I've been thinking some more about using the nCore as a transconductance amplifier - or UCD for that matter, read on - and, at least in outline, it may be simpler than I first thought
Background
Some time ago in the Hypex thread there was a discussion about the performance when feeding a moving coil drive unit with a current signal rather than a voltage signal, and how the overall distortion of the drive unit can dropped considerabaly by using this form of drive. I'm not going to go into any of the justifications in this post, but do feel free to continue the discussion about current drive and its benefits and complications within this thread
Bruno commented that he uses current output in the Grim LS1. And the Grimm LS1 uses an implementation of the nCore circuit. And he also laid down the gauntlet of inviting a DIY solution for a current feedback loop. http://www.diyaudio.com/forums/vendors-bazaar/190434-hypex-ncore-395.html#post3019048
Following that post I had a wander around the Web and made few posts about my findings http://www.diyaudio.com/forums/vendors-bazaar/190434-hypex-ncore-403.html#post3026191 and my thoughts about implementing a transconductance nCore, and my problems in trying to work out how to implement it with a balanced input
Warning:
This is a theoretical design
I welcome all feedback/evolution
I have not tried this out and do not plan to try it out in the discernable future - no time
My electronics design is very rusty - 3 decades of abstinence
There may well be mistakes in my understanding and proposals
I don't know anything about the stability of hypex modules if used in the way I'm indicating here
Not sure what the suggestions do to the CMRR
Blanced Input Transconductance nCore - Latest Thoughts
The latest posts in the nCore thread about current drive made me think about this again; in particular the fact that Bruno had openly invited us to wrap a DIY current loop around the nCore
When I last thought about it I considered the nCore as being a single big opamp implemented in a closed-loop configuration. Having thought back to some of the other information I've read on the UCD and nCore, I remembered that it's actually an instrumentation amplifier and not a single opamp block. http://www.hypex.nl/docs/appnotes/gain_appnote.pdf
Referring to the Gain application note, linked above, Rg is R141 on the nCore pcb. So we have access to the feedback loop of the nCore instrumentation amplifier. If we remove R141 we can apply additional input signals - and/or feedback signals - to the input buffer stage
Looking at the standard nCore/instrumentation amplifier, one way to consider what's going on with the gain is to consider each of the input buffers as having a -ve feedback circuit where Rf is the feedback resistance and Rg/2 is connected to from the -ve buffer input to ground (this isn't accurate in detail as the 'shared' Rg is what gives the relative difference between the common-mode and differential-mode gains) and hence the smaller Rg, the higher the input buffer gain.
Also note that there are two independent gain stages in series. What I'm proposing means that there will instead/also be an overall feedback loop to provide the current feedback/gain.
One consideration is to leave Rg (R141) alone and feed back our current measurement voltage to the external + and - inputs. This would mean they would be implmented as non-inverting summing amplifiers. I don't like the thought of this .. but I can't find any immediately justifiable reason why 🙂
I've included 2 circuit variants below which should provide the nCore with current feedback and therefore a transconductance amplifier.
nCoreTransconductance
Please note, I think this circuit is unnecessarily complicated. The simplified circuit, described below is the one to examine in closer detail. The benefit of this more complicated circuit should be that we can adjust the value of Rs in relation to Rf to let us adjust the current gain if we can't achieve enough sensitivity with the simplified circuit, below
Looking first at nCoreTransconductance the circuit is configured as follows.
The speaker is connected in series with a current sense resistor Ri.
Rg (R141) is removed and connections are made via sense resistors Rs from either side of Ri to the negative inputs of the input buffer amps
The voltage across Ri is proportional to the output current through the speaker which also passes through Ri
VRi = Ri x Ioutput
VRi is fed back as negative feedback into the input buffer stage. For simplicity's sake, effectively VRi/2 is fed into each of the input buffer amps, and the voltage gain of each input buffer stage is Rf/Rs. The output of each input buffer half will rise until the voltages on its + and - inputs are the same ie.
Vin/2 = (Ri x Ioutput)/2 /4.17 x Rf/Rs
Ioutput/Vin = Rs/Rf x 1/(4.17 x Ri)
Sanity check: If Ri increases, the same -ve feedback voltage is achieved for a lower current through Ri, so the transconductance goes down.
The switching (output differential) stage has a gain of 4.17 IIRC - Bruno mentioned it somewhere in the mega nC400 thread.
I'm not convinced about my maths concerning the gain of this implementation as I suspect I'm mixing up my interpretation of closed loop gain of the input buffer stage vs closed loop gain when we feed back across both stages
nCoreTransconductance - simplified
Looking now at nCoreTransconductance - simplified
We shouldn't really need Rs. Rs lets us apply another gain factor, but we should be able to remove Rs and simply feed back the voltage across Ri
If we drop the value of Rs to 0, then the gain of the input stage becomes infinite and now the whole of the nCore is behaving as a single opamp
In this case, the output of each buffer half will once again rise until the voltages on its + and - inputs are the same ie.
Vin/2 = (Ri x Ioutput)/2
Ioutput/Vin = 1/Ri
Sense resistor
What about a value for Ri?
Max current for the nCore is 24A (DC) -> 16.97 A rms
This is backed up by the max power into 2R being quoted as 580W, which again corresponds to 17A
Alternatively max power = 400W into 4R -> 40V rms
Voltage gain = 25.8dB = 19.5 x
so our 40V rms output is produced by a 2V rms input
We'd probably like our 2V rms input to produce our (nominal) 17A rms output
So we want a transconductance of 17/2 = 8.5A/V
and R = 2/17 = 0.1176 R -> 0.12 R
Assuming a 0.15 R resistor:
Max power through resistor = I^2 x R = 17^2 x 0.15 = 43.35 W
How/where do we choose/find a suitable resistor?
Note alternatively, that we can only get the full current into a load of 40V/17A = 2.35 R or lower. So we may decide to go for a different transconductance and therefore a correspondingly different value of Ri
I'm not sure exactly what value Rf actually is on the nCore but from Hypex's gain_appnote.pdf, on all preceding models Rf = 2k2
With an Ri of about 0.15 R any currents through Rf will have very little effect on the voltage on the -ve input of the input buffer amp.. Effectively the gain of the input buffer stage will be ~13000. This means the effects of local feedback will be 82dB down
NOTE: I/We don't know the polarity of the connections for R141
My biggest concern would be about closed loop stability. However we know from the design evolution of the nCore in particular that it is
- inherently/deliberately unstable at its self-oscillation frequency
- has very high levels of feedback and therefore very high levels of loop gain (in comparison to more conventional current (no pun intended) design philosophies) within the audio band
- and therefore the loop gain has been carfully designed to keep the oscillation feedback characteristics well away from the audio feedback characteristics
- we can always add stabilising components to the load if required
I'd welcome feedback from Bruno to see if this is going along the right lines and if there are any 'gotchas' internally within the architecture/implementation of the nCore and UCD.
Anyone fancy an experiment?
Any corrections, evolutions, comments, updates etc. all welcome
Background
Some time ago in the Hypex thread there was a discussion about the performance when feeding a moving coil drive unit with a current signal rather than a voltage signal, and how the overall distortion of the drive unit can dropped considerabaly by using this form of drive. I'm not going to go into any of the justifications in this post, but do feel free to continue the discussion about current drive and its benefits and complications within this thread
Bruno commented that he uses current output in the Grim LS1. And the Grimm LS1 uses an implementation of the nCore circuit. And he also laid down the gauntlet of inviting a DIY solution for a current feedback loop. http://www.diyaudio.com/forums/vendors-bazaar/190434-hypex-ncore-395.html#post3019048
Following that post I had a wander around the Web and made few posts about my findings http://www.diyaudio.com/forums/vendors-bazaar/190434-hypex-ncore-403.html#post3026191 and my thoughts about implementing a transconductance nCore, and my problems in trying to work out how to implement it with a balanced input
Warning:
This is a theoretical design
I welcome all feedback/evolution
I have not tried this out and do not plan to try it out in the discernable future - no time
My electronics design is very rusty - 3 decades of abstinence
There may well be mistakes in my understanding and proposals
I don't know anything about the stability of hypex modules if used in the way I'm indicating here
Not sure what the suggestions do to the CMRR
Blanced Input Transconductance nCore - Latest Thoughts
The latest posts in the nCore thread about current drive made me think about this again; in particular the fact that Bruno had openly invited us to wrap a DIY current loop around the nCore
When I last thought about it I considered the nCore as being a single big opamp implemented in a closed-loop configuration. Having thought back to some of the other information I've read on the UCD and nCore, I remembered that it's actually an instrumentation amplifier and not a single opamp block. http://www.hypex.nl/docs/appnotes/gain_appnote.pdf
Referring to the Gain application note, linked above, Rg is R141 on the nCore pcb. So we have access to the feedback loop of the nCore instrumentation amplifier. If we remove R141 we can apply additional input signals - and/or feedback signals - to the input buffer stage
Looking at the standard nCore/instrumentation amplifier, one way to consider what's going on with the gain is to consider each of the input buffers as having a -ve feedback circuit where Rf is the feedback resistance and Rg/2 is connected to from the -ve buffer input to ground (this isn't accurate in detail as the 'shared' Rg is what gives the relative difference between the common-mode and differential-mode gains) and hence the smaller Rg, the higher the input buffer gain.
Also note that there are two independent gain stages in series. What I'm proposing means that there will instead/also be an overall feedback loop to provide the current feedback/gain.
One consideration is to leave Rg (R141) alone and feed back our current measurement voltage to the external + and - inputs. This would mean they would be implmented as non-inverting summing amplifiers. I don't like the thought of this .. but I can't find any immediately justifiable reason why 🙂
I've included 2 circuit variants below which should provide the nCore with current feedback and therefore a transconductance amplifier.
nCoreTransconductance
Please note, I think this circuit is unnecessarily complicated. The simplified circuit, described below is the one to examine in closer detail. The benefit of this more complicated circuit should be that we can adjust the value of Rs in relation to Rf to let us adjust the current gain if we can't achieve enough sensitivity with the simplified circuit, below
Looking first at nCoreTransconductance the circuit is configured as follows.
The speaker is connected in series with a current sense resistor Ri.
Rg (R141) is removed and connections are made via sense resistors Rs from either side of Ri to the negative inputs of the input buffer amps
The voltage across Ri is proportional to the output current through the speaker which also passes through Ri
VRi = Ri x Ioutput
VRi is fed back as negative feedback into the input buffer stage. For simplicity's sake, effectively VRi/2 is fed into each of the input buffer amps, and the voltage gain of each input buffer stage is Rf/Rs. The output of each input buffer half will rise until the voltages on its + and - inputs are the same ie.
Vin/2 = (Ri x Ioutput)/2 /4.17 x Rf/Rs
Ioutput/Vin = Rs/Rf x 1/(4.17 x Ri)
Sanity check: If Ri increases, the same -ve feedback voltage is achieved for a lower current through Ri, so the transconductance goes down.
The switching (output differential) stage has a gain of 4.17 IIRC - Bruno mentioned it somewhere in the mega nC400 thread.
I'm not convinced about my maths concerning the gain of this implementation as I suspect I'm mixing up my interpretation of closed loop gain of the input buffer stage vs closed loop gain when we feed back across both stages
nCoreTransconductance - simplified
Looking now at nCoreTransconductance - simplified
We shouldn't really need Rs. Rs lets us apply another gain factor, but we should be able to remove Rs and simply feed back the voltage across Ri
If we drop the value of Rs to 0, then the gain of the input stage becomes infinite and now the whole of the nCore is behaving as a single opamp
In this case, the output of each buffer half will once again rise until the voltages on its + and - inputs are the same ie.
Vin/2 = (Ri x Ioutput)/2
Ioutput/Vin = 1/Ri
Sense resistor
What about a value for Ri?
Max current for the nCore is 24A (DC) -> 16.97 A rms
This is backed up by the max power into 2R being quoted as 580W, which again corresponds to 17A
Alternatively max power = 400W into 4R -> 40V rms
Voltage gain = 25.8dB = 19.5 x
so our 40V rms output is produced by a 2V rms input
We'd probably like our 2V rms input to produce our (nominal) 17A rms output
So we want a transconductance of 17/2 = 8.5A/V
and R = 2/17 = 0.1176 R -> 0.12 R
Assuming a 0.15 R resistor:
Max power through resistor = I^2 x R = 17^2 x 0.15 = 43.35 W
How/where do we choose/find a suitable resistor?
Note alternatively, that we can only get the full current into a load of 40V/17A = 2.35 R or lower. So we may decide to go for a different transconductance and therefore a correspondingly different value of Ri
I'm not sure exactly what value Rf actually is on the nCore but from Hypex's gain_appnote.pdf, on all preceding models Rf = 2k2
With an Ri of about 0.15 R any currents through Rf will have very little effect on the voltage on the -ve input of the input buffer amp.. Effectively the gain of the input buffer stage will be ~13000. This means the effects of local feedback will be 82dB down
NOTE: I/We don't know the polarity of the connections for R141
My biggest concern would be about closed loop stability. However we know from the design evolution of the nCore in particular that it is
- inherently/deliberately unstable at its self-oscillation frequency
- has very high levels of feedback and therefore very high levels of loop gain (in comparison to more conventional current (no pun intended) design philosophies) within the audio band
- and therefore the loop gain has been carfully designed to keep the oscillation feedback characteristics well away from the audio feedback characteristics
- we can always add stabilising components to the load if required
I'd welcome feedback from Bruno to see if this is going along the right lines and if there are any 'gotchas' internally within the architecture/implementation of the nCore and UCD.
Anyone fancy an experiment?
Any corrections, evolutions, comments, updates etc. all welcome
Attachments
the first issue i see is regarding the (upper left) op amp for Vin-: it's noninverting input gets the signal but it's inverting input togheter with the upper Rf are connected to GND.
I supose this will rather work as a comparator and not as an linear amplifier ...
I supose this will rather work as a comparator and not as an linear amplifier ...
The ground connection shouldn't change the behaviour of anything. The -ve input of the Vin- buffer is connected to the -ve end of Ri. It just so happens that that end of Ri is connected to ground. The 2 input buffers are looking at the differential signal across Ri.
The circuit should also be valid if Ri is connected to Vout+ and the speaker to ground. Or if you split Ri into 2 so that your current sense is going to ground. A differential amplifier is still amplifying the difference between its 2 inputs even if one of them is ground.
(It does mean, however, that my 'simplification' that each input buffer 'sees' 1/2 the signal across Ri, is just that - a simplification)
The 2 input buffers should jointly see the whole/differential signal across Ri wherever it is
Why would it be a comparator? Why won't it be linear?
i am talking only about the upper left op amp.
Look closer at the attached image.
The inverting input is held to GND. As soon as the (Vin-) non inverting input get's just a little higher than GND the output of this op amp will swing to +Ve. There is no feedback as the inverting input is still GND so the ouput will remain +Ve until the (Vin-) non inverting input get's a little under GND - then the output will swing to -Ve.
That behaviour is not linear, it's a comparator with GND as a reference.
Look closer at the attached image.
The inverting input is held to GND. As soon as the (Vin-) non inverting input get's just a little higher than GND the output of this op amp will swing to +Ve. There is no feedback as the inverting input is still GND so the ouput will remain +Ve until the (Vin-) non inverting input get's a little under GND - then the output will swing to -Ve.
That behaviour is not linear, it's a comparator with GND as a reference.
Attachments
Last edited:
I have also been thinking UCD transconductance amplifier. I don't have too much experience about analog design either, but I think that current drive should be build deeper into modulator.
If I have understood UCD concept correctly, then Rl + Cl network makes the amplifier unstable, selfoscillating.
Intuitively current drive mod works as shown in figure. I am not sure about feedback network (Lf, Rl + Ll). Just tune something there to make the design unstable at 100-500kHz..
If I have understood UCD concept correctly, then Rl + Cl network makes the amplifier unstable, selfoscillating.
Intuitively current drive mod works as shown in figure. I am not sure about feedback network (Lf, Rl + Ll). Just tune something there to make the design unstable at 100-500kHz..
rolandED
I get your point. I've been working through ideas for evolving the circuti, but I won't have time to reply/document my thoughts before Christmas
veskelin
The whole of the amp and the feedback loop isn't a single op-amp stage as you have drawn, but an instumentation amp with 3 opamp stagee. If you refer to the instumentation amp diagram, the UCD block - which includes the self-oscillating components is all self-contained in the output amp block. The feedback loop in the output block already contains the components to both create the self-oscillation and to filter out the self-oscillating part of the feedback loop from the analogue/audio band part of the loop. IIRC I think Bruno said that for the nCore it's a 5th order control loop
I believe we should therefore fundamentally treat the whole UCD instrumentation amp as something which opeartes cleanly in the audio band and that the switching feedback can primarily be ignored. When we implemnt a current feedback loop we may need to add components to reatain stability but we don't need to do anything to the self-oscillation part of the the UCD power stage
I don't believe that Bruno's invitation for the DIY community to implement a current feedback loop was an invitation to reengineer the internal UCD/switching aspects of the design. He keeps reminding us that it's an analogue amplifier and should be treated as such
I get your point. I've been working through ideas for evolving the circuti, but I won't have time to reply/document my thoughts before Christmas
veskelin
The whole of the amp and the feedback loop isn't a single op-amp stage as you have drawn, but an instumentation amp with 3 opamp stagee. If you refer to the instumentation amp diagram, the UCD block - which includes the self-oscillating components is all self-contained in the output amp block. The feedback loop in the output block already contains the components to both create the self-oscillation and to filter out the self-oscillating part of the feedback loop from the analogue/audio band part of the loop. IIRC I think Bruno said that for the nCore it's a 5th order control loop
I believe we should therefore fundamentally treat the whole UCD instrumentation amp as something which opeartes cleanly in the audio band and that the switching feedback can primarily be ignored. When we implemnt a current feedback loop we may need to add components to reatain stability but we don't need to do anything to the self-oscillation part of the the UCD power stage
I don't believe that Bruno's invitation for the DIY community to implement a current feedback loop was an invitation to reengineer the internal UCD/switching aspects of the design. He keeps reminding us that it's an analogue amplifier and should be treated as such
I would love to have a very low distortion nCore transconductance amp, but for simplicity my target is in UCD current drive.
UCD modulator is basically very simple. A comparator+drive stages+LP network+Control circuit (=just R||(Rl+Cl) network). I think than nCore is very similar, but control circuit is more complex.
I assume that that both UCD and nCore can be modelled as the “A=4.5” gain stage in your instrumentation amp fig.
There are few reason I would like to build current drive into UCD modator.
1) Shorter feedback.
2) Gain of these two gain stages is too low to make decent feedback. Removing Rg (R141) does not solve the problem.
UCD modulator is basically very simple. A comparator+drive stages+LP network+Control circuit (=just R||(Rl+Cl) network). I think than nCore is very similar, but control circuit is more complex.
I assume that that both UCD and nCore can be modelled as the “A=4.5” gain stage in your instrumentation amp fig.
There are few reason I would like to build current drive into UCD modator.
1) Shorter feedback.
2) Gain of these two gain stages is too low to make decent feedback. Removing Rg (R141) does not solve the problem.
Yes
I don't understand your point about 'shorter feedback'
With regards gain and Rg:
gain = (1 + {2 Rf / Rg) ) x {gain of UCD stage}
if you remove Rg
it means Rg -> infinity
so the gain of the input buffer stages -> 1
ie. removing Rg (making Rg bigger) decreases the gain
To increase the gain you need to make Rg smaller - replace it with a smaller resistor
Yep. You are right. I got confused about Rg. Anyway, I don't see the point why you are trying to build feedback around Rg.
Someone tested nCore in "opamp trasconductance configuration". See nCore thread posts 4246 and 4266. No success.
Was is because low open loop gain? Maybe, but I think that feedback over several amplifier stages is generally bad.
Someone tested nCore in "opamp trasconductance configuration". See nCore thread posts 4246 and 4266. No success.
Was is because low open loop gain? Maybe, but I think that feedback over several amplifier stages is generally bad.
An op amp and the assumptions of the maths of an op amp only work when it has 'infinite gain', and it's this which gives a voltage loop a zero output impedance and gives a transconductance amp infinite output impedance. With Rg in place the closed loop ncore and ucd as shipped aren't our theoretical op amp. The ncore and ucd are also both balanced. I want to retain balanced operation.
The diagram in 4246 (http://www.diyaudio.com/forums/vendors-bazaar/190434-hypex-ncore-425.html#post3042759) may or may not do something, but it's feeding back around a (voltage) loop gain of 26dB. You can try to work out it's closed loop (transconductance) characteristics if you want. It isn't balanced. It's a guess at a transconductance amp. And hence got the results in post 4266 (http://www.diyaudio.com/forums/vendors-bazaar/190434-hypex-ncore-427.html#post3044022)
Go and read Bruno's papers about the ncore and feedback wrapped around multiple stages. He has no inherent problem with it (he encourages it) but you have to be aware of loop stability
rolandED showed why in particular, but I can add a general explanation too:
Every signal pairs can be separated to common mode and differential mode components, and the behaviour of a circuit can be investigated by analysing these two modes (and not only one) separately as long as linear operation is assured. Only after linear operation is assured, and common mode signals are proved to be unsignificant/unimportant, then discussing the differential mode is enough to describe the circuit's behaviour.
These required steps were not done, hence the opposition.
Common mode amplification (as well as single ended amplification) of the 1st pair of OPA is 1+Rf/Rs, so without Rs it would be (close to) infinite. Even at both Vin- and Vout- is theoretically zero it is not a good setting, leads to unpredicable results in reality.
Two other thought:
1: Loop gain of this (outer loop) feedback must be significantly less then 0 dB (preferably under 10 dB) above LC filter cutoff freq in order not to modifying oscillation characteristics of the core amplifier.
2: Outer loop must be stable also. With high loop gain (as you wanted high output impedance) inductance of speaker and 1 pole of the power amplifier must be included in the modell at least.
If you are lucky, pole caused by speaker inductance can be enough to maintain relatively high loop gain at low freq with reasonably gain margin to avoid stability problems.
The more accurate the modell (of the speaker and power amplifier) the highest loop gain can be set based on simulation.
Kii Three
Any of you tried something like the Kii Three?
Both Voltage and Current Control = Motion Control
electronics
S
Any of you tried something like the Kii Three?
Both Voltage and Current Control = Motion Control
electronics
S
Why did you abandon the idea of Hypex transconductance amplifier? Did you solve it maybe? Please share some information for another who are not so deeply skilled in electronics. I am very interested in current drive amplifier, understand it and have some designs which comply to its specific requirements to bring only the benefits! I am not able to understand and redesign NCore feedback loop, though.
Last edited:
bimbla, did you checked it in reality? Is it stable? More info how to implement in the finished module? Thanks!
Class-D Current Driver
Windforce85,
I have not checked it in reality with Class-D, yet. I will be doing it soon and I will update my results here in this very thread.
I am jumping on the Class-D full-bridge version and implementing load current feedback is not as straight forward in full-bridge.
Either way, I'll post my results.
P.S. I have implemented this in Class-AB with tremendous results!! But that is not what you want...I know.
Windforce85,
I have not checked it in reality with Class-D, yet. I will be doing it soon and I will update my results here in this very thread.
I am jumping on the Class-D full-bridge version and implementing load current feedback is not as straight forward in full-bridge.
Either way, I'll post my results.
P.S. I have implemented this in Class-AB with tremendous results!! But that is not what you want...I know.
- Status
- Not open for further replies.