Circuit Design

Having decided to build an ultra-compact design, using a spare LM4780 seemed like an obvious plan. Having said that, I might choose a different IC if I didn't already have one to hand. The LM4780 contains two LM3886 dies (reference) giving 60 watts per channel, which is rather more than required this application. National Semiconductor make an enormous range of ICs with differing power levels and configurations, and there are plenty of possible candidates for this application - after all, we only need a few watts as this amplifier will principally be driving small speakers on the computer desk...

Still, we're not mass-producing these, so there's no harm in using a larger IC here.

Basic design requirements

I wanted the amplifier to take its main visual "cues" from the Audax mini-monitors it will be powering. So the case will have the same width and depth, although in practice I don't plan to place the amplifier on top of a speaker.

I also took some design ideas from my Musical Fidelity A1. The top panel forms the heatsink - not the most efficient arrangement, but sufficient for this application. Also, as the actual amplifier could be much smaller than one of these speakers, I decided to recess the rear panel. This means that all the connectors are hidden by the sides and rear panel - a detail of the A1 that I've always liked.

As this is primarily to be used as a PC amplifier, one input would be sufficient. But I added a second input to make the amplifier more "general purpose", should I decide to use it elsewhere.

While the amplifier will probably be switched off by the daytime breaker in the workshop, I decided that a standby or mute mode would be useful as the "outboard" power supply was originally intended to be hidden away. The LM4780 has a "mute" function, and using this is much easier and preferable than any of the alternatives.

Finally, because of the compact dimensions, heat was likely to be a problem. This means that some form of thermal cutout is required. But note that this isn't to protect the IC - these have comprehensive inbuilt protection and are more than capable of looking after themselves. Rather, this is to protect people who might come into contact with the amplifier. As mentioned, the top panel forms the heatsink so the temperature of this should be kept to safe values!

Inverting or non-inverting?

When I first built a Gainclone back in 2003, the inverting configuration was all the rage, and pretty much every forum member wrote that this was hugely superior to the more conventional non-inverting layout. Since then, things seemed to have changed, and now everyone is saying that non-inverting is the way to go. Perhaps this says more about group dynamics than amplifiers?

My LM3875 gainclones were inverting for the experiments that have been documented, but since writing up the results, I experimented briefly with non-inverting, and have to report that the differences really were marginal. To put it into context, upgrading the power supply to dual-mono operation was much more significant. Having said that, YMMV (your mileage may vary) because inverting mode presents a much lower impedance to the preamp - I wouldn't be surprised if some of the reported differences were due to this. Easily tested for by placing the input resistor from your inverting amplifier in parallel with the input to the non-inverting version.

Anyway, I chose the non-inverting configuration for this amplifier as this enables a more reasonable input impedance...

The analogue design

This schematic shows the analogue signal path:

Analogue Circuit

The input signals are applied to the source-select switch, a DPDT toggle switch with gold-plated contacts. From there, the signal is applied to volume potentiometer via an ultrasonic filter (R1 and C1) with a turnover frequency of around 2-300KHz with typical source impedances.

From the pot, the signal passes through C3 and R5 to the power amplifier IC. R5 ensures the amplifier remains stable during power up/down, and eliminating the risk of parasitic oscillations when the volume control is near or at minimum. In normal use, it plays no part in the signal path because the input draws no current at audio frequencies, therefore no voltage is dropped across it.

The feedback resistors were chosen to give a convenient value of gain (x25.54, or 28dB). It was simply a matter of choosing low values for avoid Johnson noise (combined impedances less than around 2K) while ensuring that they were components that I had in stock.

Finally, there is an output Zobel network. Generally, Gainclones are synonymous with simplicity, and this design philosophy normally results in the omission of these components. My initial prototypes managed without them, but I found that adding them improved the amplifiers immunity to RF. As always, YMMV.

Power supply

For reasons of space, the PSU is a single unit supplying both channels; having separate supplies for left and right channels is somewhat inconvenient - the point-to-point wiring will be hard enough as it is!

A star earth point is implemented immediately next to the IC. The external PSU generates a pair of supply potentials which are completely independent, and these only become a "split" supply when joined at this star earth.

During development, I performed some experiments to determine the best power supply configuration and determined the following:

  • 220uF per rail decoupling capacitors mounted directly on the chip (Panasonic FC's)
  • 6600uF per rail smoothing capacitors in the PSU, made up from 3300uF, 35V Panasonic FC's. Putting a pair of 3300uF's in parallel usefully halves the ESR and ESL of the components.
  • I found that adding 100nF per rail at the power supply was beneficial for reducing the effects of mains-bourne interference (this amplifier seems very good in this regard - much better than the Arcam Alpha 2 currently in the workshop).
  • The 120VA 18V transformer that I used for my original gainclone prototype was absolutely fine with this low-powered amplifier.

Other than initially powering the amplifier from a bench power supply, I didn't experiment further with regulation. I felt that the above measures achieved a level of sound quality that was easily in excess of my expectations and requirements!

Power Supply

The power supply has an LED, powered by an extra winding that I added to the transformer - about 30 turns of fine enamal-coated wire. However, I didn't want the LED flickering at 50Hz, so I added a simple half-wave rectifier:

Power LED

The control circuit

None of my projects are complete without a handful of transistors!

The complete circuit is neater than it first appears. There are 3 distinct sections - a temperature sensing circuit, an LED flasher, and a power supply - as you'll see, these blocks are tightly integrated resulting in a low component count.

To avoid unbalancing the supplies and inducing DC currents into the earth scheme unnecessarily, the whole circuit is placed between the positive and negative supply rails. As the mute connections of the IC need to be brought down to the negative supply rail, the temperature sensor is placed down on this rail.

Temperature Senser

The sensor is a classic 2-transistor Schmitt trigger. TR2 switches the mute current directly - this is one of only two components in the whole design that "sees" the whole supply voltage of 50-60 volts, so it's a BC546.

For reasons that will be explained shortly, the sensing circuit runs off of just 3.3 volts. RT1, an NTC thermistor, is connected in a potential divider that feeds the input to the Schmitt trigger. As the temperature rises, the potential on the base of TR1 rises. At some point, TR1 begins to conduct, and when this happens TR2 starts to turn off. As the collector current falls in this device, the potential across R17 also falls. However, the emitter of TR1 is connected to this point, so this means the VBE of TR1 is increased, causing it to conduct more. This is positive feedback, and will cause the output current to fall sharply and cleanly. The 33 ohm resistor introduces hysteresis, which means that the temperature must fall to something below the original "trip point" before normal operation can be resumed.

The mute switch is wired across the thermistor, and when activated simply shorts out the device, simulating a gross over-temperature situation which mutes the audio.

This circuit requires a tiny operating current - measured in micro-amps, rising to around 3mA when the mute switch is operated. The supply to this sensing circuit is regulated to 3.3 volts by the zener diode ZD1, with decoupling provided by C10. Current is supplied via R22, but notice that the power LED is placed in series. This is an economic way of powering both the sensing circuit and the LED. By powering this circuit between the rails, the resulting high value dropper resistor approximates current drive, ensuring that the LED brightness is relatively constant and not noticeably affected by the supply rails dropping under load.

LED Flasher

The final section is the LED flasher. The two transistors form the classic two-terminal flasher circuit, which would normally be placed in series with a load. However, I placed this circuit in parallel with the load. This means that when the transistors conduct, the LED is shorted out. When the LED is on, there is enough voltage dropped across it to power the flashing circuit (approx. 3.5 volts). Surprisingly, it works just as well when using a green or yellow LED (VF=2V approx.), and will also work with a red LED (VF=1.5V approx.) with a resistor value change.

Deciding to place the circuit in parallel was a good move. As all the components within the flasher circuit operate from the forward drop of the LED, they don't need to be rated to withstand the 50-60 volts available from the supply rails. So standard BC548/558 devices can be used, and the electrolytic timing capacitor can be a small, low voltage device. But all of this depends on the LED being present in the circuit, so for this reason, the LED must be soldered in place (i.e., not on a separate connector). Should the LED fail, it will probably go short-circuit which is perfectly safe, but if it goes open-circuit, there is a slight chance of problems. However, the 5K6 resistor limits the available current to around 10mA, which should prevent any nasty explosions.

Finally, note D1 that joins the flasher circuit to the sensor output. When the temperature is normal, and the mute switch is in the "unmute" position, the voltage at the collector of the BC556 will be close to 0V (with respect to the negative rail). Thus the diode conducts, preventing the TR4 in the flasher circuit from conducting. Therefore, the LED is on continuously. When muted, the voltage at the BC546 rises, the diode becomes reverse-biased and the LED starts to flash.

This explains why the supply to the temperature sensor is so very low (3.3V). When muted, the voltage at the base of TR4 is negative wrt to its emitter, which is held to 3.3V. Transistors have a limit on the amount of reverse-bias that can be applied to the base-emitter junction - normally around 5 or 6 volts. The zener diode ensures that the reverse-bias is held to less than around 3 volts.

So a neat little circuit that has a low component count and should be reliable as only one component generates any heat, and this has been over-specified by a factor of 4.