slyt198.pdf
(
261 KB
)
Pobierz
Texas Instruments Incorporated
Amplifiers: Audio Power
Reducing the output filter
of a Class-D amplifier
By Mike Score
System Specialist, Audio
A properly designed Class-D output filter limits supply
current, minimizes EMI and protects the loudspeaker
from switching waveforms. However, it also significantly
increases the total design cost. The current recommended
second-order output filter for the TI TPA005D02 is 30%
of the audio power amplifier (APA) solution cost. This
application note details the second-order Butterworth
filter and two reduced filtering techniques, each providing
a different price/performance node. The first alternative
to the Butterworth filter reduces the output filter by half
and the second option completely eliminates the filter. The
filters were tested using Texas Instruments TPA005D02
Class-D APA. Design decisions based on the measured
results and conclusions are drawn to provide practical
solutions for applications with different price/perform-
ance nodes.
Second-order Butterworth low-pass filter
The second-order Butterworth low-pass filter is the
most common filter used in Class-D amplifier applica-
tions. This filter uses two inductors and three capacitors
for a bridged-tied-load (BTL) output, which can be seen
in Figure 1. The primary purpose of this filter is to act as
an inductor at the switching frequency. The inductor is
needed to keep the output current constant while the
voltage is switching. If the outputs do not see an induc-
tive load at the switching frequency, the supply current
will increase until the device is unstable. Higher induc-
tance at the output yields lower quiescent current (sup-
ply current with no input) because it limits the amount of
output ripple current. The filter protects the speaker by
attenuating the switching frequency. The inductors, L1
and L2, and the capacitor C1 form a differential filter that
has an attenuation slope of 40 dB per decade. The majori-
ty of the switching current flows through C1, leaving very
little current to be dissipated in the speaker. The filter
also reduces EMI. The previously described differential
filter reduces differential radiation. Inductors L1 and L2
and capacitors C2 and C3 form common-mode filters that
further reduce EMI. See the “TPA005D02 Class-D Stereo
Audio Power Amplifier Evaluation Module User’s Guide”
1
for details on how the output filter was designed.
Half filter
The half filter (shown in Figure 2) eliminates one of
the inductors and the two capacitors to ground. The
other inductor must be doubled (for half filter C = C1 of
the full filter) to obtain the correct cut-off frequency.
This filter is still inductive at the switching frequency
because capacitor C looks like a short at the switching
frequency and each amplifier output sees inductor L. The
supply current is even lower for this application because
the total inductance stays the same as the full filter but
the DC resistance (DCR) from the inductors is reduced
causing less power loss in the filter. The speaker is still
protected in this application although one of the speaker
terminals is directly tied to the switching waveform. The
differential signal is still filtered and the audio output
rides on the switching waveform. Although this filter
attenuates the differential signal, which reduces EMI, it
does not attenuate the common mode signal, which could
cause EMI problems. Methods to reduce EMI will be dis-
cussed in a later section.
No filter
The filter can be eliminated if the speaker is inductive
at the switching frequency. How is it possible to eliminate
the filter and place a pulse-width modulated (PWM)
square wave directly across the speaker? The PWM wave-
form is the sum of the switching waveform and the input
audio signal with gain. The human ear acts as a band-pass
filter, hearing only the frequencies between approximate-
ly 20 Hz and 20 kHz. The switching frequency is much
greater than 20 kHz, so the only signal heard is the input
Figure 1. Full second-order Butterworth filter
L1
C2
C1
L2
C3
Figure 2. Half filter
L
C
Continued on next page
www.ti.com/sc/docs/products/analog/tpa005d02.html
19
Analog Applications
August 1999
Analog and Mixed-Signal Products
Amplifiers: Audio Power
Texas Instruments Incorporated
Continued from previous page
Table 1.
Quiescent current for various filter applications using
the TPA005D02 and the TPA0102
audio signal with gain. The main drawback to not using a
filter is that the switching waveform is dissipated in the
speaker, which leads to a higher quiescent current, I
DD(q)
.
This is because the speaker is both resistive and reactive,
where an LC filter is almost purely reactive. A more
inductive speaker yields lower quiescent current, so it is
better to use a two or more layer voice coil speaker for a
high inductance in this application. A concern with the
switching waveform being dissipated in the speaker is
that it may cause damage to the speaker. A primary con-
cern is speaker wear due to a rail-to-rail square wave
driving the speaker whenever the amplifier is on. For a
250-kHz switching frequency, this is not an issue because
the speaker cone movement is proportional to 1/
f
2
for fre-
quencies beyond the audio band, thus the amount of cone
movement at the switching frequency is insignificant.
2
Damage could occur to the speaker if the voice coil is not
designed for the added power. The added power to the
load is less than the added power taken from the supply
because the device cannot have efficiency greater than
100%. The added power can be calculated using
Equation 1, where
P
SW
is the added power dissipated in
the speaker,
I
DD
(
q
)(with speaker load)
is the quiescent current
measured with speaker load,
I
DD
(
q
)(no load or filter)
is the qui-
escent current measured with no load,
V
DD
is the supply
voltage and
N
is the number of channels.
L
I
DD(q)
APPLICATION
LOAD
(µH)
(mA)
Full filter
Any size resistor or speaker load
15
39
Half filter
Any size resistor or speaker load
15
42
L = DS3316-P-xxx
22
35
33
29.5
47
27
Half filter
Any size resistor or speaker load
68
25
L = DS5022-P-xxx
100
24
150
23
No filter
Notebook speaker
215
NXT speaker
199
TI PnP speaker
83
Bose 151 speaker
83
TPA0102
Any load
19
2-W amplifier. A measurement was made in each case
with the TI TPA0102 used for comparison. The measure-
ment set-up and results will be discussed for each test.
Quiescent current
The quiescent current for the full-, half- and no-filter
applications is tabulated in Table 1. The quiescent current
of the full- and half-filter applications was independent of
the load and varied greatly with inductor value, but the
no-filter application’s quiescent current was dependent on
the inductance of the speaker. The full-filter quiescent
current was measured using the filter designed for a 4-ohm
load, where L1=L2=15 µH, C2=C3=0.22 µF, and C1=1 µF,
with the components labeled as shown in Figure 1. The
quiescent current of the half-filter application shown in
Figure 2 was measured with C = 1 µF, and L was varied to
show how increasing inductance decreases ripple current
at the output. The recommended half-circuit filter for a
4-ohm load is L=33 µH and C=1 µF, which had a quies-
cent current of 29.5 mA. The recommended half filter for
an 8-ohm load is L=68 µH and C=0.56 µF, which had a
quiescent current of 25 mA. The quiescent current for
the filterless application varies with load. The TI PnP
speaker and the Bose 151 speaker had quiescent current
of 83 mA while the lower inductance flat panel NXT
speakers
3
had a quiescent current of 199 mA. A commer-
cial notebook speaker was even less inductive and exhib-
ited a quiescent current of 215 mA. The Class-AB amplifier
had a quiescent current of 19 mA, which was much lower
than the Class-D with no filter and is less than half the
Class-D with full filter, but not much lower than the
Class-D with half filter.
(1)
(
I
-
I
)
V
DD
(
q
)(
with
speaker
load)
DD
(
q
)(
no
load
or
filter)
DD
P
<
SW
N
The added power from the switching waveform can be
calculated. A supply current of 83-mA was measured with
the TPA005D02 EVM connected to the speakers available
with the TI Plug-n-Play base kit. The supply current with
no load was measured at 23 mA. A 5-V supply was used
and, because TPA005D02 is a stereo device,
N
=2; there-
fore, the maximum added switching power dissipated in
the speaker is 150 mW. Originally, 3-watt speakers were
required. The filterless solution requires 3.15-watt speak-
ers. Another concern is that not filtering causes the ampli-
fier to radiate EMI along the lines from the amplifier to
the speaker. The filterless application is not recommend-
ed for EMI-sensitive circuits.
Measured results
The quiescent current, total harmonic distortion plus
noise (THD+
N
) and intermodulation distortion (IMD)
were measured for the full-, half- and no-filter applications.
Each measurement was done using a TPA005D02 Class-D
20
www.ti.com/sc/docs/products/analog/tpa005d02.html
Analog and Mixed-Signal Products
August 1999
Analog Applications
Texas Instruments Incorporated
Amplifiers: Audio Power
Total harmonic distortion plus noise (THD+N)
Total harmonic distortion plus noise (THD+
N
) was
measured with the Audio Precision II analyzer. In each
measurement, an RC filter with a cut-off frequency of
37 kHz was added between the output to ground to filter
the common-mode signal into the Audio Precision. The
bandwidth of the Audio Precision was set from 10 Hz to
22 kHz and an internal 20-kHz low-pass filter was also
used to ensure that the switching frequency did not influ-
ence the THD+
N
measurement. Band-limiting the mea-
surement ensured that only the audible harmonic distor-
tion and noise were measured. The amplifiers were set at
a gain of 22.5 V/V. THD+
N
versus output voltage at 1 kHz
was measured with the TPA005D02 with full, half and no
filter. The same test was performed using the TPA0102.
Both can be seen in Figure 3. The Class-D full- and half-
filter applications had approximately the same THD+
N
across all power levels. The Class-D without the output
filter actually had lower THD+
N
at the lower power levels
than the TPA005D02 with the filter and the TPA0102.
The TPA0102 had approximately the same THD+
N
as the
TPA005D02 with the full and half filters at low to mid
powers, and lower THD+
N
at the higher powers.
IMD
Intermodulation distortion (IMD) occurs when two or
more signals are input into an amplifier and the sum and
difference of the input frequencies are present at the out-
put. IMD is a good measurement of linearity (the lower
the IMD, the more linear the device under test). IMD is
Continued on next page
Figure 3. THD+Nvs. V
out
8
2
THD+N
1
(%)
0.5
Full Filter
Half Filter
0.2
No Filter
0.1
TPA0102
0.05
400m
500m
600m
700m 800m
1
2
3
4
V
out
(V
rms
)
Figure 4. SMPTE IMD vs. input voltage
10
Full Filter
Half Filter
1
SMPTE
IMD
0.1
TPA0102
( % )
0.01
f1 = 60 Hz, f2 = 7 kHz
Gain = 22.5 V/V
0.001
No Filter
0.0001
20
30
40
50
70
10
60
V
in
(mV
rms
)
www.ti.com/sc/docs/products/analog/tpa005d02.html
21
Analog Applications
August 1999
Analog and Mixed-Signal Products
Amplifiers: Audio Power
Texas Instruments Incorporated
Continued from previous page
Class-D without an output filter had the lowest CCIF IMD,
which was 0.01%.
Design decisions based on results
Class-D without output filter
The Class-D without the output filter not only had
lower THD+
N
and IMD than the Class-D with the filter, it
also outperformed the Class-AB device. The two disad-
vantages of using the Class-D amplifier without the out-
put filter are high quiescent current and high EMI.
Quiescent current can be lowered using a speaker with a
high inductance, but it is doubtful whether it could ever
be lower than an application with a filter. EMI can be low-
ered using ferrite beads at the output of the amplifier,
shielded speakers and shielded speaker wires running to
the speakers. EMI can also be reduced by making the dis-
tance from the amplifier to the speaker as short as possi-
ble and by keeping the positive and negative output wires
very close together to reduce common-mode radiation. A
good application for a Class-D amplifier without a filter is
one where quiescent current and EMI are not important,
but system cost, maximum power supply and heat are
important (e.g., powered speakers). An example of this is
a powered speaker. The Class-D without the filter is less
efficient than the Class-D with the filter at lower output
levels, due to the higher quiescent current. However, the
Class-D efficiency is approximately the same with and
without the filter at high output levels (2 to 3 times more
efficient than Class-AB). A 10-W powered speaker could
use a Class-D amplifier without the output filter or heat
sink and use a lower-rated power supply than a Class-AB
amplifier. The system cost of this application is less than
the Class-D with the full or half filter because the filter is
eliminated, and less expensive than the system cost of
the Class-AB solution. The heat sink is eliminated and the
the ratio of magnitude of the sum and difference signals
to the original input signal.
IMD = [(
V
f
2–
f
1
+
V
f
2+
f
1
)
2
+ (
V
f
2–2
f
1
+
V
f
2+2
f
1
)
2
+ (
V
f
2–3
f
1
+
V
f
2+
3f
1
)
2
… ]
0.5
/
V
f
2
(2)
where
V
f
2
is the voltage at the input frequency
f
2,
V
f
2+
f
1
is
the voltage at the sum of input frequencies
f
1 and
f
2,
V
f
2–2
f
1
is the voltage at the difference of input frequencies
f
1 and 2*
f
2, etc.
The Society of Motion Picture and Television Engineers
(SMPTE) standard IMD test is the most common IMD
measurement. The SMPTE IMD test inputs a low-
frequency 60-Hz and high-frequency 7-kHz sine wave into
the device. The low-frequency sine wave has four times
the amplitude of the high-frequency sine wave.
4
The
SMPTE IMD versus input voltage of the Class-AB and
Class-D amplifier with the full, half and no filter was mea-
sured and can be seen in Figure 4. The same set-up used
for the THD+
N
measurement was used for the IMD mea-
surements. The full- and half-filter circuits had equal IMD
that was slightly higher than the Class-AB amplifier. The
Class-D without a filter had the lowest IMD.
The CCIF, or twin-tone, IMD measures IMD of the
amplifier using two high-frequency input signals of equal
amplitude. The plot in Figure 5 shows CCIF IMD versus
difference frequency. The center frequency was set to
13 kHz and the difference frequency was swept from
80 Hz to 1 kHz. The Class-D amplifier with full filter had
the highest CCIF IMD, ranging from 0.4% to 0.5%, and
the half-filter application CCIF IMD was approximately
0.1% lower over the tested frequency range. The Class-AB
amplifier had a lower CCIF IMD, which was 0.05%. The
Figure 5. CCIF IMD vs. difference frequency
10
f1 = 13 kHz
Gain = 22.5 V/V
Not Band Limited
1
Full Filter
Half Filter
CCIF
IMD
(%)
0.1
TPA0102
No Filter
0.01
0.001
100
200
300
400
500
600
700
800
900
1000
Difference Frequency (Hz)
22
www.ti.com/sc/docs/products/analog/tpa005d02.html
Analog and Mixed-Signal Products
August 1999
Analog Applications
Texas Instruments Incorporated
Amplifiers: Audio Power
power supply is reduced in a 10-W Class-D filterless appli-
cation, but the Class-D amplifier is slightly more expensive
than an equivalent Class-AB amplifier. The speaker induc-
tance must be high, and the amplifier must be close to
the speaker in this application.
Class-D with half filter
Class-D with a half filter had a lower quiescent current
and performed as good as or better than the Class-D with
full filter in THD+
N
and IMD. The quiescent current lowers
as the inductor of the half filter increases. As inductance
increases, peaking occurs at the corner frequency of the
filter. The corner frequency can be set outside the audio
band so the peaking has no effect on sound quality.
Peaking occurs regardless of LC design with a speaker
load when designing for a resistive load. The half filter
designed for a 4-ohm resistive load uses a 33-µH inductor
with a 1-µF capacitor, and the half filter designed for an
8-ohm resistive load uses a 68-µH inductor and a 0.56-µF
capacitor. Each of these examples exhibits peaking at the
corner frequency because the speaker is not purely resis-
tive at the corner frequency. An RC network can be used
across the load to reduce reactance of the load to limit
peaking. If quiescent current is very important, the
designer can increase L and lower C, decreasing the qui-
escent current and keeping the corner frequency in place.
The designer should design the filter with the speaker
load to ensure the corner frequency peaking is outside
the audio band. The only disadvantage with the half-filter
application is that it has higher common-mode EMI than
the full filter due to the filter not having a common-mode
filter. The common-mode EMI should not be a problem in
most systems if the positive and negative output signal
paths are very close together to cancel common-mode
radiation. The filter should be as close to the amplifier as
possible to reduce EMI. EMI can be further reduced with
ferrite beads, shielded speaker wire and using good board
layout. The Class-D with half filter is an ideal circuit
where battery life, heat and system cost are primary
issues. The Class-D with half filter is the most efficient
circuit at low and high powers and has a lower cost than
the Class-D with full filter. These issues make the half fil-
ter Class-D the ideal circuit for notebook PCs. Notebook
PCs are very concerned with battery life and heat, while
system costs are still important. EMI issues are well
understood by notebook designers, so the additional EMI
generated by the half-filter implementation should not be
a problem.
Class-D with full filter
Class-D with full filter had lower IMD than the Class-D
without a filter and higher quiescent current than Class-D
with a half filter. It also has a higher system cost than
either circuit. Designers should use Class-D with the full
output filter when heat, battery life and EMI are concerns.
An example of this would be a boom-box, where the ampli-
fier is in the same location as an AM receiver, where the
switching frequency is close to the band of frequencies
that the AM receiver is demodulating. Another example
would be any device that has wires connecting the ampli-
fier to the speakers. The wires act as an antenna, and if
not filtered, the switching frequency could radiate to
other devices in the vicinity.
Conclusion
It is possible to reduce the output filter in certain appli-
cations. This application note has shown that reducing
the output filter does not mean reduction in quality. The
Class-D amplifier with and without the output filter had
approximately the same THD+
N
. The Class-D amplifier
actually had lower IMD without the output filter. The
quiescent current of the Class-D amplifier was lower using
the half filter than the full filter, making the Class-D ampli-
fier even more efficient. The designer who is primarily
concerned with maximum heat and power-supply con-
straints and is not concerned with EMI and quiescent
current could use the Class-D amplifier without a filter to
save cost. A designer who is primarily concerned with
heat and battery life and has EMI as a secondary concern
could use a Class-D amplifier with a half filter. Applica-
tions that are very EMI-sensitive and/or have devices
operating around the switching frequency of the Class-D
amplifier should use a full filter.
References
1. “TPA005D02 Class-D Stereo Audio Power Amplifier
Evaluation Module User’s Guide,” Texas Instruments
Inc., September 1998, literature number
SLOU032.
2. Martin Colloms,
High-Performance Loudspeakers
(London: Pentech Press Limited, 1985), pp. 18-26.
3.
http://www.nxtsound.com
4. Bob Metzler,
Audio Measurement Handbook
(Audio
Precision, Inc., 1993), pp. 37-39
; http://www.ti.com/sc/apa
www.ti.com/sc/docs/products/analog/tpa005d02.html
23
Analog Applications
August 1999
Analog and Mixed-Signal Products
Plik z chomika:
TirNaNog
Inne pliki z tego folderu:
slyt331.pdf
(570 KB)
slyt333.pdf
(771 KB)
slyt335.pdf
(746 KB)
slyt332.pdf
(706 KB)
slyt334.pdf
(626 KB)
Inne foldery tego chomika:
ACE
AcornUser
AmigaComputing
AmigaFormat
AmigaShopper
Zgłoś jeśli
naruszono regulamin