Two Channel Subwoofer Amplifier

March 2018 update

This amp is now in service attached to the four 8 Ohm woofers (two 4 Ohm loads) of the actively crossed Jamo Oriels.  Incredible and effortless bass performance that easily rivals that of the C-J solid state “monster” (or was that “monstrosity”?) amp previously attached to them.  Unlike that vibrating bucket of incompetence (“engineered” for profit and marketed to dupe), this basic DIY amp produces no audible vibrations (because it has a decent transformer), has speaker protection, a soft starter, a safety ground loop breaker, runs only about 3°C above room temperature all day long and can be used with absolute confidence that it will never harm the speakers.  And it cost magnitudes LESS!  Below clipping (where no hi-fi amp should ever be pushed by music) it is sonically indistinguishable in bass performance from any properly engineered brand name amplifier at any price.

  • It was emphasised on the ESP site that heatsinking for this amp needed to be OTT and fan-cooling was recommended!  With the chassis shown, the heat sinks run at only ≈ 3°C above ambient all day long with music playing into 4Ω loads without a fan.  The sister 4-channel amp (50mV bias) runs at about 5°C above ambient.



In the face of so-called “ultimate fidelity amplifiers” and other unnecessary things like “super-regulation” for audio, this simple Class-B1 subwoofer amplifier is all about output power with better than adequate distortion levels still masked by those of typical subwoofer drivers themselves.  It’s a basic dual channel power amp for a pair of 4Ω subs, with outputs for up to two pairs of 8Ω subs.  With ±77V rails2 it could deliver peaks of over 600W into each of a pair of 4Ω loads.  Continuous power limits into 8Ω would be about 290W, and into 4Ω about 520W per channel.   It has a 1kVA toroidal transformer, four 18,000μF/100V capacitors in π filter configurations and a pair of ESP P68s with ten MJL4281A/MJL4302A driver/output devices per channel on 2.5Kg extruded heatsinks (about 0.4°C/W).  It has an NTC thermistor soft-starter and a speaker protection module, and is basically an animal that you certainly would not connect to open baffle subwoofers3 or tweeters!  It’s ideally suited to being fed a heavily LF-boosted/EQed signal for sealed subwoofers.

  1. It’s not really Class-B.  The pre-driver stage runs in Class-A and the first pair of output devices operate as a driver stage in Class-AB up to a fair few Volts output before the remaining 8 output devices dump current to the load and operate in Class-B.  Apparently it’s similar to a Crown DC-300A, but is more powerful.
  2. That would be if plugged straight into my house’s 250V mains, but I’m not trying my luck!  By virtue of the tight PCB layout, some of the components are limited by the available space and therefore have only marginally sufficient SOA to handle this 10% over-voltage, but if run on my usual mains bucking arrangement (to around 230V to match the nominal primary winding voltage rating of the transformer), the rails would be reduced to around the recommended maximum of ±70V for some peace of mind and performance matching the specs as published on the ESP site.
  3. That would be an oxymoron. 😀 And anyway this amp would just as soon “pole” them as look at them.


I needed a distraction from a big lawsuit that I initiated and won against someone (or something) and apart from stuffing a couple more Hypex modules into a box, (which would not be very challenging and impossible to service) I hadn’t previously built a really powerful amplifier.  This one would be something special (and serviceable) and certainly consume a little more effort and time.

Ultimately the amp will probably replace the (not so great) C-J MF2500 that’s presently connected to the 4 subwoofer drivers of the actively crossed Jamo Oriels.

Update July 2017:  New “sister” 4-channel P3A to replace the stupid valve amps which serve the mids and tweeters of the Oriels is now here.


It’s the usual P39, but with three series-connected 10Ω NTC thermistors instead of parallel fixed value resistors.  There’s no particular reason for this, but it’s something different:

Soft-start P39

And to increase the timing to about 2.5 seconds, a 220μF capacitor is added underneath and in parallel with a top-side 47μF one (the standard circuit wanted just 10μF for 100ms timing which I found to be insufficient for large toroidals):

Parallel timing cap under

The idea was to allow sufficient time for the thermistors to heat up before the second relay shorts them out – hopefully they’d be pretty much “shorted” by then anyway.  But actually come to think of it, that was wrong.  It would be correct for Class-A amplifiers that draw high quiescent current right from start-up, but the inrush peak here would be pretty short-lived and attributed almost exclusively to the toroidal and filter caps.

That module gets 9Vac from the auxiliary transformer, however there are two thermal switches – one on each of the big heatsinks in series with the supply.  If the amp is driven too hard over a continuous period, one of the switches will open and the soft starter with release the big power tranny.

Here is a little mounting trick for the auxiliary tranny:

tranny bolt through foot

A bolt through one of the chassis feet also goes through a spare rubber foot and the tranny to lift it up off some vent slots in the chassis floor. 😀

Speaker protection module

This is the usual P33 with component substitutions so it works with low voltages taken from the soft starter (12Vdc) and its auxilliary transformer (9Vac for the loss-of-AC detection).  It has been slowed down to a turn-on/unmute timing of about 4 to 6 seconds so the soft starter beats it.  It also has a Zener diode clamp (across the D9 position) to accelerate relay release should a fault be detected:

Speaker Protection

If either of the thermal switches goes open, AC to the soft starter is lost, so the soft start relays release the primary of the main power transformer.  Simultaneously the loss of AC function of the speaker protection module quickly releases the speaker mute relays placing a short across the speakers.  Due to relay coil loading, the low voltage DC supply collapses long before the main PSU filter caps can drain.

P68 Modules

ESP P68 + extra half board

These are kinda long and thin, but they’re pretty basic and I filled them up with good bits and went to a bit of extra trouble with the driver heatsinks – using copper instead of aluminium.  You can see where the extra 4 output devices are provided by a “half board” connected with solid copper wire links through Teflon tubes across the fuse clip locations.  Some pass under the 12A fast blow fuses (not yet installed in that photo).

I didn’t like the suggested aluminium driver heatsink arrangement with M3 machine screws strapped by wire to it and soldered-over.  Soldering to aluminium requires specialist solder and the whole idea looked a bit dodgy to me.  These M3 tapped mounting blocks are much better:

Copper heatsink substrate

Note:  A big mistake there!  Never use Kapton insulation for these tiny driver transistors.  There was a slight imperfection in one of the punched holes and it caused a short on applying full voltage.  Well – you live and learn.  They were replaced with Sil-Pads.

The anodised coating was filed off the back of some extruded aluminium heatsinks.  This is to provide both electrical and thermal conduction with the copper plate via a graphite gap-filler pad:

graphite gap filler

This made for a very efficient and solid driver heatsink, but shifted the plate slightly from its intended position on the PCB.  This meant that one of the resistors had to be mounted underneath the PCB.  Here is one amplifier finished and installed on its 2.5Kg heatsink with Kapton tape and thermal grease under the output devices:

Driver heatsink

The driver heatsink assembly is grounded to a gnd track of the PCB via one of those M3 mounting blocks.

Two finished and tested modules on their heatsinks (DC offset about 50mV each – within the “few hundred millivolts” target):

Two P68s

Each heatsink has a thermal cut-off switch in close proximity to the first pair of output transistors.  They’re connected in series between the auxilliary (9V) transformer secondary and the soft-starter module.  If either heatsink exceeds 65°C, then the main PSU transformer is disconnected from the mains as the relays of P39 let go.

Power Supply

Some people seem to obsess over power supplies, but I’m not interested in “fantasy improvements”.  This is quite basic and uses four 18,000μF/100V “Coke can” sized capacitors which had been sitting in a drawer for some years, so they needed to be reformed and put to good use.  Each channel only really wanted about 10,000μF per rail, so two of these would have been adequate, but four could be configured with a couple of 0.1Ω resistors in a basic pi filter arrangement to suppress high order harmonics from the bridge rectifier, so why not just use them?  There are also a couple of 10kΩ drain resistors to deplete the charge in the capacitors after powering down.  This is useful in the event that the amplifier onboard fuses blow for some reason whereupon the amplifiers themselves could no longer discharge the capacitors.

A single channel probably wanted about a 500VA transformer, and 1kVAs were readily available from the big suppliers, so I went for the heaviest one I could find (at 8.3Kg) with a pair of 50V secondaries and that was a German “Block” brand from Element14.

The 50V taps go via 12.5A slow blow fuses to a bridge rectifier:

Fuse n bridge

Fuse holders there in the chassis floor bottom right. 100nF X2 caps across each diode.  Star grounding bolt there at the left – insulated from the chassis by fibrous washers and heatshrink tube.

There is the usual (and essential) ground loop breaker as well:

ground loop breaker

That connects the star bolt to the chassis and IEC power socket ground lug.

It works

Well the amplifier works and produces a clean 225W into an 8Ω dummy load when tested on an under-voltage supply (220V Hong Kong servo-controlled power supply).

And I couldn’t see any cross-over distortion on my scope whatsoever.

I have not yet connected it to any speakers.


Why are the back two PSU caps spread apart like that?  It’s hard to see in the photos, but the mains IEC filter/socket is in the centre of the back panel so it’s just for clearance.

July 2017 test measurements (2.83V across 8Ω dummy load)

Sweep (cursor at 100Hz):

SNR at 2.83V/100Hz is 99dB.  Intended use is 100Hz down after an active LR4 crossover.  At 2.83V/50Hz, SNR is 101dB.  These are at minor noise peaks in the graph and it seems to exceed CD (which is “more than enough”). 🙂

75Hz sine wave:

IMD (30Hz and 100Hz, ratio 4:1):

Horrifying eh?

Leave a Reply

Your email address will not be published. Required fields are marked *