Hi Folks
I was asked by a magazine about Class AB amplifier design and thought it might be of interest to some on the forum.
"Hi James
I'm going to be writing up an editorial for April 1 and I wanted to focus on amplifier development/evolution, particularly as it relates to class-AB. It seems, as a non-engineer reviewer, that class-AB architecture has been taken about as far as it can go in the past 10-15 years. Obviously it has some great advantages, too, but I'm curious what your/Bryston's views on it are."
BRYSTON
Bryston has used Class AB output stage architecture for many years because it provides an excellent platform for one of the things we find most important in music reproduction: Linearity. There are numerous ways to implement a high-current output stage for an amplifier meant to drive speakers. There is Class A where all output devices carry signal current at all times; Class B where the individual output devices carry signal current 50% of the time, each on one polarity of the signal; class AB where each polarity of output device carries signal current more than 50%, but less than 100% of the waveform cycle.
There is also Class D, where the signal is converted to a varying pulse-width square-wave on a supersonic carrier, but that is beyond the scope of these comments. Suffice it to say that this can be a good, accurate way to amplify a signal, but the overall linearity is dependent on the accuracy of the conversion to an HF carrier, modulated square-wave. It also presents challenges with excluding the carrier itself from the final, delivered signal to the speakers. It can represent many hundreds of watts of RF energy, and the necessary filtering has a tendency to affect audio frequencies to some extent.
Once we have chosen maximum linearity as one of the most important parameters of an output stage, we can look at the characteristics of the power devices to see where they perform best, and where they may have problems. Power transistors have a rather broad linear region in their transfer function, from just above their 'on' threshold, up to the higher current region closer to saturation, or fully 'on'. At very low current near threshold, and at very high current near saturation, the transfer function bends toward an overall 'S' shape. Thus, it becomes obvious that for the best linearity we want to keep the output devices operating within the optimal linear region of their transfer function. We can do that by operating the output transistors as opposite polarity pairs, switching from one polarity to the other at the zero-crossing region. Class B will inevitably show a discontinuity in the zero-crossing region, since as we saw above, the transistors are not linear at those low-Voltage, low current parts of the signal. Placing a known continuous biasing current on the transistors prevents them from entering that nonlinear part of their curve, and smoothes the zero-crossing region. There may still be a small discontinuity in the zero-crossing region if the upper and lower output devices do not have exactly equal and opposite characteristics at the crossover point. Unfortunately, opposite polarity devices have similar, but not exactly equal, characteristics.
Attempting to overcome the small remaining zero-crossing discontinuity by increasing the bias current to a level that it keeps the output devices 'on' for 100% of the waveform cycle does nothing but push the devices into the nonlinear region at the high-current end of their transfer function. Distortion actually goes up, not down, all else being equivalent.
Thus, Bryston began exploring the possibility of giving the two polarities of output transistor exactly 'equal and opposite' characteristics. In that way the zero-crossing region can remain linear. The way we found was to incorporate our 'Quad Complementary' output section, discussed in the white paper sent to you by James Tanner earlier. That output configuration goes a long way toward virtually eliminating zero-crossing artifacts from the signal, and sound quality is very improved.
In the future, it remains to be seen whether Class D can achieve the overall musicality that a correctly designed Class AB amplifier can achieve. It is not there yet, but we feel it holds promise.
In today's noisy world, there are challenges remaining in terms of an amplifier's immunity to RF and other noise signals coming in on the power-supplies and input cables. Our latest 'Cubed' models have demonstrated that it is quite important to take that into account as well as overall linearity of a test signal. That's also outside of this discussion, of course, but it matters.
I hope this above is helpful, but please let me know if you have questions or comments.
