Class D Audio Amplifiers: What, Why, and How

by Eric Gaalaas Download PDF

Class D amplifiers, first proposed in 1958, have become increasingly popular in recent years. What are Class D amplifiers? How do they compare with other kinds of amplifiers? Why is Class D of interest for audio? What is
needed to make a “good” audio Class D amplifier? What are the features of ADI’s Class D amplifier products? Find the answers to all these questions in the following pages.

Audio Amplifier Background

The goal of audio amplifiers is to reproduce input audio signals at sound-producing output elements, with desired volume and power levels—faithfully, efficiently, and at low distortion. Audio frequencies range from about
20 Hz to 20 kHz, so the amplifier must have good frequency response over this range (less when driving a band-limited speaker, such as a woofer or a tweeter). Power capabilities vary widely depending on the application,
from milliwatts in headphones, to a few watts in TV or PC audio, to tens of watts for “mini” home stereos and automotive audio, to hundreds of watts and beyond for more powerful home and commercial sound systems—
and to fill theaters or auditoriums with sound.

A straightforward analog implementation of an audio amplifier uses transistors in linear mode to create an output voltage that is a scaled copy of the input voltage. The forward voltage gain is usually high (at least 40 dB). If
the forward gain is part of a feedback loop, the overall loop gain will also be high. Feedback is often used because high loop gain improves performance—suppressing distortion caused by nonlinearities in the forward path
and reducing power supply noise by increasing the power-supply rejection (PSR).

The Class D Amplifier Advantage

In a conventional transistor amplifier, the output stage contains transistors that supply the instantaneous continuous output current. The many possible implementations for audio systems include Classes A, AB, and B.
Compared with Class D designs, the output-stage power dissipation is large in even the most efficient linear output stages. This difference gives Class D significant advantages in many applications because the lower power
dissipation produces less heat, saves circuit board space and cost, and extends battery life in portable systems.

Linear Amplifiers, Class D Amplifiers, and Power Dissipation

Linear-amplifier output stages are directly connected to the speaker (in some cases via capacitors). If bipolar junction transistors (BJTs) are used in the output stage, they generally operate in the linear mode, with large
collector-emitter voltages. The output stage could also be implemented with MOS transistors, as shown in Figure 1.
 Figure 1. CMOS linear output stage.
Power is dissipated in all linear output stages, because the process of generating VOUT unavoidably causes nonzero IDS and VDS in at least one output transistor. The amount of power dissipation strongly depends on the method
used to bias the output transistors.

The Class A topology uses one of the transistors as a dc current source, capable of supplying the maximum audio current required by the speaker. Good sound quality is possible with the Class A output stage, but power
dissipation is excessive because a large dc bias current usually flows in the output-stage transistors (where we do not want it), without being delivered to the speaker (where we do want it).

The Class B topology eliminates the dc bias current and dissipates significantly less power. Its output transistors are individually controlled in a push-pull manner, allowing the MH device to supply positive currents to the
speaker, and ML to sink negative currents. This reduces output stage power dissipation, with only signal current conducted through the transistors. The Class B circuit has inferior sound quality, however, due to nonlinear
behavior (crossover distortion) when the output current passes through zero and the transistors are changing between the on and off conditions.

Class AB, a hybrid compromise of Classes A and B, uses some dc bias current, but much less than a pure Class A design. The small dc bias current is sufficient to prevent crossover distortion, enabling good sound quality.
Power dissipation, although between Class A and Class B limits, is typically closer to Class B. Some control, similar to that of the Class B circuit, is needed to allow the Class AB circuit to supply or sink large output currents.

Unfortunately, even a well-designed class AB amplifier has significant power dissipation, because its midrange output voltages are generally far from either the positive or negative supply rails. The large drain-source voltage
drops thus produce significant IDS × VDS instantaneous power dissipation.

Thanks to a different topology (Figure 2), the Class D amplifier dissipates much less power than any of the above. Its output stage switches between the positive and negative power supplies so as to produce a train of voltage
pulses. This waveform is benign for power dissipation, because the output transistors have zero current when not switching, and have low VDS when they are conducting current, thus giving smaller IDS × VDS.

Figure 2. Class D open-loop-amplifier block diagram.

Since most audio signals are not pulse trains, a modulator must be included to convert the audio input into pulses. The frequency content of the pulses includes both the desired audio signal and significant high-frequency
energy related to the modulation process. A low-pass filter is often inserted between the output stage and the speaker to minimize electromagnetic interference (EMI) and avoid driving the speaker with too much high
frequency energy. The filter (Figure 3) needs to be lossless (or nearly so) in order to retain the power-dissipation advantage of the switching output stage. The filter normally uses capacitors and inductors, with the only
intentionally dissipative element being the speaker.
 Figure 3. Differential switching output stage and LC low-pass filter.
Figure 4 compares ideal output-stage power dissipation (PDISS) for Class A and Class B amplifiers with measured dissipation for the AD1994 Class D amplifier, plotted against power delivered to the speaker (PLOAD), given an
audio-frequency sine wave signal. The power numbers are normalized to the power level, PLOAD max, at which the sine is clipped enough to cause 10% total harmonic distortion (THD). The vertical line indicates the PLOAD at
which clipping begins.

Figure 4. Power dissipation in Class A, Class B, and Class D output stages.

Significant differences in power dissipation are visible for a wide range of loads, especially at high and moderate values. At the onset of clipping, dissipation in the Class D output stage is about 2.5 times less than Class B,
and 27 times less than Class A. Note that more power is consumed in the Class A output stage than is delivered to the speaker—a consequence of using the large dc bias current.

Output-stage power efficiency, Eff, is defined as

At the onset of clipping, Eff = 25% for the Class A amplifier, 78.5% for the Class B amplifier, and 90% for the Class D amplifier (see Figure 5). These best-case values for Class A and Class B are the ones often cited in
textbooks.
 Figure 5. Power efficiency of Class A, Class B, and Class D output stages.
The differences in power dissipation and efficiency widen at moderate power levels. This is important for audio, because long-term average levels for loud music are much lower (by factors of five to 20, depending on the
type of music) than the instantaneous peak levels, which approach PLOAD max. Thus, for audio amplifiers, [PLOAD = 0.1 × PLOAD max] is a reasonable average power level at which to evaluate PDISS. At this level, the Class D output-
stage dissipation is nine times less than Class B, and 107 times less than Class A.

For an audio amplifier with 10-W PLOAD max, an average PLOAD of 1 W can be considered a realistic listening level. Under this condition, 282 mW is dissipated inside the Class D output stage, vs. 2.53 W for Class B and 30.2
W for Class A. In this case, the Class D efficiency is reduced to 78%—from 90% at higher power. But even 78% is much better than the Class B and Class A efficiencies—28% and 3%, respectively.

These differences have important consequences for system design. For power levels above 1 W, the excessive dissipation of linear output stages requires significant cooling measures to avoid unacceptable heating—typically
by using large slabs of metal as heat sinks, or fans to blow air over the amplifier. If the amplifier is implemented as an integrated circuit, a bulky and expensive thermally enhanced package may be needed to facilitate heat
transfer. These considerations are onerous in consumer products such as flat-screen TVs, where space is at a premium—or automotive audio, where the trend is toward cramming higher channel counts into a fixed space.

For power levels below 1 W, wasted power can be more of a difficulty than heat generation. If powered from a battery, a linear output stage would drain battery charge faster than a Class D design. In the above example, the
Class D output stage consumes 2.8 times less supply current than Class B and 23.6 times less than Class A—resulting in a big difference in the life of batteries used in products like cell phones, PDAs, and MP3 players.

For simplicity, the analysis thus far has focused exclusively on the amplifier output stages. However, when all sources of power dissipation in the amplifier system are considered, linear amplifiers can compare more
favorably to Class D amplifiers at low output-power levels. The reason is that the power needed to generate and modulate the switching waveform can be significant at low levels. Thus, the system-wide quiescent dissipation
of well-designed low-to-moderate-power Class AB amplifiers can make them competitive with Class D amplifiers. Class D power dissipation is unquestionably superior for the higher output power ranges, though.

Class D Amplifier Terminology, and Differential vs. Single-Ended Versions

Figure 3 shows a differential implementation of the output transistors and LC filter in a Class D amplifier. This H-bridge has two half-bridge switching circuits that supply pulses of opposite polarity to the filter, which
comprises two inductors, two capacitors, and the speaker. Each half-bridge contains two output transistors—a high-side transistor (MH) connected to the positive power supply, and a low-side transistor (ML) connected to the
negative supply. The diagrams here show high-side pMOS transistors. High-side nMOS transistors are often used to reduce size and capacitance, but special gate-drive techniques are required to control them (Further Reading
1).
Full H-bridge circuits generally run from a single supply (VDD), with ground used for the negative supply terminal (VSS). For a given VDD and VSS, the differential nature of the bridge means that it can deliver twice the output
signal and four times the output power of single-ended implementations. Half-bridge circuits can be powered from bipolar power supplies or a single supply, but the single-supply version imposes a potentially harmful dc bias
voltage, VDD/2, across the speaker, unless a blocking capacitor is added.

The power supply voltage buses of half-bridge circuits can be “pumped” beyond their nominal values by large inductor currents from the LC filter. The dV/dt of the pumping transient can be limited by adding large
decoupling capacitors between VDD and VSS. Full-bridge circuits do not suffer from bus pumping, because inductor current flowing into one of the half-bridges flows out of the other one, creating a local current loop that
minimally disturbs the power supplies.

Factors in Audio Class D Amplifier Design

The lower power dissipation provides a strong motivation to use Class D for audio applications, but there are important challenges for the designer. These include:

 Choice of output transistor size

 Output-stage protection
 Sound quality
 Modulation technique
 EMI
 LC filter design
 System cost

Choice of Output Transistor Size

The output transistor size is chosen to optimize power dissipation over a wide range of signal conditions. Ensuring that VDS stays small when conducting large IDS requires the on resistance (RON) of the output transistors to be
small (typically 0.1 ohm to 0.2 ohm). But this requires large transistors with significant gate capacitance (CG). The gate-drive circuitry that switches the capacitance consumes power—CV2f, where C is the capacitance, V is the
voltage change during charging, and f is the switching frequency. This “switching loss” becomes excessive if the capacitance or frequency is too high, so practical upper limits exist. The choice of transistor size is therefore a
trade-off between minimizing IDS × VDS losses during conduction vs. minimizing switching losses. Conductive losses will dominate power dissipation and efficiency at high output power levels, while dissipation is dominated
by switching losses at low output levels. Power transistor manufacturers try to minimize the RON × CG product of their devices to reduce overall power dissipation in switching applications, and to provide flexibility in the
choice of switching frequency.

Protecting the Output Stage

The output stage must be protected from a number of potentially hazardous conditions:

Overheating: Class D’s output-stage power dissipation, though lower than that of linear amplifiers, can still reach levels that endanger the output transistors if the amplifier is forced to deliver very high power for a long time.
To protect against dangerous overheating, temperature-monitoring control circuitry is needed. In simple protection schemes, the output stage is shut off when its temperature, as measured by an on-chip sensor, exceeds
a thermal-shutdown safety threshold, and is kept off until it cools down. The sensor can provide additional temperature information, aside from the simple binary indication about whether temperature has exceeded the
shutdown threshold. By measuring temperature, the control circuitry can gradually reduce the volume level, reducing power dissipation and keeping temperature well within limits—instead of forcing perceptible periods of
silence during thermal-shutdown events.
Excessive current flow in the output transistors: The low on resistance of the output transistors is not a problem if the output stage and speaker terminals are properly connected, but enormous currents can result if these
nodes are inadvertently short-circuited to one another, or to the positive or negative power supplies. If unchecked, such currents can damage the transistors or surrounding circuitry. Consequently, current-sensing output-
transistor protection circuitry is needed. In simple protection schemes, the output stage is shut off if the output currents exceed a safety threshold. In more sophisticated schemes, the current-sensor output is fed back into the
amplifier—seeking to limit the output current to a maximum safe level, while allowing the amplifier to run continuously without shutting down. In these schemes, shutdown can be forced as a last resort if the attempted
limiting proves ineffective. Effective current limiters can also keep the amplifier running safely in the presence of momentarily large transient currents due to speaker resonances.

Undervoltage: Most switching output stage circuits work well only if the positive power supply voltages are high enough. Problems result if there is an undervoltage condition, where the supplies are too low. This issue is
commonly handled by an undervoltage lockout circuit, which permits the output stages to operate only if the power supply voltages are above an undervoltage-lockout threshold.

Output transistor turn-on timing: The MH and ML output stage transistors (Figure 6) have very low on resistance. It is therefore important to avoid situations in which both MH and ML are on simultaneously, as this would
create a low-resistance path from VDD to VSS through the transistors and a large shoot-through current. At best, the transistors will heat up and waste power; at worst, the transistors may be damaged. Break-before-make control
of the transistors prevents the shoot-through condition by forcing both transistors off before turning one on. The time intervals in which both transistors are off are called nonoverlap time or dead time.

Figure 6. Break-before-make switching of output-stage transistors.

Sound Quality
Several issues must be addressed to achieve good overall sound quality in Class D amplifiers.

Clicks and pops, which occur when the amplifier is turning on or off can be very annoying. Unfortunately, however, they are easy to introduce into a Class D amplifier unless careful attention is paid to modulator state,
output-stage timing, and LC filter state when the amplifier is muted or unmuted.

Signal-to-noise ratio (SNR): To avoid audible hiss from the amplifier noise floor, SNR should typically exceed 90 dB in low-power amplifiers for portable applications, 100 dB for medium-power designs, and 110 dB for
high-power designs. This is achievable for a wide variety of amplifier implementations, but individual noise sources must be tracked during amplifier design to ensure a satisfactory overall SNR.

Distortion mechanisms: These include nonlinearities in the modulation technique or modulator implementation—and the dead time used in the output stage to solve the shoot-through current problem.

Information about the audio signal level is generally encoded in the widths of the Class D modulator output pulses. Adding dead time to prevent output stage shoot-through currents introduces a nonlinear timing error, which
creates distortion at the speaker in proportion to the timing error in relation to the ideal pulse width. The shortest dead time that avoids shoot-through is often best for minimizing distortion; see Further Reading 2 for a detailed
design method to optimize distortion performance of switching output stages.
Other sources of distortion include: mismatch of rise and fall times in the output pulses, mismatch in the timing characteristics for the output transistor gate-drive circuits, and nonlinearities in the components of the LC low-
pass filter.

Power-supply rejection (PSR): In the circuit of Figure 2, power-supply noise couples almost directly to the speaker with very little rejection. This occurs because the output-stage transistors connect the power supplies to the
low-pass filter through a very low resistance. The filter rejects high-frequency noise, but is designed to pass all audio frequencies, including noise. See Further Reading 3 for a good description of the effect of power-supply
noise in single-ended and differential switching output-stage circuits.

If neither distortion nor power-supply issues are addressed, it is difficult to achieve PSR better than 10 dB, or total harmonic distortion (THD) better than 0.1%. Even worse, the THD tends to be the bad-sounding high-order
kind.

Fortunately, there are good solutions to these issues. Using feedback with high loop gain (as is done in many linear amplifier designs) helps a lot. Feedback from the LC filter input will greatly improve PSR and attenuate all
non-LC-filter distortion mechanisms. LC filter nonlinearities can be attenuated by including the speaker in the feedback loop. Audiophile-grade sound quality with PSR > 60 dB and THD < 0.01% is attainable in well-
designed closed-loop Class D amplifiers.

Feedback complicates the amplifier design, however, because loop stability must be addressed (a nontrivial consideration for high-order design). Also, continuous-time analog feedback is necessary to capture important
information about pulse timing errors, so the control loop must include analog circuitry to process the feedback signal. In integrated-circuit amplifier implementations, this can add to the die cost.

To minimize IC cost, some vendors prefer to minimize or eliminate analog circuit content. Some products use a digital open-loop modulator, plus an analog-to-digital converter to sense power-supply variations—and adjust
the modulator’s behavior to compensate, as proposed in Further Reading 3. This can improve PSR, but will not address any of the distortion problems. Other digital modulators attempt to precompensate for expected output
stage timing errors, or correct for modulator nonidealities. This can at least partly address some distortion mechanisms, but not all. Applications that tolerate fairly relaxed sound-quality requirements can be handled by these
kinds of open-loop Class D amplifiers, but some form of feedback seems necessary for best audio quality.

Modulation Technique
Class D modulators can be implemented in many ways, supported by a large quantity of related research and intellectual property. This article will only introduce fundamental concepts.

All Class D modulation techniques encode information about the audio signal into a stream of pulses. Generally, the pulse widths are linked to the amplitude of the audio signal, and the spectrum of the pulses includes the
desired audio signal plus undesired (but unavoidable) high-frequency content. The total integrated high-frequency power in all schemes is roughly the same, since the total power in the time-domain waveforms is similar, and
by Parseval’s theorem, power in the time domain must equal power in the frequency domain. However, the distribution of energy varies widely: in some schemes, there are high energy tones atop a low noise floor, while in
other schemes, the energy is shaped so that tones are eliminated but the noise floor is higher.

The most common modulation technique is pulse-width modulation (PWM). Conceptually, PWM compares the input audio signal to a triangular or ramping waveform that runs at a fixed carrier frequency. This creates a
stream of pulses at the carrier frequency. Within each period of the carrier, the duty ratio of the PWM pulse is proportional to the amplitude of the audio signal. In the example of Figure 7, the audio input and triangular wave
are both centered around 0 V, so that for 0 input, the duty ratio of the output pulses is 50%. For large positive input, it is near 100%, and it is near 0% for large negative input. If the audio amplitude exceeds that of the triangle
wave, full modulation occurs, where the pulse train stops switching, and the duty ratio within individual periods is either 0% or 100%.
 Figure 7. PWM concept and example.
PWM is attractive because it allows 100-dB or better audio-band SNR at PWM carrier frequencies of a few hundred kilohertz—low enough to limit switching losses in the output stage. Also, many PWM modulators are
stable up to nearly 100% modulation, in concept permitting high output power—up to the point of overloading. However, PWM has several problems: First, the PWM process inherently adds distortion in many
implementations (Further Reading 4); next, harmonics of the PWM carrier frequency produce EMI within the AM radio band; and finally, PWM pulse widths become very small near full modulation. This causes problems in
most switching output-stage gate-driver circuits—with their limited drive capability, they cannot switch properly at the excessive speeds needed to reproduce short pulses with widths of a few nanoseconds. Consequently, full
modulation is often unattainable in PWM-based amplifiers, limiting maximum achievable output power to something less than the theoretical maximum—which considers only power-supply voltage, transistor on resistance,
and speaker impedance.

An alternative to PWM is pulse-density modulation (PDM), in which the number of pulses in a given time window is proportional to the average value of the input audio signal. Individual pulse widths cannot be arbitrary as
in PWM, but are instead “quantized” to multiples of the modulator clock period. 1-bit sigma-delta modulation is a form of PDM.

Much of the high-frequency energy in sigma-delta is distributed over a wide range of frequencies—not concentrated in tones at multiples of a carrier frequency, as in PWM—providing sigma-delta modulation with a potential
EMI advantage over PWM. Energy still exists at images of the PDM sampling clock frequency; but with typical clock frequencies from 3 MHz to 6 MHz, the images are outside the audio frequency band—and are strongly
attenuated by the LC low-pass filter.

Another advantage of sigma-delta is that the minimum pulse width is one sampling-clock period, even for signal conditions approaching full modulation. This eases gate-driver design and allows safe operation to theoretical
full power. Nonetheless 1-bit sigma-delta modulation is not often used in Class D amplifiers (Further Reading 4) because conventional 1-bit modulators are only stable to 50% modulation. Also, at least 64× oversampling is
needed to achieve sufficient audio-band SNR, so typical output data rates are at least 1 MHz and power efficiency is limited.

Recently, self-oscillating amplifiers have been developed, such as the one in Further Reading 5. This type of amplifier always includes a feedback loop, with properties of the loop determining the switching frequency of the
modulator, instead of an externally provided clock. High-frequency energy is often more evenly distributed than in PWM. Excellent audio quality is possible, thanks to the feedback, but the loop is self-oscillating, so it’s
difficult to synchronize with any other switching circuits, or to connect to digital audio sources without first converting the digital to analog.

The full-bridge circuit (Figure 3) can use “3-state” modulation to reduce differential EMI. With conventional differential operation, the output polarity of Half-bridge A must be opposite to that of Half-bridge B. Only two
differential operating states exist: Output A high with Output B low; and A low with B high. Two additional common-mode states exist, however, in which both half-bridge outputs are the same polarity (both high or both
low). One of these common-mode states can be used in conjunction with the differential states to produce 3-state modulation where the differential input to the LC filter can be positive, 0, or negative. The 0 state can be used
to represent low power levels, instead of switching between the positive and negative state as in a 2-state scheme. Very little differential activity occurs in the LC filter during the 0 state, reducing differential EMI, although
actually increasing common-mode EMI. The differential benefit only applies at low power levels, because the positive and negative states must still be used to deliver significant power to the speaker. The varying common-
mode voltage level in 3-state modulation schemes presents a design challenge for closed-loop amplifiers.

Taming EMI
The high-frequency components of Class D amplifier outputs merit serious consideration. If not properly understood and managed, these components can generate large amounts of EMI and disrupt operation of other
equipment.

Two kinds of EMI are of concern: signals that are radiated into space and those that are conducted via speaker- and power-supply wires. The Class D modulation scheme determines a baseline spectrum of the components of
conducted and radiated EMI. However, some board-level design techniques can be used to reduce the EMI emitted by a Class D amplifier, despite its baseline spectrum.

A useful principle is to minimize the area of loops that carry high-frequency currents, since strength of associated EMI is related to loop area and the proximity of loops to other circuits. For example, the entire LC filter
(including the speaker wiring) should be laid out as compactly as possible, and kept close to the amplifier. Traces for current drive and return paths should be kept together to minimize loop areas (using twisted pairs for the
speaker wires is helpful). Another place to focus is on the large charge transients that occur while switching the gate capacitance of the output-stage transistors. Generally this charge comes from a reservoir capacitance,
forming a current loop containing both capacitances. The EMI impact of transients in this loop can be diminished by minimizing the loop area, which means placing the reservoir capacitance as closely as possible to the
transistor(s) it charges.

It is sometimes helpful to insert RF chokes in series with the power supplies for the amplifier. Properly placed, they can confine high-frequency transient currents to local loops near the amplifier, instead of being conducted
for long distances down the power supply wires.

If gate-drive nonoverlap time is very long, inductive currents from the speaker or LC filter can forward-bias parasitic diodes at the terminals of the output-stage transistors. When the nonoverlap time ends, the bias on the
diode is changed from forward to reverse. Large reverse-recovery current spikes can flow before the diode fully turns off, creating a troublesome source of EMI. This problem can be minimized by keeping the nonoverlap
time very short (also recommended to minimize distortion of the audio). If the reverse-recovery behavior is still unacceptable, Schottky diodes can be paralleled with the transistor’s parasitic diodes, in order to divert the
currents and prevent the parasitic diode from ever turning on. This helps because the metal-semiconductor junctions of Schottky diodes are intrinsically immune to reverse-recovery effects.

LC filters with toroidal inductor cores can minimize stray field lines resulting from amplifier currents. The radiation from the cheaper drum cores can be reduced by shielding, a good compromise between cost and EMI
performance—if care is taken to ensure that the shielding doesn’t unacceptably degrade inductor linearity and sound quality at the speaker.

LC Filter Design
To save on cost and board space, most LC filters for Class D amplifiers are second-order, low-pass designs. Figure 3 depicts the differential version of a second-order LC filter. The speaker serves to damp the circuit’s
inherent resonance. Although the speaker impedance is sometimes approximated as a simple resistance, the actual impedance is more complex and may include significant reactive components. For best results in filter design,
one should always seek to use an accurate speaker model.

A common filter design choice is to aim for the lowest bandwidth for which droop in the filter response at the highest audio frequency of interest is minimized. A typical filter has 40-kHz Butterworth response (to achieve a
maximally flat pass band), if droop of less than 1 dB is desired for frequencies up to 20 kHz. The nominal component values in the table give approximate Butterworth response for common speaker impedances and standard
L and C values:
 Inductance L Capacitance C Speaker Bandwidth –3-dB
(μH) (μF) Resistance (Ohm) (kHz)

10 1.2 4 50

15 1 6 41

22 0.68 8 41

If the design does not include feedback from the speaker, THD at the speaker will be sensitive to linearity of the LC filter components.

Inductor Design Factors: Important factors in designing or selecting the inductor include the core’s current rating and shape, and the winding resistance.

Current rating: The core that is chosen should have a current rating above the highest expected amplifier current. The reason is that many inductor cores will magnetically saturate if current exceeds the current-rating
threshold and flux density becomes too high—resulting in unwanted drastic reduction of inductance.

The inductance is formed by wrapping a wire around the core. If there are many turns, the resistance associated with the total wire length is significant. Since this resistance is in series between the half-bridge and the speaker,
some of the output power will be dissipated in it. If the resistance is too high, use thicker wire or change the core to a different material that requires fewer turns of wire to give the desired inductance.

Finally, it should not be forgotten that the form of inductor used can affect EMI, as noted above.

System Cost
What are the important factors in the overall cost of an audio system that uses Class D amplifiers? How can we minimize the cost?

The active components of the Class D amplifier are the switching output stage and modulator. This circuitry can be built for roughly the same cost as an analog linear amplifier. The real trade-offs occur when considering
other components of the system.

The lower dissipation of Class D saves the cost (and space) of cooling apparatus like heat sinks or fans. A Class D integrated-circuit amplifier may be able to use a smaller and cheaper package than is possible for the linear
one. When driven from a digital audio source, analog linear amplifiers require D/A converters (DACs) to convert the audio into analog form. This is also true for analog-input Class D amplifiers, but digital-input types
effectively integrate the DAC function.
On the other hand, the principal cost disadvantage of Class D is the LC filter. The components—especially the inductors—occupy board space and add expense. In high-power amplifiers, the overall system cost is still
competitive, because LC filter cost is offset by large savings in cooling apparatus. But in cost-sensitive, low-power applications, the inductor expense becomes onerous. In extreme cases, such as cheap amplifiers for cell
phones, an amplifier IC can be cheaper than the total LC filter cost. Also, even if the monetary cost is ignored, the board space occupied by the LC filter can be an issue in small form-factor applications.

To address these concerns, the LC filter is sometimes eliminated entirely, to create a filterless amplifier. This saves cost and space, though losing the benefit of low-pass filtering. Without the filter, EMI and high-frequency
power dissipation can increase unacceptably—unless the speaker is inductive and kept very close to the amplifier, current-loop areas are minimal, and power levels are kept low. Though often possible in portable applications
like cell phones, it is not feasible for higher-power systems such as home stereos.

Another approach is to minimize the number of LC filter components required per audio channel. This can be accomplished by using single-ended half-bridge output stages, which require half the number of Ls and Cs needed
for differential, full-bridge circuits. But if the half-bridge requires bipolar power supplies, the expense associated with generating the negative supply may be prohibitive, unless a negative supply is already present for some
other purpose—or the amplifier has enough audio channels, to amortize the cost of the negative supply. Alternatively, the half-bridge could be powered from a single supply, but this reduces output power and often requires a
large dc blocking capacitor.

Analog Devices Class D Amplifiers

All of the design challenges just discussed can add up to a rather demanding project. To save time for the designer, Analog Devices offers a variety of Class D amplifier integrated circuits, incorporating programmable-gain
amplifiers, modulators, and power output stages. To simplify evaluation, demonstration boards are available for each amplifier type to simplify evaluation. The PCB layout and bill-of-materials for each of these boards serve
as a workable reference design, helping customers quickly design working, cost-effective audio systems without having to “reinvent the wheel” to solve the major Class D amplifier design challenges.

Consider, for example, the AD1990, AD1992, and AD1994—a family of dual-amplifier ICs, targeted at moderate-power stereo or mono applications requiring two channels with output-per-channel of up to 5-, 10-, and 25-
W, respectively. Here are some properties of these ICs:

The AD1994 Class D audio power amplifier combines two programmable-gain amplifiers, two sigma-delta modulators, and two power-output stages to drive full H-bridge-tied loads in home theater-, automotive-, and PC
audio applications. It generates switching waveforms that can drive stereo speakers at up to 25 W per speaker, or a single speaker to 50 W monophonic, with 90% efficiency. Its single-ended inputs are applied to a
programmable-gain amplifier (PGA) with gains settable to 0-, 6-, 12-, and 18 dB, to handle low-level signals.

The device has integrated protection against output-stage hazards of overheating, overcurrent, and shoot-through current. There are minimal clicks and pops associated with muting, thanks to special timing control, soft start,
and dc offset calibration. Specifications include 0.001% THD, 105-dB dynamic range, and >60 dB PSR, using continuous-time analog feedback from the switching output stage and optimized output stage gate drive. Its 1-bit
sigma-delta modulator is especially enhanced for the Class D application to achieve average data frequency of 500 kHz, with high loop gain to 90% modulation, and stability to full modulation. A standalone modulator mode
allows it to drive external FETs for higher output power.

It uses a 5-V supply for the PGA, modulator, and digital logic, and a high-voltage supply from 8 V to 20 V for the switching output stage. The associated reference design meets FCC Class B EMI requirements. When driving
6 ohm loads with 5-V and 12-V supplies, the AD1994 dissipates 487 mW quiescently, 710 mW at the 2 × 1-W output level, and 0.27 mW in power-down mode. Available in a 64-lead LFCSP package, it is specified from –
40°C to +85°C

More technical information about Class D amplifiers—including implementations with Blackfin processors—can be found in the Further Reading section.
 Acknowledgements
The author would like to thank Art Kalb and Rajeev Morajkar of Analog Devices for their thoughtful inputs to this article.

Further Reading
1. International Rectifier, Application Note AN-978, “HV Floating MOS-Gate Driver ICs.”
2. Nyboe, F., et al, “Time Domain Analysis of Open-Loop Distortion in Class D Amplifier Output Stages,” presented at the AES 27th International Conference, Copenhagen, Denmark, September 2005.
3. Zhang, L., et al, “Real-Time Power Supply Compensation for Noise-Shaped Class D Amplifier,” Presented at the 117th AES Convention, San Francisco, CA, October 2004.
4. Nielsen, K., “A Review and Comparison of Pulse-Width Modulation (PWM) Methods for Analog and Digital Input Switching Power Amplifiers,” Presented at the 102nd AES Convention, Munich, Germany, March
1997.
5. Putzeys, B., “Simple Self-Oscillating Class D Amplifier with Full Output Filter Control,” Presented at the 118th AES Convention, Barcelona, Spain, May 2005.
6. Gaalaas, E., et al, “Integrated Stereo Delta-Sigma Class D Amplifier,” IEEE J. Solid-State Circuits, vol. 40, no. 12, December 2005, pp. 2388-2397. About the AD199x Modulator.
7. Morrow, P., et al, “A 20-W Stereo Class D Audio Output Stage in 0.6 mm BCDMOS Technology,” IEEE J. Solid-State Circuits, vol. 39, no. 11, November 2004, pp. 1948-1958. About the AD199x Switching Output
Stage.
8. PWM and Class-D Amplifiers with ADSP-BF535 Blackfin® Processors, Analog Devices Engineer-to-Engineer Note EE-242.

///

How to Build a Class-D Power Amp

August 29, 2018 by Cezar Chirila

Learn how to build your own Class D power amplifier—one of the most efficient ways to listen to music.
The mighty Class-D amplifier—build one yourself and be amazed by its efficiency. The heat sink barely gets warm!
Have you always wanted to build your own audio power amplifier? An electronic project where you not only see the results but also hear them?

If your answer is yes, then you should continue reading this article on how to build your own Class D amplifier. I will explain to you how they work and then guide you step by step to
make the magic happen all by yourself.

Theoretical Basics
What is a Class-D audio power amplifier? The answer could be just a sentence long: It is a switching amplifier. But in order to fully understand how one works, I need to teach you all
its nooks and crannies.
Let's start with that first sentence. Traditional amplifiers, like the class AB, operate as linear devices. Compare this to switching amplifiers, so  called because the power transistors (the
MOSFETs) are acting like switches, changing their state from OFF to ON. This allows a very high efficiency, up to 80 - 95%. Because of this, the amplifier does not generate a lot of
heat and does not require a big heat sink like linear class AB amplifiers do. For comparison, the class B amplifier can only achieve a maximum efficiency of 78.5% (in theory).  

Below you can see the block diagram of a basic PWM Class-D amplifier, just like the one that we are building.
 

 
The input signal is converted into a pulse width modulated, rectangular signal using a comparator. This basically means that the input is encoded into the duty cycle of the
rectangular pulses. The rectangular signal is amplified, and then a low-pass filter results in a higher-power version of the original analog signal.
There are other methods for converting the signal into pulses, such as ΔΣ (delta-sigma) modulation, but for this project we will be using PWM.

Pulse-Width Modulation Using a Comparator

In the plot below, you can see how we transform a sinusoidal signal (the input) into a rectangular signal by comparing it to a triangle signal. 
 
 Click to enlarge

At the positive peak of the sine wave, the duty cycle of the rectangular pulse is 100% whilst at the negative peak it is 0%. The actual frequency of the triangle signal is much higher,
on the order of hundreds of kHz, so that we can later extract our original signal.

A real filter, not an ideal one, does not have a perfect "brick-wall" transition from passband to stopband, so we want the triangle signal to have a frequency at least 10 times higher
than 20KHz, which is the upper human hearing limit.

Power Stage—It All Sounds Good in Theory

Theory is one aspect and practice is another. If we want to put the previous block diagram into practice, we will stumble upon some problems.
Two issues are the rise and fall time of the devices in the power stage and the fact that we are using an NMOS transistor for the high-side driver.

 
  

Because the switching of the MOSFETs is not done instantaneously, but is more like going up and down a hill, the transistors' ON time will overlap, creating a low-
impedance  connection between the positive and negative power supply rails. This causes a high current pulse to pass through our MOSFETs, which can lead to failure.

To prevent this, we need to insert some dead-time between the signals that drive the high and low side MOSFETs. One way to achieve this is to use a specialized MOSFET driver from
International Rectifier (Infineon), such as the IR2110S or IR2011S. Furthermore, these ICs provide the boosted gate voltage needed for the high-side NMOS.  

 
Low-Pass Filter
For the filtering stage, one of the best ways to do this is to use a Butterworth filter.

 
  

These types of filters have a very flat response in the passband. This means that the signal that we want to achieve will not be attenuated too much.
We want to filter frequencies that are higher than 20 kHz. The cut-off frequency is calculated at -3dB, so we want it to be a bit higher in order to not filter sounds that we want to
hear. It is best to choose something between 40 and 60 kHz. The quality factor  Q = 1 √ 2 Q=12.
These are the formulas used to calculate the values of the inductor and the capacitor :

L = RL √ 22 ⋅ π ⋅ f c L=RL22⋅π⋅fc
 

C = 12 √ 2 ⋅ π ⋅ f c ⋅ RL C=122⋅π⋅fc⋅RL

Building Your DIY Amplifier (Luke-The-Warm)

Now that we know how a Class-D amplifier works, let's build one.
First of all, I named this amplifier Luke-The-Warm because the heat sink only barely gets warm, as opposed to a Class AB amplifier, whose heat sink can get quite hot if not actively
cooled.
Below you can see the schematic of the amplifier that I designed. It is based on the IRAUDAMP1  reference design by International Rectifier (Infineon). The main difference is that
instead of ΔΣ modulation, mine uses PWM.
 
 Click to enlarge

 
I will now tell you some design choices and how the components work with each other. Let's start from the left side.

Input Circuitry
For the input circuitry, I decided that it was best to use a high-pass filter followed by a low-pass filter. It is that simple.

Triangle Generator
For the triangle generator, I used an LMC555, which is the CMOS variant of the famous 555 chip. The charging and discharging of the capacitor produces a nice triangle, which is not
perfect (it rises and falls exponentially) but if the rise and fall times are equal, it works perfectly.

The values of the resistor and the capacitor set a frequency of approximately 200kHz. Any higher than this and we will run into trouble because the comparator and the MOSFET
driver are not the fastest devices.

 
 

Comparator
For the comparator, you can use whichever component you want—it just needs to be fast. I used what I had available, the LM393AP. At 300ns response time, it is not the fastest and
can definitely be improved but it does the job. If you want to use other ICs, just be careful to check that the pins match or you will have to modify the PCB design.
In theory, an op-amp can be used as a comparator, but in reality op-amps are designed for other types of work, so make sure you use an actual comparator.
Because we need two outputs from the comparator, one for the high-side driver and one for the low-side driver, I decided to use the LM393AP. This is two comparators in one
package, and we just swap the inputs for the second comparator. Another approach is to use a comparator that has two outputs, such as the LT1016 from Linear Technology. These
devices may offer somewhat improved performance, but they could also be more expensive.
These comparators are powered by a 5V bipolar supply, provided by two zener diodes that regulate voltage from the main power supply, which is ±30V.

 
 

MOSFET Driver
For the MOSFET driver, I chose to use the IR2110. An alternative is the IR2011, which is used in the reference design. This integrated circuit makes sure to add that dead time that I
talked about in the previous section.

Because the VSS pin of the IC is tied to the negative power supply, we need to level shift the signals from the comparator. This is done using PNP transistor and 1N4148 diodes.
To drive the MOSFETs, we power the IR2110 with 12V referenced to the negative power supply voltage; this voltage is generated using a BD241 in conjunction with a 12V zener. The
high side MOSFET needs to be driven by a gate voltage that is about 12V above the switching node, VS. This requires a voltage that is higher than the positive supply; the IR2110
provides this drive voltage with the help of our bootstrap capacitor, C10.

 
 

Filter
Finally the filter. The cut-off frequency is 40kHz, and the load resistance is 4 ohms because we have a 4-ohm speaker (the values used here will also work with an 8-ohm speaker, but
it is best to adjust the filter according to the speaker you choose). With this information we can calculate the values of the inductor and the capacitor:

L = 4 √ 22 ⋅ π ⋅ 40000H = 22.508 μ H L=422⋅π⋅40000H=22.508μH

 
We can safely round down to 22µH.

C = 12 √ 2 ⋅ π ⋅ 40000 ⋅ 4F = 0.703 μ H C=122⋅π⋅40000⋅4F=0.703μH

 
The closest standard value is 680nF. 

Notes on Build
Now that you know all about the inner workings, all you have to do is read very carefully the next few lines, download the files below, buy the components needed, etch the PCB, and
start assembling.

Low-Pass Filter
For the low-pass filter, you can use a 680nF capacitor to get as close as possible to the calculated value, but you can also use a 1µF capacitor without any trouble (I designed the PCB
so that you can use two capacitors in parallel to mix and match).
These capacitors need to be polypropylene or polyester—in general it's not a great idea to use ceramic capacitors with audio signals. And you need to make sure that the capacitors
that you are using for filtering are rated for high voltage, at least 100VAC (more doesn't hurt). The rest of the capacitors in the design also need to have an appropiate voltage rating.  
I designed this amplifier for an output power of about 100-150W. You should use a bipolar power supply with ±30V rails. You can go higher than this, but for voltages of about ±40V
you need to make sure that you change the values of the resistors R4 and R5 to 2K2. 
It is not necessary but highly recommended that you use a heatsink for BD241C as it gets quite hot. 

 
MOSFETs
As far as power MOSFETs go, I suggest using the IRF540N or the IRFB41N15D.  These MOSFETs have low gate charge for faster switching and low R (on) for lower power consumption.
DS

You also need to ensure that the MOSFET has an adequate maximum V  (drain-to-source voltage) rating. You could use the IRF640N, but the R (on) is significantly higher, leading to
DS DS

an amplifier with lower efficiency. Here is a table comparing these three MOSFETs: 

 
MOSFET Max VDS  (V) ID  (A) Qg (nC) RDS (on) (Ω)
IRFB41N15D 150 41 72 0.045
IRF540N 100 33 71 0.044
IRF640N 200 18 67 0.15

Inductor
Now the inductor. You can buy one already made but I would suggest that you wind your own—this is a DIY project after all.
Buy a T106-2 toroid. It needs to be iron powder; ferrite can work but it will need a gap or it will saturate. Using the said toroid, wind 40 turns of 0.8-1mm diameter (AWG20-18)
copper enameled wire. That's it. Don't worry if it isn't perfect—just make it tight.

Resistors
Finally, all the resistors, unless noted (R4, R5), are 1/4W.
 

Testing
When I designed the PCB, I made it so that it is very easy to test. The input signal has its own connector and there are two spade terminals for ground: one for the power supply and
one for the speaker.

To remove the hum noise (50/60 Hz, from the mains frequency), I used a star-ground configuration; this means connecting all grounds (amplifier ground, signal ground, and speaker
ground) at the same point, preferably on the power supply PCB, after the rectifier circuit.

The complete Bill of Materials can be found in the files below, where you can also find the PCB files both in PDF format and as KiCAD files. 

Goodies.zip

Final Thoughts
I hope that the information in this article is sufficient for you to build your own audio power amplifier. I hope it also gets you excited about building your own amplifier.

There are many things that can be improved in this project. You have all the necessary information and files, but you do not need to follow them to the letter.
You can use SMD components, improve the comparator circuit by using a complementary output one, or try the IR2011S instead of the IR2110. Just fire up that soldering iron, etch
your PCB, and start working. It does not matter if it does not work on the first try.
It's all about trial and error. When you will finally hear that crisp sound coming from your speaker, it will all be worth it.

If you have any trouble with your build, comment here or post on the forum using as much information as possible. We will work it out.
 

Give this project a try for yourself! Get the BOM.

RELATED CONTENT
 How to Ensure MultiGigabit Automotive Ethernet Compliance

 How to Defeat Heat with Peltier Devices

  How to Increase Slew Rate in Op Amps

 RF Power Ratio Conversion Calculator

 Infineon Solutions for Power Over Ethernet (PoE) Power Sourcing Equipment

 How to Use Thevenin’s Theorem

Learn More About:

 POWER AMPLIFIER
 AUDIO
 INTEGRATED CIRCUITS
 CLASS-D
 SWITCHING AMPLIFIER
 PWM

COMMENTS
75 Comments

Log in to comment

 dendadAugust 30, 2016

An interesting article, and well done to help understand these amps.

A couple of points though..
 I feel the “+12V” and “+5V2” would be better labeled as “-18V” and “-25V”.
As you have them it is rather confusing, even though you mention they are referenced to the -30V line.

 Like. Reply

o Cezar ChirilaAugust 31, 2016

You do have a valid point. However, I do not think that having -25V going out of a 5V regulator is right. I checked and the guys at Infineon (IRAUDAMP1 reference design) use the same type of notation for their
power supply.

 Like. Reply

 dendadOctober 24, 2016

It is perfectly ok to have "non 5Vs" coming out of a 5V regulator. All your voltages should be referenced to a common point. Having it as you show it, even if you are copping others, IS confusing. The regulator is
just a component, not the final design. By the argument you put forward, any variable power supply that uses, for example, a LM317 regulator would have a 1.25V output all the time as it is a 1.25V voltage
regulator. The LM317 is just a part of a circuit that happens to be a 1.25V regulator, but produces a variable output. The same way the regulators in this circuit are being used to produce "non 5V" power rails. They
are just used to get a power rail that is offset by 5V from another rail. But all rails are measured W.R.T. 0 volts.

 Like. Reply


o

Cezar ChirilaOctober 24, 2016

Thank you for your input. Sorry this took so long. I have modified the diagrams and I now hope it is less confusing to others. Again, thank you for taking your time to give your feedback, I truly appreciate it.

 Like Reply

 Q
 qbx888August 31, 2016

Great article! Please add more info about proper supply rates. You mentioned increasing input voltage to 40V but what is the lowest value? Which transformers do you recommend and what component next to it
are necessary?
Also I don’t see a potentiometer for change master value. Where it can be added?

 Like Reply

o Cezar ChirilaAugust 31, 2016

I wouldn't go lower than 20-24V for the power supply ,but at least 30V is recommended. For low power amplifier, I think a different design would be better. If you are using a transformer (I recommend toroidal
because of their size), and not a SMPS, you will need a bridge rectifier and some beefy filtering caps (I personally used 2x10 000uF per branch - positive/negative). About the potentiometer, although you could
add one before the " Input -P1" of value 100K, I suggest building a pre-amp circuit that also has tone control.

 1 Like Reply

 PickyBikerSeptember 09, 2016

This looks great. One question what is the power out one can expect from this circuit?

 Like Reply

o Cezar ChirilaSeptember 10, 2016

Somewhere between 120-140W. It varies greatly (100-150W) depending on your power supply.

 Like Reply

 C
 Colin55September 10, 2016

How do you get +12v from a -30v supply ????

 Like Reply

o Cezar ChirilaSeptember 10, 2016

It is +12V referenced to -30V. Imagine that we use a multimeter and we put the black probe (ground) to the -30V rail. If we measure the real ground (which is 0V), on the multimeter digit we would read +30V. If me
measure "+12V", on the multimeter it reads +12V. BUT, if we put the black probe on the 0V ground, on the multimeter we would have -18V. We measure the electric potential difference, V2-V1. When we measure
"+12V" rail, we consider "-30V" supply to be V1. I used this notation because we connect the COM and VSS port of the IR2110 to the "-30V" rail.

 Like Reply

 C

Colin55September 10, 2016

All voltages must be referenced from the 0v rail (called CHASSIS or earth) as this is where you will place the black probe of the voltmeter.
The voltages on the 7805 will be minus 18.6v and minus 25v.
Neither of the voltages will be stable as the minus 30v rail (line) is not stable.
The 7805 voltage regulator will have no effect on stabilising the voltage.

 Like Reply

o Cezar ChirilaSeptember 10, 2016

You are right that they will not be stable in regards to 0V, they will, however, be stable in regards to the negative rail, which is how the IR2110 is referenced.

 Like Reply
o Cezar ChirilaSeptember 10, 2016

Thank you very much for you input. After the weekend I will test again the amplifier with a reactive load (speaker) and measure the distorsions when the 30V rail is not stable (50-80Hz input signal). On my previous
test, the same as this, I didn't had any problems.

 1 Like Reply

 K

ks1233September 13, 2016

how about a nice power supply project along with a simple enclosure to complete the package?! also, is there a reputable parts supplier that you could recommend - the one that you used, maybe? thanks in
advance!

 1 Like Reply

o Cezar ChirilaSeptember 30, 2016

That is a very good ideea. If I find the time, maybe I will make a project article. About the parts, I bought them from tme,eu, which is an electronic components distributor for Europe. For US I heard about digikey,
mouser and farnell/newark but I am sure that there are more.

 Like Reply

 Edwin G. DelgadoSeptember 27, 2016

If I want to to build an amplifier in the 20 -  30 watts range, what changes should be made to the circuit?  It is just a matter of changing the supplying DC voltages?  Good circuit and a good DIY project from
you…....

 Like Reply


o Cezar ChirilaSeptember 30, 2016

If you want to build an amplifier with such a low output this schematic will not do it and it is a bit overkill I think. I would suggest to do a search for "Class D IC" and you will find some integrated circuits from ST and
TI that are great. I think the datasheet for some also provide schematic and PCB. Here are some links : http://www.ti.com/lsds/ti/audio-ic/mid-power-audio-amplifiers-5-50W-product.page and
http://www.st.com/content/st_com/en/products/audio-ics/audio-amplifiers/class-d-audio-power-amplifiers.html?querycriteria=productId=SC983 . If you need any help, please do not hesitate to contact me.

 1 Like Reply

 adgj533December 28, 2018

Hey Cezar, I had a few questions about this project: I am trying to build it at home. You mentioned that we need +-30V, but I also see +-5V going to the op amps. does that mean we need 2 power supplies? Also
Where did you connect your ground on the pcb, Im having trouble finding the common ground you talked about. One more thing, I am familiar with Eagle, is there anyway to convert KiCad files to eagle or do I
have to manually rebuild the whole thing on eagle? How would i add a volume control knob to this circuit?

 Like Reply

 J

johnl2February 25, 2017

Hello, I have been working on this circuit for quite some time. I purchased all of the parts and I just cannot seem to get the HO of the IR2110 to produce voltage. There’s V going into the HIN, -30V. I went over the
whole circuit several times to make sure everything is connected properly and it seems to be. Is this a common problem? I looked up the IR2110 HO not working and some hits come up. Or could the IC be bad?
Any help is appreciated. This thing is driving me nuts!
Thanks

 Like Reply

o Cezar ChirilaFebruary 26, 2017

Hello. First of all, did you use my pcb or did you build in on a perfboard?
  Like Reply

o
 J
johnl2February 26, 2017

I used a breadboard just to test it out and plan to move further later. I am wondering if the power supply ground is isolated from the circuit ground. I just had a cap blow in my face for the first time ever after
connecting the grounds together! My email is johnjol399@aol.com if you'd rather email me. Thanks for getting back so soon.

 Like Reply


 P
PaolaEngineerApril 01, 2021

Hello Cezar, how can I find your email or contact info? I'm working on a project for a non profit organization that educates children with audio lessons. thank you and hope to talk to you soon.

 Like

 P
PES DIGITAL AUDIONovember 11, 2017

Hello john12, really you can't get the voltage value at the HO of IR2110, unless you connect the output MOSFETS, remember its a floating ground at VS, HO voltage can be achieved when the Lower MOSFET
turns on. But there is other way to check it even though with out MOSFETS connected. Just understand how the circuit works.

 Like Reply

 P

prasad357February 27, 2017

Hi,
How can I convert this design into a constant voltage (100V),high impedance amplifier ?

 Like Reply


o Cezar ChirilaFebruary 28, 2017

To be honest I'm not really sure. It would be possible in theory if you use the rectangular signal before the filtering (the last inductor and capacitors), which can be used with a transformer, but I do not know after
that what needs to be done.

 Like Reply

 Carel ColpaApril 30, 2017

You use an audio transformer in the amplifier to bring your voltage up to 100V, this technology is used to drive speakers over a long cable.

 Like Reply

 Carel ColpaApril 30, 2017

You use an audio transformer in the amplifier to bring your voltage up to 100V, this technology is used to drive speakers over a long cable.

 Like Reply

 Pier Paolo BaldiMarch 15, 2017

Hi, all looks really great!

Can it be good, also to ampliy ultrasound?
What have I to modify to have in output an 25 Ohm impedance?
Thks

 Like Reply


o Cezar ChirilaMarch 17, 2017

Depends on the frequency, but keep in mind that this is an amplifier designed to go well with frequencies under 20kHz. With greater frequencies you would need to increase the PWM freqency to be much greater
and also to change the output filter. Generally you would like higher perfomance devices, faster op-amp, faster MOSFET driver, etc. To be honest I do not know much about Ultrasound amplifier, but I will start with
a schematic of that and see the requirements. The output can be 25 ohm impendance, no problem, you would just (again) need to adjust the output filter accordingly.

 Like Reply

 R

rezgar.sApril 05, 2017

Hi
can i use +-90 v supply?
If its possible how much power i can get?

 Like Reply


o R
RysdanaOctober 07, 2017

The thing is that class d has efficiency advantage depend on your need.

 Like Reply

 N

n289seFebruary 09, 2018

Can this amplifier deliver 60-120 watts into 16 ohms?

 Like Reply

 S

supertallarinMarch 05, 2018

 Hi, thanks for the post! I am trying to build one myself and I have a few issues. Is it possible to determine some sort of transfer function in order to get an expresion of the gain of the amplifier? For example, I want
it to be able to deliver 100W when the load resistance is 8ohms, how can I check that through an equation? Something similar happens when you try to add negative feedback to improve stability, bandwidth and
THD, I need to determine the feedback gain (and with that the total gain) by design so that I can then choose the value of certain components.
Is there anyone with some experience in this issue that can help me out?

 Like Reply

 DAEKHMarch 10, 2018

How can I transform the circuit in order to allow me to have a right and left chanel? Is it possible (because I would like to build my own studio speakers) ? Thanks in advance (if you take care of this old project)

 Like Reply

o Dinca AndreiJanuary 14, 2019

Hi,a simple way is to make 2 PCBs and use 2 bipolar power supplys...

 Like Reply

 D

danishadvanceMarch 20, 2018

Can i use CD4504 level shifter instead of this 2 trasistor 2N5401 ?

 Like Reply

 D

danishadvanceMarch 20, 2018

Can you plz share the pcb layout of this schematic..

 Like Reply

o A
artmaster547April 18, 2018

They have already been shared as KiCAD files

 Like Reply

 A

artmaster547April 18, 2018

I have a question with regards to the selection of capacitance values for C12 and C13 how did you go about selecting those values is there a calculation that I could use, as I am designing something similar
please?

 Like Reply

 Deshan RajapakshaJune 04, 2018

It is good post & good job thanks admin   i will made it

 Like Reply

 adgj533December 28, 2018

Hey Cezar, I had a few questions about this project: I am trying to build it at home. You mentioned that we need +-30V, but I also see +-5V going to the op amps. does that mean we need 2 power supplies? Also
Where did you connect your ground on the pcb, Im having trouble finding the common ground you talked about.
One more thing, I am familiar with Eagle, is there anyway to convert KiCad files to eagle or do I have to manually rebuild the whole thing on eagle?

 Like Reply


o Cezar ChirilaDecember 29, 2018

Hi adgj533! Don't worry about the +-5V, that is regulated from the +-30V power supply using the two Zenner diodes D1 and D2. You are better of using either KiCad (Which is open-source - free and easy to learn)
or start a new project with Eagle. You may run into odd problems by converting from one program to another and the time you loose looking for a fix might be greater than by just going with my suggestion. On the
PCB, The ground connectors are near the -30V connector. There are two of them: One as an input, one as an output for the speaker, although it is better to connect the speakers direct to the power supply ground,
to remove some humming noise. Take care!

 Like Reply

 adgj533December 30, 2018

Hey cezar thanks for the quick reply. I thought u wouldnt even see my post. I am trying to build this amplifier at home.I want to add a pre amplifier circuit+ a volume control circuit, where would I add these things?
Is it before the input of the amplifier? please let me know and thanks a bunch for your help

 Like Reply

 Cezar ChirilaJanuary 07, 2019

Sorry for replying to you so late. For some reason I keep getting the notification emails in my spam inbox. If you want to add those, just add them before the class-D amplifier. So you have  - - .

 Like

 adgj533December 31, 2018

Hey cezar thanks for the quick reply. I thought u wouldnt even see my post..I want to add a pre amplifier circuit+ a volume control circuit, where would I add these things? if possible could you show me how to add
it?  please let me know and thanks a bunch for your help
  Like Reply

 jbongiorno81January 27, 2019

Cezar,
What steps would need to be taken to have a design such as this push lets say 500W into 4 or 8 ohms? I am currently working on a design for school and have been looking for insight from various projects. The
amplifier would be embedded within a 2-way speaker.

 Like Reply

o adgj533February 09, 2019

look at the reference design he used from infereon, that is something you might wanna use. Cezar's design is more of an entry level or for hobbyist

 Like Reply

 adgj533February 09, 2019

Hey Cezar, I had few more questions, I substituted 2n5401 with MPS751 bc it is unavailable in the market now. Also could you link me the heatsink you used?, I opened the files in kicad and it said that a lot of the
libraries you used are missing, is it possible to send me your kicad libraries in a zip? just add at gmail to my username, that is my email. I appreciate you uploading the design so that others can work on it and
thank you for your replies.

 Like Reply

 S

sdsdsApril 19, 2019

Why power the input side of the IR2110 with -30 and -25 V? Since the comparators at the input are powered by +/-5 V, why not also power the IR2110’s input with that and avoid the level shifter?

 1 Like Reply
  A

AndesatJune 04, 2019

Hi Cezar, i have a power supply of +100v 0 -100v would i be able to use this circuit if i use IRFP260 .  what modifications will be necessary.

 Like Reply

 A

andershilmerssonJuly 01, 2019

Hi,
How much distorsion do you get at different input frequencies? Which over tune frequencies (2:nd, 3:th etc)? How much is the 200kHz signal rejected?

 Like Reply


o B
bassbindevilDecember 09, 2019

The lack of performance measurements is worrying. I'm guessing there is a whole lot of distortion, since there isn't any negative feedback to compensate for all the imperfections in the triangle wave and
comparators and output stage; it should be fine for signals with no dynamic range like a siren, or a compressed speech over a bullhorn. Adding NFB and making it work might be a worthwhile design exercise for a
student, but if you just want to listen to music, there are other proven class-D amps out there available as chips or complete modules, ready to use for less than the price of a Happy Meal.

 Like Reply

o
 B
bassbindevilDecember 10, 2019

It's not like measuring audio performance requires specialized instruments from Brüel & Kjær or Sound Technology these days. A computer sound card and Rightmark Audio Analyzer can measure noise and
distortion and frequency response with the click of a mouse button. Just watch out for ground loops and overloading the inputs; there's plans/kits out there to make sound card i/o more like a test instrument.
(google sound card buffer) http://audio.rightmark.org/products/rmaa.shtml

 Like Reply

 Amit Yadav 3September 14, 2019

 HI IS IT NECESSARY TO HAVE 220N AT +TO G AND -TO G ??? BECAUSE WE HAVE ALREADY 100N PARALLEL TO POWER CAPS. THANKS.

 Like Reply

 M

M.IniOctober 05, 2019

Very interesting, thank you. I wanted to give a look at the KiCad files but the link gives me “Error 404
The page you requested was not found”.

 Like Reply

 Mathieu LaplanteOctober 09, 2019

Just wondering, why exactly are you shifting the comparators outputs to -25V-30V with the PNP before going into the IR2110 instead of the standard 0V?

 Like Reply

 A

Ambalanduwa Kankanamge SasankaMarch 21, 2020

I have a Sony subwoofer speaker unit with 2 ohms and 180 watts RMS (Aluminium tube -dual speaker -internally wired parallel)  .Can I use your circuit for this speaker unit?.If so what changes in this circuit I need
to do ? (Meaning that I want to use this as a subwoofer amplifier)

 Like Reply

 ChristosonlineMarch 30, 2020

How comes 18v, 25v

 Like Reply
 minerwelderApril 23, 2020

Thanks for the great article Cezar, I have a question about a bipolar power supply here in the US I can only find a dual +- 20vdc, is there a project you can recommend for building a bipolar +- 30vdc power supply.
Thanks for your knowledge and have a great day.

 Like Reply

 gunterflunderJune 04, 2020

This amplifier is designed very hard to the limits of its devices. The power MOSFETs have relatively large channel resistances. There are also unexpensive devices with just 5 mOhm channel resistance in order to
improve the overall efficiency. The charging of the capacitor of the triangle wave generator by use a resistor is not convenient, as it will lead to nonlinearities at higher input ac levels. The IR2010 or IR2011 as well
as the comparator are relatively slowly devices. A PWM of 200 kHz is allready very high to operate with these devices.
I would suggest to operate with the LM5104 for the half bridge driver together with the STP16NF06 NMOS transistors to operate at 500 kHz. This half bridge driver is much faster and has less dead time for
through shoot protection. The MOSFETs have less channel resistance and less gate capacity and gate charge for very fast switching. For the voltage controlled PWM converter I would suggest not to use a
triangle converter solution. It is better to use a converter IC directly like the LTC6992 https://ibb.co/zm1s04H . But it is also possible to make a good converter by use of 555 timer ICs https://ibb.co/cDqcYyT .

 Like Reply

 James40October 19, 2020

Real nice project and well documented. I am a little confused. On the reguatlor you have got 18 volt on the input and 25 volt on the output. Shouldn’t it be the other way round?

 Like Reply

o RK37October 20, 2020

 A typical 5 V linear regulator (such as the LM7805) will make the output voltage 5 V higher than the "ground" voltage. In this circuit, the regulator's ground terminal is connected to -30 V; thus, the output is at -25 V,
because -25 V is 5 V greater than the ground voltage. A positive regulator can be used because the input and output voltages, which are labeled as negative in the schematic, are actually positive relative to the
voltage at the regulator's ground terminal.

 Like Reply

 acacacacaNovember 02, 2020

Hello, I have looked at the symmetric diagram, but I am not too sure why there is 2 set of inductors at the end of Vb and Vs port of IR2110, can anyone give me a helping hand, please:)

 Like Reply

 E

ElgitanoJanuary 04, 2021

I built this amplifier but it burns out the mosfets as soon as power is applied. All connections are 100% correct even though my PCB layout differs slightly. I used IR2113 as a direct replacement for IR2011. So I
removed the MOSFETS and measure pulses at pins 1 and 7. There is only a pulse at pin 7 but nothing at pin 1. Can you please tell me what I need to do. Thank you

 Like Reply

o Paulo RangelFebruary 11, 2021

I am having the same issue, have burned two sets of MOSFET, and 12V Zenner, and so on...I am almost giving up on this one. Any ideas on how to protect the MOSFET's? Zenners?

 Like Reply

 HackerCopJanuary 23, 2021

When there is no audio input how does this circuit prevent the carrier PMW signal from entering the speaker. If the signal is 0v and the triangle wave oscillates between +/-5v a 200KHz square wave is outputted
from the comparator. I’m guessing that because the carrier oscillates between +ve and -ve the low pass filter after the MOFSETs averages it to 0v before the speaker receives it. Am I correct?
  Like Reply

 E

ElgitanoApril 13, 2021

Why is it that no responds. No one also says that the thing works. All my connections are 100% according to the diagram. I even followed the nets on your PC board layout to see whether the connections are
correct. My components are new and reliable but 2 sets of MOSFETs already burnt out after a short while of audio. Can I trust these projects or is it just some joke ?

 Like Reply

o Cezar ChirilaApril 13, 2021

Hi, if all the voltage rails have been measured and they are correct you should use an oscilloscope to check both gate drive signals (at the gate of Q4 and Q5) with a 1kHz sine input. This would help give insight
into what is going on. This would also help you check that the gate driver IC you have selected (IR2113) has enough deadtime. To further protect the MOSFETs I would also suggest adding in a 10k resistor
between the gate and source of Q4 and Q5.

 Like Reply

o
 S
sidtronicsApril 29, 2021

Thanks sir.Iam attempting to construct this.What is the safest maximum power voltage and output power?

 Like Reply

 K

knutknutMay 28, 2021

What heat sink did you use. There is no reference to it in the parts list

 Like Reply

 E

ElgitanoJune 01, 2021

 About the construction of the inductor. Why must it be T106-2 ? Why is that particular one required ? I Used a different one because I couldn`t find the T106-2. The core that I used forced me to get as low as 13
turns to get 22uH. The Mosfets burnt out in a few seconds. Why was my inductor the wrong choice ? Please help.

 Like Reply

 S

sandeeperJuly 14, 2021

Hello sir try this circuit work very well. Now i want to make 4 channel amplifier. i use 555 for synchronized clock . my question is can i use single supply circuit for four amplifiers?? can this single circuit enough
power drive four mosfet driver??

 Like Reply


o O
oshimaptMarch 15, 2022

hello, how it sounds? is there any parasitic noises?

 Like Reply

 Noledgeispower01September 30, 2022

It’s very helpful but I need more time to understand somethings or I start from scratch. Thank you

 Like Reply

 A

AK001December 24, 2022

getting dc votage of about 3v at speaker output please sort it out

 Like Reply

 S

so-circuitousDecember 28, 2022

1) clicking on the “get the bom” link brings you to an empty BOM list.
2) this project is very interesting to me because I read a review of a very popular mini class D amp which does not have the filter, and the reviewer said that such a design makes the speaker into the low pass filter,
not so good for the speaker. None of the vendors of these mini amps publish whether such a filter is part of their design, and since they all appear to be rebranded versions of the same unit, I suspect that none of
them have it.
2) what is the per-channel output in watts, and can it be built in a 100w/channel version ? what components and values need to be replaced ?
3) those mini amps all have bluetooth connectivity, which, together with their size, makes them very popular. how would I add bluetooth to this unit ?

 Like Reply

///

Understanding output filters for Class-D amplifiers

By John Widder and Yun Tao Zhao, STMicroelectronics  01.09.2008  0
Share Post
Share on Facebook
Share on Twitter
Class D amplifiers generally use a low-pass filter to attenuate the switching noise in the output waveform while passing the audio signal to the loudspeaker, but many engineers are not familiar with the functions performed by
the various components in a Class-D amplifier filter or how to calculate the proper values. This article explains the purpose of the filter components and how to calculate their values.

The heart of a Class-D amplifier filter is a L-C low-pass filter. The corner frequency of the filter is chosen so that the filter will have minimal effect on the desired output frequency range while attenuating the switching noise
as much as possible.

Figure 1: A low pass filter for a Class-D amplifier

LOW-PASS FILTER
The optimum value for the filter inductor is

L = R  /2pf
L C

where f  is the desired corner frequency of the filter and R  is the load (speaker) resistance. Note that the inductor value is dependent on both the desired corner frequency and the speaker impedance so the inductor value
C L

will changes if the speaker impedance changes.

Practical designs require the use of standard component values so small adjustments usually have to be made to the ideal inductor and capacitor values. Rather than calculating the inductor and capacitor values independently
and then adjusting their values, it’s better to calculate the inductor value, select the closest standard inductance value, and then calculate the required capacitance using the selected inductor.

C = 1/((2pf  )  • L )

C
2

The quality factor (Q) of a filter is the ratio of the center frequency to the filter bandwidth.
Q = R  v(C /2L )
L
A high Q produces an underdamped curve and a low Q produces an overdamped curve. The Q of the filter should be in the range 0.6 > Q > 0.8 to avoid underdamped or overdamped behavior. If you use the equations above
the filter should have a Q of about 0.7, which provides good performance and allows for impedance variation in the speakers. Note that the Q of the filter will change if the speaker impedance is changed without adjusting the
filter component values, which can result in an underdamped or overdamped response.

COMPONENT SELECTION
Not only is it important to choose the correct L-C filter values, it is also important to choose the correct types of components for the class-D amplifier in order to avoid losses and minimize harmonic distortion.

The DC current rating of filter inductors must be greater than or equal to the maximum current that it will see. The change in inductance versus load current should not be more than 10%. The core material can affect the
amplifier’s harmonic distortion and should have very low hysteresis losses.

The capacitor should be a multilayer polyester, polypropylene or polycarbonate film capacitor. Avoid using ceramic capacitors in the low-pass filter. Ceramic capacitors experience large changes in capacitance as the voltage
across them changes, which can result in distortion.

SINGLE-ENDED OUTPUTS
Many designs use amplifiers with single-ended outputs because they only require half as many transistors as a full-bridge output, and integrated amplifiers with single-ended outputs only require one output pin instead of two.

Single-ended amplifiers also have a few disadvantages compared to amplifiers with bridge-tied load (BTL) outputs. First, single-ended amplifiers require either split positive and negative power supplies or DC blocking
capacitors. If DC blocking caps are used, they need to be large in order to prevent them from affecting the low-frequency performance of the amplifier. For example, an amplifier with an 8Ω speaker needs a 1000µf cap in
order to achieve a -3dB point of 20Hz.

DC blocking caps can also cause audible pops as they charge up to Vcc/2 when the amplifier is turned on. A resistor divider from Vcc to ground can be used to charge the capacitor up to Vcc/2 at a relatively slow rate when
the power is turned on, minimizing or eliminating the pop.

If the amplifier does not have feedback then PSRR might be a problem. Two DC blocking capacitors can be used to create a low-impedance AC voltage divider to improve the PSRR. If two DC blocking capacitors are used
then each cap only needs to have ½ of the capacitance because the circuit sees the parallel impedance of the two capacitors.

Figure 2a: Single-ended amplifier with a DC blocking capacitor

 
 Figure 2b: Single-ended amplifier with a resistor divider to minimize pop at turn on

Figure 2c: Single-ended amplifier with two DC blocking caps to improve PSRR
Figure 2: Single-ended Class-D amplifier filters with DC blocking capacitors

BTL OUTPUTS
Amplifiers with BTL outputs are popular because they do not require DC blocking caps even when operating with a single positive power supply. DC blocking caps limit the low-frequency response of the amplifier and can
be quite large.

BTL amplifiers have another advantage over amplifiers with single-ended outputs – the maximum peak-to-peak voltage that a amplifier with BTL outputs can apply to the speaker is twice the power supply voltage, which in
turn means that up to four times as much output power can be delivered to the load compared to a single-ended amplifier. This can be a big advantage in applications where the power supply voltage is limited, especially in
portable applications where the amplifier is operating off of a battery.

COMMON-MODE FILTERS
Common-mode filters are L-C filters with one side of the capacitor grounded. The inductor is placed in series with the amplifier output(s) and the capacitor is connected from the speaker terminal to ground. Since single-
ended amplifier filters already have one side of the capacitor grounded the low-pass filter is a common-mode filter.

When used with an amplifier with BTL outputs, the filter shown in figure 1 would be a differential filter since it filters the signal between the two outputs. Common-mode filters for amplifiers with BTL outputs are different
than the differential low-pass filter. The filter inductance for BTL amplifier outputs is normally split into two separate inductors, with one inductor is placed in series with each of the amplifier outputs. Each inductor has half
of the total inductance required for the low-pass filter.
Because the capacitance across the load is now the series combination of the two capacitors for BTL outputs, each capacitor needs to be twice the value calculated for the low-pass filter so that the total capacitance will be
correct. Note that because the inductance for each BTL output is cut in half but the capacitance to ground is twice the normal value, the resonant frequency of the common-mode filter is the same as the resonant frequency of
the differential filter.

f  = 1/(2π√(L  C  )) = 1/(2π√(½L  • 2C  ))

C T T T T

The simplest common-mode L-C filter is just an inductor and a capacitor to ground (figure 3). This produces excellent attenuation at high frequencies but this filter has an underdamped common-mode response that can
cause unwanted ringing on the speaker leads. It can also cause very high ripple current through the inductor and capacitor. The impedances of the inductor and capacitor cancel at the resonant frequency so the current at the
resonant frequency is only limited by the stray resistance in the circuit (primarily the output impedance of the amplifier and the DC resistance of the inductor).

Figure 3: A simple common-mode filter and its response

In order to damp the common-mode response of the filter it is necessary to add some resistance to the filter. Normally a resistor is added between the capacitor and ground.

However, adding a resistor between the capacitor and ground creates a zero in the filter response which can greatly reduce the filter’s effectiveness at higher frequencies. The effects of adding a resistor in series with the
capacitor can be seen in the following plot. Note that the amplitude of the resonant peak is greatly reduced but the attenuation at high frequencies is also reduced.
 

Figure 4: Common-mode filter response with damping resistors in series with the capacitors
For this reason a second capacitor is usually placed across the resistor to create a pole above the filter’s resonant frequency. This pole cancels the effect of the zero created by the resistor at higher frequencies while allowing
the resistor to provide damping at the L-C resonant frequency. Figure 5 shows the response of a filter with a capacitor in parallel with the damping resistor. Note that the resonant peak of the filter is still much lower than
without a damping resistor but the attenuation at high frequencies is much better than without the second capacitor.

 
 Figure 5: Common-mode filter response with bypass capacitors in parallel with the damping resistors
Calculating the component values for a common-mode filter is easy. The total inductance and capacitance in the filter remains the same as for a differential-mode filter:
L  = R  /(2πf  ) and C  = 1/((2πf  )  • (L  ))
T L C T C
2
T

Because the inductance for BTL outputs is divided between two inductors in series, however, the value of each inductor is equal to ½ of the total inductance
L   = L   = ½L  = R  /(4πf  )
1 2 T L C

Similarly, the capacitance for each filter is divided between two capacitors in series, so the value of each capacitor is equal to twice the total capacitance
C   = C   = C   = C   = 2/((2πf  )  • L  ) = 1/((2πf  )  • L   )
1 2 3 4 C
2
T C
2
1

The value of the common mode resistors should be

R   = R   = 1/(√2 • 2πf  C   )
1 2 C 1

This resistor value will insure that the zero is below the L-C resonant frequency and the pole is above the LC resonant frequency, allowing the resistor to damp the filter response at the resonant frequency.

Using a common-mode filter has another beneficial effect. The response of a differential-mode filter is normally damped by load (the speaker). However, if the amplifier is operated without a speaker connected the response
of the filter will be very underdamped, similar to the response shown in figure 3 . The use of a common-mode filter with damping resistors will ensure that the filter response is always well behaved, even without a speaker
attached.

HYBRID FILTERS
Metal film capacitors are relatively expensive so increasing the number of capacitors in the filter from one to four can have a significant impact on the total cost of the amplifier. It is possible to keep the filter cost close to
what a simple L-C differential filter would cost while providing some of the benefits of a common-mode filter by using a hybrid filter that combines elements of both common-mode and differential filters.

Figure 6: A hybrid output filter. C1 is a metal film capacitor and the other capacitors are X7R multilayer ceramic capacitors.
A hybrid filter has split inductors and R-C networks between the speaker terminals and ground like a common-mode filter, plus a capacitor across the speaker terminals like a differential low-pass filter. At first this would
seem to be counterproductive since it adds a fifth capacitor to the design. The total cost can be reduced by making the value of the differential capacitor significantly larger than the common-mode capacitors and only using a
metal film capacitor for the differential cap.

The other four capacitors can then be less expensive X7R multilayer ceramic capacitors. This makes the cost of the hybrid filter only slightly more expensive than the cost of a differential L-C filter while still providing some
common-mode attenuation and some damping under no-load conditions. The drawbacks of the hybrid filter are:
 1.) The differential attenuation is the same as for a normal common-mode filter but the common-mode attenuation is not as good. Because the common-mode capacitors in a hybrid filter are smaller than they
would be in a pure common-mode filter, the center frequency for common-mode filters is higher and therefore the attenuation at the switching frequency and its harmonics is lower.

Figure 7: Common-mode response of hybrid filter vs. common-mode filter

2.) The differential-mode damping of a hybrid filter under no-load conditions is not as good as a pure common-mode filter because most of the high-frequency current flows through the larger capacitor across
the speaker terminals. Normally this isn’t a problem because the speaker provides the differential-mode damping, but if the amplifier is operated without the speaker connected then the damping will not be as
good. Care needs to be taken to insure that the damping of a hybrid filter is good enough to protect the amplifier under no-load conditions.

Figure 8: No-load response of hybrid filter vs. common-mode filter

3.) The harmonic distortion with a hybrid filter will be slightly higher than with a common-mode filter because ceramic capacitors provide some of the filter capacitance, while a pure common-mode or
differential low-pass filter would normally only use metal film capacitors with much better characteristics.

Despite these drawbacks, hybrid filters can provide some of the benefits of a common-mode filter while keeping the cost close to that of a differential-mode filter.

As might be expected, calculating the component values for a hybrid filter are somewhat more complex because choosing the component values involves making performance trade-offs. In order to prevent harmonic
distortion from being a problem, the value of the ceramic common-mode capacitors should be smaller than the value of the film differential capacitor.

However, making the ceramic capacitors too small will hurt the common-mode EMI attenuation and the no-load damping. Amplifier manufacturers will normally recommend hybrid filter component values for common
speaker impedances that provide good filter performance.

HYBRID FILTERS FOR SINGLE-ENDED OUTPUTS

Hybrid filters for amplifiers with single-ended outputs are slightly different than hybrid filters for amplifiers with BTL outputs. At high frequencies the impedance of C1 is much less than the series combination of C2 and C3
so capacitor C3 is not needed.
 Figure 9: A hybrid filter for a single-ended amplifier

SNUBBERS
When the output of a Class-D amplifier switches there normally is a “dead” time between the time when one transistor is turned off and the other transistor is turned on. The dead time is necessary to insure that both
transistors are never conducting at the same time, which would cause large currents to flow from the power supply to ground through the transistors. However, the dead time causes a problem because it interrupts the current
flowing through the inductors. Snubbers are normally used on the amplifier outputs to provide another path for the inductor current during the dead time.

There are two types of snubbers. Amplifiers with BTL outputs can use a differential snubber, with a single resistor and capacitor in series in between the two outputs. Common-mode snubbers have a resistor and capacitor in
series from the output to ground and can be used with either single-ended or BTL outputs.

Common-mode snubbers for amplifiers with BTL outputs use twice as many parts as a differential snubber but they may reduce harmonic distortion. The type of snubber to use will depend on the application. The amplifier
manufacturer will normally recommend values for the snubber components.

FILTERLESS AMPLIFIERS
No discussion of Class-D output filters would be complete without talking about filterless Class-D amplifiers. The primary purpose of the output filters is to reduce EMI. However, it is possible to operate a Class-D amplifier
without any filters on the outputs. Although there is a large amount of high-frequency switching noise on the amplifier outputs, this noise is far outside of the response range of most speakers so filters are not necessary for
good audio quality. Filterless Class-D amplifiers are less efficient because the high-frequency energy that is normally absorbed by the filter is dissipated as heat and EMI.

Filterless Class-D amplifiers should have controlled rise and fall times to limit the high-frequency content in their output spectrum. Filterless Class-D amplifiers also require very careful attention to circuit board layout to
prevent EMI problems. In particular, the distance between the speaker and the amplifier must be kept as short as possible, and loop area between the amplifier output and its return path (either another output or ground) must
be as small as possible.

The wires from the circuit board to the speaker should be twisted together in order to keep the distance between them as small as possible. Of course, these measures are good practice for Class-D amplifiers with filters also.

The second article in this series offers some pc-board layout guidelines designed to help optimize the performance and reliability of Class-D amplifiers.

///

Why Class D Amplifiers May Test Well But Often Sound Terrible
By James M. Shanahan, Jam Technologies  08.29.2005  1
Share Post
Share on Facebook
Share on Twitter
Class D amplifiers are fundamentally different from analog amplifiers, not only in their circuitry, but more importantly in the way they operate. Yet, engineers continue to test Class D amplifiers using the same tests and test
procedures that were developed generations ago specifically for analog amplifiers. Because these tests were designed for analog amplifiers and are focused on analog particulars and capabilities, it is no wonder that we have
Class D amplifiers that test well in the lab, yet reproduce inferior sound under real world operation. Clearly, new tests that are specifically designed with Class D operation and behavior in mind are needed if the test results
are ever going to be relevant.
Reconsidering THD in a Class D Environment Analog amplifiers modulate power continuously in accordance with an analog input signal, and generate continuous wave (analog) outputs. Class D amplifiers generally rely on
the principle of Pulse Width Modulation, and, by virtue of their use of switching power devices, generate outputs comprised of discrete quantitized samples. So, while both amplifiers perform the same general task, the ways
they accomplish this are as different as the way LPs and CDs play back sound. As we know, LPs drag a stylus over plastic, and CDs read a digital signal with a light source.

Some of the traditional tests used to validate the performance of the venerable but ancient Class A and Class A/B amplifier architectures such as Total Harmonic Distortion plus Noise Floor (THD+N) will reveal some useful
performance or sonic characteristics in a Class D amplifier. However, these tests were originally conceived to show known common limitations of the analog amplifiers of that era. At that time, audio distortion was primarily
harmonic, and more importantly, dynamic range was limited (and defined) by the noise floor.

Digital switching amplifiers, by their very nature, provide a completely new set of strengths and weaknesses. For example, distortion may be harmonic or enharmonic in nature, due to interactions with the sampling rate.
Amplifier jitter rejection can also be critical. On the other hand, dynamic range is rarely, if ever, limited by the virtually non-existent noise floor of a digital system, especially when the incoming audio signal is digital and
therefore also virtually noiseless. To compound the issue, most audio test equipment was designed to test within the confines of analog amplification and an analog sound source, therefore, this equipment – by design and
definition – must give inaccurate results when confronted with inaudible switching frequencies.

The most significant analog amplifier measurement has now become the least reliable and least significant measurement for a digital amplifier. Since the test equipment was designed before switching amplifier systems were
conceived, THD+N readings may easily be unreliable or misleading. In other words, in the world of switching amplifiers, THD is at least not as relevant as other tests and at worst is totally irrelevant as a standard by which to
evaluate the performance of a digital amplifier. In fact, most THD measurements of digital amplifiers are no more representative of sonic performance than wow and flutter measurements are of a CD player. A single arbitrary
‘harmonic’ distortion category by itself is meaningless in the context of digital amplification without additional and more appropriate tests, such as intermodulation distortion and spectral display. Designing a Class D
amplifier solely for good THD test results does not mean it sounds good, let alone qualifies it as “audiophile”.

Dynamic Range: The Acid Test for Class D Amplifiers Dynamic range, or the ratio of the largest accurately amplified signal to the smallest, has never been more relevant than in today’s world of 16 to 24-bit digital audio
sources. A wide dynamic range is what makes music sound live, and 3-dimensional. With analog amplifiers, the popular shorthand for this measurement has become the ratio of the onset of clipping to the noise floor. This is
practical for Class A and Class A/B amplifiers because any signal below the noise floor is largely masked. This does not hold true for Class D amplifiers. Here, the noise floor is generally much lower than Class A/B
amplifiers, as a result of the noise immunity of digital circuitry. This does not mean that the amplifier can actually reproduce signals near that noise floor because most digital amplifiers, in fact, cannot. It is a limitation
inherent in the architecture itself.

The critical fact to bear in mind is that all Class D amplifiers have outputs that are comprised of discrete power increments, a notion that can more easily be thought of as resolution steps. When the definition of dynamic range
is applied in the context of quantitized increments, the result is that dynamic range of a Class D amplifier is defined as the ratio of the smallest discrete power level the amplifier can accurately output to the largest. Given this
clarity of understanding, the obvious question becomes: How do we determine the smallest output power level in a Class D amplifier? The answer can be seen in its linearity. The Key Is Testing for Linearity

The blue trace of Figure 1 shows Power Out (right Y axis) versus Input (X axis) of an E-Bridge', True Fidelity' amplifier. The red trace shows deviation (left Y axis) from linear operation over the same input range (or error
from linearity). Although the expected -30 dB output is seen with a –30 dB input, the difficult part for most digital amplifiers lies at the bottom end of the amplitude range. The further down the amplifier maintains linear

operation, the greater the dynamic range. 

Note that this test shows actual operation and doesn’t rely upon what we will call, ‘magic numbers’ (how other companies measure dynamic range often has nothing to do with the actual definition). The noise floor of this
amplifier is around –120dB, so the distinction between the two is obvious. The sonic implications of a broad dynamic range become more striking with the observation that while the upper amplitude regions impart what a
musical instrument is, the lower regions impart where it is. Spatial location of sound is the driving force behind the entire home theatre movement, which makes these low-level signals even more important. Here, the dynamic
range of this True Fidelity amplifier extends to about 102dB.

Fig. 2: Conventional Class D Amplifier Linearity

By contrast, conventional Class D amplifiers show their limitations when the same linearity test is applied. Figure 2 shows the dynamic range of a conventional Class D amplifier. Curiously, the company whose amplifier was
used in this test boasts a dynamic range in excess of 100 dB, where the graph clearly shows that that the amplifier ceases to operate linearly at levels below –80 dB. This is traceable directly to inadequate resolution of the
amplifier’s output stage. While the processing portion of the amplifier is capable of handling 16 or more bits of digitized data, it is the output stage that actually determines resolution.
Because Class D amplifiers are fundamentally different than Class A and Class A/B amplifiers, different tests need to be conducted to show performance. For more meaningful predictors of a switching amplifier, dynamic
range is the critical test in an arsenal of new tests aimed specifically at digital amplification, since it shows the amplifier’s ability to reproduce the full range of signals it receives. A simple and effective way to measure
dynamic range in a Class D amplifier is by its linearity in actual operation. With this test, you can be assured that great performance in the lab will equate to great performance under real world operation.