FET input is my new standard of all my designs. LSK170/LSJ74(A or B) are perfect here.
-91dB THD with 10Vp 10KHz output in my simulation.
-91dB THD with 10Vp 10KHz output in my simulation.
Last edited:
1. Is C4 better than a servo?
2. Transimpedance vs. frequency plot - how does it look like?
3. Do you really like R25 to reduce loop issues, based on real world results?
2. Transimpedance vs. frequency plot - how does it look like?
3. Do you really like R25 to reduce loop issues, based on real world results?
No. I turn blind eyes to the DC offset right now. Later, I tend to add a trim pot to address DC offset instead of a servo. I like the simplicity.
Could you give me a hint what issue you are suggesting.
Ground loop is unavoidable. My source would be an AV receiver. From there I break up 7.1 channels preamp output into multiple power amplifiers. The best way to address ground loop is to just use 2-prong plug for the amplifier, which is not acceptable by most of people nowadays.
From my past experience, the R25 helped the hum issue. I did mod on Hafler 9290 and Quad 405 to add this Hum Breaking Resistors.
The circuit doesn’t prevent you using lower voltage. I would say +-25V would be the minimum. Lower than that you need to remove the cascode, Q15 and Q16.
Last edited:
In my case, the VAS has no current gain. The compensation is through a single shunt capacitor C5 at VAS. Mismatching NFB paths is less an issue here.
In audio frequency, the input stage and VAS would both work in class A, as the EF3 puts very little load on the VAS. I doubt you can measure any current changes on R19 and R20.
Last edited:
Practically First Watt F5 with a Darlington OS, lots of cascoding and current sources instead of resistors, and lots of BJT instead of FETs.
What's new?
What's new?
I have miserable experience with resistor between grounds to fix the loop issues. And also a lot of damaged amplifiers seen when this resistor fails (burnt). And cmr is poor. I either use balanced input or power amp in class II and the loop issues are non existent. Otherwise it is a poor compromise and we can see many such examples here.
There was an interesting find when I was working on Quad 405. Quad 405 separates signal ground from the power ground on its PCB with a 10 Ohm resistor. However, the 2 grounds are shorted together with a wire at its DIN socket. The factory unit does not have RCA plugs. The input is through a DIN socket with a included cable. The only thing I need to do to enable the "hum breaking feature" is to cut the jumper wire at the DIN socket.
I would include this resistor in my design. We still get the option to short it with a wire later on. There are some documents suggesting what wattage for that resistor in the worst cases. I can't recall right now. For home use, a 1/4 watt resistor is absolutely fine.
I have a memory of a disaster i made many years ago with 2 parallel feedback loops.
In order to avoid it with this amplifier that i find really interesting i made a simulation.
It is a diamond buffer but with near 1x current amplification. Followed by an ideal op as current amplifier. I believe your 3 EF is near that.
Anyway in the 100 schematic the open loop gain is almost 34 dB less.
In order to avoid it with this amplifier that i find really interesting i made a simulation.
It is a diamond buffer but with near 1x current amplification. Followed by an ideal op as current amplifier. I believe your 3 EF is near that.
Anyway in the 100 schematic the open loop gain is almost 34 dB less.
Yep. I had bad experience with 2 parallel feedback, too.
The issue lays on the choice of the compensation. Miller compensation doesn't play well with 2 parallel path. The way to address that is to avoid Miller compensation. Thus, I simply use current mirrors to drive the VAS, so that I can just use shunt compensation. The voltage gain can be written down like, Rout/Rf, where Rout is the output impedance of the VAS, and Rf is the current feedback resistor. The higher the Rout is, the more gain it has.
The current mirror I used is "Wilson Current Mirror" https://en.wikipedia.org/wiki/Wilson_current_mirror
It has relatively high output impedance so that I can extract more gain from the amplifier. The same reason I use EF3 instead of EF2, to keep the impedance high.
BTW, I was looking for the "Blameless" topology for current feedback, until I found the internal topology of the current feedback opamp LT1223.
We can just swap out the output stage to scale it for audio power amplifier.
As for the input stage, using JFET just make everything simpler.
Last edited:
jxdking:
sorry for the unrelated question ...
recalling some comments from you in a Hafler amp thread I started a while back, the next time you open your 9290, I'd really appreciate some pictures. Info on the 9290 model is scarce. I guess 9290 was the last (non-Transnova) model before they moved to the (Transnova) 9300, etc.
If you just put a big capacitor ( maybe 1000uF) between the source of U1 and U2 and the problem should be gone.
Leave place for it on your test PCB and you can check the difference.
Leave place for it on your test PCB and you can check the difference.
The cap at that position, the rate of charging and discharging are not matched. The cap is more likely to be overcharged under dynamic conditions. I don’t like a cap in that position.
Although I have not test yet, I guess the improvement would be small.
1000uF 94ohm makes a time constant of 9ms. I dont think there is even the smallest danger of overheating at the start.
Hopefully the 4 resistors are perfect. In that case you will not have any improvement.
But with a dismatch of 1% the distortion will be about 20 times higher.
Hopefully the 4 resistors are perfect. In that case you will not have any improvement.
But with a dismatch of 1% the distortion will be about 20 times higher.
It was caused by the enormous input capacitance of LSJ74, about 100pf.
To fixed this, basically the C7 from the low pass filter has to be >> 100pf, so that C7 could be dominate.
I upsized C7 to 470pf. I have updated screen shorts in OP.
Looks better now, but the feedback loop will have some work to do in the most sensitive audio range (~ 2kHz).
Why this limited open loop frequency range? 100kHz is possible and avoids lots of problems.
Why this limited open loop frequency range? 100kHz is possible and avoids lots of problems.