Updated May 3. 2006


    Micro Class-D

    Class-D AMP


Prototype 1

Prototype 2

   Ribbon Speaker

  ▪  Measurement Mic.

   Mini Subwoofer

  ▪  25 W Power amp

  ▪ NEW  Home Theater


  ▪  NEW  XLS Sub


   Current projects

  ▪  Finished projects


  ▪  Amplifiers

   Ribbon Speakers

  ▪  Subwoofers


  ▪  Main page



Contact: e-mail


DIY Self Oscillating Class-D Amplifier (SODA)

Prototype no. 1



I have made a two layer PCB.

The board measures 10,0 x 6.8 cm, and is produced by Olimex. The ground plane is split into two to have a digital and an analog part.


Prototype no. 1

The prototype is based on the same diagram as the bread board muck-up. The basic of the circuit is the op-amp U2A which is configured as an integrator, and the comperator in the driver IC HIP4080A, which together forms the PWM modulator. The free running frequency of the modulator is controlled by a combination of C12 and C15 ... higher values gives lower frequency. I have chosen a frequency of around 275 kHz as this seems a good compromise between low switching losses and a carrier frequency far outside the audio band, which allows for a simple 2. order reconstruction filter. I have tried to take the feedback (back to the integrator) from the output in differential mode (based on a op-amp) but it seemed to work better and easier to control when taken from the comperator in the HIP IC, so that is why it ended with this topology.

After the input the signal passes first a high pass filter set to 159 kHz, next a low pass filter set to 48 Hz (this needs to be altered to around 15 Hz by changing C13 to 3.3 uF).

There after the signal enters a pre amp stage that both serves as a buffer and a summing amplifier for input and an overall feedback taken from the output bridge through U1A which also serves as a reconstruction filter for the feedback signal.

The output bridge is based on the HIP 4080A driver IC and 4 x IRF640 MosFets. The reconstruction filter on the output is set to 56 kHz for a 8 ohm load.

Last part of the diagram is the power supply, based on two LM7812 regulators and an op-amp which serves as a virtual ground for the op-amps, as these are running on single supply.

The new PCB populated with components.

The prototype workes fine the only problem is a kind of white noise, which is always present.

I thought that a real PCB would cure that, and it almost have, but it is still there to some extent!

When playing at very low levels the bass notes have a sound like clipping or metallic (a bit hard to explain in words)! Maybe its just the noise that is "modilated" in tune to the music

Also it is very sensitive to touch; If I place a finger on the driver IC (HIP4080), when it is not playing, the noise almost disappears, but placing a finger on one of the caps that controls the PWM freq, does the complete opposite.

Im pretty sure that the problem is to be found around the PWM oscillator.
Could also be a fundamental grounding problem (Using a ground plane split into two to kind of separate analog from digital).
I'm using a virtual ground for some of the circuit, but as can be seen on my homepage, the "noise" on this ground is not at all alarming 50 mV pp)

Have tried to use NE5532 instead of TL072, but the integrator in the PWM ran pretty hot, and it was difficult to adjust freq. down to below 300 kHz.

The amp is running on a small 16 V DC PSU (until I know it is working properly).

The amp is intended for a subwoofer, and hope to get it to deliver more than 300 W . Eventually :)

The pots for adjusting the dead time is set to 220 k for maximum dead time.

Noise measurement on Vcc3, the virtual ground for input stage, PWM oscillator and feedback amplifier, which looks exactly like the noise on the output of the 12V PSU for the same stages.

(Probe x1)


Measurement on the PWM output (pin 5 on HIP4080A) ..... quite clean

(probe x10)


1 kHz measured on the output, with a 4 ohm speaker as load: Here 8 Vpp => 2 W

(probe x10)


Here is the spectra of the same. SNR/Distortion is only 50 db down!!!


Spikes on the rail Vpp = 0,4 V (the PWM switching freq can easily be seen .. 275 kHz)

(probe x10)

8 turns around the output coil. From this I must conclude that the two coils are wound and connected correctly ... otherwise the two windings should try to negate each other with very little transformed voltage as a result!

(probe x1)

Voltage across one half of the coil. |Vripple| = Vcc.

(Probe x 1)

Test with 56 V DC Power Supply

Half of the output swing with a main power supply of + 56V DC and a secondary of 17 V DC (for the driver, modulator and input). The output coil is running very hot and the switching frequency is down to 140 kHz!!!! Something is not right.

Possible explanations:

Need for hysteresis in the comparator

Recovery Diode, Output snubber


Needed alterations:

Input filter

Needs change to cover lowest frequencies (1uF not large enough) and should limit higher frequencies to same as output filter.

Output filter.

Needs to be recalculated. Should be calculated for the half speaker impedance e.g. 4 ohm if the speaker is 8 ohm, to reflect the bridged design. If one capacitor is used instead of two this can be halved. Also when the coils is wound on same core the number of windings can be halved.

Gate diodes

After putting 1N4148 diodes in parallel with the 10 ohm gate resistors, the noise was decreased, and the heat dissipated in both MosFets, HIP4080 and the PSU for the HIP (LM7812) was diminished.

Better decoupling of driver supply

As can be seen from the picture of the underside, an addition of 470 uF across the supply for the driver, helped reduce the noise problems further.

New outout coil (low permability bobin)

According to Bruno Putzeys (The Dos and Don'ts for Proper Design Implementation) toroidal coils emit more EMI than others! ..... is that really correct?

All connections to one side (except input)

To minimize EMI (Se The Dos and Don'ts for Proper Design Implementation).

One single ground layer

To minimize EMI (Se The Dos and Don'ts for Proper Design Implementation).

Driver IC isolated from PWM modulator wrt. supply

To avoid coupling between the driver/output stage and the modulator, the PSU for the two circuits should be split. This is not really possible with the current design as one half of the modulator resides inside the driver IC.

New supply for driver

The LM7812 used to supply the driver IC runs a bit hot, and I'm wondering whether it is any good at the high frequencies where the current is drawn. Also I have had to place a large cap (470 uF) on the output to reduce noise ... this cant be good for it's ability to regulate!

Zobel network on output

When the output is not connected to a load (speaker) the amp looks into the impedance of the filter, which will be highly capacitive, the Zobel network should dampen this.

calculator - zobel neetwork

New filter coil

The used filter coil got really hot when applying 56 VDC to the power stage.

The original coil had two windings on 42 and 29 turns measuring 160 uH and 80 uH. The color of the core is light green and yellow. The size of D=1.57" and h=0.6" corresponds well with Micrometal's T157-40. This seems quite possible to be the actual core.

I had used 19 turns on the coil to give the needed 30 uH which corresponds well again with a simulation in Micrometal's PC tool. This simulation shows a loss of more than 15 W per turn in the test situation, where the switching frequency for some reason dropped to 130 kHz ..... no wonder why it felt hot after no time.

Here is a picture of the new coil. Not much room for more windings. The wire is 1.5 mm in diameter

It uses a Micrometals T157-2 core, which is a low loss core meant for HF use. In a H-bridge you actually get twice the inductance if both coils are wound on the same core!

Two times 30 turns/12 uH together with a single 1 uF cap in 8 ohm, will give a small peek of 2 db at 18.8 kHz and a - 3db point at 30 kHz. In 4 ohm no peak will be present, and the - 3 db point will be 18 kHz.

  1. Start reading this thread for information on how to design a new coil on www.diyaudio.com.
  2. Producing wound components (University of Surrey, Guildford)
  3. Discussion on coils on diyaudio.com

Where to bye coils:


Where to bye litzewire


DIYAudio thread about litzewire

Design Idears

Mixed loop feedback

DC protection


AC loss control


Current limiter

NE555 Mono stable