Project Ice Block
A high performance hi-fi amp


This is a hi-fi amplifier that I built while I was studying engineering at uni. It used a current-feedback topology with an output stage made from expensive lateral MOSFETs, and gave roughly 150 watts per channel. It was a great learning project, and I still use it every day. Those MOSFETs liked to run pretty hot, so I used a fan cooling system that inspired the "Ice Block" name.


As of 2007, I've continued to study audio, and haven't found any hard evidence that anyone can hear the difference between tube and transistor hi-fi amps, provided neither is overdriven. This amp is powerful enough that it never blinks its clipping light in day-to-day use, so I maintain that it's not overdriven, and therefore generates no audible distortion, and can be considered as an ideal amplifier. I'd still love to try making a tube stereo one day, though, to hear what a non-ideal amp sounds like ;-)

In Winter 2007, I added a 4-star freezer badge, repainted the front black, and added a series capacitor to slow the AC fan down. I also finally tested the short circuit protection, since I was feeling bored and a little sadistic. It survived 10 shorts on each channel with a screwdriver while putting out full rated power into 4 ohms. Some of them made a pretty big spark, but every time, the protection circuit operated and turned the amp completely off.

Knowing what I know now, if the fancy Exicon MOSFETs had blown up during this test, I'd just have replaced them with ordinary transistors like the MJ15024 and MJ15025. Having said that, Exicon's eval boards never included any protection circuitry at all! Exicon MOSFETs are still available too.

9 years old and still in daily use :)

Original write-up

First of all, this isn't my design: It's based very closely on the "Alexander" current-feedback amplifier. The Alexander is the closest I've seen to the proverbial 'piece of wire with gain'. A high-speed current-feedback architecture allows huge amounts of negative feedback without instability. The Class-AB MOSFET source follower power stage, with loads of idle current, generates mostly low-order harmonics which are effectively cancelled by the high feedback factor, so the distortion is very low. Because the harmonics were so small and so low-order to begin with, the usual audiophile argument of harmonic order being raised by feedback just isn't an issue.


It took much tearing of the hair and gnashing of the teeth before I got the thing to work. At first it oscillated so badly that I thought I had picked up the plans for a radio transmitter by mistake. It turned out that I forgot the Zobel network. This item has a very scientific-sounding name but is really just a resistor and capacitor in series, connected between the speaker terminal and ground. Every power amp is supposed to have a Zobel network but it had been mysteriously left out of the circuit in the AD application note. Once I put it in, it worked perfectly.

I had to modify Mark Alexander's original design a bit, because I couldn't get the IGBT output transistors he used. I tried Exicon lateral MOSFETs instead, and they work a treat. To wring a little extra power out, I added extra regulated power rails (+/-66 volt) to run the driver circuit. The extra voltage means it can really thrash the MOSFETs.

I also redesigned the protection system to make it short-circuit proof. If the amp starts to clip or current limit, it very quickly mutes the audio, and if that didn't clear the fault, it turns itself off completely. In practice, the high speed muting circuit causes vile-sounding distortion, and the amp won't stand any significant degree of clipping without shutting down.

I guess I could modify it, but I think this just hammers home the point that this amp is not intended to be overdriven, and if it deviates from its piece-of-wire behaviour, it lets you know all about it.

Technical spec


Amplifier, 19-inch rack mounting, 3U high. Fan cooling. Dual mono construction, class AB, with overload/overtemp/DC offset protection. Four unregulated and five regulated power supplies. Inputs balanced on XLR, outputs on speakon connectors.

Power (measured at 240V line voltage)

Drive capability

Noise and distortion

Dynamic response

I'm not pretending that I have a complete set of instructions for building this amp. It was a complicated experimental project because I just wanted to explore a new way of building an amp. There are much easier ways of building a hi-fi amp which will work just as well. If you are new to the craft, I'd recommend you build something from a kit.

Oh, what a lovely... amp?

Outside view

Innards. This amp has two totally separate power supplies. Each channel has its own 160VA toroidal transformer and 10,000uF filter caps, supplying +/-50 volts to the output stage. Only one output stage is visible: the other one is hanging underneath, and the two heatsinks fit together to form a letterbox-shaped cooling tunnel that accepts air from the front grill. With a dissipation of around 50W per channel at idle, it sure needs it :-/

Close-up of the output stage. The two complementary pairs of Exicon MOSFETs were the most expensive and hardest to get parts of the project. There is about 250 USD worth of MOSFETs in the amp. Note the electrolytic (red) and polypropylene (yellow) caps that give extra decoupling for the +/-50V rails, and the low inductance wiring methods to try and avoid parasitics. I never found any need for an output choke, but maybe that was just because I never formally tested it with extreme capacitive loads.

The preamp compartment. The volume control is actually a 21 position, 2-pole rotary switch with a resistor network. It gives 3dB per click and channel balance of 1%. I did it this way because I had one of those switches in my junk bin looking for an application. There's plenty of room left in there for something nice like a DAC, or maybe even a tube distortion simulator ;-)


Here are the schematics (as huge GIFs)

The original Alexander design

Driver and output stage (Spot the "deliberate" mistake)

Power supply


Protection 1: the Destructo circuit

Protection 2: logic board

Deliberate mistake?

The junction of C2 and C3 is shown connected to the junction of R6 and R7. It should be connected to the output of U1 instead. Whoops


I built this amp in summer 1998. As of spring 2002 it's still working perfectly and nothing has blown out. I still haven't tried shorting the output to test the protection- and it hasn't got shorted by accident. It's about to be borrowed as the testbed for Project Boots where I will use it in bridged mode to drive my bass guitar cabinet. (edit: it never did because life got in the way of my plans for a giant bass rig)

I also read Douglas Self's Audio Power Amplifier Design book. Apparently I wasted my money on MOSFETs and current feedback. The book also includes a dual- slope protection system which works the same as the Destructo but uses half the components. Bummer!

In Winter 2002 I added a speaker relay because I finally got fed up with the massive turn-on/off thud that it produced. I had to add a couple of transistors to drive the relay. So I suppose this is Rev. B.

It's January 2005 now and the amp is still in daily use. It still hasn't been short-circuit tested.


The driver board circuitry is a slightly modified version of the 'Alexander' current-feedback design. This was invented by Mark Alexander and is owned by Analog Devices, Inc. It is reproduced here with Analog's permission.

You are allowed to build this amplifier only for your personal use. If you want to produce the amplifier commercially, you must take out a license from Analog Devices. I think.

The original Analog Devices application note is AN-211. You can find it on their website at

I used to link directly to it but they keep moving it! Their search function should find it.