New approaches to switched-mode audio power amplifiers (Part 1)
The Class D amplifier has been the usual switched-mode answer to linear amplifiers for three decades. Class D amps have a clear efficiency advantage over their linear alternatives, which, over the years, has led to much effort being invested in improving the linearity of switched-mode designs.
Even so, the harmonic distortion of switched-mode amps remains inferior to that of linear power amplifiers. A new approach promises better efficiency than Class D amplifiers and better performance than linear amplifiers.
The Basic Class D Amplifier
The basic Class D audio amplifier is shown in Figure 1. Either the upper or lower switch in the totem pole is ON at any time, except for a brief transitional period which prevents both switches conducting at the same time. The two switches operate alternately to produce an average output that remains between the Power Input voltage and ground. The time ratio of the ON time of the two switches determines the average output voltage.
For example, a 50% duty cycle will produce an output of half the Power Input voltage. The series capacitor removes the DC component. The series inductor and shunt capacitor form an LC low pass filter. The switches operate at a frequency at least least ten times higher than the maximum output frequency to allow the filter to remove the chopping frequency from the desired output signal.
A simple amplifier of this sort is entirely adequate for voice applications, but there are implementation details that need careful attention for more demanding uses. A cell phone audio amplifier is typically a Class D amp. These amplifiers were originally thought of as needing only to meet the standards for voice communication. Increasingly, phones and handheld devices are being called on to produce high fidelity audio. With care to minimize distortion, Class D amplifiers can still meet that need, though their limitations are too often plainly audible.
There are three main sources of distortion in Class D Amps. First, the best FET switches, driven optimally, still spend a significant time in the linear region when switching. That means dead time must be inserted to avoid upper and lower FETs conducting at the same time (shoot-through). Dead time causes non-linearity in the output which is difficult to correct for. The brief period of linear switch operation is a source of asymmetry which contributes to distortion, as well as reducing efficiency.
Second, Class D amplifiers do not provide power supply rejection. Audio frequency noise on the supplies will appear only slightly attenuated at the output. That requires extra care in regulating and filtering the power supply. The task is further complicated by the way Class D amps return inductive energy to the rails, sometimes called bus pumping. Bus pumping can cause a power rail to rise above its regulation point, leading to distortion.
Third, the switched inductor in a Class D amp interacts with the speaker's inductance (or capacitance) in non-linear fashion. Care must be taken to avoid resonances and beat frequencies that can be heard in the audio range.
Faster clocking decreases the size of filter elements and extends the upper range of the bandwidth, but it also increases switching losses. As the clock rate increases, dead time becomes a higher percentage of the cycle time, which increases non-linearity. This effect can quickly become the largest source of distortion.
Take the example of a Class D amplifier running at 1 MHz with 25 ns of dead time. At a 50% duty cycle, after subtracting for the dead time, the upper switch is on for 475 ns and the lower switch is on for the same 475 ns. The theoretic output is then exactly where it should be, at 50% of full scale. The situation is different at 25% of full scale. On time is now 175 ns and OFF time is 775 ns instead of 200 ns and 800 ns. That puts the theoretic output at 175 / 775, or 22.58%. There is almost 2.5% distortion before the non-idealities of the power switching are taken into account.
While it is possible to reduce dead time below 25 ns, extreme care must be exercised. Temperature and aging effects can shorten dead time, causing shoot-through. Even a little shoot-through disturbs the power rails and generates distortion. A little more shoot-through can be destructive.
A related issue is the difficulty in producing extremes of duty cycle that still exhibit symmetrical positive and negative edges, and that do not dwell too long in the linear switching region. The problem stems largely from the limitations of switch drivers.
One answer is to use Sigma-Delta modulation instead of Pulse Width Modulation. With Sigma-Delta modulation, periods of ON or OFF are integer multiples of the clock period. Linearity can then be improved, but at the expense of a higher clock rate. Sigma-Delta systems generally need to run at least 64 times the maximum signal frequency. There are significant efficiency penalties associated with clocking that much faster.