Kumisa II Headphone Amplifier.

by Benny P. Jørgensen


The goal of this second version of my headphone amplifier is of course to get a very good sound and still keep it easy to build. Easy to build is an argument for keeping the number of components as low as possible. In this version I have removed the DC servo and used global feedback instead of local feedback. In this way I have remove 8 components from the signal way. The constant current generator is made faster and more precise.

The amplifier contains a very fast video opamp (AD844) and a push/pull output stage with two equally fast transistors (2SD1763 + 2SB1186), both of which are biased heavily into class A operation. All of this is supplied with current from a fast and stable power supply. I can’t see any reason to make it harder than it really is.

For me the worse type of distortion is the cross over distortion. It occurs when the two transistors in a push/pull stage are switching the current flow. The only way to prevent this type of distortion from mingling with our signal is to make sure that all transistors are either on or off. Switching between on and off is not allowed. This leads to pure class A. The same problem is in the opamp, and here the problem is solved by drawing a constant current out of the opamp and thereby making sure that cross over distortion does not occur here either.

I also demand that the amplifier have low noise, be fast and be able to supply the headphone with any given voltage regardless of the current need of the headphone. This leads to a big, but still fast, power supply and a stable design. Please note that there are no parallel components (except in the bypass capacitor where I use 100 nF||10 uF) since no components has exact the same data and thereby there will be two different signal ways. Not good.

I have made the opamp even slower than in the original version. The AD844 has a good sounds but it tends to be a little aggressive and I have found that it helps to make it slower in bandwidth and not in slew rate, as we will see later on.

The Amplifier

The AD844 runs in pure class A, as long as it is guaranteed that the current is only flowing either in or out of the opamp. In the Kumisa this is about 1 mA. I have chosen to draw much more current, so that the output transistors have something “to live off”. The smaller the current, the harder it is for the constant current generator to draw the output transistors down to ground. Therefore, I have used 1 V / 68 ohm = 14.3 mA.

It’s the largest amount of current that the opamp can deliver constantly without getting hot and not pressing its output stage. Since the transistors only use up to 7 mA (at 800 mA out), the opamp always operates in pure class A. There is nothing that can bring the opamp out of class A. Not even a short circuit!

The 2SC2705 holds constant the voltage (Vce) over the 2SC2389 to make the constant current generator more precise and faster by neutralizing the internal capacitance in the 2SC2389. The 2SC2705 is a 1 W transistor while the 2SC2389 is only 0.4 W. Using two transistors here prevents the 2SC2389 from getting hot. The 2SC2705 is, because of the way it is being used, less sensitive to temperature variations. The collector current through both transistors is about 1V/68 ohm = 14.3 mA. If it is too hot, then decrease the idle current to 25mA by increasing the 68 ohm resistor to 82 ohms.

The BD135 transistor and other components enclosed in the blue square in the schematic form a temperature-sensitive zener diode of around 1.6 V and adjust the constant current generator to eliminate terminal runaway. The components in the blue square are shared between the 2 channels.


The BD135 is placed on the heatsink with the output transistors. When the heatsink gets very hot, the voltage of the BD135 drops to under 1.4 V. The Vbe of the two power transistors is about 0.66 V (35 mA at room temperature) but when they get hot, the Vbe falls to under 0.62 V and thereby increases the idle current the hotter it gets. Using the BD135, the bias voltage on the output transistors drops as the temperature increases, so the idle current is constant and not increasing. The collector current on the BD135 is about 8.2 mA.


The idling current in the output transistors of the Kumisa is about 38 mA when they are at room temperature, and about 35 mA when they are hot. When the amplifier is cold, the bias voltage on the power transistors is 2.12V and about 1.98V when they are hot. Then the idling current is:

Cold : 2.12 – (2 * 0.66) – 0,04 = 0.76V and 0.76V / (2 * 10 ohm) = 38 mA
Hot : 1.98 – (2 * 0.62) – 0.04 = 0.70 V and 0.70 V / (2 * 10 ohm) = 35 mA

It is possible to adjust the idle current of the output transistors by changing the 68 ohm resistor. The following table roughly shows the effects of changing the 68 ohm resistor (approximate because of the variations in the transistor data):

100 ohm 10 mA
82 ohm 25 mA
68 ohm 35 mA
47 ohm 50 mA

Do not change the other resistors, because it might withdraw some of the “magic” sound quality around the 2SD1763 and 2SB1186, which can get a little aggressive.


The entire amplifier uses 120 mA (2*38 mA output transistor + 2*14 mA constant current generator + 2*4.5 mA to the opamp itself + 8 mA through the BD135). I am aware that it’s not quite class A but its very close and the few switches of the leading output transistors will be very smooth because of the 10 ohm emitter resistors, which is a quite high value. The two 15 uF polyester capacitors insure that the needed current can be delivered very fast.

The Power Supply 

The power supply is regulated for the opamps (±15VDC) and not more. Because of the resistors and the big capacitors, there are not big demands on the PSU. The output stage is decoupled from the ±22VDC power supply with two 15uF poly capacitors and two 10,000 uF electrolytics. The 4.7 ohm and the 10000uF make a low pass filter at 3.3Hz and help to reduce any 50 Hz line noise. The 47 ohm is there to isolate the “signal ground” from “real ground”. This value is not critical. I have tried with resistors up to 100 ohm and that too works fine. There is under 2 mAac through this resistor.


If I had put a 7815 regulator in the ±22VDC supply, I would have added a current limiter. By the way, the amplifier is capable of delivering up to 800 mA in a matter of nano-seconds. I don’t think any regulator can do this. There is no reason to try to give the ±22VDC supply more stability than it already has. I have measured under 10 mVac on the power line while it was playing louder than I like to hear. The two 15uF poly capacitors and two 10000uF do their job well.


Here are the basic circuit voltage measurements:

Output transistors at room temperature: Output transistors are hot
(do not touch more than 10 sec.):
Voltage over BD135 1.62 V 1.55 V
Vbe of BD135 0.65 V 0.62 V
Voltage over 68 ohm 961 mV (14.1 mA) 898 mV (13.2 mA)
Voltage over 150 Ohm 2.12 V 1.98 V
Total Vbe loss on 1763+1186 0.66 + 0.66 = 1.32 V 0.62 + 0.62 = 1.24 V
Voltage over 120 ohm 0.04 V (=120 ohm*0.038 mA/110 Hfe) 0.04 V (=120 ohm*0.035 mA/110 Hfe)
Idle current (2.12-1.32-0.04=0.76)/20 ohm = 38 mA (1.98-1.24-0.04=0.70)/20 ohm = 35 mA


As the above graph shows, there is a 0.8dB lift in the frequency response around 1 MHz and down 0.4dB at 2 MHz. The output impedance up to 100 KHz is below the equipment range. At 1 MHz it is only 4 ohms.


A simple 1 KHz sinus wave output at 10.5 Vrms. The gain is 6,96 times = 16,9 dB.


The noise measured through a home-built power amplifier with a gain of 77.7 times. The home-built power amplifier has a noise output at 10 mVpp. I can only guaranty that the noise of the Kumisa II is under 190 uVpp.


With a load of 100 nF the square wave output looks like this. The peak output current is around 800 mA. It does not look very good and we will look more at the reason for this later on.


The opamp has a slew rate of 2000 V/uS. A 500 W / 8 ohm power amplifier has to have a slew rate at 56 V/uS to be able to reach 100 Khz. Here vi see a slew rate on the positive flank at 833 V/uS and on the negative flank only 375 V/uS. At the same time we can see that there is a delay at 60 nS from input to output. This delay can cause problems at around 8 MHz. Remember the funny notch earlier?


I have tried to make the amplifier become unstable by forcing a 500 KHz square wave at the output and look what I found. I do not think this will cause any problems, because as you can see, the oscillation is damped very quickly, and again this is not a natural situation.


View full size PC board design and component layout
Download PDFs of Kumisa II schematic and PC board design

I know that I use some components that can be difficult to get. I bought my special components at LC Audio Technology. Otherwise you can experiment with other opamps such as OP176, AD845, OPA134 or AD825 (the AD825 is available in SMD only). Its possible to use a BC635 instead of the 2SC2389 and 2SC2705. If you want to use the BD135 instead of the 2SD1763 and the BD136 instead of the 2SB1186, be careful to observe the pin layout of these transistors.


All transistors are from Toshiba. I have used Philips model 370 polypropylene capacitors. Polypropylenes are not as fast as ceramic, but they are less aggressive. The 10 ohm resistors are 0.5 W, and the rest are 0.25 W metal film. I have been told that carbon resistors have higher noise and better sound, but I haven’t tried them. The volume pot is an Alps 10 Kohm log.


The heatsink should be bigger than 6 K/W. The BD135 (only 1 for both channels), 2SD1763 and 2SB1186 transistors are mounted on the heatsink. The enclosure is plastic container from the supermarket used to store cream puffs. Although it is not metal, there is no noise (f.x. 50-60 Hz) from outside that I can hear. The inside of the enclosure can reach 90 degrees Celsius. I have drilled holes in the plastic cover above the heatsinks for ventilation.


The Results

My wife (my non-technical, interested helper, who studied music in high school) has told me that the Kumisa II is a bit more true towards the music. It doesn’t add charm to the music. If it’s a bad recording you will hear it, but the same way around, if it is a good recording, you will be blessed with music. For example, it cannot play Geri Halliwell in a way where she sounds good, and therefore, I’m happy. The meaning is for the amplifier to be neutral. I think tubes add to much to the music, and irrigate the music with charm, and therefore play the music in a way other than how it was recorded.


If a recording/artist has atmosphere, charm or other feelings involved, then the amplifier has to lead them out in a true way. Therefore not all people will be totally happy with using this amplifier. Most of this was true with the first edition of the Kumisa also, but I think it just got a little better this time and is better driving low impedance headphones. It’s more stable too.

c. 2000, Benny Peter Jørgensen.
From Benny’s Web Site. Republished with permission.


Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s

This site uses Akismet to reduce spam. Learn how your comment data is processed.