Need help with Giesberts HEXFET AMP


2004-10-03 1:22 pm
Hi everyone! :cool:

I have a project with this amp, i was nearly done with it, but then this problem came out! :bawling:

So i made everything ready and started to test the amp... Everything was fine until i put FETs fuses to the holders and bang.. every IRF9540N and IRF540N shorted and every fuse blew. Then I thought that the problem was gate voltage so I removed the wrecked fets and put multimeter over R24 and R27. I adjusted voltage as low as it´s possible, 2,8V in my amp.

I soldered a new fets in their places and put only 400mA source fuses in holders. Then I turned the amp on. Everything seemed to be under control, so I adjusted gate to source current to 200mA. It was very steady and thought that everything worked fine.

So i shut it down and put the normal 2.5A fuses back to their holders and turned the amp on again. BANG! every fuse blew and the fets were gone.

So I came here to look if someone else have samekind of problems with that amp. I found that someone here have the amp which is working and some have the same problem with their IGBT version.

I don´t have every value of the components just right because couldn´t find those and thought that those would work fine. Powersupply is 100% same as in original. Here is the values that I have used to countervail some components on the apmlifier PCB.

T12 IRF9540N (original IRF9540)
T13 IRF540N (original IRF540)

R6 82.0 ohm 2% (original 84.5 ohm 1%)

C3, C4 2.2nF 63V (original 2.7nF)
C5 470pF styroflex (original 330pF styroflex)
C9 1uF MKS4 100V (original 1uF polypropylene)

Ok I know that C5 should be really close that 330pF but can´t get a right component yet. And isn´t it just affect to maximum frequency of the amp. (so 470pF should decrease the maximum frequency?)


So if you guys could help me out of this situation i would appreciate that very much! :)

I put links to the schematics here:


2003-01-18 9:50 am
The problem is IMO not the different components you used.
If the amp died immediately after power-on I suspect a bad connection somewhere. Give it a rest and check your solderings thoroughly.
Use a lightbuld in series with the mains to reduce current if you don't have a variac at hand.

/Hugo :)


2004-10-03 1:22 pm
Thanks for your reply!

I checked my solderings, but couldn´t find anything suspisious.. maybe because I used so much time to get good solderings :)

I try to get a variac from somewhere... let see what is says then. :devilr:

But i can only say that it was absolutely steady when i chanced the bias current from 0mA to 200mA and kept it there.
That´s why i thought that the problem would be at starting the amp.
Wacky circuit.

The output stage has gain, although I'm not sure why. Normally this is only needed if you are running an op-amp front end with +/- 15 volt rails. Seems goofy to me.

The feedback network provides the same gain at DC as at audio. This means that the amp has DC gain. There is a circuit at the input to inject some current to adjust the overall DC offset. Again it seems goofy to me. The maximum input offset current is only about 1.5 micro-amps.

Was a load connected to the output of the amp when it failed? If so, then maybe there is a DC offset problem. If not, then I would suspect parasitic oscillations.

Good luck...
Zimo said:
I don´t have every value of the components just right because couldn´t find those and thought that those would work fine...

T12 IRF9540N (original IRF9540)
T13 IRF540N (original IRF540)

This is likely your problem - not that there is anything wrong with the FETs but some components around them need changeing, I will get to that a bit later on.

R6 82.0 ohm 2% (original 84.5 ohm 1%)
C3, C4 2.2nF 63V (original 2.7nF)
C5 470pF styroflex (original 330pF styroflex)
C9 1uF MKS4 100V (original 1uF polypropylene)

These substitutions are unlikely to be a problem, even the 470pF cap instead of 330p, and 2n2 instead of 2n7.

It is very likely you have an oscilation problem, and possibly a bias problem as well.
This amplifier has two problematic ponts - one major and one slightly less major:

The slightly less major one is the use of a CFP BJT-MOSFET pair with gain in the output in order to avoid the use of separate higher supply rails for the input and driver portion of the circuit. These sort of circuits can be prone to instability and require good PCB layout and decoupling techniques, even under ideal conditions.
Your substitution of standard fets with ones with the N at the end will make this problem quite a lot worse. The IRF540N is quite a bit faster in this circuit than it's non-N cousin, while the IRF9540N (which is an even worse complement to the 540N than the 9540 is to the 540!), is not THAT radically different compared to the 9540 without N. In particular, because the gate-source and gate-drain capacitance is reduced in the N versions (quite signifficantly in the 540N), the resistors in series with the gates are too low. I would go with at least 47 ohm even for the original FETs, more (100-150) for the N versions.
Furthermore, the 540N has a substantially lower threshold voltage than the non-N version, 9540N also has a lower threshold but not a lot compared to the 9540 without N. In order to bias these correctly, you will need to change R24 and R27 to a lower value (100 ohms or so), or, even better, R25 and R28 to a higher value (22 ohms or so). In any case, your lowest bias voltage between G and S of the MOSFETs whould be well below 2V.

The first major problem this amp has is also the reason for the problem not being obvious - the fuses are in the source line of the MOSFET, so they act as source resistors, lowering the gain of the MOSFET. This is why you could use 400mA fuses but not 2.5A fuses - the 400mA have a larger resistance at 200mA, and you get a lower bias current. You would have not seen such a discrepancy with the original FETs, the N versions have a lower threshold and higher gain so the problem is magnified.
Putting fuses in the source line is, IMHO a rather bad idea. Fuses are quite non-linear, and in this application they represent local feedback for the FETs. Granted, there is a secondary feedback between the FETS and the BD139/140 drivers, still, why make things worse? Further, some slow blow fuses are constructed as small coils. Keeping in mind that these MOSFETs can easily osvillate at tens of MHz, putting any kind of inductance into the source line is asking for trouble. I don't know what the layout of the PCB looks like, C11 and C12 should be very close to the FETs, and IMHO they need a good quality non-inductive foil cap in parallel. At the frequencies where the FETs can oscillate, electrolytics alone may as well not be there - they will provide almost no decoupling unless they are special low ESR kinds, and even them it will be inferior to a non-inductive foil (not to mention tantalum or ceramic, but I would rather not use these here).
Still, the resistor changes should do the trick. It should be noted that you really cannot rely on any kind of idle current measurement in this circuit, if you take out a fuse and measure in it's place. The fuse type and rating is one of the things that determines the idle current in this design, which i find a very bad idea.
Hi !

Charles, it's completely correct that the outputstage has gain,
this way this amp can deliver nearly rail to rail outputswing.

Zimo, this c5 increased to 470pf is very critical, this can bring
the amp into oscillation which will blow up everything...
But you should follow hugo's suggestion to recheck every
connection and so on. Just for testing, check if you get a ceramic
330pf, or maybe only 220pf. If it stops blowing...
As you reduced c3 and c4 also the amp already had less compensation.
The reason c5 is so critical, is because this cap reduces closed
loopgain for higher freqs and thus increases feedbackfactor.
This can lead to a point where openloopgain is still above
closedloopgain at the critical resonancefreq, then the amp

It's possible that the resistance or inductance of the smaller
fuses prevented the amp from oscillating as they are directly
connected to the sources of the mosfets.

I think this is a nice circuit !

The MOSFET has only two failure modes:

1) Too much current (power) will raise the die temperature to the point where it melts. This could happen if the bias were too high.

2) Excessive gate-source voltage will puncture the oxide insulation. This could happen if it oscillates.

So it has to be one of these mechanisms. Since it worked fine with the small fuses, it may be an intermittent problem, perhaps due to a bad connection. Or it could be related to oscillations. Sub-optimal layout can cause it to oscillate.

Good luck...


2004-10-03 1:22 pm
WOUH!!! :eek:

THANKS to everyone of you guys!! :) :) (specially to you ilimzn, I think you´re really a guru with amps :D)

I would never have a glue that those IRF9540 and IRF9540N (and 540/540N) have so much differences... And I was quite sure that they would fit well. But hey thanks a LOT of that!

So now I will change definedly those resistors and that 330pF styroflex in the feedback.

And i forget to put here the link to layout of that amp. So here it is: H...cuments and pics of the original/pcb_igbt.gif

I change the components in few days and will be back here to tell how it works after these modifications.

Be back soon ;)
Zimo said:
WOUH!!! :eek:
THANKS to everyone of you guys!! :) :) (specially to you ilimzn, I think you´re really a guru with amps :D)

Not a guru for sure, just had similar experience with IRFxxx and IRFxxxN, IIRC so did MikeB but in class D...
To be perfectly honest, neither IRF540/9540 or 540N/9540N strike me as the ideal devices for this amp, especially since the layout offers a possibility of using larger ones. My favorites for this sort of application would be IRFP240/IRFP9140, without N at the end. Similar current handling, as good a complementary maych as you can get within the IRF series, and more robust. They do cost a bit more - about 2-3 EUR normally.
The layout strikes me as rather sparse, long tracks tend to be problematic with MOSFETs. One small suggestion: although it may seem unimportant, it may help to exchange places of R26 and the jumper next to it. With MOSFETs it is always good practise to put gate stopper resistors as close to the MOSFET as is feasible.

I would never have a glue that those IRF9540 and IRF9540N (and 540/540N) have so much differences... And I was quite sure that they would fit well. But hey thanks a LOT of that!

It is a good idea to download the relevant datasheets from and compare the Vgs vs Id @ Temp curves. It is not always easy as they are seldom in the same scales, but even a few points will give you an idea. Next, compare capacitances in the parameter table. The problem with this amp is the gain, which svales up the reverse transfer capacitance by the gain, so at first glance low value gate stopper resistors are a good idea. However, in the N ch N suffix versions, Crss is lowered quite a lot, so the gate stopper may become ineffective. In order to truly get the proper value, you would need to experiment, which implies access to a signal generator and scope :(

MikeB may also be right, though at first glance, I don't see the capacitances being a huge problem, unless the design is not overcompensated at all (designs usually are overcompensated a little, at least by the tolerance of the caps, though doing it JUST right improves performance, but requires said signal generator and scope).

The problem of gate overvoltage is a very real one in this amp. There is no protection, and if the amp oscilates or has a shorted output, the gate voltage may become excessive. At some point you may want to add zeners across G-S of the MOSFET. 6V or so will also limit maximum current somewhat, but in this instance, the gate protection is of more importance.
Elektor carried two versions of this amp. One was called '60 watt HEXFET amplifier' and the other was called '90 watt IGBT amplifier'. The PCB for both amps are the same and that is why you will find a larger pitch pad for the output devices even with the 60 watt amp.

This is one amp that I have found almost impossible to stabilise. I did have one running for quite a while. It sounds pretty laid back and relaxed. The resolution is very good. But it lacks a sense of liveliness or 'bite'. I have four of these boards completely assembled and unstable. I wonder if it is worth pursuing this since there are too many amps out there that will outperform this one in terms of stability and sound quality.
If I was trying to 'fix' this design, I would probably do the follwing:
1) Use IRFP devices even for 60W (you do want these to survive the fuses melting in the event of problems)
2) Put the fuses into the D line of the MOSFETs, not S.
3) Reroute power supply and ground for the output stage - the fuses in the D lines would aid to thie by acting as large jumpers, freeing space to get the ground close to the S lines for proper decoupling using a pair of noninductive foil caps in parallel with 100uF or so electros.
4) Possibly insert low ohm NON INDUCTIVE source resistors for the MOSFETs, on the order of 0.15 ohms. Usually I try to avoid these in a one output transistor pair amp, especially as MOSFETs do not like inductive resistors in the source, but in this case they may alow better control of the idle current and less problems with oscilation since getting this at the expense of the gain is not a problem in a CFP configuration, and neither is the small voltage drop a problem (~1V).
5) Include proper gate protect / overcurrent limiter zeners.

Since this would need a redesign of the PCB, it would be a good oportunity to get rid of the rather large amount of jumpers. Experience has tought me that ground plane PCBs are a more than worthwhile investment - no need tor fancy through hole metalisation, this makes producing the PCB using DIY techniques quite easy, you just don't etch the ground plane side, and then 'blank' off the parts of it around the holes that are not connected to the ground plane with a large drill bit. THIS design I believe would benefit substantially from such a PCB.


2004-10-03 1:22 pm

You know what?

My amp works now like a beast! :hot:

I did the component changes like ilimzn told me to do.. (thank you very much of that!)

So I made these modifications to get it work:

R25 and R28 from 15ohm TO 22ohm

R26 and R29 from 150hm TO 120ohm

I move the places of that jumper and R26 nearby IRF9540N

C5 from 470pF styro TO 332pF styro 1% (a suggestion of MikeB)

I put 100nF capasitors beside C11 and C12 (10000µF)

I let those C3 and C4 to be 2.2nF (originals should be 2.7nF)

After those modifications we measured the bias voltage and the lowest value was about 1.8V.

We adjusted bias current to 100mA and used multimeter 10A line to avoid the effect of multimeter own fuse, and measured it with the 2.5A fuse like ilimzn advice said (we were not sure what should be the bias current cause the amp is using fets and in the quideline with IGBT it was 200mA after 30min.)

Then we let it to warm up about 30 min and checked and adjusted the DC offset, it was only about 0..4mV in both of channels! (with shorted line in)

Then we took and oscilloscope and an audiogenerator...

At 1kHz we couldn see no distortion, but when oscillossoce was in 10mV section we could see some distortion... so the circuit is a little bit tricky one and really choosy with components!

My friend who have repaired many amps at his job, wanted to test the circuit a little bit more.

So we tested how its frequency responce says at frequencies like from 20kHz to 60kHz. What we saw was unbelievable! It can handle frequencies to 60kHz like nothing! Also with squaremode of audiogenerator, it didn´t smooth the edges of square much even at 60kHz!

And last we took a resistive 4.7 ohm load (no inductive reactance).
With left channel it gave about 100W power at max point, and with other channel 90W. So we thought that the differences between the channels would cause by differences between bias currents...

Im now very satisfied of that amplifier! :)

We were going to add those zeners across gate-source. But didn´t know which direction is right :xeye:

To Samuel Jayaraj!

If you have four of those completely assembled I would advice you to check what components YOU have countervailed!!!

Maybe that little oscillation in my amp now is from those fuses which have a little bit inductive construction! I will change them and maybe put them before the power rail comes to PCB...

But now Im going to listen it some and put circuits against DC. :cool:

See ya :)

Zimo said:
You know what?
My amp works now like a beast! :hot:

Excellent, I'm happy to read about it!

So we tested how its frequency responce says at frequencies like from 20kHz to 60kHz. What we saw was unbelievable! It can handle frequencies to 60kHz like nothing! Also with squaremode of audiogenerator, it didn´t smooth the edges of square much even at 60kHz!

Keep in mind that amplitude and load will change this somewhat, but even so, from the schematic it is a fairly fast amp. It would actually be prudent to limit it on the high end with an input filter, to prevent RF pickup generating intermodulation products with the audio signal. It already has a filter made out of R1 and C2, but it's actual cut-off depends on the source impedance. A signal generator usually has 50, 75 or 600 ohms output which is a low impedance and will result in a high cut-off frequency. In a regular application, depending on what you use to drive the amp, it may be lower. If you are using a low impedance source, you may want to increase somewhat R1 and C2.

We were going to add those zeners across gate-source. But didn´t know which direction is right :xeye:

A of zener1 to G of T12
K of zener1 to S of T12

A of zener2 to S of T13
K of zener2 to G of T13


2004-10-03 1:22 pm

Maybe I was too close to success :(

I add those 6.8V protectionzeners to the finalstage fets today and listened maybe an hour... Then I shut it down and turn it on couple of times. When loudspeaker-relays were powered I heard awfull buzzing from left loudspeaker :xeye:

Shut it down as fast I could but I have lost the game anyway!

I can´t get any explanation to this, but I thought - would those zeners cause this.

I didn´t check bias current after adding those zeners :rolleyes:

Have to check it from the other channel to avoid that it doesn´t burn...

Maybe I just buy next those IRFP models if those would be better and more steady. And if I add IRFP should I again change some other parts or should only bias adjustment take care of that...

And maybe I should take still those fuses off from that source line and just put it before rails come to PCB or don´t put those anywhere...

This isn´t building an amp anymore - this is more like art or nuclear physics :whazzat:
Zeners should not present a problem. You may still need to increase the gate resistors some (I used a similar CFP with gain approach with IRFP240/9140 and put 68 ohms in the gate line).
You may also have instability into a reactive load - when you next do some tests, it would be a good idea to look up some of the 'standard' reactive loads used for amp testing, and test with that. Pure resistive loads nearly never match a real speaker (one exception are some of the magnepans but even they have a resonant peak). You may want to look at using those 2.7nF caps instead of the 2.2 that you replaced them with, though to be honest, I don't think it will make a huge difference, you may just get across the 'marginal' limit of stability but I myself prefer rock stable ;)
Regarding IRFP9140/240, they are more similar to the original IRF parts, just much better complementaries. You need about 3-3.5V Vgs to get 10-200mA range of bias current. It should still be reachable with the component changes you made, if not, you change them back to what they were for the IRF540/9540.

Regarding fuses, use fast fuses with a straight wire inside and quality fuse holders. The layout of the amp is a bit odd, it could have been better, but debugging these sort of issues is way beyond text-only messages :(

Regarding art, well, building amps is an art ;)


2004-10-03 1:22 pm
It´s been a while since last "report".

I have now changed my fets to IRFP240/9140 models. And I have done some changes also to fuses. I took off the fuseholders which were in Source line and put jumpers there. Now the fuseholders are between the cables which comes to main boards. This sould make outputstage more stability and (hope so) reliable.

So in this solution I have now one 10 000µF capasitor after fuse and ofcourse I have to change fuserating. I decided to try with 6.3 amp slow ones first. I don´t know is that a serious crime to leave one big cap after fuse... ok. It may cause a problem if fet shorts, then the whole capasity of that 10mF cap is going to discharge to fet.

I haven´t got those 2.7nF caps yet, but I will. I have also finished DC protection circuits. I decided to do a protection led to show when DC protection is blocking. I had to use a SN7432N 2 input OR-gate to get 2 channels to one LED. :D

I´ll inform you how it will make this time when I get my few components. Thanks for everyone... I couldn´t make it alone ;)


2004-10-03 1:22 pm
Ok. So you´re suggesting that those two resistors are unnecessary or what do you mean with "need to be linked out.."? (sorry that I didn´t understand that). Well I have no exact clue why those are there, but I quess taking those off would make the differential pairs whole different. Could you please tell me more about what does those two resistors do there and what does it change if I took them off or connect them otherwise.. (sounds so radical :D)

However I know that R3, C3, R4, C4 limits differential amps maximum operating frequency to 6.5MHz and now I have maybe 10MHz rating there because of wrong values of C3, C4. I´m just asking because I think Giesberts have not put anything there without a reason. :rolleyes:

BTW have you much expierience of this circuit? If have I would like to share it with you :)

I appreciate your answers. Thanks :)