1|Page

VERSION 2

Part 1
IRS-900D Class-D Amplifier Tutorial
Oleh: kartino@yahoo.com

The concept of Class-D Amplifiers

In the last decade, in the world of audio power amplifiers, commercial products have been available for a new
type of audio amplifier, the Class-D Amplifier. In contrast to Class-A, B, AB, where the incoming signal is
immediately strengthened in the form of an original voice signal, in a Class-D Amplifier, the incoming voice
signal is enumerated but still remains in the analog method, then the signal is amplified. At the final level of the
amplifier, the enumerated signal is returned to a sound signal using an inductor-capacitor (L-C) filter.
The Class-D concept that will be discussed here adjusts to the scheme that we will take as an example of the
next design, the Class-D design with a block diagram is as follows:

Figure.1 - Class-D Block Diagram

In short, Class o D self-oscillating works with an internal feedback loop, where the feedback loop has an R-C
circuit that will make the circuit oscillate itself, becoming a saw signal generator. This saw signal will be
compared with the audio input signal by Comparator IC TL071.

The output signal from the Comparator IC is an audio signal that has been chopped by a saw signal, so that it
becomes a slope signal with the composition of the ON-OFF time adjusting to the sinusoidal level of the input
signal. This slope signal voltage is referenced to ground (GND) and its characteristics still do not meet the input
criteria towards the gate driver IC. While the gate driver IC uses a voltage reference to -VCC. So for that we
need a level shifter in the form of 2N5401 PNP transistor and IC CD4049 Invert logic.

While the gate driver also requires two input signals namely high-gate and low-gate. Comparator ICs that have
one output, by Invert logic IC duplicated into two outputs, one of which is inverting to produce a low-gate and
high-gate output that is suitable for IC gate driver input. Gate driver IC is ready to move both sides of the high
and low power mosfet alternately.

The output of the power mosfet is ON-OFF identical to the Comparator output. And already explained before the
comparator output itself adjusts to the sinusoidal audio input. So from the comparator to this mosfet on paper
there will be no distortion. There are no special effects that need attention that can cause sound defects caused
by signal amplification. While the Comparator, Level shifters, Logic ICs, and gate driver ICs work at 12 Volt
voltage, the voltage of the mosfet can be very varied and very wide, free of the SOA mosfet voltage. In the end,
this final voltage limit becomes the clipping limit of a Class-D amp.

The final output of this mosfet in order to return to a sound signal is then done with a Low Pass Filter (LPF).
With the right composition of the large inductor and capacitor, the chopped slope signal can be converted back
into sinusoidal. It's not perfect like sinusoidal in Class-A or Class-AB. However, it must be remembered that the
speaker is a dynamic mechanical device. The physical effects of the speakers' mechanics such as the spring
1
2|Page

and the inert mass of the speaker leaves themselves actually help to improve the sound output so that the
defect, ie the remaining ripple of the enumeration frequency filtering becomes meaningless

2
3|Page

Schema ImplementationAda
many Class-D Amplifier schema designs can be found from the internet. This time what will be discussed in
this paper is the Self Oscillating Cass-D Amplifier IRS-900D. This scheme is circulating on the internet, and is
pretty good, can be implemented and has an output capability of up to 900 Watts at 4 ohms, a capability that is
quite large. This scheme was chosen because it adjusts to the availability of components and PCB technology
available in Indonesia.

To make a good and high-quality Class-D very good quality components and PCB layouts are appropriate for
high-frequency switching. However, even with this limitation, even though the quality cannot be produced in
accordance with industry criteria, it is still very satisfying when compared to the Class-AB assembly classes
that are widely sold in the market.
The following is a scheme that applies the design and concept explained above. Larger image resolutions are
enclosed on the back page along with the PCB layout. This scheme has been modified by the author to be more
reliable and safe. The red values below are replacements from the original schema value and there may not be
a path on the PCB. So it must be soldered under the PCB.

Figure.2 - Audio Amplifier Schema Class-D IRS-900D

How to work the above scheme as explained before. The above circuit uses a power supply +/- 100V DC.
Besides that, a separate 12 Volt power supply is needed to drive the IR2110 IC gate driver. The consideration is
made separately is because it will further simplify the scheme. With the above circuit, suitable for allotment from
subwoofer to mid-hi. The designation is not disappointing, considering the 900 Watt level assembly kit market is
generally intended for Subwoofer applications until Mid-Hi. Modifications can be made to improve performance.
For example by increasing the oscillation frequency capability. In Class-d self oscillation, the oscillation
frequency speed is determined by the on-off speed effect of closed loop feedback which involves all
components in the feedback loop. Improvements can be made by replacing the OP Amp comparator with higher
capabilities, for example from Burr-Brown. And the main thing is the power mosfet, which can be selected
products with low gate capacitance from IXYS or IRFP4227, or IRFB4227, or other brands with more advanced
generations. What must be considered is the smallest gate capacitance and highest switching speed, and of
course a large SOA

LPF Filter
LPF filters that must be considered are the core must be suitable for high frequencies. For a Class-D amplifier the inductor
without air gap must be of iron-dust or iron powder type. Iron powder has a magnetic saturation value at high currents with
low losses. Globally the LPF criteria must meet the following criteria:
1. The inductance value is like a scheme, in this case 22-30uH
2. The magnetic core does not saturate at the highest load
3. Core losses must be low so that excessive heat does not occur when the load is high

3
4|Page

The following examples are iron powder cores that can be used to assemble class D.

T-157, Kool Mu, 40 - 50 lilit

T157 Amidon, 45 – 55 lilit

T157 Micrometal, 45 – 55 lilit

MS157 Arnold, 24 - 28 lilit

MS130 Arnold, 16 - 22 lilit

4
5|Page

The alternative is a ferrite core that can be obtained from the PSU core or filter. PSU filter output can also be
used. The type of ferrite core must be made air gap to increase the ability to magnetize, namely by breaking the
toroid circle with a hacksaw or cutting the core for EI Core. Because there is no manufacturer data, to determine
the optimal number of turns can only be done by trial and error. The number of turns with the air gap ferrite core
should be tried from highest to lowest between 14-22 times with a 1-1.5mm diameter guard.

The function of the air gap is to increase the magnetic capacity of the core to increase the saturation value. If
the saturation value of the iron core is not large enough, the inductor coil and the core will become very hot at
high loads. However, the presence of this air gap also opens up opportunities for electromangetic leakage (EMI)
that may get out into the network. The picture below is an example of material that can be used:
Gambar.3 – Alternatif Inductor

For capacitors because they work at high frequencies, non-polar types with high charge-discharge capability
and high working voltage are chosen. Its capacity ranges from 470n - 1uF to non-polar 600V.

Gambar.4 - Output Capacitor dan Inductor

The recommended types of capacitors are polyester or polypropylene such as LLC grade capacitors or MKP.

WIMA MKP Capacitor

5
6|Page

Polyester based capacitor untuk LLC resonant (biru)

MEX-X2 MKP capacitor

6
7|Page

Attachment

Schematic PCB Layout

Layout Components

7
8|Page

R0Of6V3

10 1K5
5W

10

s,e

8
9|Page

9
10 | P a g e

Part 2
Class D Frequently Asked

Class D amplifier fails to work?

Most failures in assembling a class D amplifier are due to the selection of parts below quality standards.
Generally, caused by the replacement of components that are not in accordance with the original scheme.
The amplifier component must be compatible with a high enough working frequency. Not just any part can be
used as a substitute. The key components are the active part and the LPF filter design.

Failure can also be due to the layout design that is not in accordance with the design rules for switching at high
frequencies. A good layout must be as thin as possible for weak signals, sufficient distance for the difference in
voltage and current direction, weak current and strong current signals must be kept so that large currents do not
induce weak signals. The weak high frequency signal path must be as short and as low as possible from the
PCB ground plane. Ground planes may not form close loops.

LPF Design?
LPF design, determine the type of inductor, type of core, number of turns and large capacitors see Part 3.
Class D sounds raspy and rough, treble too sharp?
Hoarseness and rough voice are usually caused by LPF cut offs that are too high, causing the remaining PWM
frequency to be carried to the speaker. Generally accompanied by complaints of a hot inductor. The solution is
added by the number of turns. When checked with an oscilloscope, PWM residues are cultivated for less than
2Vpp in the built-in speaker state (Figure-1). But the addition of winding also reduces PWM accuracy.

Figure-1 shows the PWM residue, measured at the output terminal in the built-in speaker state
The number of turns is short so that the inductance is generated around 22uH to 30uH or more depending on
PWM switching speed. The faster the smaller the inductance and vice versa. Of course the cut off will be higher
but that's good. The real purpose is to make the PWM frequency output signal low in PWM frequency. The
faster the closer the PWM, the easier it is to make it smooth, ripple-free.

Rough noise can also be caused by LPF capacitors that are too small, dry or dead.

Hot mosfet?
Hot mosfets can be caused by lack of dead time resulting in sufficient cross-conduction to cause direct current
to flow through the high side and low side mosfets. The solution is to increase the resistance at the gate, which
ranges from 22 to 47 ohms. For light mosfets such as IRFP4227 R gate can be chosen up to 66 ohms. The
greater the Qg mosfet the smaller the R gate and vice versa.

Hot mosfets can also be caused by too many windings so that the mosfet becomes more difficult ON and OFF
because of the induction of resistance from the inductor due to storage energy stored too large

10
11 | P a g e

Brek-brekkk when the clip?

Brekk brekkk when the clip is normal, and is characteristic of class D. Brekk brekk occurs when the clip, PWM
stops for a moment, and causes the speaker to be exposed to DC current.

Brek-brekkk before the clip?

Brekk brekkk before the clip because the supply bias is too weak, it could be because the bias voltage is
lacking, the bootstrap diode is ineligible, fast or not large enough. Bootstrap diodes must be at least 3A with a
reverse recovery time (TRR) of approximately 30ns.

The sound of the treble isn't as clear as being cut off?

The treble sound is reduced due to the LPF cut off being too low. Second, due to the lack of PWM speed. To
increase the speed of a PWM switch, the circuit must be agile and lightweight. For this reason, components that
have a low miller effect must be supported, for example the use of mosfets with a Qg of less than 70nC, and
just one set of mosfets such as IRFP4227 without totems. To produce good treble sound a switch speed above
300kHz is required. To be faster, it can be combined with iron core cores such as the red T157 from Micrometal.

The sound at the start is good, a few moments later it is bad?

This problem is caused by the use of saturated cores because it is too small or too high core losses so that it is
hot. When the core heat AL value goes down and causes the inductance (uH) to also fall.
The solution is to select cores that are large enough and are not easily saturated like blue from Arnold products
or red from Micrometal. Test it in the field. If when the full load core is hot, so it is not strong enough to hold it
means that it is not big enough and saturated.

Saturation is a condition where the magnetizing value of the core is not going to go up again because it is
saturated due to the current that flows around is quite high.

Is the amp hard to start?

Difficult amps are usually easy when lured with an input signal input. Or by filling in the bootstrap by adding R
15K from plus to bootstrap high and Zener 12V from boostrap to the prefilter path.

Noise when the volume is closed (zero)?

Squeaking amps or noise occurs when the volume is closed to zero due to disruption of the PWM frequency
due to the position of the potential cause the feedback loop is interrupted. To overcome this, the input R is
raised so that there is a minimum value when the volume is closed up and the high frequency triangle on the
input cannot be passed to ground via the potentiometer.

11
12 | P a g e

Amps coming out of smoke at the back of the resitor from output to ground?
This generally occurs due to the resonance that occurs when the output is not connected to the speaker when
the amp is turned on. Because there are no speakers, there is an equilibrium (resonance) of L and C from the
LPF so that there is very high efficiency, so that the energy entering the LPF as it is accommodated rises
causing the output voltage to be very high so as to burn the resistance earlier.

The fix is to install an anti-resonant resistor with a large enough wattage, 1K5 5W and install an ultrafast diode,
as shown below. When a resonance occurs when the output voltage exceeds the supply voltage because the
current is small enough, by the ultrafast diode the LPF charge will be shed to the supply elco so that the
maximum voltage is limited to the same as the supply voltage. So it's safe.

Mosfets often die when ON, without cause?

Bypass mosfets suddenly on, suddenly, and often happens because the gate driver is floating and induced so
that the mosfet is ON by itself and continuously. Secondly, the Vgs voltage exceeds the requirements.

When mosfets die suddenly, generally the GDS junction is melted so that all three will be short. If this happens
then it could be a totem pole, and maybe the gate driver IR2110 or IR2092 also caught fire. So it must be
replaced too.

To overcome the floating gate driver, it is necessary to add 4K7 resistance and also Zener 12V from the Gate to
the source mosfet.

12
13 | P a g e

Low PWM switching speed?

PWM switching speed is determined by the size of the miller cap of the entire feedback loop from beginning to
end as shown below.

So that each component influences speed. PWM speed is significantly affected by mosfet selection. The
selection of low Qg mosfets such as IRFP4227 significantly increases the PWM speed. Also a concise layout
might support high PWM speeds.

The number of turns of the inductor (or high-low inductance - uH) also affects speed. The higher the inductance,
the lower the PWM speed. But the lower the ripple remaining PWM from the output, so it's like a scale.

To increase the speed of PWM, you can choose mosfets with low Qg and reset the number of turns. So that the
ideal speed obtained from a class D amp.

What mosfets are right for Class D?

The correct mosfet for class D is adjusted to the allotment of the amplifier. The designation of the amplifier
determines the level of PWM switching speed. PWM switching speed determines the type of mosfet used. The
higher the PWM speed the more susceptible to interference.

For amplifiers that are applied for Low and sub to mid, no high speed is needed. This amp can be set at 200-
250kHz. Mosfet can use IRFP260 or IRFP250. Because Qg mosfets are large, totem poles must be used

13
14 | P a g e

For full range amlifiers, PWM speeds should not be less than about 300kHz. To be reliable at these speeds a
lightweight Qg mosfet must be chosen such as IRFB4227 with only one set without totem pole

How many watts of Class D Amplifier output?

Large Class D amplifier outputs are as usual formulas only differ in efficiency that can be up to 90%.

P = (V2 / R) x 0.9 x 0.707 for half bridge

P = (V2 / R) x 0.9 x 1,414 for full bridges and BTL

But because the output current must pass through the inductor and mosfet, the two components are the limiting
of the output according to the above calculation.

Why does class D amp sound so powerful?

Many opinions in the field that Class D feels powerful even though the calculation is no different from the amps
of other classes. The specifications are indeed the same wattage. But feeling more powerful generally refers to
dynamic conditions (lay njedug - agile). To produce a dynamic and powerful sound that requires a very high
damping factor. While wattage measurements in general are usually average or peak regardless of dynamic
conditions.

Very high damping factor occurs when the overall internal loop resistance of a large current amplifier includes
both PSU and amplifier amplifiers. In the damping factor amplifier is greatly influenced by the final power
amplification. In class AB or H, or linear amplifier, the final stage function is like a faucet with dynamic variable
resistance to the rhythm of the music. So the final stage works hard like a water dam that keeps the floodgates
open to the rhythm of the music in a linear fashion, by adjusting Vce. Whereas the nature of BJT is still subject
to the nature of general physics, where there is certainly a lag, which is generally the speed of work expressed
in the slew rate. For large amps it is very difficult to make high slew rate amps. Is high slew rate unnecessary at
low output? Very necessary, because when large currents automatically naturally slew rate will slow down.
Practically it is very difficult to make large amps with high slew rates.

In class D amplifiers, the power is regulated by opening and closing time and energy storage in the inductor coil.
This is like a dam with an open and close door, which when opening and closing adjusts the input signal. In
front of this dam there is a small dam, in the form of an inductor with a door that always opens fitting to the
discharge request, which is the load speaker. So that any debit request is not a problem because the mosfet
does open and close very quickly. There are no obstacles or small
i.e. Rds ON mosfet and DC resistance of the inductor. Very small in total. So the damping factor is very high
and very dynamic memorable.

Damping factor is very high when using SMPS and class D. So this combination is very high damping factor.
And really feels powerful.

Many people think the sound of "njedug" in general is due to distortion so that the clipped peak gives the

14
15 | P a g e

impression of "njedug". Not so, because precisely class D is very easy to reproduce low sounds

Part 3
Class D Amplifier LPF Design

LPF Filter is a very crucial part of class D Amplifiers. The selection of components and design must be done
correctly because this LPF filter determines whether or not the sound produced by the amplifier. An important
part of this design is choosing the type of inductor core, the number of turns, the physical size and determining
the capacitors. Many class D amplifiers sound bad because they fail in designing LPF filters. In this section, it is
explained simply, more towards the results of practice, not in detail to make it easier for beginners. Detailed
calculations can be obtained from the reference.

The number of turns of the inductor and the size of the capacitor depends on how much the LPF wants to cut off
frequency. But in class D, LPF functions differently in passive crossovers. In addition to cutting PWM
frequencies, LPF must ensure that the output signal is clean enough of the remaining PWM. The higher the
PWM frequency, the easier it is to make the remaining PWM ripple smaller, and vice versa. To do LPF settings,
tools such as LC Meters to make inductance (uH) are needed. If there isn't, the only way is by trial and error.
The amp is turned on and the residual voltage is checked with an oscilloscope. Even if there is an LC Meter, a
PWM residue checking is still needed.

15
16 | P a g e

Figure-1 Vout shows 1 Vpp PWM residue, measured at the terminal output in the built-in speaker state

The detailed calculation to determine the UH size of the inductor must refer to each manufacturer's website
such as:

http://www.micrometals.com/downloads/MicroRelease_March2010.exe
http://www.micrometalsarnoldpowdercores.com/upload/ArnoldDesigns.exe.zip http://www.mag-
inc.com/design/software

The amount of inductance depends on the PWM switching speed, as follows:

PWM Besar kapasitor

150 – 200 kHz 40 – 60uH
200 – 250 kHz 30 – 40 uH
250 – 325kHz 22 – 30 uH
325 – 400kHz 18 – 22 uH

Whereas the output size (Watt) determines the physical size of the inductor which determines the saturation
value limit. The greater wattage means the greater the output current until the higher saturation value limit is
needed. So what determines the size of the core is the current not wattage. To find out for sure you have to use
the help of core manufacturer software like the link above. As an initial reference based on the results of
practice, here are estimates of the use of cores that have been tested in the field

16
17 | P a g e

Core Max. Arus Output Rata-Rata

MS130 5-10A
T130 5-10A
MS157 15 – 20A
T157 15 – 20A

The third thing that is not less important is the characteristics that determine the properties that need
attention in design. Not that which is good and bad, but the selection is adjusted to the application. These
properties determine the number of turns, physical size, permissible heat levels and losses. So it should
always refer to the factory datasheet. Example of Kool Mu factory data.

http://www.mag-inc.com/products/powder-cores

Kool Mu MPP High Flux XFLUX AmoFlux

Perm 14-125 14-550 14-160 26-60 60
Core Loss Low Very Low Moderate High Low
DC Bias Good Better Best Best Better
Saturation
1.0 0.75 1.5 1.6 1.5
Flux Density (Tesla)
Curie
500 460 500 700 400
Temperature (°C)
Operation
-55~200 -55~200 -55~200 -55~200 -55~155
Temp. Range (°C)
60u, u flat to... 900 kHz 2 MHz 1 MHz 500 kHz 2 MHz

The following is a list of the number of turns for a particular core at a PWM frequency between 250 s.d.
300kHz and also capacitors from the results of the practice so far:

Core uH Jumlah lilitan

Arnold MS130 35 – 22 uH 26 - 30 lilit
Arnold MS157 35 – 22 uH 24 - 28 lilit
Arnold MS130 dobel 35 – 22 uH 16 - 22 lilit
Micrometal T130 35 – 22 uH 45 - 55 Lilit
Micrometal T157 35 – 22 uH 45 - 55 Lilit
Kool Mu T157 35 – 22 uH 40 - 50 Lilit

For the amount of LPF capacitors depends on the high and low PWM switching because the function of the
capacitor to cut the remaining PWM ripple. The lower the switching speed, the greater the ripple and vice
versa. As a rule for the following table LPF capacitors:

PWM Besar kapasitor

150 – 200 kHz 1 uF
200 – 250 kHz 680 uF
250 – 325kHz 470 uF
325 – 400kHz 390 uF

Referensi:
http://www.ti.com/lit/an/sloa119b/sloa119b.pdf
http://www.irf.com/product-info/audio/classdtutorial2.pdf Chapter 4.

17