Clipping at 20kHz. The positive clipping is good, but there is still evidence of a slight 'latch-up' effect on negative clipping. This is of no real importance, clipping sounds bad with or without such effects, and should be avoided. (Actually, clipping at 20kHz generates only harmonics at higher frequencies which are inaudible.)
20kHz square wave 100mV input, 7.5ohms load. The shape is determined almost entirely by the low-pass filter at the amplifier input.
20kHz square wave 100mV input, 7.5 ohms parallel with 2uF load, measured across the load.
As above, but measured before the output inductor, showing that the ringing is just the effect of the inductor interacting with the load capacitor, not an amplifier stability problem.
This is the current through the bottom half of the output stage with a 20kHz sine wave signal. This half supplies most of the output at this signal level, but cuts off above a positive output current of about 100mA (the value of the quiescent current), and then the top half takes over to supply the remaining part of the signal:
The current through the top half is just the 'error' signal which needs to be added to the current from the bottom half to give an undistorted sinewave output current into the load. The yellow line is the zero current level, demonstrating that the top half does not switch off at this signal level, remaining in class-A.
A 20kHz square wave signal produces the lower trace at the base of the first input stage transistor. The overshoot is not the result of the high frequency compensation which can have this effect in a more conventional circuit, but is caused by the 2p7 compensation capacitor plus base-collector capacitances of transistors loading the high impedance output of the input stage, and this is why low capacitance transistors are specified. The overshoot amplitude is what matters, and is just 2mV peak for a 100mV input square wave, while the input stage is adequately linear for inputs of 250mV peak, so this is not important. For normal audio signals rather than 20kHz square waves the input stage signal will be far smaller. The signal at the input transistor base with sinewave input 20kHz at 300mV rms is about 0.35mV rms. From 5kHz down to 20Hz with the same level input it is only about 0.1mV rms.
Distortion at 20kHz, input 300mV rms, load 7.5ohms. The distortion includes an obvious second harmonic component, with much of the remainder being just noise. Normal music signal inputs can be expected to have a peak level about 100mV at this frequency, so even this level of distortion is unlikely to occur in practice. Second harmonic distortion is estimated at 0.006%, (-84dB), but the lack of any obvious crossover effects is the important feature. The improvement described in the simulation section was aded for this test, Iq being 120mA and the base-emitter resistor of the lower power transistor is increased to 100R. With the original 47R a regular sharp edge becomes visible on this distortion trace. With the signal level reduced to 100mV the distortion is almost entirely hidden below the noise.
Again, distortion at 20kHz, input 300mV, but this time the load is 7.5ohms with 1uF parallel capacitance. This value is chosen to give about equal resistive and reactive components, (i.e. a phase angle of 45deg.). The distortion certainly looks different, but the total load impedance is now lower, and also the harmonics are shifted in phase by the capacitance, and so little can be deduced from this. The distortion is estimated at 0.015%, showing that distortion is a little increased with this lower and reactive load, as expected. Again, reducing the signal level to 100mV reduced the distortion, and it was almost hidden by noise. This is not really a good test of distortion into reactive loads because the added load capacitance will attenuate any high order distortion components. A more complex load to more accurately reproduce the impedance of a typical speaker may give more meaningful results, but there appears to be no good theoretical reason why reactive loads should seriously affect the crossover reduction mechanism in this design.
A PC spectrum analyser with zero test signal, including noise and interference components from the distortion extraction circuit. The large component near 6kHz is possibly picked up from the PC or a nearby tv. The PC sound card is a cheap (under £10) type, and is only just adequate for these tests.
Distortion from 2.5kHz sinewave input to amplifier at 300mV rms, taken from the signal nulling distortion extraction circuit, with further 40dB gain before feeding to PC sound card input. An uncancelled component can be seen at 2.5kHz, and distortion components at 5kHz and 7.5kHz. The 5kHz second harmonic is at -102dB (0.0008%) and the 7.5kHz third harmonic at -110dB (0.0003%). Any higher harmonics are somewhere below the random fluctuations, the high frequency peaks constantly changing in level with or without the test signal.
This is an unusual test, to try to detect intermodulation distortion from two signals both very close to 20kHz and equal amplitudes. (Signal cancellation is not very good in this photo.) The point of this is that the sum varies as the signals drift in and out of phase. The sum is large when they are in phase and small when of opposite phase, so in effect we have a 20kHz signal varying in amplitude at a rate which is adjustable by changing the frequency difference, and any intermodulation will be at a low frequency which should be easily visible as a blurring of the high frequency distortion signal. (The photo includes the effect of a long exposure). The distortion remains sharp and clear, simply varying in amplitude as the signal amplitude varies, whatever the rate of change. The maximum peak to peak distortion amplitude also remains constant whatever the frequency difference, which would not be so with significant added distortion. This is as expected, of course, there is plenty of negative feedback at low frequencies to reduce distortion. The reason for being interested in the result is the claim made by Levardin, if I understand it correctly, that thermal memory in an amplifier input stage will add distortion to a varying level signal. Although it was not designed for that purpose my input stage has the property of low thermal modulation from signal level changes, and this test was designed to see if any effect could be seen. This may be just another 'miracle ingredient' from an amplifier manufacturer, but it is reassuring that nothing could be found. Whatever effect there is must be below the normal distortion and noise levels.