The amplifier consists of four identical channels, each using half of an OPA2134 as an input amplifier which drives both channels of an LM4780 working in parallel mode.

Inverted mode

The power amplifier works in inverted mode. I chose this because I've found that inverting mode results in lower and more dependable DC offsets - essential in parallel operation. Of course, this can readily be achieved in non-inverted amplifiers with an electrolytic capacitor in the feedback network, but this is normally omitted in minimalist Gainclone-style amplifiers.

Inverted mode results in an increase in noise and a reduction in input impedance. The latter can cause problems with some preamplifiers, although not mine because it is capable of driving low impedance loads. The increased noise floor can be a real problem, particularly for people using sensitive speakers, but again this isn't too much of an issue for me with my inefficient speakers. But both of these factors are worth remembering if you're considering a similar design...

Input amplifier

The normal reason for adding an input buffer is to resolve the issue with low input impedance when using inverted mode. While this isn't a concern for my preamp, some kind of inverting buffer is required to drive the power amplifier connected to the negative speaker terminal when using bridged mode operation...

I spent a long time thinking this through. I needed a simple way to switch between normal and bridged operation, while maintaining a consistent signal path in all channels and in both modes - of course one of the potential problems of switching to balanced operation is that extra circuitry can introducing its own sonic character.

My solution final solution is very simple and neat - simply add a differential amplifier to the input:

Block diagram of the final solution (8K)

In normal use, the signals are applied to the inverting inputs of the differential amplifiers, which of course cancels the inversion of the power amplifier, thus avoiding having to reverse the polarity of the loudspeaker terminals. Also, the input op-amps work in shunt feedback mode which eliminates the common-mode distortion that some FET-input op-amps such as the OPA2134 suffer from when the inputs are fed from non-equal source impedances.

For bridged operation, signal A is applied to the non-inverting input of the second amplifier. This means that overall the channel is now inverting. No extra components are in the signal path, and no complicated switching is required.

This arrangement offers an important benefit. The use of differential amplifiers offers the possibility of hum-free operation. While this can be a difficult problem in any amplifier, earth loops could form intractable problems in a four-channel unit where different pieces of equipment could be driving the inputs.

Final solution

This schematic shows the adopted solution. For clarity, power supplies and muting arrangements have been omitted.

Schematic of the analogue sections (10K)

The mode switch is slightly more complicated than above, but as you can see it was a simple matter to provide a couple of extra options. Here are the modes, starting with the switch in the top position and working down:

  • Normal

    Signal B simply passes straight through the switch and is applied to the differential input amplifier in the same way as channel A.

  • Bi-amp

    The signal applied to input A is applied to amplifier B.

  • Bridged

    Again the signal from input A is fed to amplifier B, but the polarity is reversed.

  • Single

    The input to amplifier B is simply shorted to ground. A spare contact on the mode select switch mutes the LM4780.

The input grounds are connected to earth via 110 ohm resistors, bypassed with 100n capacitors. These components tie the source equipment ground to mains earth, but prevent an earth loop forming should the source already be earthed. The input ground is connected directly to the differential input amplifier, meaning that any hum voltage developing across the 110 ohm resistor will be rejected.

The unity gain differential amplifier is made using 0.5% resistors, hand-matched to within a few ohms for best common-mode rejection ratio. This is important to maximise the hum-rejection described above. The value of these resistors is a compromise between input impedance and thermal noise, 10K is a good choice.

The 220pF capacitors provide high-frequency roll-off. The -3dB point is around 70KHz with the current values - perhaps slightly lower than the nominal 100KHz that I might normally aim for, but I had some nice 220pF polypropylene capacitors in stock. 150pF would give 100KHz, but I only had ceramic devices in that value - obviously, these would be absolutely fine in this position, but I wanted bragging rights! (Yes, I realise these only work effectively in inverting mode due to a zero appearing in non-inverting mode. But it's better than nothing.)

The inputs to the differential amplifier are DC-coupled. As the amplifier has a differential gain of unity, this doesn't cause any problems - any DC offset present on the source will simply be presented to the input of the following stage (albeit inverted in polarity). This is blocked with a capacitor between the input and power amplifier stages - this is a good place to install such a component as the surrounding impedances are well defined and the low frequency roll-off can be dependably placed below the audio band. The correct low frequency rolloff will have been set in the pre-amp, using non-electrolytic capacitors.

The resistors surrounding the LM4780 give a voltage gain of 22.6, or 27dB. I chose 0.5% resistors here as well because in parallel-mode operation it's vitally important to match the gain of the two amplifiers to ensure equal current sharing. As before, the resistors were hand-selected precisely. Incidentally, I paid a premium to buy 0.5% RC55 resistors, but could have spent even more to have 0.1% devices. Happily, all of the 0.5% devices were comfortably within 0.1%, so I was pleased I didn't spend more!

As discussed on the gainclone experiments page, and op-amps for beginners page, the resistor connecting the non-inverting input should be equal to the parallel combination of the two resistors connected to the inverting input - around 6.5K here. However, I prefer to minimise the value of this resistor in a power amp because the thermal noise in this resistor is amplified by 1+(Rf/Ri). Some people omit it altogether, connecting the non-inverting input directly to ground, but I believe this is a bad idea because unpredictable currents can flow into the inputs during power up/down. Using 1K is a good compromise between noise and DC offset, but be prepared to change this if the DC offset is too high for your application. Here, what matters most is having near-identical DC offsets in each half of the IC.

Worrying about noise in the IK resistor is clearly unnecessary here because the noise in Rf dominates! Lowering Rf is of course desirable, but maintaining the required voltage gain requires a corresponding fall in Ri, and the resulting reduction in input impedance has a knock-on for the OPA2134 buffer. So the chosen values are a compromise between minimising noise while not presenting too much load to the OPA2134. With the values shown, the op-amp "sees" an output load of around 2.5K - any lower and the distortion of the op-amp would start to rise.

The outputs of the two halves of the LM4780 are joined by a pair of 3 watt 0.1 ohm wire-wound resistors. These are standard 5% devices, but hand-matched to within 0.5% (this is tricky to do, but an accurate current source helps here).

Each amplifier has a Zobel network, which ensures stability into reactive loads, and helps to improve r.f. immunity.

Op-amp power supply

The OPA2134 is powered from split 15 volt supplies. As the unregulated supplies to the amplifier modules are on the limit for standard 7x15 regulators (around ±36V), I chose a simple shunt power supply using two 12mA current sources and a pair of Zener diodes. The noise left at the Zeners is filtered by a low-pass filter formed by a 15 ohm resistor and a 47uF capacitor, which turn over at around 200Hz. This frequency is comfortably low enough to ensure that the op-amps PSRR is sufficiently high to provide good rejection. Finally, a low-ESR 1uF capacitor is connected between the power pins of the op-amp.

Preamp PSU

While it looks simple, this power supply is well-suited to this application. The current sources provide excellent isolation from the broadband noise and ripple on the unregulated supply rails. While the Zener diodes might not be the last word in DC precision, they are far better than required here.