Both circuits require a Zobel network across the output. I didn't use them on my early prototypes, and they were fine with my speakers and cables. But be aware that differing load conditions might make their presence mandatory, and assuming you implement them correctly they can't have any negative effects on the sound quality...
Here is a schematic of the prototype "PCB":
As you can see, there are some options here - along with choosing component values, you decide where you connect the ends of R2 and RI to. To make a non-inverting Gainlone, use the blue option - and make R2 a wire link. For an inverting Gainlone, connect R2 to ground, RI to the input point, and omit the blue components. Simple!
There are some extra components on the input - the 33k is just to ensure that C1 doesn't develop a charge across it - this is just good practice. There is also an ultrasonic filter which is normally omitted. Personally, I believe in such measures - as well as swamping the cable capacitance, it helps to stop noise from phones and the like getting onto the audio signal.
I chose the following values for the prototype:
| C1 | 10µF, 50V | Non-polarised Nichicon Muse | DC Block |
| R2 | 1k | 1% 0.25W MF | Connects +ve input to ground. |
| RI | 4.7k | 1% 0.25W MF | Sets input impedance |
| RF | 120k | 1% 0.25W MF | Sets gain to 25.5 (28dB) |
This is my first working Gainlone, built back in March 2003:
As you can see, each channel shares a transformer - this one has dual 18V secondaries, but is only 120VA. I actually had a pair of these transformers and planned to use one for each channel, but one of them buzzed mechanically in a rather worrying manner!
Unsurprisingly, low-level 100Hz charging pulses found their way onto the output when the two amplifier channels are connected together by the preamp - the earthing scheme was designed for dual-mono operation. But, this is inaudible with your ears more than 12 inches from the speakers, and disappears when two transformers are used.
The fused and switched mains inlet assembly came from an old PSU, complete with the rather convenient bracket. I have 5 of these, they'll be finding their way onto the final amplifiers... Also, the heatsinks are about the same size as I plan to use for the final versions...
As mentioned, these boards were designed on paper and cut through to the copper. The excess copper was removed using heat from a soldering iron. This works quite well, and isn't as time-consuming as you might think for such a simple project. You can see how I removed the "NC" leads from the LM3875T, and formed the remaining leads so that they could reach the copper...
Unfortunately, I reversed the paper template here, and produced a mirror-image board. This isn't a problem - you get a better view of the components, as well as the layout. Actually, I prefer this because the large capacitors get a better contact with the copper, and it should enable a tighter component layout. The rectifiers were meant to be mounted on the PCB, as you'll see shortly - this makes the extremely layout compact.
This image shows the capacitors and LM3875 power IC, held with a clamp to give better heat transfer. Also, there is no insulation washer but this means the heatsink is at -30VDC (clearly not a problem with our pine chassis). Both these features reduce the thermal resistance between the IC and heatsink, meaning that ultimately a smaller heatsink is required...
Initial listening tests
The prototype worked perfectly first time, with no hint of instability or oscillations. It powers up and down cleanly with no pops and clicks, DC offset is around 15mV and the signal-to-noise ratio is subjectively fine. And, it sounded rather good.
At the risk of sounding like What Hi-Fi, here's how things went: Compared to my Musical Fidelity A1, the Gainclone was cleaner at the top end, seemed more detailed, and was more dynamic. It had less midrange "glare" at higher levels. The imaging was excellent. Despite the extra detail, the amplifier is easy to listen to, and doesn't become tiring. The bass is clean and tight, and makes the A1 sound a bit "tubby". It clips rather less gracefully than the A1, but that's the only criticism.
At this point I quickly realised that was easily good enough for surround applications, and can even be used as a short-term upgrade for the A1 (until the "monster" monoblocs get off the drawing board). This was a genuine surprise - whenever I tried different amplifiers in the past, the A1 has always been the better amplifier - which is why I've stuck with it for so long.
Dual-mono power supplies
Encouraged by the success of the initial trials, I ordered a pair of 160VA torroids - after some careful rearrangement, here's how things looked:
I had to move the PCB's outwards to accommodate the new transformers - the pine chassis makes life really easy ;-)
The rectifiers have been relocated to their intended position on the PCB - this means that the power-supply wiring is nice and compact, which is how it should be. One thing that worries me about the original GainCard is the distance between the rectifiers which reside in the "Power-Humpty", and the supply capacitors in the GainCard chassis. As any first-year EE student should know, the 100Hz charging pulses are large in magnitude, and therefore can radiate nicely from the wiring, especially if it's not been tightly twisted.
This close-up shows how the rectifiers have been bodged onto the prototype. They are just standard International Rectifier KBPC601 types (6 amps @ 100V - Farnell part number 438-029) which happened to be in my "stores". They don't require any heatsinking - even after long periods of use, they don't feel even slightly warm. This is because of the dynamic nature of music - naturally, steady-state sinewave testing does cause them to warm up...
The transformers are from Farnell (part number 306-8894) - from the same range I used in my preamp. Despite the low price (£14+vat back in 2003), they feel like good quality devices and the mechanical noise is nice and low. Indeed, all the 30VA models I've used are totally silent, and of the two 160VA units, one is silent and the other hums just slightly. As I'll eventually buy another pair, I'll simply select the two quietest models to use for the main channels - when watching films, we tend to wind the level up, so a slightly humming transformer or two won't bother us...
These new transformers have 25V secondaries, which means higher DC supplies compared to the 120VA 18V model used initially. These transformers have a quoted regulation figure of 7%, which means the AC output should rise to around 26¾V off-load, translating to 37.8V DC, neglecting rectifier losses. But the measured voltages are now around ±38½V with no signal applied. This is higher than expected. Why?
You can blame EU harmonisation for the difference. We're supposed to be using 230V these days, but in practice none of the mains supply voltages actually changed. Rather, the tolerances were altered so that everyone magically became "in-spec". But the primary windings of transformers bought today are designed with 230V in mind. The mains here measures 237V, 3.3% too high, meaning that the off-load secondary voltage is (7%+3.3%) higher than 25V, or 27.58V. This translates to 39V DC, again ignoring rectifier loses - close to the observed value...
All of this is worth bearing in mind when designing power supplies, but what about the performance?
First, the easy objective stuff. As expected, the 100Hz residual has gone. Managing charging currents and earth loops is a difficult topic, and there's nothing like separate transformers for making life easy!
The higher power supply rails obviously bring more power - I didn't measure it, but it should be around 50W per channel. 50W is 20VRMS, or 28V peak, so even allowing for quite large drops in power supply voltages under load, this is quite plausible. Before, we had about 20-25W, so while this is only a 3dB increase, it's definitely noticeable.
The sound is surprisingly more dynamic. With my inefficient ATC's, you normally hear a low-powered amplifier clipping well before the sound gets uncomfortably loud, but this is less true now - the amplifier plays much more loudly than the raw numbers would appear to suggest. It is much less constrained, and can generate sound levels well in excess of what I would want to listen at. Perhaps the monster monobloc plan will be abandoned after all?
There seemed to be an improvement in the stereo imaging. Perhaps I'm falling foul of some sort of "experimenter-expectancy" effect here, but I was fairly convinced of this. Voices are much more sharply placed in the centre of the soundstage, and don't move about or lose focus as the music varies in level or complexity. This is probably the nearest I've got to hearing a truly convincing 3D sound-stage in my system.
So, this little experiment usefully demonstrated the importance of the power supply. OK, this is hopefully something that every engineer knows about already, but it's nice to demonstrate it now and again...
Alternative rectifiers
Inspired by the results of the transformer upgrade, I decided to perform some more tests with the power supply. At the time I was reading through some listening tests of rectifier diodes on diyAudio.com, and found the discussion highly amusing - apparently the MUR860 ultra-fast recovery diode is the best sounding device!
But upon investigation I found that it wasn't difficult or expensive to try this for myself. The recommended diodes seemed quite expensive at just under a pound each (Farnell part number 931-020), but if you buy 25 or more, the price comes down to 45 pence each. It's surprising how quickly they add up - you need 8 per channel, and I'm planning to build 4 monoblocks - that's 32 diodes, or £14! But when I looked up the details of the standard bridge rectifiers that I'd used above, I found that the IR KBPC601 is £2.10+vat each, working out to £16.80 for a set of 4 monoblocks. So the MUR860 diodes aren't as expensive as I thought..
So I figured that they can't hurt the sound quality, and as I already have most of the other components, a little extravagance here won't matter. At the very least, it will be an interesting experiment that might teach me something...
Using a scrap of Veroboard, I built a pair of rectifiers, ready for mounting onto the existing prototype PCBs. I haven't cut the legs short as I'm hoping to lay them flat on the final version of the PCB.
The Veroboard is simply attached to the PCBs with solder. This is perfectly adequate for testing, but I wouldn't regard this as good enough for long-term use. Lots of solder was used, and the unused tracks are soldered to the board for maximum strength.
Was there any change in the sound quality?
This was a tricky one. Audio memory is notoriously short, and it took me over an hour to fit these rectifiers. And, it's easy to hear differences when you are expecting them. But, I didn't know what to expect - I found to hard to image what differences the diodes could make. In theory, the fast turn-off characteristics of these new diodes should make the power supply cleaner - reducing the energy present in any spikes that occur at 100Hz intervals - but with my high-quality low ESR smoothing caps, I couldn't see any trace of these to begin with...
But, I thought I heard a subtle difference; there seemed to be more top-end detail and "air" around voices and percussion instruments. This subtle increase in low-level detail helped to make the stereo image even more convincing because subtle acoustic cues are more clearly heard. The sound seemed fractionally more "focused", but still not harsh or tiring.
So, not a big difference, but no obvious negatives from using these diodes. I'd like to conduct a blind AB test to be surer about the differences, because in the years since doing this, I've learnt a lot about the hearing system and how untrustworthy it is - especially when it comes to audio memory and "experimenter expectency", so treat my impressions with caution!
Conclusion
This was really great fun - and convinced me that modern power amplifier ICs are capable of brilliant results. I've since built a number of amplifiers that use them, and strongly recommend them as an alternative to conventional discrete designs.
There are a few more experiments I'd like to do, as these days I have access to much better audio test gear, but life is just too busy at the moment for this sort of thing. One day things will be calmer, hopefully...