How to maximise power efficiency and minimise noise when using Class-D amplifiers

READ THIS TO FIND OUT ABOUT: The trade-offs involved in choosing and designing with Class-D amplifiers Design tips and recently released components that can reduce noise and improve performance of Class-D amplifier designs.

Class-D amplifiers have gained acceptance in the market because they operate much more efficiently than earlier technologies such as Class-A, Class-AB and Class-B amplifiers. But Class-D amplifiers achieve their efficiency because they operate in switched mode, not the linear mode used by Class-A and Class-B. This switched mode introduces sources of noise that can have surprising and dramatic effects and these will not necessarily be familiar to designers who have only previously had experience of the older types of amplifier. Andreas Koch, Technical Solutions Manager, Future Electronics (Germany) explains.

This article addresses the main sources of noise or distortion that arise in Class-D applications, including:

• Parasitic components that cause ringing.
• Influence on the power supply of reactive power from the load.
• Non-linear output filtering.
• Unwanted properties of switching MOSFETs.
• The modulator generating a non-linear PWM signal.
• Timing errors such as dead time, on/off and rise/fall time introduced by the gate drivers.

Previously, designers using Class-D amplifiers have been required to implement many difficult techniques to mitigate these sources of noise. Now, however, a new generation of components is available that reduces the amount of error that these noise sources produce, and make it far easier to implement noise-reduction techniques. Many do this by integrating noise-reduction features that designers would previously have had to design themselves using discrete devices.

 

 

Class-D amplifier operation

In a Class-D amplifier’s basic operation, the input signal modulates the duty cycle of a specific switching frequency much higher than the highest input frequency (see Figure 1). This PWM signal drives a power stage usually composed of a half-bridge or full-bridge MOSFET.

The output signal can then drive a speaker via a low-pass filter. In some low-power applications, the low-pass filtering is done by the inductance of the speaker itself resulting in a so-called filterless design. In this arrangement, every inaccuracy of the components results in Total Harmonic Distortion and Noise (THD+N).

Several approaches to reducing the influence of the undesirable behaviour of real-world components have emerged. Adding feedback is the most common means of enhancing the audio quality of Class-D amplifiers. This closed-loop method can be accomplished both by complete analogue designs and by those which use digitally-controlled power stages aided by Digital Signal Processors (DSPs).

 

How do noise and distortion arise in a Class-D amplifier?

The modulator has to translate the input signal into a corresponding duty cycle. Both analogue and digital modulators have their disadvantages.

The analogue technique needs a triangle oscillator and a comparator. The waveform of the oscillator is the reference for the PWM. Every deviation from the ideal duty cycle or linearity of the triangle results in non-linearity at the output. The comparator also introduces errors to the modulated signal. Offset can easily be cancelled out, but non-linearity and noise are directly fed into the output signal.

Digital modulation poses a wholly different challenge: the limited resolution of the PWM. The duty cycle is derived from a clock signal, which cannot be arbitrarily high. For example, to achieve 16-bit resolution at a PWM frequency of 500kHz, the clock frequency has to be:

On the other hand, at a given clock frequency of 100MHz, the resolution will be less than 8 bits:

The feedback circuit also introduces distortion, arising from the noise and non-linearity produced by the operational amplifiers that such circuits use. In addition, phase shift due to the low-pass filtering of the switching frequency hampers stability.

The power stage is critical because of the introduction of timing errors. If the high-side and the low-side transistors are in the on-state at the same time, a shoot-through condition occurs. This results in at least excessive heating, at worst in the breakdown of the amplifier. To prevent this, dead time is added to avoid shoot through. Unfortunately, dead time results in the non-linear behaviour of the duty cycle of the PWM and, therefore, a non-linear output signal. The output filter is an important circuit that the signal has to pass through. Its most important parameter is frequency response. As the cut-off frequency and the filter characteristic depend on the load impedance, poorlymatched loads might hamper the flatness of the audible frequencies.

In addition, passive components are not free of distortion. In particular, powdered iron or ferrite cores introduce distortion due to hysteretic effects.

 

 

More efficient power stage

It is apparent, therefore, that there are many potential sources of noise in Class-D amplifier systems. But three significant advances in Class-D amplifiers are promising substantial reductions in THD+N. These latest components provide a more efficient power stage that reduces dead time, integrated closed-loop feedback circuits that save the user from having to design these difficult and time-consuming functions, and digital modulation to reduce the use of lossy analogue components.

Timing is essential for good THD+N performance and for reliability. To this end, International Rectifier has introduced a new Class-D audio chipset comprised of the IRS20955 200V digital audio driver and the IRFI4024Hx series of digital audio MOSFETs (see Figure 2). The driver was designed for Class-D audio amplifiers up to 500W; while the MOSFETs reduce the power-switch part count of the Class-D stage by 50% for the entire mid-voltage range of mid- and high-power amplifiers. This can be significant in home-theatre applications, professional amplifiers, musical instruments and car entertainment.

Crucially, this new device offers preset internal dead-time generation to enable accurate and stable gate-switch timing, while delivering optimum dead-time settings for improved THD performance and high noise immunity.

 

Integrated noise-reduction features

In the field of analogue modulation, National Semiconductor now offers a variety of integrated low-power Class-D amplifiers. In the LM4675, a filterless design is implemented to realise extremely simple and cheap designs for lowpower applications such as mobile phones, PDAs and other portable electronic devices.

 

 

The filterless circuit reduces errors added by the imperfect passive components in the output filter. The electrical model of the speaker (see Figure 3) shows that it behaves as a low-pass filter. In addition, due to the characteristics of the human ear, which cannot recognise any frequencies above 20kHz, there will be no audible noise emissions caused by the switching frequency.

Beyond eliminating the output filter, early Class-D amplifiers typically operated in open-loop mode (that is, with no feedback path). In open-loop operation the system cannot cancel out any error introduced by comparator offset, device mismatch, oscillator jitter or finite rise time of the Class-D output, along with noise contributed by the power supply.

Thus, these amplifiers exhibited poor THD+N of greater than 0.5%, and almost non-existent, 0dB, AC Power Supply Rejection Ratio (PSRR). Because Class-D is a switching amplifier topology, the feedback path consisted of more than just one or two passive components. Operational amplifiers with extensive RC networks were required for filtering and differential-to-single-ended conversion. Unlike the signal path of a linear amplifier, a Class-D signal path includes a delay associated with the conversion from a linear input to a PWM output. This conversion delay further complicated the design of the feedback loop.

Previously, the burden of applying feedback was placed on the system designer. Amplifier manufacturers suggested externalfeedback architectures and provided guidance, but the success of the final amplifier rested on the expertise of the system engineer, not that of the amplifier manufacturer. Such external feedback topologies, though effective, increased component count, board space, and cost, as well as system complexity.

 

 

The latest Class-D amplifiers feature integrated feedback. The LM4675, in particular, features global feedback (see Figure 4). The input signal to the error amplifier is taken after the H-bridge therefore reducing any mismatch, jitter, finite rise/fall time, or supply noise present at any point in the amplifier signal path.

It also allows the introduction of spread-spectrum techniques to reduce Electro-Magnetic Interference (EMI). This means the switching frequency varies by ±30% about a 300kHz centre frequency, reducing the wideband spectral content, thus improving EMI emissions from the speaker and associated cables and traces.

 

 

Digital modulation eliminates lossy analogue components

In the digital domain, Freescale Semiconductor has launched the new Symphony digital audio-amplifier controller, which features an innovative digital feedback technology. The company’s complete Class-D solution consists of the Symphony FSA95601 digital-amplifier controller chip and the Freescale MC33851 output power-stage chip (see Figure 5).

The Symphony FSA95601 is a high-performance, six-channel Class-D digital amplifier controller that can process up to six channels of Pulse-Code Modulation (PCM) digital audio input data to produce the corresponding PWM outputs to drive external power stages. The Symphony solution also accepts a digital input directly from the DSP, which eliminates the need for a DAC for each channel of audio. This can significantly improve audio quality by removing unnecessary conversion steps, and can help reduce cost by eliminating components.

 

 

The FSA95601 controller feedback system compares the output of the switching power stage with an internally-generated ideal PWM signal for the corresponding channel to identify errors (see Figure 6). The resulting error is converted into the digital domain by an analogueto- digital converter. Compensation for the digitised error component is mixed with the original PWM signal to make any necessary corrections for timing or amplitude errors.

The Freescale MC33851 two-channel H-bridge power stage is designed for applications requiring 50W to 100W (bridged) per channel. When used with the Symphony FSA95601, the MC33851 slew-rate control can reduce EMI significantly with no audio-quality degradation.

Also offering integrated closed-loop feedback, Direct Digital Feedback Amplifier (DDFA™) technology from Zetex enables switchingamplifier solutions capable of producing a sound quality to challenge that of the best linear amplifiers.

 

 

Comprising a multi-channel digital modulator (ZXCZM800) and a feedback processor (ZXCZA200), the DDFA™ chipset (see Figure 7) combines with Class-D output stages to create a high-performance amplifier achieving a THD+N of less than 0.004% and dynamic range of 120dB with frequency.

The DDFA™ architecture compensates for the limitations of FET output bridges at high powers, accommodates non-linearities in output filters and is tolerant to high levels of power-supply noise, allowing the use of unregulated supplies.

 

Conclusion

To improve the performance of Class-D designs in terms of THD+N, two major developments have become extremely influential. The first is the use of improved power stages: new drivers offer reduced tolerances, and MOSFETs offer lower gate-charge and stray-inductance specifications. The second development is the use of advanced feedback techniques. Particularly significant is the introduction of highly-integrated feedback mechanisms, which have taken a large and technically difficult task away from the system designer.

As a result, using the latest Class-D amplifiers allows designers to create systems that provide higher audio quality than ever before in less time than was previously possible..