Design Article
Efficient amplification for real-world audio
James Lewis, HiWave Technologies
4/25/2012 11:53 AM EDT
Various techniques have been used to improve Class D amplifier efficiency, but these have only addressed the mismatch between best-case scenarios and real world usage at a very broad level.
Many amplifiers use a comparatively high supply voltage. The output stage, whether based on a linear or switching topology, then scales this supply voltage to the required output signal level. These high-supply architecture amplifiers tend to be very inefficient in the face of high LPAPR.
Power consumption in amplifiers tends to scale with the square of the rail voltage. But the operating voltage used is generally very much higher than the average output voltage so, except for the very rare occasion when a transient demands the maximum output power, large losses occur.
Some amplifiers use low-supply architectures, with a supply voltage much closer to the average output level. For the rare occasions when a high voltage peak is needed, a boost converter switches in to provide that. As the converter is used only for brief periods, its conversion losses have little impact on the amplifier's overall efficiency.
Rail tracking offers a finer-grained alternative to boost conversion. This technique varies the output-stage rail voltage by following the audio envelope or, in simpler implementations, by tracking the user volume level.
One issue for rail tracking can be the speed with which the rail voltage is varied. Most circuits have relatively large decoupling capacitors on the output rails and considerable currents are needed to change rail voltage rapidly. Switching between two parallel rails provides one answer to this. A lower-voltage rail is used most of the time, switching to a higher voltage rail only when the output signal exceeds the lower rail's available voltage.
Both rail tracking and switching have been used for some time in high-power linear amplifier designs, such as Class G and Class H variants, but transferring these approaches to a switching amplifier presents significant challenges. For example, the switching algorithm must be designed to avoid any objectionable clicks, or other artefacts, as the amplifier moves from one rail to another.
One of the biggest sources of efficiency loss in Class D amplifiers is high-frequency switching. These losses could be reduced using a slower switching rate, but this will result in audible degradation of the signal if used indiscriminately, especially as the switching frequency gets closer to the human hearing range.
However, it is possible to use the attributes of the audio to inform the switching process and adjust the modulation rate dynamically. Such a technique ensures the amplifier output stage only switches when necessary. The modulator continually analyses the incoming audio and uses its characteristics to decide when best to switch the pulse-generating transistors on the output stage.
The most efficient solution is to combine all of these techniques, and in doing so obtain a large overall reduction in power wastage. The results of doing so can be seen in Figure 3, which compares the efficiency of a conventional Class D amplifier IC and one that uses these new techniques, to give a real world example we've used measurements taken from our HiWave Audium amplifier IC.

In both cases peak power output capability is 100W. Equivalent SPL is shown below the x-axis, assuming a single speaker that has an output efficiency of 90dB(SPL) at 1W and a distance of 1m. At the typical listening level of 73dB(SPL) the minimised quiescent power means such amplifiers consume one twentieth of the power of a conventional Class D amplifier.
By understanding the characteristics of an audio signal, together with normal consumer listening levels, amplifiers are being designed to maximise efficiency in real-world audio and not just test signals. This shift in mindset reduces average amplifier power consumption by several orders of magnitude and brings significant benefits to a wide range of consumer audio systems.

About the author:
James Lewis is CEO of HiWave (www.hiwave.com), a UK based manufacturer of audio and haptic feedback devices.


Tony Casey
4/26/2012 4:48 AM EDT
This article contains an number of errors. The theoretical maximum efficiency of a class A amplifier is 25% for a singled-ended stage, and 50% for a push-pull stage. Similarly, it can be shown that the maximum efficiency of a class B output stage is about 78%.
Sign in to Reply
bcarso
4/26/2012 5:40 AM EDT
I believe there are some significant infringement issues here, considering John French's class HD patents.
Sign in to Reply
GordonScott
5/2/2012 9:19 AM EDT
In fairness to the author, he is rightly pointing out that maximum efficiency is pretty much irrelevant, certainly in audio-type applications.
He's also rightly pointing out that class-D isn't necessarily as efficient in real-world applications as the marketing/hype would suggest.
In the real world, class-G, especially with the inner transistors in class-B, can be surprisingly efficient and vastly lower distortion than any class-D.
FWIW, various combinations of class-D and switched/pumped/tracking supplies have been around for decades in things like loud-audio radio-pagers.
Sign in to Reply