I believe you are talking about the comparator in the servo part ?
Some explanations.
The cap C31 in the input signal path is there to detect if any source has DC in it in case somebody want to get rid of the C26 input cap of the amp. It will lead to a big offset and the protection will fire instantly.
There is no risk to inversion of DC. No DC from the input signal, because of C31 and DC from the amp output is always inverted by the fist OPAMP U1, and reinverted by U3. So it will be applied in phase at the feedback point.
I suggest you eliminate 2N3904 from the BOM (and the schematic), replacing it in position Q15 by an NPN that is used elsewhere in the circuit. I suggest either BC546 (TO-92) or 2SC3503 (TO-126).
The Fairchild Korea fab makes KSC3503, which I used in the Mountain View board in the DIYA store. It has the advantage of being in active production: https://www.onsemi.com/PowerSolutions/product.do?id=KSC3503link , (whereas the 2SC3503 is not), and its TO-126 package allows you to bolt it directly to the heatsink, if you ever want to experiment with that option.
The Fairchild Korea fab makes KSC3503, which I used in the Mountain View board in the DIYA store. It has the advantage of being in active production: https://www.onsemi.com/PowerSolutions/product.do?id=KSC3503link , (whereas the 2SC3503 is not), and its TO-126 package allows you to bolt it directly to the heatsink, if you ever want to experiment with that option.
I use LTSpice VII. Working.
Tried LTSPICE IV, got an error: Aborting "Unknown schematic syntax V" with no more help ;-(
Then, it close.
I have to figure out where is this unknown "V".
Sorry for the inconvenience.
1- Yes thanks, The best components had not been chosen for the moment. Just the schematic.
Misses a lot of details, like caps exact parts, powerrate of the resistors etc.
That explain why I change some active parts randomly with no care to see how it behaves with different models.
2- As the power mosfets has this nice temp behavior to be positive behind 100-150mA, and negative above, and run hot, I think it would be counter productive to bolt it on the main cooler. In fact, they are here to sense the temp in the enclosure, and even over compensating, that is a good thing ?
There are many options for tempco. The proposed solution on the table should be bread-boarded as suggested by Terry, while other avenues are investigated.
Also using the gear option for simulation is not recommended. LTspice uses a modified version of trap innately which damps ringing. The gear option is for aligning with other simulators which produce erroneous results.
Also using the gear option for simulation is not recommended. LTspice uses a modified version of trap innately which damps ringing. The gear option is for aligning with other simulators which produce erroneous results.
Sorry, my poor English, "brea-boarded " ?
Where do-you find a problem with the actual temp behavior ? It is overcompensating.
Sorry Krisfr, tried the front end with different devices: the problem of
disparities of VBE between NPN & PNP seems not to be solved in a satisfactory way with all the pairs. I have to work a little more on it.
Very often a good strategy is 'divide and conquer'. It could be worthwhile to just concentrate on simulating the temp and matching effects in the input part, without dragging along the whole thing as a big distraction. Just a thought.
Jan
I don't understand what you mean. The problem is very simple. It works, and there is no more problem with temperature effects, as far as I can see.
The problem is that I think it is too tricky for the moment to tune the input stage depending on the VBE differences between VBE of NPN and PNP devices.
Because any change of current affect both the Offset and the second stage currents in the same time. I have to try various solutions. And each part affect all the balance of the others.
It is not simple. If you tune the input stage alone, incorporating it in the whole circuit makes an offset that destroys its balance.
And i'm open to any of the solutions you can imagine.
The first thing should be some suggestions for best devices for this input stage.
Last edited:
What I mean is, will all due respect, if your input stage 'works' or 'works not' due to different small signal transistors, there is a big oversight in the circuit. Changing between, say BCxxx types or 2N34xx types can make some small differences but should not upset the system. And I would think that can be solved in isolation.
Same thing with small Vbe differences. IRL differences may easily be 50mV or 100mV from one build to another. Your circuit should work with high and low values.
Make sure that the input stage does what you want it to do, before hanging a complete amp on it. Try it with a load that acts like the rest of the amp.
In a sim you can do things in analysis that are impossible to do IRL.
Jan
I believe there is a way forward. Krisfr has the right idea. Dual devices are now available at low cost. BCM53/56 and DMMT5401/5551. Some gain may have to be given up in the input stage. Since more gain was acquired with the Schottky in the VAS, the net change might be zero. Early experiments give a positive outlook, but as alluded to previously, the homework must be done.
Again, with all respect, if it needs that specific device to work, I would not go with such a circuit.
Jan
Jan, it is not that it doesn't work with various devices, even non paired. It is just, as I said that it is difficult & painful to adjust. At least in sims.
I'm thinking to the real world. And I have to find the limits to be safe whatever the transistors and nobody to be be obliged to change a resistance on the bench, because, in his situation, it will be, bad luck ?, out of range.
My remark was addressed to Krisfr: I don't want him losing his precious time on something that can be modified after he will have finished his board design. And I still have to experiment one or two ideas to see what is the best: I'm not fully satisfied for the moment.
Working on the input stage in isolation is what I do yet. Reason why I had to
replace back the offset adjustment. But I don't think it should be a good idea to have to tune the input stage with VAS & output stage non populated in real world.
Spladski, I made for you my last version that seems to work more or less under LTSPICE IV.
Attachments
I am more than okay with the progress, I am working on output end and power distribution aspects of the board. If I had not just finished moving and had access to all my bench stuff I would bread board it in a day or two or three. hehehe. Keep up the good work, the rewards are for those that don't quit and give in easily. I have some work completed on the op amp area and it will be finalized later. Keep GOING.
Thanks again, Krisfr, or your nice help and participation.Keep GOING.
May-be some explanation is needed to explain the project to some readers.
First, there is some kind of a war between VFA (Voltage feedback amplifiers) VS CFA (Current feedback amplifier).
Without entering in controversies and detailed explanations, just let's say that they differ, in their behaviors, by two main aspects.
The first is speed (High slew rate) of the CFAs, and a closed loop bandwidth that remains near unchanged whatever the amount of feedback applied, while in a VFA, we have a bandwidth that increase with the amount of feedback: low bandwidth at low feedback, higher while feedback is reduced.
The second is currents on demand for the CFAs, that offers an expansive behavior, while VFAs are compressive.
VFAs are the most popular and generalised technology. Most of the time based on an LTP (Long tailed pair) topology or the input stage. LTP brings high gain in the input stage and balance, because both side of the input stage are driven with the same current source.
In fact, looking closer to the difference, we could say that in both cases, the feedback is applied to the emitter of positive input transistor. But, while in CFA, the F.B. current is taken directly from the output power transistors by a low impedance network, IE with no current limitation because power stage offers A LOT of current, in a LTP, the feedback voltage is (in a way) transformed in current by the - input transistor acting here like a emitter follower while its current is limited by the current source itself.
By nature, the CFA do not offer easy gain in the input stage. The result is the whole amp will have low loop gain with a flat bandwidth up to, say, 10 KHz, while VFA will have high loop gain, usually beginning to decrease at low frequencies. Low distortions VF speed.
The attempt of this project was to see what happens if we succeed to build a high gain input stage to a CFA, making-it closer to a VFA.
As you can see, we now have a high loop gain that look similar to a VFA. But we keep the current on demand behavior. The slew rate is now limited, for stability reasons, but still higher than a VFA, closed loop bandwidth looking identical.
This limitation explain some of our choices: Cascoded VAS, high speed drivers, Lateral FETs for output power.
All the difference reside in the input stage. Because, by essence, there is a voltage difference between +Input (the base of the first transistor) and the - input in a CFA, due to the VBE voltage of this transistor, it is hard to get a DC working amplifier. Hence the idea to use some kind of diamond input, using the NPN/PNP VBEs to compensate this DC difference. But giving gain to the second stage of the diamond.
It works, but, because in our imperfect world there is no free lunch, comes with some drawbacks that we will explain in the next message, this one is yet too long and I'm not good in technical explanations ;-)
Last edited:
Why do people call you "Professor"? All of the professors I know are above average at making technical explanations; most are excellent. After all it is part of the job.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.