The Awesome Cordell Amp!

Jump to Latest
#21
Rikard Nilsson said:
Thanks for your reply Bob! The prototype I built use MJE340/350 as I had some lying around, but it doesn´t seem to work 100%, I have tried using trimpots but still cannot find a "dip" in the distorsion. Perhaps these trannys are simply not good enough, or I have missed something. As I understand I should get a well defined "error correction balance" tuning the pots?

🙂
Click to expand...

Don't forget, everything but the output devices should run in class A. Otherwise you may get a significant dip only at small output levels.
 
#26
Absolutely! If the bias is out off class A range or the positive and negative feedbacks are not in balance the error correction won´t be effective. I´ve even seen it can make the amp perform worse than without EC included.

I´m not finished yet, but once I´ve verified the bias levels and made sure it runs in class A mode all the way up to full swing, then finding the EC balance is the next step. A clearly visible drop in THD is the receipt it works.

I know overall amp performance will be awesome once the EC network is working, as I´ve tried running the output at insane bias currents. It produced H2 better than -90dB and H3 below the noise floor.

🙂
 
#27
Rikard Nilsson said:
Absolutely! If the bias is out off class A range or the positive and negative feedbacks are not in balance the error correction won´t be effective. I´ve even seen it can make the amp perform worse than without EC included.

I´m not finished yet, but once I´ve verified the bias levels and made sure it runs in class A mode all the way up to full swing, then finding the EC balance is the next step. A clearly visible drop in THD is the receipt it works.

I know overall amp performance will be awesome once the EC network is working, as I´ve tried running the output at insane bias currents. It produced H2 better than -90dB and H3 below the noise floor.

🙂
Click to expand...
does that mean class A is out & class B is out.
there is a 'sweetspot' of class AB only?
 
#28
Hi

I think he is referring to the EC devices always operating in class A. It is important that they stay in thier linear region, full 360 degrees, in order to acheive better bandwidth, and thus less error. As for the outputs, class AB. Part of the idea is to use the error correction to reduce crossover components from the AB bias. Crossover components contain very high order non-linearities as do Vgs vs tranconductance. So the faster, the better.🙂
 
#30
Yes, that´s correct, I was referring to the EC transistors that has to operate in class A, never moving out of the "range".

I did some more testing today, and after having replaced two of the feedback resistors with trimpots I had no problem finding the "null" point of optimal error cancellation. However, the output stage with IRFP240/IRFP9240 seems very temperature sensitive, and the EC performance is almost never at it´s best.

I let things stabilize at about 200 mA idle, and put 10 watts of 1 kHz into a 4 ohm load. After optimizing the EC balance I got fairly good H3 figures, (but not as good as the 2SJ201/2SK1530 amp without EC I am also fiddling with). After pausing the signal generator for a few seconds and then enabling it again the H3-figure was much worse! I din´t touch anything else!

So, I assume this is because the temperature increase at load moves the balance point of the entire EC network, making it virtually impossible to really make this work in a every day amp.

I even noticed that increasing the output could result in better distorsion, simply because I entered the "EC sweetspot".

I hate to admit it, but as much as I´d like the EC approach to work this makes me seriously doubt I will ever achieve it. If the circuit is this sensitive to bias, drive, temp and even frequency I siimply cannot rely on it to really reduce distorsion once in a finished amp.

I have used the original values on everything, and checked that the operating point is near "center" (from what I understand at least). The voltage across the 680 ohm resistors are 4,5 volts, and the voltage between the gates is about 8 volts. Input voltage to the EC depends on temperature as I put a multiplier in series with the zener diode and reduced it to 15 volts + the multiplier, about 20 volts total. Perhaps this is the culprit? The multiplier makes the biasing fairly stable, but is it also what makes the EC balance move all over the place?

:smash:
 
#32
I guess I will go back to the old approach, regular driver stage and loads of global NFB.
Click to expand...
😱....:dodgy:



m2003br said:
Unlike bipolar output, Mosfets don't have a sweet spot bias. the more, the best.
Click to expand...

True, but at what cost in efficiency? Even 300mA bias would not get rid of the crossover components because those mosfets have very non-linear transconductance vs Vgs in that region and down to cut-off. A transistor can be bias in class B (or AB) and still never turn completely off. If it is working less than 360 degrees, there will be some crossover components in the mix. This topic popped up in another thread some time ago, here. 🙂
 
#33
Well, as I have unwillingly convinced myself the HEC doesn+t "fly" in real life, what´s the alternatives? Of course some other kind of local feedback around the output stage could be a solution to minimize crossover distorsion. But this doesn´t mean I won´t be using large amounts of global NFB as well to minimize what´s left of the OP distorsion.

I already have very good results with a non EC amp built with Toshiba 2SJ201/2SK1530 (much better than a bipolar version I also tried), so my hope was it could be improved further by using the Cordell EC circuit.

🙂
 
#34
Hexfet is just too good a device for driving a nasty reactive load like a speaker. I've got a couple of ideas I want to try before I completely give up. Bob mentioning something about using an opto-isolator, got me thinking.🙂 For example, a circuit to sense and differentiate between the common mode DC current and the AC signal current then translate the common mode current signal into a voltage. This voltage drives a linear opto-coupler LED, like a IL300, which controls the conductance of the Vbe multiplier. I find in general the multiplier idea does work, but it's setting will fight with the positive feedback node in the HEC, re-biasing the EC transistors. This makes the sensitivity manageable, but like you stated, keeping proper balance in the EC transistors is important for the circuit to function, so the multiplier idea can make things a little tricky. If you can control the bias via the common mode current, then for the most part temperature would be out of the equation, particularly the temperature of the error amplifiers. This sounds like a worthy goal to go after, however, I doubt it will be as simple as it sounds. :dodgy: I have some time tomorrow, maybe I'll burn off a few brain cells at it.:smash:
 
#35
Rikard Nilsson said:
Well, as I have unwillingly convinced myself the HEC doesn+t "fly" in real life,
Click to expand...
Your conclusions are going to make a few people on this forum very happy. 😀

However, HEC does fly in real life, see the PGP amp on my web site. What you are experiencing sounds like you are having stability issues.

The minimum is indeed bias and thermal dependent but also frequency dependent. The higher the frequency the smaller and broader the dip, you should set for a THD minimum at 20KHz or so.

Transistors types are critical as well, anything like MJE in the EC loop won't do properly.
 
#36
syn08 said:

Your conclusions are going to make a few people on this forum very happy. 😀

However, HEC does fly in real life, see the PGP amp on my web site. What you are experiencing sounds like you are having stability issues.

The minimum is indeed bias and thermal dependent but also frequency dependent. The higher the frequency the smaller and broader the dip, you should set for a THD minimum at 20KHz or so.

Transistors types are critical as well, anything like MJE in the EC loop won't do properly.
Click to expand...

I use 2N5401/5551 for the two error amps, but it might be that the others have to be replace as well with better transistors. Wouldn´t I notice some HF on the scope if there is a stability problem?

Otherwise, I can´t see why some people would be happy if the EC doesn´t work. It means one less option to get rid of distorsion....nothing good about that! So of course, I would still like to try any suggestions to make it work. I could try removing the cascode FETs I use in the driver to eliminate any possible instability there. Has anybody else built a working Cordell EC amp?
 
#37
Rikard Nilsson said:
Yes, that´s correct, I was referring to the EC transistors that has to operate in class A, never moving out of the "range".

I did some more testing today, and after having replaced two of the feedback resistors with trimpots I had no problem finding the "null" point of optimal error cancellation. However, the output stage with IRFP240/IRFP9240 seems very temperature sensitive, and the EC performance is almost never at it´s best.

I let things stabilize at about 200 mA idle, and put 10 watts of 1 kHz into a 4 ohm load. After optimizing the EC balance I got fairly good H3 figures, (but not as good as the 2SJ201/2SK1530 amp without EC I am also fiddling with). After pausing the signal generator for a few seconds and then enabling it again the H3-figure was much worse! I din´t touch anything else!

So, I assume this is because the temperature increase at load moves the balance point of the entire EC network, making it virtually impossible to really make this work in a every day amp.

I even noticed that increasing the output could result in better distorsion, simply because I entered the "EC sweetspot".

I hate to admit it, but as much as I´d like the EC approach to work this makes me seriously doubt I will ever achieve it. If the circuit is this sensitive to bias, drive, temp and even frequency I siimply cannot rely on it to really reduce distorsion once in a finished amp.

I have used the original values on everything, and checked that the operating point is near "center" (from what I understand at least). The voltage across the 680 ohm resistors are 4,5 volts, and the voltage between the gates is about 8 volts. Input voltage to the EC depends on temperature as I put a multiplier in series with the zener diode and reduced it to 15 volts + the multiplier, about 20 volts total. Perhaps this is the culprit? The multiplier makes the biasing fairly stable, but is it also what makes the EC balance move all over the place?

:smash:
Click to expand...


Hi Rikard,

I'm sorry you're having so much trouble with HEC, both in terms of bias stability and stability of the HEC distortion null point. It sounds like something is just wrong, but I'm not sure what it might be. My experience was that biasing the HEXFETs in the EC arrangement was quite stable - certainly more forgiving than bipolars. My experience with the HEC null also was that it was quite stable.

I wonder if you might be having some high-frequency parasitic oscillations. They can play havoc with bias points and increase distortion - and sometimes even blow up the amplifier. HEXFETs are fast devices, and can get into parasitic oscillations more easily than BJTs.

Cheers,
Bob
 
#38
Thanks Bob for popping in! 🙂

I also think it might be something like that. I have checked everything thorougly to rule out any silly connections I might have done, but ut looks ok. As soon as I get some spare time I will try removing my cascoded VAS and run the simpler version I know is very stable just to see if this helps the HEC. It could very well be a parasitic causing all this. (really hope so!)

By the way, how did you calculate the 270 ohm and 330 ohm resistors in your original schematics?


Rgs Rikard
 
#39
Hi

I had similar problems, but this was before I installed the gate Zobel filters. Very effective in reducing Ft to a manageable realm.🙂 I remember seeing a small RF oscillation once when I had trouble getting the circuit to balance. It was one of the outputs, but the gate Zobel fixed it.

In my case, the amplifier driving the HEC circuit has no GNF loop that surrounds it, so it is less likely that instabilities of the HEC is compounded by the rest of the amplifier via a fb loop. I think this may also help with the stability issue.:dodgy:
 
#40
CBS240 said:
Hi

I had similar problems, but this was before I installed the gate Zobel filters. Very effective in reducing Ft to a manageable realm.🙂 I remember seeing a small RF oscillation once when I had trouble getting the circuit to balance. It was one of the outputs, but the gate Zobel fixed it.

In my case, the amplifier driving the HEC circuit has no GNF loop that surrounds it, so it is less likely that instabilities of the HEC is compounded by the rest of the amplifier via a fb loop. I think this may also help with the stability issue.:dodgy:
Click to expand...


Hi CBS240,

Yes, the gate Zobel resistors I used are very important. I actually felt that their introduction was one of the creative points in my original amplifier. I actually had done quite a bit of research to better understand the causes of oscillation in the HEXFETs. I really could not stand the idea of the usual big gate stopper fix, as it wastes precious speed that is needed for the most optimum HEC performance.

The key is in understanding the parasitics in the HEXFETs and in the surrounding circuitry, and the need to reduce the Q of the resulting resonances. There is a section in my original MOSFET power amplifier paper that deals with this at some length. Without care and understanding, one can make a great Colpitts or Hartley oscillator with these HEXFETs. Their equivalant ft is over 100 MHz and rises as current increases.

Also notable is the fact that HEXFETs can get blown by parasitic oscillations that cause the gate voltage to be exceeded even if there are zeners put on the gates. This is apparently because with stray internal inductances the magnitude of these oscillations at the gate inside can be larger than at the gate terminals, and the gate can only withstand 20 volts.

When designing and building MOSFET power amplifiers it is fairly important to have an oscilloscope that goes out to 100 MHz or more. I recall a time when I was fooling with my amplifier I knocked a TV station off the air in my home.

Cheers,
Bob
 
You must log in or register to reply here.
Top Bottom