If anyone wants the layout files, just ask.

Did anyone else build it? Any advice maybe?

http://ole.ine.c...mponentside.fig
http://ole.ine.c..._solderside.fig

The EMI filter on the board has the same circuit as those available integrated into power sockets. I just had the components lying around, so I put them on the board.

I didn't finish the amp yet, so the layout is not really verified.

For the placement of the components, please refer to the original layout by Benny on Headwize. It's nearly same. Btw, I found one mistake there - the polarity of the 47uF electrolytics should be reversed, I think.

http://ole.ine.c...kumisa3_ole.jpg

It looks a little messy, as I had to make some corrections on the PCB layout after it was already completed.

The ideas behind the Kumisa III design are compelling and very well worked out indeed, plus the designer is apparently a true music lover - something that is not easily said about some people who are into DIY electronics. The result is no less than Benny Jorgensen already described in his article. Using my HD580, my first impressions are very deep and perfectly controlled bass, absolutely authentic voices, superb soundstage and generally lots and lots of more details than I remember even from my favourite jazz recordings. The Kumisa III has a sound that simply "flows" - no better way to describe it. It's neither harsh/stiff nor smooth/soft, it doesn't make me think of transistors, opamps or tubes, just about the music I'm listening to.

I have built several amps in the past, tube, transistor and opamp based, but nothing really compares to this one. Sure, differences between amps are usually subtle, but in this case rather obvious.

I strongly recommend anyone who thinks about building an amp - go for this one! You won't regret it.

The one that connects to Q5 should have the - pin connected to signal ground, the one that connects to Q6 should have the + pin connected to signal ground. On the original site at http://hjem.get2...I_headwize.html , the voltages printed in the circuit diagram also suggest this. However, on the layout at http://hjem.get2...Printlayout.gif the polarities are reversed.

Also, it should be mentioned that the 500 Ohms pots in the circuit are used to adjust the idle current of the output transistors. While obvious for some, the average DIYer might run into problems. They should be adjusted in such a way that the voltage drop across the 3.3 Ohms resistors is 150mV (as shown in the original circuit diagram).

My amplifier tends to oscillate with a very, very low frequency around 0.1-0.2 Hz or so if a low-Z headphone is connected, in my case 32 Ohms. The amplitude constantly increases until after 1-2 minutes it reaches dangerous levels of several volts that could kill the headphones. It immediately returns to normal when the phones are disconnected. Apparently, the output voltage and also the DC offset of the output stage drops too much if a low impedance load is connected, so the servo reacts with too much delay. To compensate, I decreased the resistor value connecting from the amp's output to the servo from 33k to 10k. This fixed it, however I'm not sure what effect on the sound quality this may have when using high impedance phones. I didn't notice anything obvious, however. The oscillation is still there, but only in the order of 1-2mV. Maybe someone else got a better idea how to fix this?

It sounds like you have the same problem as I did, but to a lesser degree. I wrote about it in this thread:

headwize.com/ub...fdays=20&page=1

As I wrote then, I shorted the 47 ohm resistor in the power supply to fix it. I can't hear any problems.

I think you should check the output DC after a longer time with your high impedance headphones plugged in, or a higher value resistor. I found that the instability was still there with 100 ohms, but took a long time to get even to 100mV.

Congratulations on your amp. It's very neat, more so than mine. I won't post pictures, because they won't be flattering in comparison. I second your opinion about this amp - I definitely don't regret it.

Stephen

http://ole.ine.c...kumisa3_ole.jpg (PCB top view)
http://ole.ine.c...sa3_ole_ext.jpg (amp with HP890 and HD580 headphones)

steveh, I just decreased that resistor to 34 Ohms, and it did become more stable. Still, I left the resistor at the servo at 10k. Thanks for the hint.

Hmm...have you read my posting above in this thread? I cannot come up with anything much more, I think. It's the best I've ever heard, but that still won't give you an idea of how it sounds :-)

The amp is very quiet - well, it should be. It gives you punches when appropriate, and doesn't color anything as far as I can tell. It sounds...well, think of the performance of a 12 cylinder car engine. It makes you forget that it is there, there is only well controlled power in abundance, no matter if it's up- or downhill, it delivers effortless. No need to think about dB, impedance, distortion. A good way for me to confirm the sonic quality is the fact that I'm usually listening at too high volumes with this amp - to jazz music. I just can't do that with other equipment. It would hurt my ears. With this amp, it doesn't. It's just plain fun. It sounds good at low levels, the details are all there - but live music is LOUD, and this amp does give me this feeling.

Really - give it a try, but don't try to save money on the power supply. It's a bit overspec, but for sure contributes very much to the overall result.

I didn't use anything Zobel like. Zobel is supposed to present a low impedance load to the amplifier at high frequencies, compensating for inductivity in the cables and in the speaker's coils. Still, since it is basically a low pass filter, it does mean a high frequency rolloff, taking into account the output impedance of the amplifier - which should be, but isn't quite zero. Do you really think it will improve anything?

[Edited by olesth on 07-28-2004 at 02:42 PM.]

Since the amp does not use feedback, how do you control the gain? Since you would normaly use a couple of resistors, and feed it back into the negative input of the amp to set the gain.

And if you can't control the gain, do you have to match the transistors? I am worried if i build the amp, I will have one channel louder than the other.

I don't think it is quite accurate to say, however, that it has no feedback. The amp has quite a lot of feedback at subsonic frequencies (as Benny notes in the article). Without the servo the amp's response is down to DC because it is all direct coupled. When the servo is attached, with the small value capacitors, it provides the NFB necessary to remove the very low frequency response. Apparently, however, according to several posts above, there is still a bit of a low frequency instability. I've looked at this in some sims and there is a small bump below 10Hz in the response curve. I don't know if this is enough to cause this behavior, but in general things like this are worse in practice than in the sims.

So, here is another variation of the Kumisa. I've called it the Kumisa IV (hoping that Benny will pardon me for this ).

First, I've changed the output stage topology to a CFP (Sziklai). I think that this will give better results. But, only building one will tell. The sims show much less distortion with this arrangement and although sim distortion results must be viewed carefully, the drop in THD is significant enought, that this might actually be better in practice.

I've changed the servo arrangement. The AC response for this arrangement has no low frequency bumps and drops montonically below 10Hz.

I've eliminated the gain from the current mirrors because the amp now has more gain than it really needs.

Would be interested to know what Benny thinks of these changes. Or anyone else who might be reading this thread.

[Edited by runeight on 07-02-2006 at 09:14 PM.]

I once planned to build a Kumisa III but shifted the project back the queue prefering other projects. Last time a looked the transistors were very hard to source, we'd need to search for some alternatives.

Building the amp should be easy and cheap, I'm considering to revive my old plans ... anyone else interested ? May someone give a rough sonical description / comparison of the amp ? Or know something through hearsay ?

Just one thought about the thermal coupling of the vbe multiplier: long traces to the vbe transistor may lead to instabilities, instead of mounting the BD135 to the heatsink of the output stage one could consider to put some TO-220 diodes in series with the 1k8 resistor and mount the diodes to the heatsinks.

cmoy, how would you rebalance the current mirrors?

I plan to build the amp very soon, Steinchen. I cannot get the transistors either so I have been trying to find some alternatives. I have come up with BC550 and BC560 so far - they are ment to be low noise, small signal audio types. BD135 is available here.

As for the sound of the amp, have you seen post #5 of this thread? That is the only kind of review of the kumisa III i have seen. The amp reminds me alot of the Dynalo. I would be very interested in a direct comparison.

It is possible to sub other bjts and the results should be close to the original, maybe just as good. The output devices can be BD139/BD140. These are easy to almost anywhere. And BC550 and BC556 are good candidates for the small bjts. They would work anywhere in the circuit.

I have not built the Kumisa III so I can't compare them.

[Edited by runeight on 07-02-2006 at 09:23 AM.]

My guess is that this amp will perform as well as the original. Of course, no one has tested this theory and at least two other folks have built the original Kumisa III and like it very much.

So, here it is should you be willing to give it a try.

[Edit] This is not the way to do it. See next post.

[Edited by runeight on 07-02-2006 at 11:29 PM.]