Chapter 1 1

Chapter 1

Introduction
Fundamentally, only little has changed in the final stages of the audio reproduction chain
for decades. The electrodynamic transducer as invented by C.W. Rice and E.W. Kellog in
1925 still forms the basis for the majority of loudspeakers in use today and the principle
has only seen marginal changes within nearly 75 years. The widespread use can not be
justified by superior performance, in fact the principle of electric-acoustic conversion is
limited by numerous fundamental problems, that makes this ultimate stage in the audio
chain the weakest – by far. One essential limitation is the striking inefficiency. Generally, a
given amount of acoustic power requires orders of magnitude higher power input delivered
by the power amplifier. The power amplifier has the task of amplifying the audio signal to
a level that, combined with sufficient current to move the coil, produces the desired
acoustic level from the loudspeaker. The poor loudspeaker efficiency is very unfortunate,
since power amplifiers generally have to be capable of delivering large amounts of
undistorted power, to produce the subjective levels demanded by the consumer.

The field of audio power amplification has equally suffered from a lack of real break-
through inventions for decades. Thus, sound reproduction today is founded on a few power
amplifier principles that are characterized by a linear operation of the output transistors.
The advantages include topological simplicity and good performance, but the linear
amplifier principles suffer from low efficiency, which is critical since the power amplifier
handles considerably amounts of power. Accordingly, power amplifiers are in general
provided with massive heat sinks of extruded aluminum to cope with the heat
development. Negative side effects of inefficient power amplification include high volume,
weight, cost and reliability problems. Moreover, the power amplifier has low energy
utilization, which is clearly not an attractive feature in this energy-conscious area.
2 Introduction

PM PC Analog PL PA
DC
Power
Power Supply
Mains Amplifier

Fig. 1.1 Power flow in audio reproduction chain.

Consider the model of essential power flow in a typical audio system, shown in Fig. 1.1.
To illustrate the low power utilization, a typical 100W power amplifier is considered. The
power flow at two specific output levels is given below.

Situation PM PC PL PA
Typical 15W 10W 100mW 1mW
1/3 of max. 115W 90W 30W 300mW

Clearly, the transducer is the fundamental source of the efficiency problems, i.e. an
efficiency improvement by and order of magnitude would virtually eliminate the need for
power amplification, as we know it today. However, most of the power is dissipated in the
power amplifier due to low efficiency in this stage.

The primary objective of the research resulting in the present dissertation has been to
invent practical power amplification methods with significant improvements in efficiency
at all levels of operation – without any compromises on audio performance. Improving
power amplifier efficiency isolated will have significant influences on overall system
efficiency, especially at lower levels of operation where much can be gained.

It might seem paradoxical, that this level of efficiency is tolerated in the industry,
especially since a low efficiency only has negative side effects. There are several answers
to this apparent paradox. Principles for more high efficiency power amplification – so
called Class D or swithing power amplifiers - have been known for decades, however
previous findings have shown several problems in terms of achieving the e.g. desired
efficiency and audio specifications. On the other hand, methods to achieve sufficient levels
of performance with linear power amplification are well known, and linear power
amplification have over several decades of time established a reputation for good quality.
A second problem is, that the audio community is highly conservative, often dominated by
religious belief rather that scientific documentation and objective evaluation. Furthermore,
only few have been interested in environmental issues despite the potential for dramatic
improvements. Finally, amplifier weight, volume and energy consumption (!) has actually
been considered a quality parameter (the larger and heavier the better), although there has
never been any scientific documentation for any correlation between sound quality and
these parameters. It is to expect however, that environmental issues will also reach
consumer products. A high efficiency could become an attractive parameter in the future
especially since labeling or standardization in consumer electronics is on its way [Ni95].

1.1 Audio power stage topologies

A brief introduction to commonly used circuit techniques for audio power amplification is
given in the following with focus on their efficiency characteristics. It is common to
differentiate between the different amplifier principles are by their class. A list of used
Chapter 1 3

Class Characteristics
A Conducts signal current throughout the cycle of the signal waveform (3600
conduction)
B Conducts signal current exactly for one-half of the cycle of the input-
signal waveform (1800 conduction)
AB Class B with bias to avoid crossover distortion.
C <1800 conduction with resonant loading. This method is primarily for RF
frequency, and is only rarely used in the audio frequency range.
D 00 conduction. The power stage transistors are switched which in theory
prevents the system from entering the active region.
B2, G, H Extension of class B where more complex power supply circuitry is used
to improve efficiency.
Fig. 1.2 Amplifier techniques reviewed.

classifications is listed in Fig. 1.2, with a short description of the characteristics [Be88].
The Class AB and B output stage topology, shown in Fig. 1.3 (top), forms the basis for the
majority of power amplifiers today, and design techniques to realize high quality class B
amplifiers have been known for decades. However, since the output voltage is derived
form the supply voltage via the output transistors, it follows that the difference between the
output voltage and the rail voltage must be dropped across the output transistors. This
results in a wasteful dissipation of energy in the output transistors. The effective voltage
drop across the output stage transistors can be reduced by the complex class B2 or class G
configuration shown in Fig. 1.3.

1.1.1 Power and energy efficiency

Since amplifier efficiency is an essential parameter throughout the thesis, the efficiency of
the most widely used power amplifier principles is investigated and compared. It is trivial
to derive analytic expression for the efficiency vs. relative output level as shown in
Appendix A for the three amplifier principles. The power efficiency is the ratio of utilized
power over the supplied power:

PL ( x ) (1.1)
ηP ( x) =
PS ( x )

Where x denotes the relative output level. The amplifier output power is:

V2
PL ( x ) = x 2 (1.2)
2 RL

As shown in appendix A, the efficiency for the three output stage configurations are:

π
ηB ( x) = x (1.3)
4

ìπ x
ï4 α (x < α)
ïï
η B 2 ( x, α ) = í π x 1 (x ≥ α) (1.4)
π
ï4
ï α + (1 − α ) ò 2 α sin(ωt )d (ωt )
ïî α sin( )
x
4 Introduction

V
V vL(t)

vL(t) 1
t

-V

Class B

 V

V
vL(t)
V V

1
t

V V

-V

 V

Class B2

V V(t)
vL(t)
V(t)

vL(t) 1
t

V(t)

-V

Class G

Fig. 1.3 Three widely used output stage configurations Class B, Class B2 and Class G.
Chapter 1 5

π 1
η G ( x, β ) = x ( )
(1.5)
4 β + x π4

Fig. 1.4 shows the ideal efficiency vs. relative output level for the three amplifier
principles, and Fig. 1.5 equally shows the relative power dissipation vs. x . Class B2 and G
and achieve reasonable efficiencies at higher output powers, but all topologies still have
significant power dissipation at all output levels. In more aspects, these idealized models
are not sufficient to estimate the real output stage efficiency especially at lower output
levels. There are practical limitations further decreasing the efficiency of the output stage
topologies, such as the necessary quiescent current to linearize the output stage and the
saturation voltages in the output stage transistors. Subsequently, more realistic models of
the three output stage topologies have been investigated that incorporate these important
effects. Fig. 1.6 and Fig. 1.7 shows the efficiency and relative power dissipation of these
more realistic output stage topologies. Note how the typical efficiency of all topologies is
extremely low, primarily due to the significant power loss at quiescence.

1.1.2 Energy efficiency considerations

There is only little correlation between power efficiency and the amplifier energy
consumption, since the power efficiency is typically specified at the level of maximal
power dissipation or at the maximal output level. An alternative efficiency measure - the
energy efficiency - is defined in the following. The basis is investigations of the general
consumer behavior in terms of an average time distribution of volume control positions.
Such a distribution will vary as a consequence of e.g. loudspeaker sensitivity, room size,
user age and numerous other parameters. However, it is possible to generalize [Ha94b] and
define four subjective listening levels as given in Table 1.1. The distribution should be
interpreted as: In 89% of the time, the average user listens to background music with an
average output level of -40dB etc. Since the distribution might vary dependent on
application (e.g. in professional systems the distribution would be different), a general time
distribution is considered:

(n1 , PL,1 , PS ,1 ), ( n 2 , PL,2 , PS ,2 ),K( n N , PL, N , PS , N ) (1.6)

( n j , PL, j , PS , j ) refers to that the output power in average is PS , j in n j percent of the time
and PS , j refers to the supplied power at the given output power. The energy efficiency is
derived as the ratio of the average output power and the average supplied power:

N
å ni ⋅ PL,i
i =1
ηE = N (1.7)
å ni ⋅ PS ,i
i =1

Distribution in time Output level (rel.) Subjective level

0% 0 dB Clipping
1% -9 dB Party
10 % -24 dB Listening
89 % -40 dB Background
Table 1.1 Average time distribution of volume control positions.
6 Introduction

100

90

80

70

60
Efficiency (%)

Class BG
Class
50

40
Class B2
ClassGB
Class
30

20

10

0
−40 −35 −30 −25 −20 −15 −10 −5 0
Relative output level x (dB)
Fig. 1.4 Efficiency vs. relative output level for ideal power amplifier output stages.

0.5

Class B
)

0.4
L,max
Power dissipation (P /P
D

0.3

Class B2
0.2

0.1

Class G

0
−40 −35 −30 −25 −20 −15 −10 −5 0
Relative output level (x)
Fig. 1.5 Power dissipation vs. relative output level for ideal power amplifier output stages.
Chapter 1 7

100

90

80

Class G
70

60
Efficiency (%)

50

Class B2
40

Class B
30

20

10

0
−40 −35 −30 −25 −20 −15 −10 −5 0
Relative output level x (dB)
Fig. 1.6 Efficiency and power dissipation for realistic models of class B, B2 and G power amplifier
output stages topologies.

Class B

0.5

Class B2
Power dissipation (PD/PL,max)

0.4

0.3

Class G
0.2

0.1

0
−40 −35 −30 −25 −20 −15 −10 −5 0
Relative output level (x)
Fig. 1.7 Efficiency and power dissipation for realistic models of class B, B2 and G power amplifier
output stages topologies.
8 Introduction

25

Class G
20

Energy efficiency (%)

15
Class B2

10

Class B

0
−3 −2 −1 0
10 10 10 10
Quiescent power loss factor
Fig. 1.8 Energy efficiency of Class B, Class B2 and Class G.

The parameter can be used to minimize the energy consumption in any system. Let the
quiescent power loss factor λ be defined as:

PS (0)
λ= (1.8)
PL,max

Fig. 1.8 illustrates the energy efficiency for the realistic models of class B, B2 and G
power output stages, as a function of λ . The typical energy efficiency for a class B output
stage is only:

n E ≅ 1 − 2% (1.9)

This holds for λ between 0.1 and 0.2, which is typical for a class B output stage. Note that
the energy efficiency for an ideal class B amplifier is only 7.1%, due to the inherent losses
bound to the linear operation of the output transistors. Since the power amplifier operates
in the “background music” mode most of the time, the important parameters in terms of
energy efficiency are not surprisingly the power dissipation at low output levels and
especially the quiescent power dissipation. The energy efficiency of class B2 and class G
can be made considerably better (see Fig. 1.8) if the λ is sufficiently low. However, this is
not easy to obtain in any of the two alternatives. For class B2, lowering α will compromise
the efficiency at higher output levels (no gain compared to class B) and for class G, the
switching power supplies will cause the quiescent power loss λ factor to be high. To
conclude, all output stages have their limitations, and more than a doubling in energy
efficiency by optimization cannot be expected.

1.1.3 The switching (Class D) power stage topology

Based on this initial investigation of output stage topologies it is concluded that
considerable improvements in efficiency cannot be achieved with output transistors
operating in the linear range. A completely switching power stage is needed. This approach
Chapter 1 9

DC power supply

Analog or Pulse
digital input Modulator Power Switch Demodulation

Control

Fig. 1.9 General Pulse Modulation Amplifier (PMA) utilizing a class D power stage.

has the inherent advantage of a 100% theoretical efficiency. With the continuing
improvements in power switching devices, the efficiency will converge towards this
theoretical optimum as time progresses. Throughout the years the fundamental principle of
power amplification using switching technology has been called class D, switching power
amplification, digital power amplification and PWM power amplification. Recently, what
is believed to be a more suitable and general designation – Pulse Modulation Amplifiers
(PMA) – has been introduced by the author [Ni97a]. This general designation will be used
henceforth.

1.2 The Pulse Modulation Amplifier (PMA)

Fig. 1.9 shows the general PMA topology. The four fundamental blocks are the pulse
modulator, the switching power stage, the demodulation filter and the error correction
block. The pulse modulator may be based on either analog or digital pulse modulation
techniques, correspondingly referenced to as analog PMA and digital PMA systems. The
power switch converts the pulse modulated signal to power level. Following, the power
pulse modulated signal is feed to a filter to reconstitute the modulated signal. The control
system serves to compensate any errors that are introduced in each of the three essential
blocks of the system.

The pulse modulator is the heart of the PMA system. Two analog pulse modulation
methods for PMAs are PWM and PDM, shown in Fig. 1.10. The methods may be
implemented in both the analog and digital domain. The modulator output generally
contains three distinct elements:
• The modulated signal.
• Distortion components related to the modulated signal.
• A high frequency spectrum.

The high frequency output is composed of either discrete components related to the carrier,
noise shaped noise or a combination of both. The pulse modulator can be based any
scheme performing a coding of the modulating signal to a pulse modulated form. One of
the objectives of this thesis is to research in coding schemes that provides optimal PMA
performance.

Power amplification based on pulse modulation techniques in combination with a

switching class D power stage have been known for decades. Some of the first designs
were already presented by e.g. Sinclair and Johnson [Jo67], [Jo68] some 30 years ago.
However, until recently the research has been quite modest with only a few noteworthy
publications as e.g. [At83], [Ha91]. Within the last decade, the field has “resurrected” now
with focus on digital pulse modulation methods for digital PMA systems. An exhaustive
10 Introduction

t t

t t

Fig. 1.10 PWM and PDM pulse modulation methods.

set of publications exists on this specific subject, e.g. [Sa86], [Le91], [Go91], [Go92],
[Ha92] and [Hi94]. The digital PMA approach was considered a break-through in audio
power amplification. Nevertheless, it has proven very difficult to realize the acceptable
audio performance in both analog PMA and digital PMA system. The problems relate to
the non-linear power conversion and demodulation.

1.2.1 Design problems and challenges

PMA systems and audio power amplifiers in general are complex non-linear systems.
Thus, power amplifiers are subject to varying input signals, generally drive reactive loads,
are most often supplied by non-regulated power supplies which supply the other channels
for sound reproduction. Furthermore, power amplifiers might be subjected to overload
situations if not prevented. No power amplifier can be optimized without a simultaneous
consideration of a broad range of desired specifications. Furthermore, the tolerance of
these specifications has to be considered. The essential issues to consider when designing
and evaluating power amplifier systems are outlined below.

Gain
A specification of the insertion gain of the system typically specified in dB.

Frequency response / Bandwidth

The ability of amplifier to amplify signals over a range of frequencies, with defined
source and load. Specifications are generally a –3dB bandwidth, and a tolerance on
the deviation from the desired response at any frequency up to the bandwidth limit.

Harmonic distortion / Intermodulation distortion

The non-linear behavior of the amplifier causes harmonic distortion (THD) and
intermodulation distortion (IMD). Moreover, the distortion will in general depend on
parameters as signal level, frequency and load parameters. Distortion has to be well
controlled within this parameter space. Distortion is normally specified in
percentage or dB. Various IMD measurement methods exist as the two-tone CCIF,
SMPTE or Transient Intermodulation Distortion (TIM).

Noise / Signal-to-noise ratio / Dynamic range

All amplifiers have internal noise sources that contribute to the output noise. Typical
specifications are the residual noise referred to the output with terminated input or
the Signal-to-noise ratio relative to a given output level e.g. 1W. Also frequently used
is the dynamic range, which is the relationship between the maximal RMS voltage
output before clipping and the RMS of the residual noise.
Chapter 1 11

Output impedance / Loading

The load impedance is generally frequency dependent with resonant peaks etc. The
influences of a variable load in system frequency response should be minimized.
Furthermore, the amplifier output impedance should be as low as possible to cope
with variations in nominal load impedance.

Power Supply Rejection

The power amplifier has to cope with the inevitable power supply perturbations. The
amplifier should be able to suppress these perturbations such that the output is not
influenced. A widely used specification is the power supply rejection ratio (PSRR),
which is the sensitivity of the output to perturbations on the power supply. The
rejection of such perturbations has to be controlled over the complete bandwidth.

Stability
A control system is generally required to secure robust performance for the power
amplifier. This introduces a potential risk of instability. The amplifier should be
prevented from instability under all circumstances, since this will generally have
dramatic consequences as a burn out of the speaker of the amplifier itself.

There are no definitive margins between what is required to be acceptable and

unacceptable. However, a set of parameters corresponding to satisfying performance in
most applications can be specified:

Parameter Condition Value

Bandwidth -3dB > 60KHz
Power bandwidth -3dB > 20KHz
Frequency response 20Hz – 20KHz <± 0.2dB
THD @ 1W / 1KHz < 0.01 %
THD 20Hz-20Hz (complete range) < 0.05 %
IMD CCIF two tone < 0.01%
SNR A weighted @ 0dBW >90 dB
Dynamic range A weighted > 110dBA
Load range 2-16 Ω
PSRR All frequencies > 40dB

A power amplifier with a reasonable power handling capability (e.g. 100W) is considered.

1.2.2 Other considerations

The application range for high efficiency power amplification is very broad. Active
speaker systems with dedicated speakers and power amplifiers for each frequency band is a
specifically interesting application. This enclosed environment offers some special
advantages:
• The load and amplifier can be matched perfectly. Load variations (typically 2, 4 and 8
ohms) do not have to be considered.
• Connection wires from amplifier to speaker can be minimized.
• Bandwidth limited amplification can be utilized to optimize the efficiency.

The general spectral amplitude distribution of music material only emphasizes that the
technology has certain advantages in active speaker systems. Almost independent of music
material, the average acoustic power in the tweeter band 4KHz – 20KHz is much lower
12 Introduction

Tw eeter
(10% )
A m plifie r

M idrange
Input (45% )
A m plifier

Woofer
A m plifier (45% )

Fig. 1.11 An active speaker system (here 3 way) based on separate amplifiers for each band.

than in the other bands. Fig. 1.12 illustrates a more specific analysis. This will be reflected
in the power amplifier size for each band as illustrated with a typical active 3-way speaker
system in Fig. 1.11. Thus, the power handling capability is only around 10-15% of the total
power in the tweeter band. Optimization of energy efficiency in active speaker systems can
as such be carried out by improving efficiency in the lower 20% of the frequency band
only. In this particular environment, the PMA design challenge is clearly somewhat
different from the general case.

1.3 Thesis structure

The primary objective of the research resulting in the present dissertation has been to
invent practical power amplification methods with significantly improved efficiency at all
levels of operation – without any compromises in terms of audio performance. It will
become apparent when reading the thesis, that this has required multi-disciplinary research
involving such diverse fields as e.g.; analog and digital modulation theory, power
electronics, DSP, semi-conductor physics and control systems.

The thesis is divided in two volumes, the main part consisting of 11 chapters divided in
four parts, and a second volume consisting of the three appendices A, B and C.

100

90

80

70
Percentage of time

60

50 Woofer

Lower mid
40

Upper mid
30

20
Tweeter
10

0
−70 −60 −50 −40 −30 −20 −10 0
Relative level
Fig. 1.12 An example of the distribution of signal levels on a CD in four bands. The bands are
divided at 650hz, 1500hz and 4.5KHz. The programme material is "The Division Bell" by Pink
Floyd.
Chapter 1 13

Chapter 1 has introduced the motivating factors and the foundation for the research.

Part I (Chapters 2-3) presents a comprehensive analysis of analog and digital pulse
modulation methods suitable for analog PMAs and digital PMAs, respectively.

Chapter 2 is devoted to analog pulse modulation methods. A broad set of pulse

modulations methods are subjected to a fundamental analysis of their suitability in analog
PMA systems. The chapter contributes to fundamental modulation theory by introducing a
novel family of modulation methods – Phase Shifted Carrier Pulse Width Modulation
(PSCPWM).

Chapter 3 is devoted to digital pulse modulation methods for digital PMAs. Previously
presented methods are reviewed. A simple design methodology is presented for digital
PWM modulators.

Part II (Chapters 4-5) continues to the second major block of the PMA - the power stage.
Power stage structures are synthesized and analyzed, and methods for optimal power stage
implementation are devised.

Chapter 4 is dedicated to a fundamental analysis of error sources within PMA systems. It is

shown how the power conversion stage seriously effects all the important parameters of the
system, i.e. linearity, noise and efficiency. Modulator error sources are also investigated.

Chapter 5 is devoted to efficiency optimization in the power conversion stage of PMAs.

Starting with a simple switching leg, the analysis extends to the general multi-level
PSCPWM power stage topologies.

Part III (Chapters 6-9) continues to the third major block of the PMA – control system
design.

Chapter 6 investigates the application of robust linear control to analog PMA systems. A
methodology for control system design is introduced. Three fundamental linear control
methods are investigated, and robust case example designs are synthesized and analyzed.

Chapter 7 is devoted the presentation of a control method, dedicated to solve the

fundamental problems in analog PMA systems – Multivariable Enhanced Cascade Control
(MECC).

Chapter 8 investigates the application of non-linear control methods for analog PMAs. The
focus in on a new non-linear modulator/controller structure – Three level One Cycle
Control (TOCC). Advantages and disadvantages compared with linear control are
emphasized.

Chapter 9 is dedicated to the complex issues of error correction in digital PMA systems. A
new control method for enhanced power amplification of a pulse modulated signal is
presented – Pulse Edge Delay Error Correction (PEDEC). PEDEC based digital PMA
topologies are presented and evaluated.
14 Introduction

Part IV (Chapter 10) is devoted to implementation and evaluation.

Chapter 10 is dedicated to the practical evaluation of all investigated principles and

topologies. Performance specifications for the various prototypes developed during the
project are presented.

Chapter 11 summarizes the essence of the thesis and draws the essential conclusions.

Volume II contains three appendices:

Appendix A supplements Chapter 1. Investigations of power and energy efficiency for

various amplifier concepts are carried out.

Appendix B supplements Chapter 2. Details of the derivation of analytical double Fourier

series expressions for a broad range of analog pulse modulation methods are given.

Appendix C is a complete reproduction of the conference and journal papers that have been
published at during the project.