Use of Cherry compensation results in much lower distortion at high frequencies(20 kHz) and insensitiveness to change of the bias current.
Simulation of the triple BJT OPS used beffore but compensation cap connected to the VAS output(classic Miller) instead to the amp output.
THD20k is 15 ppm at 120W/8ohm at optimum bias of 80 mA.
Compensation cap connected to the amp output(Cherry)
THD20k is 0.42 ppm at 120W/8ohm at optimum bias of 80 mA. This is 35 times or 30 dB lower. I never succeeded to get close to that result with the Hawksford EC(best was 20 dB) in my simulations.
With Cherry compensation THD20k was 0.83 ppm at the bias of 31 mA, 0.42 ppm at optimum bias of 80 mA and vent up to 1.04 ppm at bias of 340 mA.
That shows how the bias setting is not critical and even bias current fluctuation with increase of the output transistor is not critical.
BR Damir
Simulation of the triple BJT OPS used beffore but compensation cap connected to the VAS output(classic Miller) instead to the amp output.
THD20k is 15 ppm at 120W/8ohm at optimum bias of 80 mA.
Compensation cap connected to the amp output(Cherry)
THD20k is 0.42 ppm at 120W/8ohm at optimum bias of 80 mA. This is 35 times or 30 dB lower. I never succeeded to get close to that result with the Hawksford EC(best was 20 dB) in my simulations.
With Cherry compensation THD20k was 0.83 ppm at the bias of 31 mA, 0.42 ppm at optimum bias of 80 mA and vent up to 1.04 ppm at bias of 340 mA.
That shows how the bias setting is not critical and even bias current fluctuation with increase of the output transistor is not critical.
BR Damir
Petr, the link you provided gives me a 502 (Bad Gateway) error. Can you provide a working
link or post the paper here as an attachment ?
link or post the paper here as an attachment ?
Hi Petr,
Normally, I'm called Edmond Stuart. 😉
(btw, the article is also on my website)
Cheers, E.
Not to rest on his laurels, Ian is now working on a 'cube law' error cancellation power amp, to appear in a future Linear Audio.
Class A all the way, but 1/4 of the idle dissipation 😉
Jan
Jan, I very much like Hegglun's works. I am a bit sad not to have read any comment about his article "Towards a reconciliation of measurements with listening tests" (Linear Audio volume 4) as it concerns both objective and subjective approachs to amplification.
Get this. Short free version here.
0.8mm width of 30um thick PCB trace on 1.6mm FR4 over a ground plane has 5.3nH/cm and 0.6pF/cm.
15cm of PCB trace resonate with 1pc of IRFP240 power mosfet Ciss (1300pF) at about 16MHz. With 5 IRFP240 in parallel, resonance is at about 7MHz. This is a very rough calculation, in reality the Ciss' are distributed across the PCB trace.
You have to use gate resistors that create poles in the forward path at significantly less than these resonance frequencies, to isolate them.
Last edited:
So piecing together my limited knowledge on the subject. This statement could be wrong. For two topologies, one with ops local feedback and one with only GNFB, both using the diamond front end and the same output devices the overall Return Ratios would be comparable. This is assuming similar specs for THD etc.
See your point now. What a rookie error! But that's what I am unfortunately.
so? use 8 mm wide trace - if you want RF performance you have to design with RF trace
http://www.ixysrf.com/index.php?pag...ategory_id=16&option=com_virtuemart&Itemid=35 ("only" 4.5 mm gate lead)
more on package parasitics: http://www.ixysrf.com/index.php?opt...ckages-in-high-speed-high-frequency-operation
Last edited:
It sounds to me like it would be best done in a single package.
Maybe even interlaced devices on a single die.
I think that you complicate to much, I used similar simulation for my VFAs with TMC and bult a few of them. All of them were very stable and good sounding. Did you ever build any real amp, if yes please show some to us please?
Even in a VFA I like to keep the feedback node impedance in the 250 to 500-ohm range. It is easy to do, but one must be aware of the dissipation in the feedback resistor when such low impedances are used.
Cheers,
Bob
Let's not forget the LT1166 chip, which implements essentially the same concept as the squre law approach, although I don't recall if it is mathematically identical. The LT1166 works well when properly applied, but such proper application is not really described in the app note. I do discuss it in my book. With the LT1166, neither output transistor ever completely turns off.
Cheers,
Bob
I'm afraid you quoted me out of context. This was about how many mosfets you can parallel until the law of diminishing returns apply.
With 8mm gate traces the resonance increases by a factor of 2.5. It also adds about 120pF of capacitance from the gate to the ground (it's actually in parallel with Cgd (of the same order of magnitude), for the upper half of the output stage).
There are many different ways to use the IRFP240/9240, and I'm not surprized if some of the designs do not sound good; this is true of most devices. I believe that they are capable of superb sound and measured performance at the same time.
Cheers,
Bob
I should note that I discussed vertical MOSFET resonances and parasitic oscillations in my MOSFET power amplifier with error correction article a long time ago. In that article I explained the use of gate Zobel networks to mitigate the problem. Dealing effectively with parasitic oscillations or instabilities in vertical MOSFETs without totally killing the HF performance with brute force gate stopper resistors is key to making the best use of verticals.
Cheers,
Bob
Thanks for this Waly. This is an excellent document and worth study on many fronts.
Wouldn't it be better to have ju..ust sufficient gate resistance AT the MOSFETs to critically damp the resonance?
This would introduce the least amount of evil phase shift. I don't think its sensible to try to isolate them. That would be a brute force solution.
Bob, do you think gate Zobels can totally remove the need for stopper resistors?Bob Cordell said:
My concern is that the Zobel introduces further 'capacitance'.
Last edited:
Ferrites work wonders here in the mosfet gate leads (also works for bipolars as well). There are cyclindrical types that slip over the device leads and you can also use the SMD versions. You need need to select the correct values though - I spent some time with a Murata kit a while back.
No, the Zobels just reduce the amount of gate stopper resistance needed. I think I used 100 ohms for the gate stopper in the MOSFET amp with error correction.
Cheers,
Bob