Circuit Surgery
Regular clinic by Ian Bell
Class-D, G and H amplifiers
R
egular PE author Julian
Edgar suggested this month’s topic
after reading an article by Mike
Spence on Texas Instruments’ support
forum discussing ‘look-ahead’ voltagerail boosting in class-D audio amplifiers
(these circuits are also known as class-G
and H amplifiers). There was also a post
on amplifier classes recently from Tom
Mot on the PE forum: ‘I’m fairly new to
speaker amplifiers, but I understand the
different classes (ie, Class-A, Class-AB,
Class-D). But how would I identify the
class of an amplifier, for instance if I was
buying an amplifier online and the class
was not mentioned (which often seems
to be the case, from my experience) –
how could I know the class of the amp?’
In direct response to Tom’s post we
cannot say much beyond the reply on
the forum from Richard Gabric; that
if the information is not supplied it
would be difficult to know, but class D
is widely used in modern designs. This
may be particularly likely for lower-costs
designs, and portable personal electronics,
but class D is also used in high-quality
amplifiers. In this article we will look
briefly at the other classes used in audio
(that is A, B and AB – C, E and F are
mainly used for radio-frequency signals)
before concentrating on D, G and H.
Classes A, B and AB
A class-A amplifier is based on a single
transistor operating as a linear amplifier
– for example, the circuit in Fig.1. The
circuit requires biasing (not shown)
to put the transistor into its forward
active mode, and set up at about half
the supply voltage at the output with no
signal present. This is a relatively simple
circuit, but the bias conditions require the
transistor to be conducting continuously
with no signal, which wastes a significant
amount of power in proportion to that
provided to the load. This is characterised
as amplifier efficiency – the ratio of the
power actually delivered to the load
(the loudspeaker in an audio amplifier)
to that taken from the supply. Class-A
amplifiers have poor efficiency.
To improve efficiency it would be
possible to bias a single transistor so that
it was not conducting when there was no
signal – increasing input voltage would
switch the transistor on, allowing it to act
as an amplifier. Unfortunately, this would
result in only half of the signal being
amplified – a change of the signal in the
opposite direction would not switch the
transistor on, and no amplification would
occur. This problem can be overcome by
using a pair of complementary transistors
– one conducts for the positive half
of the waveform and the other for the
negative. This is the class-B amplifier,
which is shown in Fig.2. The circuit is
much more efficient than the class-A
amplifier, but suffers from a problem
called ‘crossover distortion’. When the
signal voltage is small both transistors are
off, so there is dead band in the output
for low signal voltages, which means that
the amplifier does not fully reproduce
the signal waveform shape at its output.
Class-AB amplifiers overcome crossover
distortion by applying a bias voltage to each
transistor in a circuit similar to class B. The
bias ensures that both transistors are just
+ V C C
conducting with no signal. Even a small
change in the signal away from zero will be
amplified by one of the transistors, reducing
crossover distortion. The no-signal bias
level can be much smaller than for class-A
amplifier, thus providing better efficiency.
Class AB is a compromise between classes
A and B and can be designed to set the
desired level of trade-off between efficiency
and linearity.
Note that the circuits in Fig.1 to 3
show the basic circuit topology, but not
details of the circuitry required to create
a working design.
Efficiency and class D
Class-D amplifiers are more efficient
than class A, B and AB. This has been
the key driving force in the significant
increase in their use in recent years.
The need to maximise battery life in
devices such as mobile phone, and the
increasing awareness of the need for
greener technology, both favour energyefficient approaches to circuit design.
Efficient electronics also tends to be
physically smaller and lighter, which is
another significant advantage in portable
devices, and in achieving modern slim
aesthetics in mains-powered products.
Amplifiers waste any power that is
dissipated in the output transistors. Ideally,
all the power from the supply should go
to the load; however, linear amplifiers (A,
B and AB) involve some voltage being
dropped across the transistor (collectoremitter (CE), or drain-source (DS)) while
they are conducting, which leads to power
dissipation in the transistor (equal to VCEICE
or VDSIDS). As we have seen, depending
+ V C C
+ V C C
Out
In
V bias +
Out
In
In
Out
+
V bias
C C
Fig.1. Basic class-A amplifier architecture.
40
Fig.2. Basic class-B amplifier architecture.
C C
Fig.3. Basic class-AB amplifier architecture.
Practical Electronics | June | 2020
2 5 %
duty cycl
e
5 0 %
duty cycl
e
7 5 %
duty cycl
e
Fig.4. Fundamentals of duty cycle.
on the circuit design, the transistors may
also have a bias current that means they
are conducting with no signal present –
this leads to continuous power wastage
even when no signal is present.
Class-D amplifiers achieve their high
efficiency by switching their output
transistors (which are usually MOSFETs)
fully on and fully off. In the off state,
very little current flows through the
transistor – IDS is close to zero, so VDSIDS
is very small. A MOSFET operated as a
switch has a low voltage across it in the
on state – VDS is close to zero, so VDSIDS
is again very small.
The switching nature of the class-D
amplifier means that its output is like a
square wave, or more specifically a train of
pulses, so how can it reproduce an audio
(music or speech) signal which is nothing
like a square wave? The answer is that the
pulses are produced (the transistors are
switched) at much higher frequencies than
the audio signal and are manipulated in
such a way that if the pulses are passed
through a low-pass filter, which removes
frequencies at the pulse-switching rate,
then the output of the filter will be the
audio signal. The manipulation of the
pulses is referred to as ‘modulation’.
There is more than one modulation
technique that can be used for class-D
amplifiers. Two widely used approaches are
pulse-width modulation (PWM) and pulsedensity modulation (PDM). These names
are descriptive of what happens. In PWM,
the width of pulses at a fixed frequency is
varied in sympathy with the audio signal.
In PDM, the number of fixed-width pulses
per unit time is varied. In both cases, for
the time). After 30ms the
duty cycle increases to
50% (equal on and off
times) and at 60ms it
increases to 75%. The
other trace (blue) shows
an average value obtained
by low-pass filtering the
pulse waveform. This
represents signal power
Fig.6. LTspice schematic for a PWM modulator concept circuit. delivered to the load in
a PWM system.
As can be seen in Fig.5, when the
a fixed pulse height (voltage), the average
PWM duty cycle changes, the averaging
value of the modulated signal (the pulse
process means that it takes a while for
train) is equal to the modulating signal (the
the power level to settle down to the new
audio) and can be obtained by a suitable
value – this is normal for a low-pass filter
low-pass filter. We will look at PWM in
with a step input (the duty cycle has
more detail as it is the simplest case,
undergone an instantaneous changes in
although many modern class-D amplifiers
this example, whereas most audio signals
actually use PDM.
would change more smoothly). There
is also a very small amount of ripple
Pulse-width modulation (PWM)
in the level with each pulse in Fig.5.
The width of pulses produced at a
The exact waveforms in a real system
fixed frequency (repetition rate) can be
will depend on pulse frequency and the
described in terms of their duty cycle (see
properties of the filtering process. Fig.5
Fig.4). Duty cycle is the proportion of the
is for illustrative purposes only and does
cycle period that the pulse is high for. A
not aim to represent a specific system.
pulse train with a 50% duty cycle is an
ideal square wave, which is on for exactly
half the period. Lower duty cycles have
PWM uses and filtering
proportionally shorter pulses and higher
In addition to class-D amplifiers, PWM
ones have longer pulses, as shown in
is used in a wide variety of applications
Fig.4. A pulse width modulator generates
including switch-mode power supplies,
a train of pulses such that the duty cycle
and motor speed and LED brightness
of each pulse is set by the modulating
control. In some cases the load itself
signal value at the instant each pulse is
low-pass filters the pulses to provide the
produced. Duty cycle has a limited range
averaging function. A simple example
of 0 to 100%, so excessive signal values
of this is applying power to a heating
will run out of duty cycle range – this is
element where the element and/or the
similar to a linear amplifier clipping large
item being heated has a thermal time
signal peaks due to its output range being
constant much longer than the pulse
limited by the supply voltage.
duration. For control of LED brightness
As indicated above, the modulating
for human observation, the LEDs are
signal can be recovered from the PWM
actually switched on and off (flashed)
pulse train by averaging the pulse signal
with the PWM pulses, so the averaging
voltage over time (low-pass filtering).
processing occurs within the observer due
The waveforms in Fig.5 illustrate the
to persistence of vision. More specifically,
process of averaging a PWM signal. A
the PWM pulses to the LED must be faster
constant frequency (1kHz) and voltage
than the flicker fusion frequency, which
(1V) pulse waveform (red trace) starts
is the frequency at which a modulated
with a 25% duty cycle (the pulse is 1V
light source appears to have constant
for 25% of the time and 0V for 75% of
intensity to a human observer.
Fig.5. Example PWM waveform.
Practical Electronics | June | 2020
41
Fig.7. Relationship between the audio, triangle and PWM signals for a PWM modulator
(simulation of the circuit in Fig.6). Each triangle cycle lasts 30µs.
Fig.8. Simulation of the circuit in Fig.6 showing a complete cycle of the audio sinewave.
For class-D audio amplifiers the
loudspeaker and human ear are able to
provide the filtering required to recover
(demodulate) the original audio signal.
An important design requirement is that
to avoid potentially damaging effects,
loudspeakers should not be driven with
high-amplitude square waves in the audio
range (20Hz to 20kHz). Fortunately, PWM
pulse-train frequencies of hundreds of
kilohertz to megahertz cause very little
direct cone movement, and speakers will
respond to the average signal level – that
is the original audio signal. Furthermore,
V D D
the human ear also acts as a filter for
signals above about 20kHz, which is well
below the pulse frequency. However,
class-D amplifiers using relatively low
pulse frequencies may require an LC
filter for demodulation.
Filters may also be required on class-D
amplifier outputs for reasons other
than demodulation. For example, highfrequency switching can result in an
amplifier producing EMI (electromagnetic
interference) if an unfiltered PWM signal
is sent down relatively long speaker wires
without filtering. Often, ferrite beads can
be used for EMI suppression rather than
larger and more expensive LC filters.
Class-D amplifier circuit structure
Out+
Osci llator
In
C lass- D
m odulator
V D D
Out–
Fig.9. Outline schematic of a filterless
class-D amplifier.
42
The basic structure of a class-D modulator
is shown in the LTspice schematic in
Fig.6. This is not intended as a practical
design – it illustrates the basic concept,
and LTspice provides a convenient
means of drawing the waveforms. The
circuit comprises a comparator driven
by the audio signal and a triangle wave
– both waves have zero DC offset. The
audio signal is represented by a 1kHz
sinewave of ±1.8V amplitude. The triangle
waveform has the same amplitude as the
supply voltage, ±2.5V. The triangle wave
frequency is 33kHz, much lower than a
typical real class-D amplifier switching
frequency, but better for illustration of
the general shape of the waveforms.
The comparator will switch on (produce
a high output voltage) when the audio
signal is more positive than the trianglewave voltage. So for more positive audio
voltages the output will be on for a greater
proportion of the triangle wave cycle (see
Fig.7). Thus, the output duty cycle will
vary from 0% for an audio input voltage
of −2.5V to 100% for an audio input of
+2.5V. The example audio signal has a
lower amplitude than these extremes, so
that the modulator switches throughout
the audio waveform cycle (see Fig.8).
Fig.9 shows an outline schematic of a
filterless class-D amplifier. Typically, the
modulator has complementary outputs
which drive two MOSFET push-pull
switches in an H-bridge arrangement, as
shown. In practice, the circuit may be more
complex – for example, with feedback from
the output to an error amplifier before the
modulator in order to improve output
signal quality. If a demodulation filter is
used it will typically be a balanced secondorder LC filter, as shown in Fig.10. Note
that the circuit in Fig.9 is a fundamentally
analogue design – the ‘D’ in ‘class D’
does not mean ‘digital’. In digital audio
systems the PWM (or PDM) output can
be digitally generated, however the audio
may still be converted to analogue (by a
DAC) for modulation because fine control
of the pulse modulation requires digital
clock frequencies many times faster than
the pulse rate.
Class-G amplifiers
At a given supply voltage, the efficiency
of all the amplifier classes discussed
above reduces for low signal levels – a
significant proportion of the wasted power
is independent of the signal level, but is
dependent on the supply voltage, so a
relatively high proportion is wasted at low
signal levels. For typical audio signals,
such as music, the level may be relatively
low for a significant proportion of the
time, and, in typical usage, amplifiers
may be run at relatively low volumes
much of the time; only being turned up
loud occasionally. For the lower level
signals the amplifier could operate at a
lower supply level, and would be more
efficient, but then would not be able to
Out+
L 1
C 1
C 2
Out–
L 2
Fig.10. Typical class-D amplifier output filter
(if used).
Practical Electronics | June | 2020
Fig.11. LTspice
schematic to
generate illustrative
waveform for a
class-G amplifier
based on a
class-D amplifier.
in different contexts, but often the
distinguishing factor is that class G has
two fixed supply rails and the amplifier
switches between them, whereas class H
has a single variable power supply. Thus
class H is like driving the supply rail of
one amplifier from another amplifier.
Class-G and H ICs
correctly reproduce louder passages of
music, or satisfy the user when high
volume was required.
A solution to this supply/efficiency/
output issue is to run the amplifier at a
relatively low supply voltage most of the
time, but switch the supply to a higher
voltage (‘boost’ it) when a higher output
power is needed. Amplifiers operating
in this way are referred to as ‘class G’.
Class-G amplifiers can be based on classAB or class-D circuits. The version based
on class D is sometimes called class DG.
Fig.11 shows a version of the circuit
in Fig.6 in which the supply is switched
between two levels depending on the
audio signal level. Again, this circuit is
just for showing the basic concept and
for producing illustrative waveforms – it
is not a practical design. As before, the
pulse rate is lower than in a typical real
design to provide a clearer waveform
drawing. The supply voltage switching
is modelled using a pair of voltagecontrolled voltage sources (E1 and E2)
with a table function to set the ranges of
audio voltages to control the two supply
levels. The triangle waveform amplitude
is matched to the supply voltage using
a behavioural source (B1) to multiply a
1V waveform by the supply voltage. The
PWM waveform for a single cycle of the
audio signal is shown in Fig.12.
Class-H amplifiers
Class-H amplifiers extend the concept
of controlling the supply voltage by
providing either a continuously variable
or stepped supply voltage matched to
the signal. Specifically, for signals above
a certain level, the supply is set to a
voltage just above the signal output level
and sufficient for amplifier operation.
Again, this can be applied to both
class-AB and D amplifiers. Example
waveforms for a class-D version are
shown in Fig.13. The LTspice circuit
in Fig.11 was adapted to produce this,
with the continuously variable supply
being modelled using two behavioural
sources in place of the E1 and E2 tablebased controlled sources. The behavioural
source voltage was set using the equation
V=max(3,abs(V(audio))+1.4). The
pulse frequency is a little higher than the
previous examples to help emphasise
the pulse amplitude shape, although the
individual pulses are less clear.
The difference between class G and
H can be confusing or inconsistent
Fig.12. Amplifier waveforms from the circuit in Fig.11.
Design of class-D amplifiers operating
in class-G or H modes is far from trivial.
Fortunately, some ICs are available which
implement most of the functionality. For
example, the MAX98307 from Maxim
Integrated is a ‘3.3W, Mono Class DG
Multilevel Speaker Amplifier’ and the
TAS2562 from Texas Instruments is
‘6.1W Boosted Class-D Audio Amplifier
with IV Sense’. Both devices cost in the
order of a pound or two for one-off (GBP,
Mouser UK prices at the time of writing)
and are aimed at applications such as
mobile phones, tablet PCs, Bluetooth
speakers and consumer audio devices.
The MAX98307 has an audio input and is
relatively straightforward. The TAS2562
(the subject of the article which prompted
Julian to suggest this topic) is a more
complex beast, with digital audio input at
up to a 96kHz sample rate via I2S/TDM/I2C
interfaces. It has a large number of control
registers to customise device configuration
(the datasheet runs to 114 pages). Texas
Instruments provide software that can
calculate register values and programme
the device for development purposes.
The TAS2562 can operate in both
class-G and class-H modes with the
objective of extending battery life in
portable/personal electronic devices with
audio output. It uses an integrated DC-DC
converter to produce the ‘boost’ supply
voltage needed when higher output power
is required. All class-G and H amplifiers
require a means of measuring the signal
and using this to adjust the supply ready
for the signal level being amplified. This is
relatively straightforward for the TAS2562
because it receives a stream of digital
audio and holds multiple samples in its
internal digital signal processing logic.
Therefore, while a given sample is being
applied to class-D modulator, the power
supply control system can ‘look ahead’
at the later waveform samples to decide
what to do with the supply voltage. The
look ahead period can be in the range 1
to 20 samples, with something around
13 being typical at a 48kHz sample rate.
This value is set by a control register.
Simulation files
Fig.13. Illustrative waveform for a class-H amplifier based on a class-D amplifier.
Practical Electronics | June | 2020
Most, but not every month, LTSpice
is used to support descriptions and
analysis in Circuit Surgery.
The examples and files are available
for download from the PE website.
43
