Class D Amplifiers – Not 'Just Audio'
by Steve Somers, Vice President of Engineering
Yes, I'm a video guy. You've probably heard one of us (not me) disrespectfully downplay the ubiquitous field of aural engineering as "it's just audio". This article is about audio, but don't energize your shields just yet. I'm on your side... veteran project manager of an audio system having CD quality signal-to-noise ratios and immeasurable harmonic distortion. Though I am fundamentally programmed to refresh my memory at 30 frames per second (or my mind blanks), I really have embarked on occasion to that holiest of realms in search of greater dynamic range.
Today, audio is about more than dynamic range or Total Harmonic Distortion (THD). It's also about power... more of it for less. Class D audio systems have come into their own as a practical choice whether you need more audio power from a small system or more audio power from a small battery.
There are five "classes" of amplifiers: A, B, AB, C, and D. It's helpful to know where we've been in order to understand where we are going. Let's review. The class A amplifier is the traditional, fully linear amplifier with active circuit elements biased into their linear operating region. This means that the region must have enough voltage range to encompass the entire dynamic excursion — amplitude — of an incoming signal in order to reproduce it without clipping or compressing at either extreme. For this reason, the amplifier's output power supply voltage must equal, roughly, 200% of the maximum output signal swing expected. Signal amplitudes reaching the nonlinear region become distorted. This method of operation is pure, but inefficient. Class A amplifiers rarely exceed 20% efficiency in terms of power consumed (converted to heat) versus power delivered to the load.
Class B amplifiers are somewhat more efficient by utilizing two drive elements operating in a push-pull configuration. On the positive excursion of the signal, the upper element supplies power to the load while the lower is turned off. During negative going signal excursions, the opposite operation occurs. This increases operating efficiency, but suffers from the nonlinear turn-on, turn-off region created where the driver elements switch from their ON state to their OFF state. This switching error creates a condition commonly called cross-over distortion.
Class AB amplifiers remedy cross-over distortion to a great degree by combining the best features of both classes. The push-pull drivers are carefully biased just above their fully OFF state so that the transition between drivers is smoother. Therefore, each driver is never completely turned OFF. This alleviates most of the cross-over distortion at the expense of efficiency. A temperature-compensated bias network is required in close proximity to the output devices. An AB amplifier is still more efficient (60 - 65%) than a Class A amplifier. However, amplifier efficiency numbers are usually derived from application of steady-state sine wave tones having a low crest factor. When taking the crest factor (ratio of peak signal to rms signal) of real signals into account, the efficiency of either class A or AB drops to barely 20% at best.
Class C amplifiers, biased at or below cutoff, are commonly used for certain types of RF transmission, but not commonly used in audio applications. Therefore, this article will not dwell on class C.
Figure 1. Class D amplifier compares analog audio to triangle wave to create pulse width modulation.
D Does NOT Equal Digital
Class D amplifiers are not digital in the true sense. They are not driven directly by coherent binary data. They do act digitally in that the output drivers operate either in the fully ON-region or fully OFF-region. Think of Class D amps as being similar to a switch-mode power supply, but with audio signals modulating the switching action.
A switch-mode power supply uses pulse-width modulation (PWM) to control the on/off duty cycle of the power switching transistor(s) providing power to a load. The efficiency is high because there is little voltage drop across the switch transistor during conduction. This means very low power dissipation in the switch while virtually all the power is transferred to the load. During the OFF period, there is essentially zero current flow. The quality and speed of MOSFET (metal oxide semiconductor field effect transistor) devices has led to compact, efficient, high frequency power supplies. Switch-mode power supplies are more efficient at high frequencies. At higher operating frequencies, components may become smaller and the power supply becomes very compact for the power delivered. In addition, the output filter components may be much smaller. Today, switching frequencies over 1 MHz are not uncommon. But, as you probably know, switch-mode supplies generate considerable noise.
What does this have to do with audio? Audio signals can be used to modulate a PWM system to create a high power audio amplifier at nominal voltages using small components. Class D audio utilizes a fixed, high frequency carrier having pulses that vary in width based on signal amplitude. Class D amplifiers reach efficiencies as high as 90%. This is of great importance to portable applications relying on battery power. Class D portable, battery-powered audio gear may have battery life extended by 2.5 times or more.
Saving electrical power is now becoming a concern. Equipment utilizing Class D systems save significant operating power. For equipment having a limited power budget or available voltage range, Class D can get the job done without redesigning power supplies for more signal headroom. Sound like a system fraught with poor performance? I think you'll be pleasantly surprised at the quality. Let's take a closer look at how it works.
Time to Check Under the Hood
In its most common implementation, the incoming audio signal amplitude is compared to a triangle waveform operating at the intended switching frequency. The comparator circuit switches its output high or low based on the incoming audio's threshold against the reference amplitude and frequency of the triangular wave. As the audio signal rises above the comparator threshold, the comparator switches on and remains on for the duration of time where the audio exceeds the reference level, thus creating a wide, positive-going pulse width. Conversely, while the audio signal is below the comparator's reference level, the negative-going portion of the output pulse duty cycle is wider. Refer to Figure 1 to see this relationship.
Some describe this conversion method as a 1-bit A/D converter. A big advantage this conversion method has is linearity. The relationship of audio amplitude to the variable width pulses in this system is perfectly linear. This fixed frequency pulse train becomes a carrier for the audio.
The outputs of the comparator connect to gate drive circuitry for the MOSFET output transistors. Typically, the comparator has complementary outputs and drives two sets of "totem-poled" transistor switches. This configuration along with the point of connection for the loudspeaker describes an output drive topology known as an "H" configuration or a "bridge-tied" load. See Figure 2 for a more simplified diagram of Class D topology.
Figure 2. Class D topology
The final, and no less important, section is the output filter. Most Class D designs utilize a Butterworth filter scheme for simplicity and low cost. The output filter is essential for low pass filtering, or integrating, the carrier's varying pulse duty cycle into the original audio content while attenuating (absorbing) the switching carrier frequency. Selection of filter component values is very important and essential to maximizing efficiency.
Dynamic range is attained by selection of the carrier switching frequency. A factor of at least 12 times the upper audio cutoff frequency is recommended. This means that the minimum switching frequency will be about 250 KHz. At rest (no signal), the duty cycle of the switching frequency is 50%, or evenly divided between ON and OFF. Interestingly, the no signal input state is the most stressful for a Class D design. Positive going signal peaks drive the duty cycle one direction and negative going peaks drive it the opposite direction. Thus, the higher the switching frequency, the more "bits" of resolution available for signal reproduction.
Interesting...How About Quality?
Class D amplifiers have been criticized as lower quality than Class AB systems with use limited to lower performance applications such as public address systems. Through recent advances in power semiconductor devices and the need for better efficiency under battery power, Class D now sees a resurgence of interest. It is now possible to develop a Class D design that rivals most AB amplifiers. For example, look at the frequency response of the Extron Class D system in Figure 3. Now, compare its signal-to-noise performance against that of a typical class AB amplifier (Figures 4 and 5). Note the closeness of performance between the two classes while the Extron design pumps 67% more power into the same load. Finally, Figure 6 illustrates very respectable total harmonic distortion (THD) performance that is very competitive with class AB. It's also interesting to note that, at full output power, the Class D output switch transistor heatsink is just warm to the touch. Its power supply voltages are one half the level needed by the Class AB device.
Class D Advantage
The largest advantage is in efficiency. Improved efficiency translates into lower system cost, lower operating temperatures, lower power supply voltages, and lower power consumption. In addition, Class D building blocks are readily available along with significant design support for rapid implementation into new product designs. While real operating efficiency in Class AB amplifiers struggles around 20%, Class D systems attain 75% efficiency without significant effort. Higher efficiencies are possible depending on details of the design with higher power (around 100 watts or more) amplifiers actually attaining higher efficiencies than their low power relatives.
Figure 3. Bandpass response at level 25 watts into 8 ohms.
Figure 4. Class AB S/N performance at 15 watts into 8 ohms.
Figure 5. Extron Class D S/N performance at 25 watts into 8 ohms.
Figure 6. Total harmonic distortion (THD) at 25 watts into 8 ohms. Note that below 49 Hz amplifier bandpass is intended to roll off.
OK, What's the Catch?
Competing with Class AB designs in the name of efficiency does carry a few caveats. Of three critical design features, the output filter rates number one. The output filter reconstructs the original audio signal and attenuates the switching carrier frequency. It also sets the amplifier's -3 dB bandwidth. In designing the output filter, it is important to select filter topology and component values such that the switching frequency is sufficiently attenuated while the audio band is not significantly affected. Some residual carrier is always present after the filter. The newcomer to Class D will not see a quiet, no-signal condition at the speaker terminals. Some of the efficiency loss in Class D is the result of the output filter design.
Because of high frequency operation, power supply decoupling is very important. The switching carrier must be removed from all supply voltages to prevent it from degrading circuit operation. Finally, good high frequency circuit board layout technique is essential to minimizing EMI generation. As power level increases, switching currents traveling in high impedance board traces will generate significant electrical noise.
A Brave New Class
Advances in the electronic art are more interdisciplinary than ever before. Class D audio applications require a wide breadth of design knowledge and technique. Maintaining audio signal purity is the Holy Grail during this orchestration of high-speed energy from the chaotic into the realm of orderly, efficient signal reproduction. Class D audio systems are rapidly approaching the expectations of the audiophile and winning the praises of energy efficiency experts alike. Today, should any design engineer dismiss audio systems as unchallenging? I think not.