Class G Amplifier
A Class G amplifier is a Class AB output stage with a second trick bolted on: instead of running its output devices off one fixed high-voltage supply rail all the time, it keeps a lower rail connected for most of the signal and switches to a higher rail only during the signal's rare loud peaks, which cuts the wasted voltage-drop-times-current heat that a fixed-rail Class AB stage burns constantly, since most real audio spends the overwhelming majority of its time at a fraction of peak amplitude.
The efficiency gain comes entirely from how audio signal statistics actually look, not from any change to the amplifying devices themselves. Real music and speech have a high peak-to-average ratio — the loudest transient moments can be 10-20 dB above the typical level — so a fixed-rail Class AB amplifier sized to reproduce those rare peaks without clipping is forced to run its output devices at that same high rail voltage even during the 99% of the time the signal sits well below peak, dissipating (Vrail minus Vout) times current continuously; Class G simply recognizes that most of that dissipation is wasted and removes it by keeping the rail low until a peak actually demands the headroom.
Rail switching has to happen without the listener hearing it, which is the real engineering problem. The switch between rails has to complete faster than the audio signal changes and without introducing audible glitches, clicks, or crossover-like artifacts at the switch point — this is typically handled with fast diode or transistor rail-switches and careful timing margin so the switch always completes slightly ahead of when the low rail would actually clip, which is a different and arguably harder problem than anything a plain Class AB design has to solve.
Two rails is the traditional implementation, but the number of rail steps is a design choice, not a fixed rule. Classic Class G designs from manufacturers like Crown and QSC used two supply rails (for example plus-or-minus 20 volts and plus-or-minus 50 volts), but some designs add a third intermediate rail to smooth the efficiency curve further — each additional rail step buys a bit more average efficiency at the cost of more rail-switching circuitry and more opportunities for switching artifacts, which is why most commercial Class G designs settle on two rails as the practical sweet spot.
It sits as a middle ground between Class AB's simplicity and Class D's full efficiency, which is exactly where high-power audio amplification wanted something. Class G can't match Class D's 90%+ efficiency because the output devices are still operating in their linear region and still dissipate real power during normal operation, even if less of it than a fixed-rail design; but it preserves Class AB's output linearity and avoids the switching-frequency noise and filter design burden of Class D, which is why high-power professional touring and installed sound amplifiers used Class G for years before Class D's measured performance caught up enough to take over most of that market.
Read a Class G amplifier through a *statistics-of-the-signal* lens rather than an *amplifier-topology* lens: the circuit is exploiting the fact that real audio spends almost all its time well below its own peak level, and recovers efficiency simply by refusing to pay the high-rail dissipation cost except during the rare moments the signal actually needs that headroom — without changing anything about how the output devices amplify the waveform itself.