But slew rate is very low, about 15V/usec, and I don't know how to improve it. Probably this type of the CFA is not good enough.
I wonder if someone has done a comprehensive noise analysis of CFAs as yet. I'd expect it to exist by now --- maybe Franco?
In any case, although I prize low noise, the perspective to bear in mind is the noise already in the best sources. It will still usually dominate, unless the amp is really noisy.
In any case, although I prize low noise, the perspective to bear in mind is the noise already in the best sources. It will still usually dominate, unless the amp is really noisy.
You could for a start remove 150pF and do a miller compensation instead.
Remember that the slewrate is determed by the current in the feedback network and compensation cap.
So more signal applied to the amp will reveal higher slewrate.
It is like charging a Capacitor through a resistor .. The timeconstant will never change regardles if you are charging with 1V or 100V.
I tried compensation like in Alexander amp but got to much peaking, only shunt compensation worked.
Noise in Current Mirrors
I can think of two separate considerations and there are perhaps others I haven't considered. What did you have in mind?
Hope I haven't been tedious with my very special requirements for direct driven horns.😉 I doubt it is a problem even for very efficient conventional, mono-amped speakers.
Best wishes
David
I can think of two separate considerations and there are perhaps others I haven't considered. What did you have in mind?
Hope I haven't been tedious with my very special requirements for direct driven horns.😉 I doubt it is a problem even for very efficient conventional, mono-amped speakers.
Best wishes
David
Hope you had a nice rest. Some seem obsessed with miller, its inappropriate for CFA, there is a reason all opamp companies use shunt with CFA. CFAs can theoretically deliver very high currents to the vas even go into class AB mode if need be. Shunt can be made even with large values and the topology will still benefit as you saw in your experiment. Only one particular design by Gosner uses any other compensation, and shunt is used in combination.
Dont forget to update the other thread. 😀
No Dadod has a current gain of ~5x in the current mirrors. If you use the miller caps instead you will only use a little portion of the current (20%) for compensation and the rest can be used for drive the next stage. This reduces the distortion level.
If I can remember accurately 35 years ago when i was bread-boarding the direct-coupled CMFB amp topology idea... I only got peaking when the gain was too low or the R's used from output to - input and to ground were too low value. What values are you using??
Thx-RNMarsh
Thx-RNMarsh
Last edited:
You dont seem to understand the topology. It can theoretically supply any current you wish. This does not affect performance. What does affect performance is that with shunt one has wider openloop bandwith and more feedback available to reduce distortion. Look at DADODs results, even higher compensation values resulted in lower distortion irrespective of the current.
Forget the 'what values are you using' to mean... change the values themselves and see if you can get a condition without peaking.... not related to C comp et al.
-RNM
Dadod, wrt the compensation, I tend to think that Miller is not the way to go. If you do use MC In CFA, you need very small values if you want to preserve the SR (2-3 pF) and there are other issues as well.
Here's how I'd approach comping your design ( my view, others please chip in )
1. Remove the shunt comp network
2. Close the loop at c. 3 MHz max, keeping the feedback resistors as low as possible
3. Apply some Alexander comp (try 33 or 50 pF to start with)
4. Go back and adjust the feedback resistor so the loop closes again at 3 MHz. You may have to repeat this process a few times
5. Do a closed loop bandwidth plot after you have completed 1-4
6. Highly likely you will have gain peaking
7. Apply a bandwidth limiting filter at the input to tame the peaking.
Comping a CFA is a reiterative process, but for max bandwidth you want the feedback resistor as low as possible and the Alex comp cap as low as possible consistent with a ULGF of 3 MHz. If you were using slow output devices, then the ULGF would have to be lower of course due to OP stage phase shift.
One final point, I did not check your TIS current, but I run mine at 1 mA - I think you need to run it at least at this level or higher (2 or 3 mA) if your design can handle it.
Anyway, good luck - lets try to get this thing to > 200 V/ us!
Here's how I'd approach comping your design ( my view, others please chip in )
1. Remove the shunt comp network
2. Close the loop at c. 3 MHz max, keeping the feedback resistors as low as possible
3. Apply some Alexander comp (try 33 or 50 pF to start with)
4. Go back and adjust the feedback resistor so the loop closes again at 3 MHz. You may have to repeat this process a few times
5. Do a closed loop bandwidth plot after you have completed 1-4
6. Highly likely you will have gain peaking
7. Apply a bandwidth limiting filter at the input to tame the peaking.
Comping a CFA is a reiterative process, but for max bandwidth you want the feedback resistor as low as possible and the Alex comp cap as low as possible consistent with a ULGF of 3 MHz. If you were using slow output devices, then the ULGF would have to be lower of course due to OP stage phase shift.
One final point, I did not check your TIS current, but I run mine at 1 mA - I think you need to run it at least at this level or higher (2 or 3 mA) if your design can handle it.
Anyway, good luck - lets try to get this thing to > 200 V/ us!
Last edited:
You should be able to target >200 V/us for SR. For the small signal rise/fall times I measured about 40 ns IIRC on my CFA's - so very fast indeed (2V pk- pk)
off-topic digression
However, the numbers really didn't seem to add up --- there was a much larger improvement than could be reasonably accounted for just by the removal of the feedback R thermal noise. Kern and Mackenzie at Bell Labs conjectured that it was getting the FET out of contact with the lossy borosilicate sealing glass of the TO-18 package that was the main reason for the improvement, and went on to have JFETs packaged in lower-loss headers (BeO, BN, and Al2O3 iirc). They used a large feedback R as in a conventional design, I think 2Tohm. They got equally low noise. 😀
The Bell Labs environment was evidently a very productive one. In far-flung fields the people there frequently either pioneered, or tellingly corrected the work of others. One example was in charge preamps for nuclear science, which I was looking at for photonic detectors circa 1973. A group at Lawrence Berkeley, Landis, Goulding, Pehl, and Walton, decided to try a scheme to eliminate the high-meg feedback resistor in such a preamp, needed to discharge the feedback capacitor after unipolar charge pulses were deposited on the input. Since the drain-gate photocurrent in an N-channel JFET was of opposite polarity to the particular detector, they illuminated the FET chip with pulsed LED light. To do this they removed the 2N4416 from its case, still attached to the gate lead. They got a dramatic improvement in noise charge which they attributed to the absence of the feedback resistor. The whole scheme made for quite a lot of favorable publicity in the community.
However, the numbers really didn't seem to add up --- there was a much larger improvement than could be reasonably accounted for just by the removal of the feedback R thermal noise. Kern and Mackenzie at Bell Labs conjectured that it was getting the FET out of contact with the lossy borosilicate sealing glass of the TO-18 package that was the main reason for the improvement, and went on to have JFETs packaged in lower-loss headers (BeO, BN, and Al2O3 iirc). They used a large feedback R as in a conventional design, I think 2Tohm. They got equally low noise. 😀
Dadod, I see you are running your front end at 2.7 mA. That should be plenty fast enough to support very high SR's.
my old amplifier, mostly current feedback solid state driver, and a EL84, so to say half circlotron as output stage;
I do understand it fully.
I have designed current feedback amps since 2002. Before just step on me, you could try throw it In a simulator to see it.!!!
I would not suggest anything i already has tested.
Actually it is 2.9 mA, and I tried 4.3 mA (reducing R4 and R7 from 220R to 150R) and got reduction in distortion a bit, but slew rate did not change(increase). For the moment I don't know where the slew rate bottleneck is. Here is zip file if someone whants to play with.
BR Damir