simulations
Hello,
here are the LTSpice schemes (where I compare the C9-R10-R7 branch and the active version. Please, do not consider LT1001 as a final choice: it is just one of the first I got on LTSpice... things can be improved):
input impedances (blu lines are the "active" version, green lines are the original one)
--
Daniele
Hello,
here are the LTSpice schemes (where I compare the C9-R10-R7 branch and the active version. Please, do not consider LT1001 as a final choice: it is just one of the first I got on LTSpice... things can be improved):
An externally hosted image should be here but it was not working when we last tested it.
input impedances (blu lines are the "active" version, green lines are the original one)
An externally hosted image should be here but it was not working when we last tested it.
--
Daniele
Side Topic
I don't want to distract the discussions on the servo design, but I don't have much to add to that part of the discussion. There were a few basic questions left unanswered, at least I didn't understand the answers, so I did a little more experimentation.
Klaus had suggested balancing the input impedance to reduce the need for a servo or for C9. Mark had suggested a difference from another circuit where a 10k resistor was added to balance the inputs. Looking at the schematic, I found that I could connect test leads at the output of C13 (junction to the input of R12) and at the input of R10. Remember that the FE switches the order of C9 and C10, so that R10 is closer to the LM318 and C9 is between R10 and audio ground.
I used the same FE RC module as in the earlier measurements. Of course, different day, different measurements.
With the FE as built and shorted inputs, I now had 1.0 mV (previously 0.6 mV) at the output. When I jumpered C9, the measurement became 11.5 mV (previously 20 mV). I admit that I gave the unit more time to stabilize than before, but this result is still a little puzzling. I'm not surprised that the basic offset might vary over time, but I expected the ratio between jumpered and normal to remain close to the audio gain of 30:1. Oh well, another mystery.
When I added the 10k resistor between C13 and R9 with C9 still jumpered, the output jumped to 3.3 V and the relay protected the speaker. This implies that this approach to balancing the inputs does not reduce the DC offset, but makes it significantly worse. I don't know why.
I don't know what it means to look at this circuit without C9 jumpered because C9 will support DC voltage between R10 and ground, but I measured it anyway. In that condition, the output DC was 119 mV.
I guess my conclusion is that these results help explain why Mauro removed that 10k resistor, even if I can't figure out why it didn't help. Also, it support the idea of the DC servo as an alternative to C9 for DC offset. I hope it means even more to you guys and can help explain how you would implement a DC servo.
Jac
I don't want to distract the discussions on the servo design, but I don't have much to add to that part of the discussion. There were a few basic questions left unanswered, at least I didn't understand the answers, so I did a little more experimentation.
Klaus had suggested balancing the input impedance to reduce the need for a servo or for C9. Mark had suggested a difference from another circuit where a 10k resistor was added to balance the inputs. Looking at the schematic, I found that I could connect test leads at the output of C13 (junction to the input of R12) and at the input of R10. Remember that the FE switches the order of C9 and C10, so that R10 is closer to the LM318 and C9 is between R10 and audio ground.
I used the same FE RC module as in the earlier measurements. Of course, different day, different measurements.
With the FE as built and shorted inputs, I now had 1.0 mV (previously 0.6 mV) at the output. When I jumpered C9, the measurement became 11.5 mV (previously 20 mV). I admit that I gave the unit more time to stabilize than before, but this result is still a little puzzling. I'm not surprised that the basic offset might vary over time, but I expected the ratio between jumpered and normal to remain close to the audio gain of 30:1. Oh well, another mystery.
When I added the 10k resistor between C13 and R9 with C9 still jumpered, the output jumped to 3.3 V and the relay protected the speaker. This implies that this approach to balancing the inputs does not reduce the DC offset, but makes it significantly worse. I don't know why.
I don't know what it means to look at this circuit without C9 jumpered because C9 will support DC voltage between R10 and ground, but I measured it anyway. In that condition, the output DC was 119 mV.
I guess my conclusion is that these results help explain why Mauro removed that 10k resistor, even if I can't figure out why it didn't help. Also, it support the idea of the DC servo as an alternative to C9 for DC offset. I hope it means even more to you guys and can help explain how you would implement a DC servo.
Jac
I haven't actually tried that 10k resistor shunt on a physical Rev C, but I did look at it in simulation, with a topology similar to the Musical Fidelity A370, but with different opamp models (LT1361, LME49860, etc.) rather than the LM318.
I found that the effect of the 10k is sensitive to the bias current of the opamp model. For the LT1361, the output offset nearly halved with the 10k resistor. At about 2.2k, the output offset became nearly 0 mV. For lower values, it changed polarity and started climbing again. Hence, there is an optimal value of that resistor for any specific input bias current (which will vary a lot between different opamp types, as well as different instances of the same opamp type).
For the LME49860, the output offset was much lower, and it increased by about 20% with the 10k resistor. So it's probably useful only with specific opamp, which has already been measured and a median value of the input bias current already determined.
I'm guessing that it's more trouble than it's worth, which is why Mauro abandoned that approach for the Rev C, and also went with a DC servo for the Evolution.
No - the simplest approach requires changes to the feedback network that will alter the AC response significantly. Replacing 12k/390 with 100k/3k3 shifts some poles by 8x, etc.
There are alternative biasing arrangements that allow the AC impedance seen by each input to stay the same as it is now, while balancing the DC impedance seen by both inputs. However, these require large electrolytics, maybe even larger than C9, for creating suitable bootstrapped nodes, so there's probably no advantage there either - we'd ideally like to get rid of large electrolytics.
The easiest way to avoid large electrolytics in the signal/feedback path is with a DC servo.
There is one other approach using Vgs-matched JFETs as an external, discrete LTP with the drains connected to the offset-trim pins of the LM318 (pins 1 & 5), thus using the remaining internals of the LM318, as well as its external compensation networks. However, it will still require compensation changes, because the gain of the JFET LTP isn't going to be the same as the darlington LTP of the LM318.
I did similar simulations with the MF 370A 10K resistor and replacing R12/R13 with 1K/10K and both reduced the offset with a 8 ohm load.
Two additional problems that I can see are that the offset of the My_ref is very complex and also influenced by the connected load.
Each opamp generates its own offset and the offset of the LM318 is inverted and amplified by the LM3886, then returned to the input of the LM318, only then is it blocked by C9 to unity gain.
As I understand it the voltage gain of the LM3886 is influenced by the connected load. If this is correct then both opamps have variable gains and when C9 is not used, a not easily to predict offset is then the result.
I did some simulations of the single opamp servo.
Two things that stand out most are that the servo output starts at full negative voltage, but I don't know if its not a simulation error and that the input filter did not remove all of the audio signal from the input of the servo opamp.
Two additional problems that I can see are that the offset of the My_ref is very complex and also influenced by the connected load.
Each opamp generates its own offset and the offset of the LM318 is inverted and amplified by the LM3886, then returned to the input of the LM318, only then is it blocked by C9 to unity gain.
As I understand it the voltage gain of the LM3886 is influenced by the connected load. If this is correct then both opamps have variable gains and when C9 is not used, a not easily to predict offset is then the result.
I did some simulations of the single opamp servo.
Two things that stand out most are that the servo output starts at full negative voltage, but I don't know if its not a simulation error and that the input filter did not remove all of the audio signal from the input of the servo opamp.
Last edited:
Thanks again. I had been thinking about the 100k/3k3 approach and then changing C9, C10, and C32 to hold their RC constant. But I am starting to understand that it isn't that easy. I think I understand now that, because the load is part of the feedback, changing the impedance of the feedback resistors will change more than desired. In other words, it's a bad idea unless I want a major compensation tuning project on something that I don't understand well enough.
As you say, it seems that the best alternative to our current C9 is the DC servo and that's challenging too.
I have been watching this thread with great interest!!
My plan is to build a PA200 with servos that has a composite front end like the My_ref amp.
Here is a thread from a while back that may be of some interest to you,
http://www.diyaudio.com/forums/chip-amps/166722-problem-parallel-lm3886-pa-100-a.html#post2188413
There was another thread that seemed to be an extension of this discussion and if I run across it again I will post the link for you.
A short time ago I made a DCservo amp out of a LT1210 and a LT1007 and it can be found here,
http://www.diyaudio.com/forums/soft...re-capacitor-impedance-graph.html#post3848338
I basically just used the example found in AN-1192 and it worked flawlessly.
I made the LTspice file after I had made the working circuit and it modeled nearly exactly as the real thing.
While I was going through my opamps to check for the lowest DCoffset I could find, I tried the LME49860's, and the samples I have are not at all matched!!
So I used the LT1007 for this circuit.
One of the opamp's offset in the LME49860 was quite typically about 1-2mv I think if not better and the other one in the same package was about 9mv.
These were within TI's spec but I was surprised to find such a difference in the same package.
I tried two of the three samples that I have and they both had the exact same offset parameters per section.
Doing some more searching I have decided to give the OPA1664 a try for my amp as the guarantied offset is something like below .5mv per section with no more than worst case +/-1.5mv.
Being that it is a quad opamp the circuit may end up a PA150 My_ref.
I am still new to this type configuration but I hope that my little experiment will help to give you some insight.
I have not been able to get any further on this because I have misplaced my sample OPA1664's so until I find them and/or order some more I am just going to keep reading.
Although, I do have more LT devices to work with for now.
Here is the data sheet for the OPA1664,
http://www.ti.com/lit/ds/sbos489/sbos489.pdf
FWIW
Cheers !!
jer
My plan is to build a PA200 with servos that has a composite front end like the My_ref amp.
Here is a thread from a while back that may be of some interest to you,
http://www.diyaudio.com/forums/chip-amps/166722-problem-parallel-lm3886-pa-100-a.html#post2188413
There was another thread that seemed to be an extension of this discussion and if I run across it again I will post the link for you.
A short time ago I made a DCservo amp out of a LT1210 and a LT1007 and it can be found here,
http://www.diyaudio.com/forums/soft...re-capacitor-impedance-graph.html#post3848338
I basically just used the example found in AN-1192 and it worked flawlessly.
I made the LTspice file after I had made the working circuit and it modeled nearly exactly as the real thing.
While I was going through my opamps to check for the lowest DCoffset I could find, I tried the LME49860's, and the samples I have are not at all matched!!
So I used the LT1007 for this circuit.
One of the opamp's offset in the LME49860 was quite typically about 1-2mv I think if not better and the other one in the same package was about 9mv.
These were within TI's spec but I was surprised to find such a difference in the same package.
I tried two of the three samples that I have and they both had the exact same offset parameters per section.
Doing some more searching I have decided to give the OPA1664 a try for my amp as the guarantied offset is something like below .5mv per section with no more than worst case +/-1.5mv.
Being that it is a quad opamp the circuit may end up a PA150 My_ref.
I am still new to this type configuration but I hope that my little experiment will help to give you some insight.
I have not been able to get any further on this because I have misplaced my sample OPA1664's so until I find them and/or order some more I am just going to keep reading.
Although, I do have more LT devices to work with for now.
Here is the data sheet for the OPA1664,
http://www.ti.com/lit/ds/sbos489/sbos489.pdf
FWIW
Cheers !!
jer
Last edited:
Low offset voltage and low drift of offset voltage are the two very important parameters of any FET input opamp to be used for DC servo duty.
the 1664 has an input impedance of 170k. it's +-1mV & 8uV/c cannot make up for the low input impedance.
the 2604 although +5mV and 8uV/C and should perform worse, has 10^12ohms input impedance.
the 1664 has an input impedance of 170k. it's +-1mV & 8uV/c cannot make up for the low input impedance.
the 2604 although +5mV and 8uV/C and should perform worse, has 10^12ohms input impedance.
Last edited:
Hmmmmmmmm.......Interesting thanks for pointing out the low input impedance of the OPA1664, AndrewT!!
I had not noticed that.
The search for a suitable quad opamp continues, I may be stuck with using the single or dual types.
I was hoping to keep the cost down by using a quad type.
The LT1007 has an input impedance of 5 to 7 Gigohms.
The TL07x's are rated at about 10 to 12 gigaohms, but most all of the ones I have in my pile as tested have an offset above 5mv or so to 10mv, I have only seen a few samples that were round 1.5mv to 3mv.
the LME49722 may be a good candidate for a dual type.
http://www.ti.com/lit/ds/symlink/lme49722.pdf
jer
I had not noticed that.
The search for a suitable quad opamp continues, I may be stuck with using the single or dual types.
I was hoping to keep the cost down by using a quad type.
The LT1007 has an input impedance of 5 to 7 Gigohms.
The TL07x's are rated at about 10 to 12 gigaohms, but most all of the ones I have in my pile as tested have an offset above 5mv or so to 10mv, I have only seen a few samples that were round 1.5mv to 3mv.
the LME49722 may be a good candidate for a dual type.
http://www.ti.com/lit/ds/symlink/lme49722.pdf
jer
Last edited:
The offset of the LF353 is typically 5mv to 10mv as do the TL07x's.
The offset of the LF412 that is used in the AN1192 example is typically .5mv to 1mv or so.
http://www.ti.com.cn/cn/lit/ds/symlink/lf412-n.pdf
For my particular app. I have just been looking at the OPA4227 and OPA4188 that may be some viable quad types for just the DC servo.
http://www.ti.com/lit/ds/symlink/opa2227.pdf
http://www.ti.com/lit/ds/sbos641b/sbos641b.pdf
The OPA4188 seems to be the best one of the two but is a bit costly at aproximately 3X that of the OPA4227.
This is still somewhat cheaper than some of the single and dual type that I have been looking at including the aging LF412's.
I was hoping to find a suitable HQ audio type that could be used for both the servo and the input stage.
Maybe a OPA1654 could work.
http://www.ti.com/lit/ds/symlink/opa1652.pdf
FWIW, Carry On!!
jer
The offset of the LF412 that is used in the AN1192 example is typically .5mv to 1mv or so.
http://www.ti.com.cn/cn/lit/ds/symlink/lf412-n.pdf
For my particular app. I have just been looking at the OPA4227 and OPA4188 that may be some viable quad types for just the DC servo.
http://www.ti.com/lit/ds/symlink/opa2227.pdf
http://www.ti.com/lit/ds/sbos641b/sbos641b.pdf
The OPA4188 seems to be the best one of the two but is a bit costly at aproximately 3X that of the OPA4227.
This is still somewhat cheaper than some of the single and dual type that I have been looking at including the aging LF412's.
I was hoping to find a suitable HQ audio type that could be used for both the servo and the input stage.
Maybe a OPA1654 could work.
http://www.ti.com/lit/ds/symlink/opa1652.pdf
FWIW, Carry On!!
jer
Last edited:
In the spirit of the the World Cup, I'm calling a time out in the game. The detail and technical considerations in this discussion is truly impressive and fun to observe, but you guys left me in the dust (locker room ) way back in the first period.
I suspect others watching the developments have the same question so I'll pose it here. "What is the target/goal in this search?" Is it directly related to a describable/discernible improvement in the quality of the output, or more directed at stability within a particular segment of the circuit(or both)? I do understand Dario's comment in the threads first post suggesting the eventual elimination/replacement of C9. I thought I would gather a better insight as to why/how a servo could be a better option as the project developed, but that is still not clear to me and probably several other non-designer MyRef fans who are following along.
Could someone translate what's happening here in simpler terms that would help bring the goal in focus for us observers.
(I'm fine with being directed to sources to read, but don't have a clue where to start.)
Thanks.
The detail and technical considerations in this discussion is truly impressive and fun to observe, but you guys left me in the dust (locker room ) way back in the first period. I suspect others watching the developments have the same question so I'll pose it here. "What is the target/goal in this search?" Is it directly related to a describable/discernible improvement in the quality of the output, or more directed at stability within a particular segment of the circuit(or both)? I do understand Dario's comment in the threads first post suggesting the eventual elimination/replacement of C9. I thought I would gather a better insight as to why/how a servo could be a better option as the project developed, but that is still not clear to me and probably several other non-designer MyRef fans who are following along.
Could someone translate what's happening here in simpler terms that would help bring the goal in focus for us observers.
(I'm fine with being directed to sources to read, but don't have a clue where to start.)
Thanks.
Last edited:
opa4134 is available.
As good as lf411 and lf412
There are many low offset and low drift of offset FET input opamps that can perform better and some much better than an lf411 at very much greater cost.
But, for an audio amplifier DC servo, is better output offset control required?
As good as lf411 and lf412
There are many low offset and low drift of offset FET input opamps that can perform better and some much better than an lf411 at very much greater cost.
But, for an audio amplifier DC servo, is better output offset control required?
Last edited:
Thanks, I understand that, but is C9 (as an electrolytic) contributing a distortion element that a servo would not? The servo appears to introduce a higher parts count so its hard to see how that could lead to a cleaner signal. Why/what does a servo (in this part of the circuit) do that prompts C9 as the target part. In other words - could the suspected/proposed advantages of a servo be beneficial to a part like C13?
Last edited:
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.