AB2ECV4. Improved version of AB2EC.
By tying the collector of Q9 to the emitter of Q10, and vice versa, instead of tying them to the output (as in fig.1, Q5 & Q6), the distortion form the bias
circuit itself has been further reduced. As a consequence, the constant current sources (CCS) has to be changed as well. Adding two more transistors
(Q5 & Q6) allows to use CCSes drawing the same amount of current. Opposed to fig.1, where a ratio of I1 : I3 = I2 : I4 = 3 : 4 has to be maintained, that
is, regardless of the temperature, it is easier to maintain the correct balance by keeping them just all equal, i.e. a ratio of 1:1.
Also R8 and R9 has been added. These resistors determine the minimum drain current (Imin), in this example 28mA. If R8=R9=0 then -theoretically- Imin
will also be zero. In practice however, slightly more or even worse, slightly less, i.e 'beyond zero'. Since the latter is impossible, the bias voltage -v(a,b)-
will collapse instead. As a result: far more distortion. Needless to say that this should be avoided at any time. For the same reason I3 (or I4) should not be
smaller than I1 (or I2), though tiny deviations, less than say 5%, doesn't harm that much.
Furthermore, R8 and R9 have also a pronounced effect on the temperature coefficient (TC), that is, quiescent drain current (Iq) and minimum drain
current (Imin) versus the temperature of the bias circuit, thus not the temperature of the output devices, which has no effect. Compared to the previous
circuit, TC-Iq dropped from +0.3% / K to +0.23% / K, while TC-Imin got even slightly negative, it dropped from +0.3% / K to -0.14% / K. That's nice of
course, but the downside is a greater dependency on the CCSes. An increase of 100% (that's extreme, btw) of the CCSes for example, will result in an
increase of Iq by 19% and an increase of Imin by 96%. Nevertheless, this is not a serious problem, since the CCSes should be well defined anyhow.
Fig. 10a. Harmonic distortion products at 20kHz of amplifier AB2ECV4. THD = 3.6ppm (bandwidth = 500kHz)