Cavalli Kan Kumisa III

An All Discrete, Complementary Heaphone Amplifier

The Kumisa III

The Kumisa III amplifier was originally designed by Benny Jørgensen. Benny's article on this amp can be found in the projects section of Headwize. The Kumisa III is the third in a series of Benny's amplifiers. The first two used Opamps and in the Kumisa III Benny wanted to create an all discrete design.

As he describes in his Headwize article, Benny thought through the Kumisa III pretty carefully, including ensuring that it has a wide frequency response. His original power supply design is an unregulated 24V bipolar supply with various techniques applied to also give the power supply a wide bandwidth.

A small group of diyers built Kumisa IIIs based on Benny's original design. One review was posted by olsethand a working design thread was started by peter counsell.

The Cavalli Kumisa III

I started reading the design thread during one of my visits to Headwize. A few of the builders were having difficulty with the servo scheme that Benny had designed. Their Kumisas were showing subsonic oscillation that was changing the DC at the output of the amp. As I studied Benny's design I thought that I could improve on the servo scheme. At first we tried changing the component values in the original servo. Then we tried running the servo back to the input to get more control over the amplifiers offset. Several of the builders prototyped this version.

Running the servo back to the input, however, has the effect of lowering the input impedance of the amplifier and puts the servo into the audio signal path. So even after the new prototypes were built and worked, I continued to look for another, better way to servo the amp. In the end, the new servo scheme turned out to be very simple and is the scheme that is now in the CK2III amplifier.

CK2III boards and support are available from AMB Labs. Everything you want to know about building this amplifier can be found on AMB's website. After I had finalized what I thought was a really good revision of the Kumisa III I asked AMB if he would be willing to make, sell, and support boards. Diyers in the small community of headphone amp builders are very familiar with the exceptional job that AMB does creating new amplifier designs and supporting them. AMB was kind enough to take on the CK2III even though he was in the middle of introducing another new amplifier of his own design.

The Cavalli Kan Kumisa III

I sent my original Express PCB design to AMB and we then collaborated to make the final board layout. During the PCB collaboration AMB and I also collaborated on finalizing the design. We each ran extensive simulations on the amp alone and on the amp and power supply in combination. I noticed that AMB really liked the Toshiba jfets in his σ22 amp and I suggested that we could substitute these devices for the input bjts without any modifications to the design. AMB agreed. This was the final change before going into production with the new boards.

The currently active CK2III thread is here.


The Original Kumisa III

This is the schematic for the original Kumisa III (click to enlarge).

Original Kumisa III

The Original Kumisa III

The theory behind the design is explained in Benny's article at Headwize. The amplifier has a BJT input stage, a current mirror gain stage, and a Darlington output stage. Bias is set using a conventional Vbe multiplier. In Benny's design and build the Vbe device is attached to the same heatsink as the output devices. The purpose of this is to allow the Vbe device to compensate for the temperature rise in the output stage.

The servo uses a multistage filter to an inverting opamp. The opamp is connected to the Vbe multiplier section where it acts on the inputs of the output devices to zero the DC offset.


The Cavalli Kan Kumisa III

My modifications to the design included:

  • Replacing the Darlington output stage with a Sziklai pair to get better temperature stability.
  • Replacing all of the small signal transistors with commonly available BC550/BC560 devices.
  • Replacing the output transistors with commonly available BD139/BD140.
  • Modifying the servo scheme to adjust the two BJT voltage references, creating a strong servo mechanism while keeping the servo completely out of the audio loop.
  • Running the amplifier from a simpler 15V regulated, split supply.
  • Designing a single PCB that included both channels and the power supply.
  • Replacing the input devices with 2SK170/2SJ74 complementary Toshiba jfets, increasing the input impedance of the amp.


This is the schematic for the Cavalli Kan Kumisa III showing these changes(click to enlarge).

CK2III Jfet Input Amplifier

The Cavalli Kan Kumisa III

The changes to the design are outlined in the History section. There are three significant changes to the amp.

Output stage

The Darlington output stage was changed to a Sziklai (Complementary Feedback) Pair. The Sziklai is a bit more temperature stable because the bias current is more dependent on the driver devices. Because of this the CK2III can operate at its nominal 30mA idle current without heatsinks on the output devices.

Input stage

The input BJTs have been replaced with the 2SK170/2SJ74 complementary JFETs. These Toshiba devices offer low noise and high input impedance. Their bias point in this design is approximately the same as the BJTs, requiring no changes for their substitution.

The Servo

The biggest change to the amp is the servo. Benny's clever design includes two voltage references for the emitters (sources) of the input devices. These references are set by the BE drop of the reference transistors taken from a voltage divider from either rail to ground. The midpoint of the rail-to-rail voltage divider offers the best place to attach the servo. Furthermore, attaching the servo here keeps it entirely out of the audio signal path since the voltage reference transistors are simply that - voltage references. The capacitors at the bases of the reference devices have been increased from 47u to 1000u to ensure that the references stay constant well into subsonic frequencies.

Placing the servo in this new position necessitates changing the polarity of the opamp from inverting to non-inverting. These changes are obvious in the diagram.


The changes to the Kumisa III have made it a stable, low distortion, high input impedance, fully discrete, no NFB headphone amplifier that anyone can build. In this regard, this little project can be considered successful. :-)


Getting the new CK2III boards to market would not have been possible without the interest and efforts of AMB. His support for the amp on the various forums is second to none. Many thanks!!