HELP!Problem with noise

[Manowar1985]

Hi,all!
I desperately need your help!
I am designing a descrete transistor power amp.
For now I am working on the preamplifier section and I have connected the output of the preamp to the feedback resistor.
My problem is located in the second stage(the voltage amplifier stage)
At first I use a constant current source of about 15mA.Everything seemed in my oscilloscope to work just fine for the case of simple transistor and for the case of darlington.
When I use darlington with cascode the noise goes so high that I can even see the waveform with that bias current of 15mA.The noise is reduced to a reasonable level when the current is
reduced to about 2.8 mA but this current is insufficient to drive the output stage with minimum distortion.
What are your suggestions.If the noise cant be reduced ,can I use the stage with current of 2.8mA and then design a buffer for next stage to "lighten" the load seen by the voltage amplifier stage.
I am adding images of the schematics for each case
Thank you.

[nickb]

There looks to be a very large gain in the design so are you sure it's noise and not instability? Adding Q8 will increase the gain by a few hundred more and so would make instability more likely. How about you post a scope screen shoot maybe?

What is the intended function of R7, C3 and R2?

It seems odd to my mind to use a cascode which improves the HF response along with a constant current load which has a high impedance for the output stage. The output impedance even with feedback would be high enough to be severely affected by the capacitance of the load. It seems to to be asking for unpredictable stability problems. Maybe the intention of the cascode is to reduce distortion due to collector-base junction width modulation?

[nickb]

Looking at this a bit more, the choice of Q1-4 has a great effect on the noise. You can certainly use much lower voltage types with a better noise figure for these. That said, the noise on the output should be around 10uV so hardly a problem. OTOH I do see a huge resonance in the open loop response around 500Hz that would worry me from a stability point of view. It might tell us something if you could look at the spectrum of the noise.

PS:
1) The resonant peak I was seeing is just because of the way I was measuring the open loop response. With that fixed I see 18uV of noise.
2) The open loop gain is 11K without Q8 and a massive 650K with it. The difference in the noise is minimal. Is there any chance the Q8 you are using just happens to be bad i.e. noisy?
3) Observation. The 2.2K feedback resistor will sink almost the entire output current at the maximum output voltage. Good for noise though.

[Manowar1985]

Hi Nickb.
Let me answer your questions from the beginning.

1)The open loop gain is very large indeed(I would be grateful if you tell me how you calculate the open loop gain).Why does this cause oscillations because I have applied the global negative feedback.

2)R7 C3 R2 work as follows:

I chose the feedback resistor to be low for low noise and for minimum dc offset I choose the same value for input resistance. Because the input resistance may be yoo low for our source I used a bootstrapping method so the DC input resistance will be:1.1+1.1=2.2k and the ac input resistance is a lot higher because of the bootstrapping (I don't know how to calculate it)

What should I aim for to reduce instability(Q8 is not faulty)

Thank you very much

[Manowar1985]

I have uploaded the photos of my oscilloscope. Measurements have been made with different bias generator voltages with no load and with a resistive load of 9.4 ohms(4.7+4.7 ohm)
the dvm means that the multimeter is attached to the vbe multipler to measure the bias voltage .If don't understand why the waveform clears with the dvm on and why the waveforms is awful when loaded.

Thank you again for your time

[nickb]

Thanks for the info on the bootstrapped input. I had missed that.

Roughly, the voltage gain is
S1 * B2 * RL

S1 = transconductance of first stage approx 1/110/2 ( 110 ohms is the total effective emitter resistance)
B2 = current gain of darlington pair say 250 x 250
RL = load resistance, I took it as 2K

so Av =1/100 * 250 * 250 * 2000 = 568182. So you see the open loop gain is highly dependent on the load resistance

The output current source is 15mA so we can only drive 150mV into a load resistance like 10 ohms. I doubt that is what you wanted. No way is this a power amplifier.

Sadly I have no idea of the test conditions for each scope shot do it doesn't tell us much. I still think instability. Don't apply an input signal and speed up the timebase until you can see what is really there. You can Use the FFT function to get a spectrum of the signal when it's happening. A fairly narrow bandwidth for the unwanted signal will be a big clue. What were the test conditions when the trace cleaned up?

Feedback is the primary cause of instability. To solve you can try to lower the open loop gain, reduce the feedback and /or use a bigger compensation cap than the 100pf you have right now.

A better approach might be to drive the output with an emitter follower to make the gain less dependent on the load,

[Manowar1985]

Nickb thanks a lot