Apr 26 2011
Introductory Aside: Class-A inefficiency?
There are many who say that Class-A is a gross waste of energy – that over 75% of the energy is “wasted” as heat. Well it depends on how you look at it. If the windows are open, then of course the heat is wasted, but what about in winter time – say when the amplifier is used in a closed house having central heating running? In this case the heat generated by the amplifier can all be used to supplement the central heating system to warm the house. Is it then a 75% efficient heater (with 25% “wasted” on music)? Or is it 100% efficient in some way?
Having lost the earlier “Silicon Chip” Class-A tweeter amp and the ESP AB midrange amp to my brother’s stereo system, it was time to build a pair to keep. The midrange amp sounded so good that I decided to base this new ”tweeter amp” on the P3A as well, but to do it a bit differently. The complementary ”lateral MOSFET” (P101) midrange amplifier is on another page. It worked flawlessly from Day 1. This one was a bit more effort, but they now work beautifully together with a pair of 800W UcD subwoofer plate amps in an active 3-way speaker system.
The black anodised aluminium chassis came from China. It has efficient extruded heatsinks. I just added taller rubber feet to the floor, step-drilled a larger front panel hole to accept a mains voltage power switch, added a separate front panel LED and removed a silk screened “fake” logo from the front.
See that badge. It’s a telescope of “aural” with my name “Ian” and a contraction of “Australian” and I claim the trade mark rights!
This one was built using the same type of bisected P3A PCB as the earlier midrange amp, but is biased into Class-A operation following the P3B project article. In hindsight I should probably not have chosen this board for bisection and installation at opposite sides of a Class-A stereo chassis as they might not be properly suited. See “buzz problems” below.
Each channel draws 1.5 Amps quiescent current from a lower voltage power supply to produce up to 25W (8Ω) in pure Class-A (unlike the Silicon Chip 20W amp which I am advised goes AB relatively early despite what they might say). Incidentally, this amplifier can deliver 9W in pure Class-A to a 4Ω load. Accordingly when paired with the efficient ScanSpeak tweeters chosen for the active speaker system, 100 dB is available (at 1m) without any crossover distortion. 50W is avaiable in Class-AB to 4Ω, but my ears don’t really require that kind of drilling!
It would want to run quite hot, but the extruded heat sinks are up to it and adequate heat transfer is assured by adhesive-free Kapton tape, thermal paste and clamping bars at the output transistors.
The project article called for a CLCCc power supply with six 10,000μF caps and a pair of 10mH inductors. An Altronics/Silicon Chip Class-A power supply PCB was a good starting point, but it had to be butchered to add the filter chokes after the first pair of capacitors.
Jaycar 9mH laminated core passive speaker crossover inductors were close enough to spec. and were tested for over-heating using the DC bench supply:
3 Amps (continuous current of two channels) was too much for these chokes to bear (around 8.4W dissipation) and they would rob around 2.8V from the rails, so I decided to duplicate the supplies so that each would pass just the 1.5A current of one channel. The inductors then drop only 1.4V from the rails and dissipate only around 2.1W which is easily managed by natural convection via the chassis floor and lid slots. As the filter capacitor banks are mounted directly over the inductors, 105°C/15,000μF Nippon Chemi-con caps were used instead of the standard ones supplied with the Altronics PSU boards. A total of 180,000μF is a ludicrous amount of capacitance for a low powered stereo amplifier, but extremely low ripple voltage and noise were the aim and I was not interested in building a “capacitance multiplier” circuit on VeroBoard. These supplies are estimated to have less than 20mV P/P ripple at 1.5A loading:
The resin laminated cores are islolated from the stand-offs with double heatshrink tubing. There are two reasons for this: One is that the heatshrink provides good grip for the cable ties; but much more importantly, the cores have an “infinite gap” by virtue of their free ends. Shorting both ends to the chassis via the stand-offs could cause the cores to saturate with consequences that I don’t want to know about. Apparently, there is no sonic benefit in providing dual power supplies in a stereo Class-A amplifier because the current demands on the PSU are constant compared to the dynamic demands of stereo music signals in an AB amp in which reduced channel interactions can be achieved. Here the dual supplies simply serve to increase overall capacitance and to share the current loading.
The transformer is a Harbuch brand made-to-order 625VA with 20V secondaries – a significantly more substantial transformer than the unit used in the first tweeter amp and weighing in between 5 and 6 Kg:
The transformer has bifilar secondary windings ending in common PVC tubes:
Almost like having two transformers in one. The isolated windings were originally extracted as separate circuits to left and right bridge rectifiers:
I was subsequently advised by the transformer manufacturer that this was not a good idea due to the single PVC tubes. Although it worked and both channels drew the same current at all times, I later stripped back the heat shrink and combined the bifilar windings with solder (see below in relation to a minor buzz issue).
The four centre taps are tied to that central chassis-isolated stainless steel star grounding bolt:
… which ties to chassis ground (and the mains socket earth pin) only via the safety ground loop breaker. The whole thing ended up somewhat like an over-the-top development of the dual power supply described here.
The dual power supplies left little floor space free, so the muting relays went on the back panel and the speaker protection module (of course ESP P33 – tuned to around 300Hz cut-off) ) had to tuck in under the right PSU board with its screw terminals protruding sufficiently for screwdriver access :
Here’s a mistake. I stupidly increased the ballast resistance on the soft-starter from the recommendation – mistakenly thinking that because the toroidal was bigger and the filter capacitance larger, higher ballast resistance was required when in fact lower is correct. The tranny growled slightly after the second relay activated with these in place:
Easy fix was to piggy-back another 220Ω on top for 55Ω and a much softer start.
This and the speaker protection module are each connected to the centre bolt and powered by an EI auxiliary transformer as before:
The auxiliary transformer has the “engine mount” suspension to minimise mains vibration noise:
It had to be strapped to ground in case its frame contacted a stray active wire during testing!
The P3A PCBs were populated as before, but with 0.75W metal film resistors and heat sinks on the driver transistors:
Those driver heatsinks took some data sheet searching. Like all ESP PCBs, high-performance necessitates a compact layout, so there is a premium on real estate. Screwdriver access to the bias trimmer was required with the amp stabilised and warm, so the heatsinks must be in place during adjustment.
25μm Kapton tape (about ten times thinner than sil-pads or mica washers):
Thermal paste both sides:
These tended to bend plastically if tightened too much (which wasn’t much), so they were subsequently replaced with stronger 8mm bars made from a door handle shaft:
The heat shrink provides a gap under the bar centre for even force distribution and grip/protection just in case the grounded bar loosens and twists into any component legs.
Being a Class-A amp, the chassis maintains a constant temperature above ambient once stabilised and warm. A temperature-regulated variable speed cooling fan such used in AB amps with under-sized heat sinks wouldn’t be suitable. However since excessive chassis heat could develop on a hot day, I installed a 60°C thermal switch on the back of each heat sink. These are 250V-rated switches, but rather than running mains voltage leads all over the place inside the chassis, I placed them in series with the 9V low current auxiliary AC supply to the soft-start circuit. If either switch reaches 60°C, the soft-start relays open and everything in the amp (except the auxiliary transformer) switches off including the front panel power indicator LED. The speaker protection curcuit’s “loss of AC” function activates immediately to release the speaker mute/protection relays. The idea is not the cycle the amplifier between ON and OFF states. With around a 10-20°C drop required to reinstate the ON condition, there is sufficient hysteresis to get up and turn the thing off at the power switch and come back another day. So far this has not happened.
Powering up and Calibration:
This was practically the same as for the previous AB version – using the dual bench supply on each out-of-chassis module - heatsinks attached. At ±25V, one channel’s DC offset was a mere 10mV and the other a miserly 0.9mV – both well under the 100mV target. After attaching the heatsinks to the chassis, adjusting the bias current was straight forward, but I gave the heatsinks a bit more time to warm before fine tuning. Drift was negligible. The amp definitely runs warmer than the standard version – about 25°C above ambient.
The power supply produces the target figure of ±25V DC at the rails. The transformer has a 240V primary winding and 20V secondaries, but the amp is fed with 225V mains, the inductors take around 1.4V and the amps themselves create a sag of a few Volts with the high bias.
Well Rod Elliott was interested in the chassis, so I lent him the amp and he was very kind to measure it’s performance. It has less than 0.02% distortion (limit of his equipment) from 10Hz to 100kHz with a slew rate of 6V/μs. For 25W, you only need 2.2V/μs at 20kHz so it’s the perfect tweeter amp!
Although it was suggested to me that the amp would sound the same as the standard AB version of P3A and that Class-A was over-hyped, I was quietly enthusiastic despite an underlying reservation about the likelihood of power supply hum with all that current drawn at idle.
Anyway, for the initial music test it replaced a borrowed Audio Research D200 in the upstairs system - after the 2-way active crossover to power the midrange and tweeters (which were passively crossed). This is where the original AB P3A was tested, so it’s a direct comparison with identical equipment albeit some months apart.
Thankfully noise was very low and indeed below the noise floor of the valve preamp (which is itself below the noise floor of any recording). Just a slight general noise with ear to midrange driver and a slight buzz in the tweeter. Probably about the same as the AB version.
The first impression with music was that it blows the AR D200 out of the water, but that was to be expected. I’d love to say that it sounds better than the standard P3A, but overall it is about equal. The sound is pretty much the same on orchestral music etc. at ordinary listening levels. It gives the same great 3D impression.
Downside: Lower power means that at higher listening levels it does not have that characteristic lock-jaw dynamic grip on the music as P3A in standard form.
Upside: It is cleaner and even more 3D on voices and the treble (tweeter) is cleaner. Nat King Cole is right there in a room bigger than my listening room even more so and Joan Sutherland sounds like she is singing from the heavens. Please excuse my tastes in test music. I may be strange. However to anyone that suggests that Class-A is irrelevent, I disagree. To me, it atually sounds less distorted in the high frequencies.
It will make a perfect tweeter amplifier and thrashes the Silicon Chip Class-A amp in the midrange department hands down.
This infrared thermometer is calibrated to a surface having an emissivity coefficient of 0.95, so it’s not perfect for black anodised aluminium (about 0.88) but “close enough”.
I measured a few critical places on the amplifier chassis. The following temperatures were read:
- Ambient metal temperature (before powering up): 23°C.
Then after a good long warm-up period:
- Base of heatsink between fins directly behind any output transistor: 54°C (+31°C)
- Between fins directly behind 60°C thermal cut-off switches: 48°C (+25°C).
- Middle of front panel: 42°C (+19°C).
All seems OK – protection circuit might cut in on a 35°C day.
Addendum (removal of the slight buzz):
Each channel of the amp was completely silent when a source was connected individually to just one of them. Now used solely for the tweeters in the new tri-amplified speakers, when both channels were connected to a stereo source, there was a very slight mains-related buzzy noise with ear close to either tweeter. It was below the noise floor of recordings, but annoying nonetheless and presented a worthwhile challenge, so I had a little fuss.
Some investigation suggested that the strong magnetic field radiating from the power supply AC cabling (around the bridge rectifiers) due to the high quiescent current may have been passing the shielding of the thin internal signal coax cables and/or entering the amplifier modules directly and combining with the signal. A major strip-down and overhaul was carried out.
OK, not knowing for sure what the cause was, I made a list of things to do:
- Combine the bifilar secondary windings of the toroidal for single circuit extraction to the “stereo” rectifiers. This would remove any unwanted inductive coupling across the bifilar windings previously extracted separately to the left and right power supplies.
- Add a steel disc under the transformer for further magnetic absorption (the chassis floor is just aluminium).
- Use large/heavy steel (high magnetic permeability) plates to isolate the amplifier modules from the AC wiring and rectifiers.
- Move the input RCAs away from the PSU to alongside the heatsinks.
- Use Belden/Blue Jeans (brand) double braid low capacitance coax instead of the thin Teflon coax.
- Pass the coax through heavy steel tubes en route to the modules.
- “Crack” the rear corners of the chassis to prevent circulating currents near the coax cables. This required Kapton tape insulation and the substitution of Nylon screws for the steel ones joining the panels.
- Add copper plate to under-side of chassis lid as HF barrier (it should be grounded but isn’t – yet).
Some progress pics:
Combining bifilar secondary windings (I think this was crucial as it removed channel-to-channel inductive coupling within the transformer):
A weapon (12mm OD 2mm wall thickness steel tube):
Not intended as RF shields, these need not be grounded to the chassis. They simply divert any impinging LF magnetic field lines through their ferromagnetic/permeable mass en route to the opposite pole – i.e. around (not through) the coax.
A bit of trouble getting the crazy amount of copper in the double braids down to a maneagable terminal-attaching size:
Adhesive Kapton tape also at rear corners of floor and lid (else circulating currents around the side walls might simply cross the floor and lid corners):
New CMC (brand) RCA – better than the old Neutriks. See Nylon (non-conductive) screw to ensure a proper electrical break:
RF copper shield for chassis lid (held on with thin thermal transfer adhesive pad):
Larger 3mm thick mild steel plates now bolted down by the rectifiers (and tranny bottom disc laser-cut to size):
Spray painted and installed:
Although the steel panels are grounded electrically to the aluminium chassis floor, this is purely for safety. They do not pass LF magnetic flux to the aluminium (magnetically reluctant) chassis floor. They simply provide a ferrous mass to “grab” incident LF magnetic fields (which might otherwise pass straight through to the amplifier modules) en route to the opposite pole.
While waiting for some parts to arrive, I was looking for other possibilities as to why the amp might buzz only with both channels connected and came across the Silicon Chip article for the original Class-A amp that I built. That amp sounds nothing like this one, but has no such problem. I read/remembered that the amp has an optional “ground lift” resistor of 10Ω at the input of each PCB. It is to be replaced by a wire link in a monoblock version and the resistor is said to “reduce circulating currents in a completed stereo amplifier” and to improve channel separation! So I studied the Silicon Chip circuit diagram and decided to modify the ESP PCBs to do likewise (it turns out that the high-power ESP A/B amps exploit a similar “lift”). That it is not included in the PCB is beyond me. Perhaps being designed as a single stereo board, circulating currents are not an issue and the bisection to mount on opposite sides of a chassis was not properly considered. Who knows?
Here is my hack on one module. The other got the same (I also think this was crucial):
Anyway after all of that this boat anchor now weighs 19Kg but it’s silent! At first powering-up I thought I must have stuffed up and the amp was no longer working since there was complete silence. But no – it was the perfect result – an inaudible noise floor with both channels connected and with ear at either tweeter! Sorry I can’t say for sure which modification did it, as they were all done at once to avoid repeat strip-downs.
Caveat: The tweeter has 74μF of series “protection capacitance” forming a high pass with -3dB somwhere around 530Hz, so if the amp does still produce any noise around say 100 Hz (my guesstimate of the original buzz) it would be suppressed somewhat by that.