The switching amplifier, or class D amplifier, has risen quickly to prominence in consumer audio applications, from MP3 devices including mobile phone handsets to games consoles, LCD-TVs, and home cinema. The ace in the deck for class D is its vastly superior efficiency, which can be as high as 85 to 90% in practice. A linear class-AB implementation will usually achieve around 25% at typical listening levels.

In handheld applications, the low power dissipation of class D allows designers to combine high audio performance with a long battery-recharge interval. Battery life is a key figure of merit for all personal communication and audio devices. For mains-powered equipment, such as audio-visual (AV) products and games consoles, the high power efficiency of class D brings the advantages of smaller power supplies and reduced heat dissipation. Hence, designers can specify smaller heatsinks to achieve lower profile styles as well as lower bill-of-materials and assembly costs. In fact, careful design of the power supply can allow heatsink-less operation up to several watts per channel of output power.

Amplifier Chip Solution
In a class D amplifier, the audio signal is compared with a sawtooth wave switching at a frequency very much higher than the audio range (Fig. 1) . The result is a pulse-width-modulated square wave of period equal to that of the sawtooth. The pulse width represents a sample of the audio signal. The PWM square wave — and its inverse — then drives a MOSFET output stage (usually an H-bridge) to create an amplified version of the square-wave sample. After that, the sample is filtered by a low-pass filter to regain the amplified audio signal.

Increasing the switching frequency incurs greater losses in the output stage due to MOSFET gate capacitance, but brings advantages by reducing the output filtering requirements and increasing the audio signal-to-noise ratio (SNR). This is because a higher switching frequency increases the effective resolution of a PWM modulator, much the same way as oversampling in a sigma-delta modulator. Further performance improvements can be achieved using noise-shaping techniques. With the Wolfson WM8608 class D amplifier IC, for example, signal -to-noise ratios greater than 100 dB (A-weighted) have been demonstrated at a pulse frequency of 384 kHz — eight times the 48-kHz sampling rate.

Keeping the internal clock clean is crucial, as jitter causes random variations in the timing of the PWM signal edges that produce noise in the analog output. The clock is therefore generated from the system master clock by an on-chip, low-noise phase-locked loop (PLL). This will remove most jitter, but the master clock must also be reasonably clean. Therefore, generating the master clock inside the class D amplifier IC is advisable, too. It will prevent possible corruption by interference from the switching output stages or other sources by keeping the connection between the oscillator and PLL on-chip. This also eliminates external PLL filter components. A decoupling filter inserted close to the supply pin keeps noise out of the 3.3-V analog supply powering the PLL.

Designing the Power Stage
The design of the power bridge (Fig. 2) depends upon the desired output power of the amplifier. For instance, class D ICs are available with headphone drivers or with drivers for loudspeakers — the design of the output stage is one of the key differences between these configurations. Amplifiers designed for use with loudspeakers can produce from less than 1 W up to several watts of output power without requiring a heatsink. These ICs enable a single-chip solution in many consumer applications, from portable media players to games consoles and some LCD-TVs. In most of these applications, particularly handheld products, a single-chip solution is essential.

However, for very high output power, a class D modulator IC can be combined with an external output stage built using fast-switching power MOSFETs. These may be discretes, or integrated in a separate IC. The modulator must provide a suitable pre-driver, and the output-stage MOSFETs must be optimized for digital audio operation. The "on" resistance (R <sub>on</sub>) of the power MOSFETs causes heating and lowers power efficiency, and should therefore be as low as possible. The MOSFET gate capacitance should also be small to minimise power dissipation and heating in the level shifter driving the MOSFET. For the same reasons, it's important that the level shifter's input capacitance be small as well. A high gate capacitance will also cause RC delays that slow down the switching of the transistors.

A less obvious potential issue is the matching of switching characteristics between transistors. If, for example, an NMOS device turns on significantly faster than its PMOS counterpart switches off, the "on" times of both devices may overlap for a short time on signal edges. With both devices conducting, the power supply is essentially short-circuited, leading to reduced power efficiency, increased heat dissipation, and possibly a dip in the supply voltage, which would distort the audio signal. To preserve signal integrity, the switching delay of the output stage (power MOSFETs plus level shifters) should be small compared to the minimum PWM pulse width.

Several manufacturers offer integrated output stages that can be connected directly to the outputs of a class D modulator IC. The stages, which generally include four matched power MOSFETs per channel, also manage the level-shifting of the PWM signal from 3.3 V at the amplifier outputs to a higher voltage capable of switching the power devices.

Filtering
The output from the class D MOSFET H-bridge is a square-wave representation of the audio signal. The switching frequency and its harmonics must be attenuated to prevent interference and ensure that the end product will pass electromagnetic-compatibility (EMC) certification. Low-pass filtering, with a cut-off frequency just above the audible band, is required. For a given filter design, a higher switching frequency will be more thoroughly suppressed. Alternatively, increasing the switching frequency makes it possible to achieve the same level of attenuation with smaller, less-expensive filter components.

On the other hand, capacitive MOSFET losses scale with switching frequency, thereby driving down efficiency and leading to increased power dissipation and associated thermal-management issues. Hence, in practice, the design of a class D amplifier IC's output stage is predicated on fabricating low-loss MOSFETs and setting the switching frequency low enough to meet the specified target for electromagnetic interference (EMI).

Filter-less connection to a loudspeaker, such as a cellular-phone speaker, is a distinct advantage in size- and cost-sensitive applications. When the class D output is physically close to the speaker, the parasitic resistance and inductance of the speaker coil may be used as a suitable R-L low-pass filter. As such, one inductor and one capacitor can be eliminated from each output connection. If the distance from the amplifier output to the speaker is longer, a small amount of additional inductance, in the form of a ferrite bead, will be required to improve EMC performance. Figure 3 compares class D output spectra with and without a ferrite bead.

Power-Supply Design
Unlike linear amplifiers, designers using class D amplifier ICs also must pay closer attention to the effect of power-supply behavior on audio-output quality. Because the class D output is a switching stage, effectively connecting the supply rail directly to the audio output, audio-band fluctuations in the supply will modulate the output signal directly. Therefore, designers must ensure high load-regulation in the audio band, or take steps to eliminate the effects of mains or audio-band ripple.

If load regulation needs to be improved, a number of manufacturers provide floating regulators that can be added to existing supplies. Using a separate regulator for each amplifier output has the extra benefit of reducing crosstalk between audio channels. However, an additional regulator, or pair of regulators, raises the overall cost of the implementation. Also, power dissipation in the voltage regulator offsets the efficiency gains that are the key justification for a class D implementation.

Alternatively, increasing the amplifier's power-supply rejection ratio (PSRR) reduces the effect of load regulation on the audio output signal. Adding feedback from the PWM output to the analog audio input raises the PSRR by compensating for supply voltage variations. This can achieve a PSRR of up to around 80 dB, which is very close to the PSRR of a differential class-AB amplifier for portable applications. If the class D input signal is digital, however, this technique can't be applied without first converting it to the analog domain. The PSRR of an all-digital class D amplifier is 0 dB, so the designer must ensure close regulation of the supply voltage.

The power supply's transient performance should also be considered. To reproduce the PWM waveform accurately, the power supply must be able to react quickly to sudden changes in current draw. A linear amplifier is much less demanding in this respect, since the bandwidth of the output stage is limited to the audio range. In a power supply for a class D amplifier, voltage fluctuations outside the audio band, resulting from poor transient response, will modulate the PWM signal and thus introduce harmonic distortion that can be heard in the audio output.

Of course, high-value capacitors can be used to deal with these fluctuations. But physically large capacitors aren't desirable in handheld products. On the other hand, high-value capacitors in small-outline packages are expensive.

A helpful technique is to have the MOSFETs in the different output stages switch at different times, thus reducing the peak supply current. For example, the Wolfson WM8608, a 5- to 7.1-channel digital power-amplifier controller, has a built-in "PWM output phase" function that introduces a short delay between the PWM signals for each output channel. This has the effect of spreading the switching transients around the PWM cycle (Fig. 4) , although the added delay is far too short to make an audible difference to the output. In a multichannel system with six channels, this technique significantly diminishes the maximum instantaneous load current and reduces crosstalk.

Switching Supply, Switching Amplifier
One potential concern with switched supplies is EMI, caused by the rapid switching of large currents. This problem is exacerbated when a switching supply and switching amplifier operate in the same system at different switching frequencies. Intermodulation produces tones that may be audible in the output. Synchronizing power-supply switching with that of the class D PWM modulator can eliminate this effect.

Alternatively, a class D amplifier may be powered from a regulated linear supply. This may be attractive where extremely low-cost targets dominate design, but switching power supplies are usually preferred for their high efficiency and small size.

Careful design of the power supply, with consideration for the amplifier PWM and output stage, can achieve SNR and THD performance comparable to most consumer analog amplifiers. With its inherent efficiency, size, and thermal-management advantages, as well as further developments that will enhance performance in the future, the class D amplifier will extend its domination in consumer audio products.