Super GainClone with Klever Klipper
The Super GainClone was designed by Bob Cordell and has gathered some popularity in DIY audio circles. The amplifier design is covered in his Designing Audio Power Amplifiers book (2nd ed. pp. 686-687). The circuit topology is illustrated below.
The output stage of the Super GainClone is an LM3886 power amplifier IC. Unlike the regular GainClone, the Super GainClone has the LM3886 connected as an inverting amplifier with a gain of -20 V/V (26 dB). This offers one key advantage: The elimination of common-mode distortion.
To fully understand this, let's explore the innards of the LM3886 a bit. A simplified schematic of the LM3886 is shown below.
The input signals to the LM3886 are applied to the bases of Q1 and Q2, which act as emitter follower buffers for the main differential pair, Q3 and Q4. When a common-mode signal is applied to the input of the LM3886 the Early effect in the input pair, Q3 and Q4, will cause a single-ended error voltage to develop on the output node (OUT) of the input stage. This error voltage is the cause of the common-mode distortion.
Here is how the inverting configuration reduces common-mode distortion: Due to the high loop gain of the LM3886 and the negative feedback applied to it, the LM3886 will strive to keep its input differential voltage near zero. As the non-inverting input of the LM3886 is connected to ground in the inverting configuration, this means the inverting input of the LM3886 stays near ground as well – a concept known as virtual ground. This causes the common-mode voltage of the LM3886, i.e. (Vp-Vn)/2, to remain near zero throughout the audio band, which minimizes any common-mode distortion caused by the LM3886.
I should note, however, that common-mode distortion is mostly a thing of the past. The Early voltage of devices in a modern semiconductor process is much higher than that of the devices available in the early 1990s when the LM3886 was designed. I have not been able to measure any common-mode distortion in modern opamps. In fact, modern opamps often provide better performance in the non-inverting configuration due to the higher loop gain available in this configuration. Anyway...
The drawback of the inverting output stage in the Super GainClone is that it has a rather low input impedance (1 kΩ in Cordell's design). Most also prefer to have their amps preserve the phase of the incoming signal, so the Super GainClone adds an inverting driver stage based on an LM4562. An LM4562 buffer is used to drive this inverting driver stage.
Performance of the Super GainClone
I chose a Super GainClone implementation that recently surfaced on DIY Audio. The PCB layout is designed by Mark Johnson. I could nitpick a few details in the PCB layout, including some traces that run very close to the board edge and a trace that ends in an unterminated stub, but overall the layout is pretty decent. Most importantly, the LM3886 section of the board appears to have a good layout. In fact some of you may recognize the component placement and selection from the Neurochrome Modulus-86 boards. The fully built Super GainClone with Klever Klipper is shown below. Note that the Klever Klipper can be disabled by removing a jumper.
For the performance test, I powered the amp by a pair of HP 6643A lab power supplies providing ±30 V. The two plots below show the THD+N vs output power for 8 Ω and 4 Ω loads, respectively. I added a measurement of a conventional LM3886-based amp for comparison.
As seen in the graphs, the Super GainClone is a bit noisy (80 µV, RMS, unweighted, 20 Hz – 20 kHz) but it does provide lower distortion than a conventional LM3886 amp at higher output power. The difference is 1.3 dB, which isn't earth-shattering, but it is definitely quantifiable and worth taking. Both the Super GainClone and the conventional LM3886 amp provide 45 W into 8 Ω (and 70 W into 4 Ω) at clipping.
The intermodulation distortion (IMD) of the Super GainClone is a bit lower than that of the conventional LM3886 amplifier. The multi-tone IMD measurement for the Super GainClone is shown below. The reference (0 dB) level is 40 W into 8 Ω. The tallest IMD component measures -118 dBr.
For comparison the same measurement performed on the LM3886DR is shown below. The tallest IMD spur reaches about -112 dBr.
Some argue that the input stage of a power amplifier should clip before the output stage. One way to accomplish this is to limit the output voltage of the driver stage and Cordell proposes the Klever Klipper for this purpose (2nd ed., pp. 512-522).
The Super GainClone lends itself well to the addition of the Klever Klipper. The input impedance of the inverting driver stage is split into two segments and the clipping circuit is added to the node in between as shown below.
On the surface, the Klever Klipper appears to be just another diode clamp, but it's actually pretty clever. In the Super GainClone with Klever Klipper, the reference voltage (VREF) for the clipping circuit is derived from the negative power supply rail. This is an excellent choice as the LM3886 enters clipping earlier on the negative swing than on the positive. Furthermore, deriving the reference voltage from the power supply of the output stage causes the reference voltage to track any variations in the supply voltage, such as those caused by ripple voltage and variations in mains voltage.
Selection of the Reference Voltage
The output dropout voltage of the LM3886 depends on the supply voltage and load impedance. Thus, the optimal reference voltage of the Klever Klipper will depend on these parameters as well. In the Super GainClone with Klever Klipper, the reference voltage of the clipper is adjustable.
The LM3886 has higher dropout voltage with a 4 Ω load. I, therefore, recommend adjusting the Klever Klipper reference voltage with a 4 Ω load connected to the amp. This will allow the amp to work well with both 8 Ω and 4 Ω loads.
The first step is to drive the amp to clipping as shown below.
As seen above, the Super GainClone actually clips pretty cleanly. Some effects of device saturation are seen when the output approaches the negative rail. These effects cause the sharp discontinuities observed when the LM3886 enters and exits clipping on the negative half-cycle.
Adjust the reference voltage on the Klever Klipper until the clipper just engages and smooths over the discontinuities as shown below.
The amplifier is now ready for use.
Performance of the Super GainClone with Klever Klipper
The graph below shows the THD+N vs output power for the Super GainClone with Klever Klipper. I included a measurement of the Super GainClone without Klever Klipper and the basic LM3886 for comparison.
The Super GainClone and the Super GainClone with Klever Klipper perform identically until the clipping circuit engages. Unfortunately this happens already at 12 W and the THD+N reaches 0.1% at 25 W. The distortion caused by the clipper is primarily odd-order, thus, I would expect the clipper to cause some harshness in the perceived sound quality at higher volume levels.
The THD+N vs output power into a 4 Ω load is shown below for the three amps. The Super GainClone with Klever Klipper reaches 0.1% THD+N at 50 W, whereas the amps without the Klever Klipper provide nearly 80 W into 4 Ω at 0.1% THD+N.
The intermodulation distortion is rather high as well as seen in the multi-tone IMD measurement below. The reference (0 dB) level is 40 W into 8 Ω.
The parts for a basic LM3886-based amplifier such as the LM3886DR will set you back about $25. The added complexity of the Super GainClone nearly doubles the parts cost to $47. Adding the Klever Klipper to the Super GainClone adds about $3 for a total parts cost of $50.
The added cost and complexity of the Super GainClone and Super GainClone with Klever Klipper begs the question whether better performance can be had at the same parts cost. For example, a composite amplifier such as the Modulus-86 would be the same complexity. The parts cost for a Modulus-86 is $47.
The performance of the Modulus-86 is compared against the Super GainClone and the Super GainClone with Klever Klipper in the graph below. The Modulus-86 provides lower noise and nearly 20 dB better THD+N than the Super GainClone. In fact, the measurement of the Modulus-86 is limited by the THD+N of the Audio Precision APx525 audio analyzer used in this test.