Four Channel Power Amplifier

March 2018 update:

This amp is now in service attached to the mids and tweeters of the Jamo Oriels.  Absolutely flawless performance that completely trashes that of the highfalutin C-J valve amps previously attached to them.  And unlike certain “monuments to incompetence”, there’s no audible hum or hiss whatsoever from any speaker driver until after the active crossover is switched on to produce the inevitable but whisper quiet white noise perceptible only with ear super-glued to a tweeter!  And there’s no shut-down thump either!  The lower noise floor provides greatly improved dynamic range.  Sonically, this amplifier is indistinguishable from the very best of the competent brand name amplifiers at any price and it cost SFA!


The stereo P3A built for my brother about 6 years ago is so much better¹ than any “audiophile” amplifier that it simply isn’t funny.  It easily surpasses even the ridiculous C-J “Premier Eight XS” valve amps which regrettably I still own.  Those hideous monstrosities stand in plain view as irrefutable testament to a past life in Audiofoolsville.  This 4-channel P3A together with the dual channel P68 subwoofer amp will eventually see all that garbage presently attached to the Jamo Oriels thrown on the tip where it belongs.  The amp is specifically designed and optimised to serve the stereo midrange and tweeter drivers in the actively crossed system.  It has a pair of two-channel P3A (now superseded Rev. 2) PCBs at opposing sides of the chassis and OTT linear power supplies inbetween made using PCBs from eBay with a few sensible mods.  Each channel can deliver about 70W into 8Ω (a lot more than the tweeters need but hey).

  1. “Better”: simple circuits, much lower distortion and noise despite use of standard parts, less expensive, stable, has useful rather than useless features and can be repaired if need be at the blink of an eye – having built it I know what everything is!

Power supplies

The usual Silicon Chip “Class-A” power supply filters¹ from Altronics were no longer available, so I found alternatives.  These came as bare PCBs.  Apart from the rectifier diode heatsinks and quick-connect tabs, all components came from Element14 and RS to avoid “fake” problems that plague eBay.  Despite a few strange errors in the screen printing and missing “thermals” around the diode through-holes (requiring that the heatsinks be attached to the diodes prior to soldering), the boards are actually of pretty decent quality as it turns out.  A few mods were needed though:

  1. There was nothing about those Silicon Chip filter boards to actually render them ideal for a Class-A power amp.  There was no provision for chokes for example.

The only through-hole diodes to fit these PCBs are ultra-fast common cathode/dual anode 3-legged types which are completely unnecessary and ridiculous for linear power supplies operating at mains frequency.  Such things are revered however by DUI audiophiles, so that’s where the market has gone unfortunately.  The through-holes for these might have been bypassed with wire links for use of a simple off-board monolithic bridge rectifier bolted to the chassis floor, but floor space was at a premium in the tight chassis and there’s no reason these woudn’t at least work, so I found some soft recovery ones from Fairchild.

The output end:

The axial 1μF caps might be a bit questionable, but the drain resistors are certainly handy.

The diodes are bypassed with multi-layer ceramic capacitors (under the PCBs) to help suppress unwanted high order switching harmonics which might either back-feed through the transformer as conducted emissions down the power cable, or pass to the amplifiers:

The photo only shows half of the ceramic caps finally installed.  There were four more added to each board so that both anodes are bypassed directly to the centre (cathode) pin of each diode.  It’s stupid because the PCB combines the anodes anyway.

And stop screaming at me!  X2 or X3 caps were too big to fit.  Due to the use of ceramics, slow blow fuses are included after the transformer secondaries just in case the primary fuse at the IEC socket takes too long. 😛

The PCBs are double-sided, so both top and bottom sections of the wide ± rail tracks were cut between the first and second filter caps so those 0.1Ω resistors could be fitted.  These form π filters on each rail to further suppress high order harmonics generated by the diodes.

Main Power Transformer

This is a stock standard 625VA dual 28V secondary tranny from the nice man at Harbuch in Sydney.  With its 240V primary, it provides around ±38Vdc (unloaded) at the output end of each filter PCB when connected to a 230V mains supply.  It’s about 3V over ESP’s design spec. but it’s perfectly safe when connected to nominally 8Ω speakers.

The Harbuch transformers are very heavy for their VA rating.  This 625VA weighs more than a 1kVA RS/Nuvotem for example.  And they do not emit an audible hum like the cheaper ones from the major online suppliers.

There was a left-over steel transformer canister, so it got used – not that there’s anything particularly useful from it in a Class A/B amp, but it is pretty.

It’s a bit of a monster and there are lots of caps to charge, so a soft-starter is essential.

Soft starter module

It’s the usual P39, but due to the excessive capacitance (120,000μF) across both boards, I used 3 series-connected “20Ω” NTC thermistors and increased the short-out timing to about 2.5 seconds.  This necessitated several cuts in the auxiliary PCB tracks:

Why series-connected?  Three by 20Ω thermistors in series gives the same starting point as three parallel standard 180Ω resistors.  Since an NTC’s resistance drops with increased temperature and they’re not precision devices, if connected in parallel, one could carry the majority of current and heat far more than the others and you might have a situation with two of them doing SFA.  In series the same current must pass through all three.

  • Aside.  The thermistors are shorted out of circuit once the second relay activates.  There is a cheap and stupid thermistor arrangement promoted by the “First Watt” community in which the thermistors remain in circuit.  Yeah – it’s for a Class-A amp with constant heavy quiescent draw, but so what?  It’s gross incompetence IMO as the thermistors will go through continuous thermal cycles regardless!  They MUST be bypassed!

Slower timing is due to extra capacitance including this electrolytic hanging off C2’s legs under the PCB:

Speaker protection modules

This is where the amp is “optimised” for my speakers.  They’re the usual P33 PCBs and they each take 12Vdc from the P39 module.  Each of a pair of P33 modules serves two of the four channels.  Rather than dedicating one module to the tweeter channels and the other to the midrange channels, I preferred a simplified chassis layout with more symmetric internal wiring.  That is, one module serves the right channels (mid and tweeter) and the other serves the left.  This simply required the installation of different LF cut-off tuning caps on each module – seen at top left of each PCB with flipped positions.

They’re set at about 100Hz and about 2kHz for midrange and tweeter channels respectively.  This suits the Jamo Oriel speakers that the amp is intended for.  Moreover, any significant amplifier output from fault DC up to the tuning threshold frequency will trigger the relays to release (and short-out) the speakers.

Also the D9 position has a series 1N4004 and 56V Zener for faster relay release in the event of a DC fault (bottom right of each board above).

A couple of further component substitutions ensure that the speakers only unmute about 5 seconds after powering up.  This gives the soft starter time to do its thing first.

The modules are tucked under the main filter banks with the screw terminals protruding for screwdriver access.

Each P33 controls a 15 Amp DPDT speaker relay bolted to the rear panel at each side.

Amplifier modules

IMO the ESP P3A amplifier is as good as any hi-fi amp needs to be for moderate power requirements.  Simplicity, reliability and great sound.  I actually prefer P3A over P101 despite P3A’s lower power output and less impressive PSRR – hence the crazy power supplies above.

A little effort in matching the transistors of the differential pairs (Q1 & Q2) saw an equal 15mV of DC offset in each channel once the quiescent current was set.  Suggested target was “within a few hundred millivolts”.  The contraption below was used to at least measure β and  Vf or a whole bunch of BC546s.

The only downside that I ever found in P3A (and it only revealed itself in the high bias P3B Class-A stereo implementation of it with the PCB dissected to opposing sides of a common chassis) is the lack of a ground lift on the signal input, so I added them under the PCBs:

That Teflon insulated wire is wonderful stuff.  Soldering never melts it.  I use it everywhere – especially for coax signal connections.

The 10Ω lift resistors are each bypassed with a 100nF cap.

  • Aside:  This is not one of those ridiculous modifications to P3A being tossed about at that imbecile forum.  Those idiots are bypassing the rail fuses for reasons only a lunatic might explain, suggesting sh:tty PCB layouts, installing long aluminium “thermal strips” across expanses of PCB real estate to tie transistor pairs that couldn’t care less, positioning driver transistors on the main heatsink, using gigantic polypropylene capacitors for fantasy reasons, and the greatest insanity so far is a suggested PCB layout with a ground loop “antenna” the whole way around it.  Cretins!  Just buy the genuine ESP boards!  My mod is made because there is more than one amplifier PCB sharing a chassis and it addresses only a potential problem of circulating chassis currents and resultant hum at the speakers when all the signal inputs are connected to the active crossover.  The ground lift circuits are identical to those which are standard in P101 and P68.

Completed modules on their heatsinks with a mica card under each to prevent shorting of the added components (not that they would on an anodised surface but it was no trouble):

Insulators are Kapton tape with thermal grease. Each bolt has a spring washer.

No fancy parts (well maybe a few polystyrene caps).  Basic BD139 and BD140s, BC546s etc. and the output devices are just NJW3281 (NPN) & NJW1302 (PNP) which are fairly recent On Semi equivalents of the Toshiba devices around which I believe P03 was originally designed.

The chassis

It’s a cheap one from Alibaba (Chinese “Dodgy Brothers” – even worse than eBay).  Same as the one used for the two-channel subwoofer amp.  The seller was called “Breeze Audio” and the 10mm thick front panel arrived with a badly dented corner – obviously dropped at the factory prior to anodizing.  Seller claimed it was damaged in transit which was nonsense of course, but sent another one – with a dent in the same corner!  Maybe a whole pallet load was dropped.  Maybe that’s why it was cheap, but please say so!  Anyway the second one wasn’t as bad, so I filed it down and touched it up.  Also 95% f the screws were too short and of the wrong type.   Don’t go there.


The soft-starter, power supplies and mute relays were tested first.  A problem.  A loud turn-on chatter from the muting relays.  This was essentially eliminated by adding extra capacitance under the auxiliary 12Vdc supply terminals of each P33 board:

Main PSUs deliver about ±37Vdc (on test bench 220V mains) so all was OK.  The P3A boards were then connected with heavy gauge (20A) wires stripped out of some electrician’s house wiring and the bias set to 100mA in each channel (≅67mV across the specified resistors →).  DC offset 15mV for all channels.

The back panel:

Goodness me – little people. 😆

Time to stick some sine waves through to an 8Ω dummy load.


Here is the minimum sine wave I could put across the dummy based on the minimum output level setting of my old analogue tone generator:

It’s 75mV across the load (0.7W) and a true RMS reading at 250Hz.  No apparent PSU noise, no oscillation and extremely clean compared to say Class-D at a similar level.  That’s Hypex.  →

And nothing obviously resembling cross-over distortion (anyway I think I’d have to use a much higher frequency test tone for that).

The grounding clip of the oscilloscope probe is not connected to anything.  Doing so could damage the ‘scope and amplifier.  The ‘scope and the power amp power leads are connected to the same power strip and that is the Earth reference to which everything is tied.

And this one is just before clipping:

Same 250Hz (so as not to trip the speaker protection on the midrange output tuned to 100Hz) and at 22.6V across 8 Ohms that’s about 64W. That’s on the 220V HK supply.  Actual use will be at around 230V.

Since I used the basic BD139 & BD140 driver transistors as recommended, there is no +ve rail sticking evident whatsoever.

REW measurements

Disclaimer:  I don’t have expensive lab equipment to measure distortion properly.  What I do have is a laptop and a Focusrite Scarlett 2i2 USB sound card.  The sound card’s base performance using a loopback cable is like this:

So at 1kHz, the “test equipment” has 0.006% THD and 0.002% noise.

Here’s the test set-up/procedure:

The dummy load is a string of fourteen 11W 2R2 wirewound resistors in 7 cross-tied pairs.  The DC resistance of the full array is 8Ω and it can handle around 150W if that fan (top left) is blowing across it.  For this amp the fan is not needed however.

The amplifier and DMM are connected across the whole array, but the Focusrite 2i2 USB-powered sound card (which is this):

… connects across only the last pair of resistors using alligator clips:

The last pair of resistors is at the amplifier ground end of the array.

The idea is that the dummy load itself is used as a voltage divider so the sound card only sees one seventh of the voltage applied by the power amplifier across the array.  It’s input attenuator comes in handy too.

The laptop PC to which the sound card is connected via a USB cable runs off its own batteries and is not earthed via a battery charger.  Just like the DMM, the PC must float with respect to the power amplifier Earth.  This is a Sony Vaio that I used.  It has REW installed:

Unless these precautions are observed, something will blow up!

My procedure was to use the REW generator to emit a 100Hz sine wave (50Hz might trip the speaker protection relay – it’s a midrange/tweeter amp after all).  The output level (from one of the midrange channels) is then brought up using the 2i2’s front panel monitor level control until the voltage across the dummy load reads what you want on the DMM.   The monitor level control is not touched again.  The input gain control on the 2i2 is set below a level that clips the sound card.  The tone is set at -12dB in REW.  It’s then switched off and a measurement sweep is performed after checking levels with pink noise and adjusting only by the input gain control of the 2i2 if needed.  The actual sweep level remains at the -12dB setting made previously.


This is just a “nervous ninny” test done for 1W (2.83V across the load) and limited to a sweep between 100Hz and 20kHz:

  • The cursor is at 1kHz.  THD (including that of the sound card) is 0.011% at 1W.  Even with my amateurish measurement set-up, this is more than four times better than Nelson Pass’s Class-A “First Watt” F7 which measures (by his own publication) at 0.048% at the same point!  And P3A is an AB amp with no pretensions whatsoever!  Guess you can fool some of the people some of the time – and more of them when you’re at!

SNR at 2.83V/1kHz is 101dB (which is better than CD which is “more than enough”).

Some PSU rectifier switching noise (100Hz) and harmonics (200Hz and 300Hz) there in the noise floor.  Maybe those ceramic caps and the π filter in the PSU actually work as any higher orders are buried.  The only reason that I didn’t start from 20Hz is that (again) P33 might have tripped the mute relay.

  • Note:  A similar sweep at the P68 page where a standard monolithic bridge rectifier was used shows no such harmonics whatsoever.

The fundamental is that fine red line up at the top (“100%”).  It’s ruler-flat because the chart is normalised to 100%, but the line was ruler-flat before normalising anyway:

That’s the so-called “first Watt” and includes harmonic distortion and noise (above) of the very cheap sound card (and who knows what else?).

  • Actually come to think of it, the above measurement only has about 0.4V presented across the sound card input.  That is, the noise floor might have read somewhat lower again had I connected the card to say a 330Ω/1kΩ voltage divider across the load rather than that dodgey connection along the resistor array.

Taking the 1kHz THD reading of 0.011% from the above graph and subtracting 0.006% (from the sound card graph) leaves 0.005% THD at 1W/1kHz/8Ω.

SNR (1W/1kHz/8Ω) is about 93dB (including the sound card).  Even that PSU noise at 100Hz it at -88dB, so there’s no audible noise from the amplifier when connected to speakers in a room.

Anyway when as low as this, distortion measurements of amplifiers alone are pretty much irrelevant.  Once a speaker is connected, measurements done acoustically will show that speaker distortion seriously swamps it.  For example, my upstairs active midrange (an 8Ω Scanspeak Illuminator in a critically damped sealed encosure) measures around 0.15% THD by Earthworks microphone at 1kHz and a 75dB level.  That’s between about 10 and 25 times worse than the amplifier alone depending on how it’s measured/calculated.

And with the above in mind, I see little point in making any high power tests.  Although I’ve gotten away with things so far, I fear blowing something up.  For instance the sound card inputs could easily be damaged unless I set up different voltage dividers, or a pot of some kind, and I really couldn’t be bothered.

Although the low frequency clipping test above is safe, it is very risky practice to test power amplifiers at high power levels and high frequencies.  The output transistors could be damaged.  Regular music content does not contain high energy content at high frequencies anyway (see typical spectrum to right →), so the results (if attainable) would be pretty meaningless.


To establish a baseline I did a couple of sound card loop-back measurements for the American SMPTE (probably not so relevant for Australian or European 50Hz mains) and German DIN standard dual tone settings.  The two plots are here (Hann FFT window):

So they basically show the IMD contribution of the 2i2/Sony Vaio USB.  I accidentally had the frequency axis on linear there, but it isolates the sum and difference spikes quite well, so I’ll stick with it.

At 1W output (2.83Vrms across the load) here are the SMPTE and DIN FFTs:

The legend for the frequencies at which the figures apply is here →

  • Note:  I’m not sufficiently expert to know how decent these measurements are and the fact that the power amp measures slightly lower than the sound card has me a little confused, so I include the graphs for fun only.  Indeed for the intended use of the amplifier after an active crossover to drive midrange (100Hz to 2.8kHz) and tweeters (2.8kHz up), both the SMPTE and DIN standards are irrelevant since the respective stimuli would not be amplified by the same channel anyway.  One of the great advantages of actively crossed systems.

And while I was at it, I did a single tone 1kHz FFT for harmonics and here’s the graph:

It seems to roughly corroborate the 1kHz (cursor position) figures in the sweep graph above.

  • Note: The second harmonic is at -79dB (-92 +12.9 since the fundamental is at -12.9dB).  The third is at -80dB.  Pass’ “fancy priced” F7 has a -67dB 2nd (11dB worse) and a -77dB 3rd, claiming some big deal about the 3rd being 10dB lower than the 2nd.  Big whoop!  IMO P3A is clearly superior in every way, not the least of which is its efficiency.  Of more importance is that the dissonant 7th harmonic (conveniently cropped from his graph) is at -100dB and P3A has almost triple the power output!  And my chassis has heatsink fins that extend in a useful direction! USD3000 indeed – give me a break! :mrgreen:

More photos

A neighbour from across the road had a look and took the following photos with his phone camera (which is much better than my camera camera).  Some of them show the safety ground loop breaker and floating star grounding bolt that I didn’t bother mentioning, but are standard in all of my amps:

4 thoughts on “Four Channel Power Amplifier

  1. Dear Ian. You should be the top amp designer in a big Audio company, to teach those silly amp designers the way to make an excellent unit. Providing speaker protection and a well set up slow star power on. You’ve done again. Kind regards,Elson.

  2. Wow Elson. What a nice thing to say. Thanks so much for your kind comment as usual, but the credit for P3A must go to Rod Elliott. I just put them together with a nice chassis and a logical layout. 🙂

  3. Hello, Ian! Thank you for your nice blog!
    I’m also building a 3 way active system now, so exciting reading for me.
    You wrote that you had to add groung separating 10 Ohm resistor at the pcb which seems to be a good idea, of course.
    But as I could see on some pictures, P3A pcb has only one GND terminal (for power supply?), correct?
    How did you connect “minus” wire of the speaker – to the power supply GND?
    May be I’m wrong understand something but usually it is recommended to connect speaker “minus” to the GND on amp pcb and then amp pcb to connect to PSU GND.
    Could not it also be a cause of some humm you have got?
    Best regards, Anton

  4. Hi Anton and thanks for posting.

    The 10R ground lifts were just a precaution. Necessary for a high Class-A bias version, but only precautionary here with standard low bias as there are four channels spread around a common chassis and there’s a potential for circulating currents around the chassis.

    Yes the P3A PCB has only a single power supply ground connection. Well actually the PCB from ESP has two, but only one is used unless the board is cut in half. That single GND lug ties back to the zero volt line at the output end of the PSU filter bank. The speaker return must also connect directly back to the same point. Assuming that you are not bridging two amplifiers, the speaker return must never be connected to the amplifier directly. Heavy return currents could modulate the GND reference potential of the amplifier – especially if only skinny wires connected the PSU to the amplifier module. I know that many designers do it, but it’s bad practice. Hum? Me? no way. 🙂 None of the power amplifiers on this site have audible hum. Well the C-J ones did, but that’s another story. 🙂

Leave a Reply

Your email address will not be published.