The phase shift from the output through the feedback network and back to the input needs to be kept under 90deg at all frequencies and with all reasonable loads. There is another phase shift in the feedback loop caused by the internal high frequency compensation which will be around 90 deg at high frequencies, so these two effects added together must be less than 180deg to give unconditional stability. The feedback network is shown next, with a 2ohm open-loop amplifier output resistance, 8ohms in parallel with 2uF load, and a 600ohm input source resistance. V2 is the voltage at the base of the input transistor:
The 2uF in parallel with 8ohm is not a typical speaker impedance, though it is similar to some electrostatic speakers, which is the worst case commonly encountered. Designing for stability with this or lower capacitances should ensure stability with almost any speaker and cable. The phase shift of V2 relative to the voltage source was obtained by a Spice simulation. The first phase diagram is for the above circuit diagram:
Only a few of the components can conveniently be changed, but one useful possibility is the resistor R8 in parallel with the inductor. This has the effect of damping the series resonance of the inductor and load capacitance. Reducing R8 to 1ohm next:
Then R8 reduced to 0.5ohms:
If reducing R8 improves the phase shift perhaps zero resistance is better, in which case the inductor will have no effect and can be left out altogether. Unfortunately if we do this the phase shift goes beyond 80deg and stays there up to 10MHz and beyond as shown next, and therefore is still a problem near the unity gain frequency where an extra 80deg phase shift is definitely undesirable.
Leaving out the inductor is not a good idea. The 0.4uH value used here will have little effect at audio frequencies, but the reduction of high frequency phase shift round the feedback loop is highly beneficial.
With R8 = 1ohm and the load capacitance increased to 4uF the phase shift is still below 90deg.
Greater values of capacitance will eventually take the phase shift beyond 90deg, and to pull this back the inductor would need to be increased.
The phase shift at low frequencies was also checked to see the combined effect of the 4400uF output capacitor, 2.2uF input capacitor, and 10uF in the dc feedback path, and the effect was fairly small, with phase shift round the feedback loop varying from -12deg to +30deg, as shown next.
Of course the closed loop phase shift of the amplifier is a different matter, it is only loop stability which has been considered here.
It is sometimes suggested that the output inductor serves another purpose, to help reject radio frequency interference picked up by the speaker and its cable. The rejection is of course critically dependant on the interference source impedance, which will be a complex function of frequency. For a simple assessment a not entirely realistic impedance of 50ohms was used, and breakthrough at the amplifier input transistor base simulated for a 10V rf input up to 100MHz. First with an 8ohm load and with 1ohm in parallel with the inductor:
The high breakthrough at lower frequencies is because we have ignored the effects of negative feedback, which reduces the amplifier output impedance. Including this effect will take the breakthrough down to very low levels in the audio frequency range, but with reducing effect as we go up to 100kHz, and only a little effect beyond 1MHz. The remaining problem is therefore the high frequency rejection, and we may need to improve this. It may be thought that 1ohm in parallel with the inductor will seriously reduce its effect on rf breakthrough, but increasing this to 7.5ohms gives the next result:
There is only a little improvement, so returning the resistor to 1ohm and instead adding a 0.1uF capacitor across the amplifier output gives a far more useful rejection:
Increasing the capacitor to 0.47uF:
If interference pickup is a problem the addition of a capacitor across the amplifier speaker sockets is clearly worth trying, but as mentioned earlier the assumption of a 50ohm interference source impedance makes the result inaccurate, so some experimentation may be needed. Using screened (coaxial) cable to connect the speakers is an alternative method of interference rejection.
The 6 transistor board layout is easily modified for the 7 transistor version, and again a sketch on graph paper is shown with 0.1 inch squares, using thick widely spaced tracks suitable for a simple production method with an etch-resist pen. Transistors are shown with a thick line at the metal side. Black dots show the three earth wire connections to be taken to a single earth point, and also the output connection. The component side view is shown. (A mirror-image copy helps with drawing the layout on the copper side if the etch resist pen method is used.)