Andy_c wrote:
So what?
Why is this a benefit?
Usually,when designing a NFB system I either decide a value of dc NFB loop gain (feedback factor) or I decide I want as much as possible for a given unity gain bandwidth/phase margin. I don't see any reason to have this as an adjustment.
The test bench observation that the distortion can be minimized by adjustment is true but is irrelevant. As far as I can see it is just a gimmick. The distortion is NOT being reduced MORE than it would with a conventional NFB loop at all. As I see it, it is like designing a NFB loop with a certain loop gain and bandwidth and distortion and then adding a pot to be able to "spoil" it.
What's the benefit?
In short, the HEC adjustment allows you to change the dc NFB factor whilst keeping the roll-off slope constant-ish.
So what?
Why is this a benefit?
Usually,when designing a NFB system I either decide a value of dc NFB loop gain (feedback factor) or I decide I want as much as possible for a given unity gain bandwidth/phase margin. I don't see any reason to have this as an adjustment.
The test bench observation that the distortion can be minimized by adjustment is true but is irrelevant. As far as I can see it is just a gimmick. The distortion is NOT being reduced MORE than it would with a conventional NFB loop at all. As I see it, it is like designing a NFB loop with a certain loop gain and bandwidth and distortion and then adding a pot to be able to "spoil" it.
What's the benefit?
Re: Re: hec != hoax ?
Hi Ovidiu,
Your points are well-stated.
Certainly the lack of a trimpot in Edmond's true feedback approach is an advantage unless one is willing to settle for an approximate 40 dB EC effect, in which ordinary 1% resistors will do just fine.
One certainly needs to be wary of design by simulation only. However, I agree with Edmond that SPICE is extremely valuable in exploring the design space and converging quickly to a good solution. While SPICE is certainly no substitute for building the device in practice, it does allow the kinds of comparisons of techniques that Edmond and I have been making. I also agree with Edmond that the iterative simulation approach, in the right hands, can provide very good insight into how the circuits work.
I also am a bit irritated by Edmond's use of the term "hoax". There has never been an intention of dishonesty about EC from Hawksford himself onward. Nor is it appropriate to assert that Vanderkooy and Lipshitz "de-bunked" HEC. Hey, the circuit works! I think some people have gotten too obsessively hung up in the semantics of HEC, which were presumably chosen by Hawksford to help people visualize in a certain way what goes on with that circuit. No matter what it is called, it is a nice, elegant circuit. It's too bad we can't just leave it at that.
At the end of the day, what matters is which circuit approach is best for making the best amplifier. Maybe they are equally good, and that is a good thing for all of us. In my opinion, the value of Edmond's contribution here lies not in debunking HEC, but rather in his showing a real alternative to the HEC architecture that may be just as good.
Cheers,
Bob
syn08 said:
Hi Ovidiu,
Your points are well-stated.
Certainly the lack of a trimpot in Edmond's true feedback approach is an advantage unless one is willing to settle for an approximate 40 dB EC effect, in which ordinary 1% resistors will do just fine.
One certainly needs to be wary of design by simulation only. However, I agree with Edmond that SPICE is extremely valuable in exploring the design space and converging quickly to a good solution. While SPICE is certainly no substitute for building the device in practice, it does allow the kinds of comparisons of techniques that Edmond and I have been making. I also agree with Edmond that the iterative simulation approach, in the right hands, can provide very good insight into how the circuits work.
I also am a bit irritated by Edmond's use of the term "hoax". There has never been an intention of dishonesty about EC from Hawksford himself onward. Nor is it appropriate to assert that Vanderkooy and Lipshitz "de-bunked" HEC. Hey, the circuit works! I think some people have gotten too obsessively hung up in the semantics of HEC, which were presumably chosen by Hawksford to help people visualize in a certain way what goes on with that circuit. No matter what it is called, it is a nice, elegant circuit. It's too bad we can't just leave it at that.
At the end of the day, what matters is which circuit approach is best for making the best amplifier. Maybe they are equally good, and that is a good thing for all of us. In my opinion, the value of Edmond's contribution here lies not in debunking HEC, but rather in his showing a real alternative to the HEC architecture that may be just as good.
Cheers,
Bob
traderbam said:
That's not what I'm saying at all. This is not related to the EC adjustment, which has almost no effect on the unity loop gain frequency of the EC loop. That adjustment increases the DC loop gain of the EC loop while decreasing the loop bandwidth, keeping the unity loop gain frequency almost constant.
The idea is that if you implement the output stage with a 1DOF feedback circuit and you want the output stage to have, say, a 10 MHz bandwidth, you'll need to make its unity loop gain frequency around 10 MHz. But such a circuit is likely to be pretty twitchy with capacitive loads, even with an output inductor. To make it more stable against such loads, the best way is simply brute force - by reducing its unity loop gain frequency. Let's say you reduce it to 1 MHz. Now the closed loop bandwidth of the output stage itself will be 1 MHz also - the same as its unity loop gain frequency. When you put this output stage into the global feedback loop of the amp, the global unity loop gain frequency must be less than 1 MHz, or else the phase margin will be degraded by the 1 MHz bandwidth of the output stage.
Now take a source follower with a 10 MHz bandwidth and put an EC loop around it. Compensate that loop so its unity loop gain frequency is 1 MHz. This should have similar distortion to the output stage discussed in the previous paragraph. However, its bandwidth will still be 10 MHz. Now when you put this stage into a global feedback loop, the global unity loop gain frequency can be made larger because the bandwidth of the output stage is 10 MHz instead of 1 MHz. You'll get lower closed-loop distortion.
Andy, I misunderstood your original point. In your example, I think what you are saying is that you want the OS NFB loop to become benign above 1MHz leaving the original follower to operate up to 10MHz unity-gain bandwidth. Ok.
But I don't see why this cannot be accomplished in a single-loop NFB system. Again, imagine a unity gain follower with bandwidth 10MHz. You insert a gain element A in front of it. A has dc gain 40dB and rolls off to unity gain at 1MHz. A has unity gain above 1MHz. Subtract the output from the input signal and feed the result into A in the conventional way.
Erratum: that's not quite right. The output of A should be the input plus the difference between input and output put through a rolled-off gain function. So that above 1MHz the output of A is primarily the input signal. Same number of feedback loops: one.
But I don't see why this cannot be accomplished in a single-loop NFB system. Again, imagine a unity gain follower with bandwidth 10MHz. You insert a gain element A in front of it. A has dc gain 40dB and rolls off to unity gain at 1MHz. A has unity gain above 1MHz. Subtract the output from the input signal and feed the result into A in the conventional way.
Erratum: that's not quite right. The output of A should be the input plus the difference between input and output put through a rolled-off gain function. So that above 1MHz the output of A is primarily the input signal. Same number of feedback loops: one.
I'm only understanding part of what you're saying. You have a gain block A (presumably with differential inputs?) with a DC gain of 100x and a pole at 10 kHz, such that its magnitude would approach 1 at 1 MHz in the absence of other poles or zeros. But you then introduce a zero at 1 MHz into A, such that the asymptotic gain as frequency becomes large is equal to 1. Then the output stage contributes a pole at 10 MHz. So as frequency increases from DC, you have pole at 10 kHz, then a zero at 1 MHz, then the output stage pole at 10 MHz. Okay so far?
Assuming that's what you intended, I'm a bit confused by the final hookup you proposed. Could you sketch something and scan and post it to clarify?
Assuming that's what you intended, I'm a bit confused by the final hookup you proposed. Could you sketch something and scan and post it to clarify?
Andy,
Something like this. "A" has a low pass filter response just like a standard single pole op-amp whose unity gain f = 1MHz. G represents the output follower transistor(s) with 10MHz BW.
Would this produce the sort of response you were talking about?
Edit: Oops...sorry, the summer should have + signs on both inputs.
Something like this. "A" has a low pass filter response just like a standard single pole op-amp whose unity gain f = 1MHz. G represents the output follower transistor(s) with 10MHz BW.
Would this produce the sort of response you were talking about?
Edit: Oops...sorry, the summer should have + signs on both inputs.
Attachments
Andy, you might have noticed that if you were to connect the + input of A to Ve instead of Vi you end up with the familiar HEC topology, provided A = 1. You can see HEC as being an alternative way to generate the gain of A by using a regenerative loop.
I've mentioned in past posts that using a PFB loop to produce forward gain has pros and cons. One pro is that you can generate very large gains at low frequencies. One con is the amplification of any distortion within the loop.
Brian
I've mentioned in past posts that using a PFB loop to produce forward gain has pros and cons. One pro is that you can generate very large gains at low frequencies. One con is the amplification of any distortion within the loop.
Brian
Hi Brian,
Your posts just jogged my memory from about a year ago. I believe it was Bonsai who posted a similar circuit. I don't know if he ever ended up building it though. I believe he had some voltages bootstrapped +/- V from the output. He biased an op-amp integrator from these, with a configuration that was probably very similar, if not identical, to your block diagram.
Your posts just jogged my memory from about a year ago. I believe it was Bonsai who posted a similar circuit. I don't know if he ever ended up building it though. I believe he had some voltages bootstrapped +/- V from the output. He biased an op-amp integrator from these, with a configuration that was probably very similar, if not identical, to your block diagram.
Andy,
I haven't seen Bonsai's circuit but it sounds similar. I've attached a conceptual circuit to help illustrate the topology using Bob's OS. This model provides a similar amount of correction and bandwidth and has the same property that if the fb is disabled the circuit behaves as a follower.
I haven't seen Bonsai's circuit but it sounds similar. I've attached a conceptual circuit to help illustrate the topology using Bob's OS. This model provides a similar amount of correction and bandwidth and has the same property that if the fb is disabled the circuit behaves as a follower.
Attachments
Re: Re: Re: Re: Re: hec != hoax ?
Hi Edmond,
In referring to "his" temperature compensation above, are you referring to Hawksford's version or mine?
Mine works fine, as shown in Figure 12 of my original paper. The Q22-23 error amplifier transistors set the bias voltage across the emitters of the Q26-27 drivers of the MOSFET gates, so no other junctions are involved here. This is accomplished by feedback action.
In practice, for thermal feedback, I mounted Q22 on the heatsink. This gave just about the right coefficient of thermal compensation. See Figure 4. In doing so, I used an MPS-U06 transistor with a tab for convenient mounting.
There are, admittedly, two things that are sub-optimal about this approach. The first is that often transistors with mounting tabs are medium power transistors with a lower ft that can be gotten otherwise. Secondly, this mounting arrangement on the heatsink tends to increase the lead lengths on Q22 in the error amplifier. Ideally, other means can be used to introduce the desired heat sink temperature dependence, but they will tend to add some complexity.
Cheers,
Bob
Edmond Stuart said:
Hi Edmond,
In referring to "his" temperature compensation above, are you referring to Hawksford's version or mine?
Mine works fine, as shown in Figure 12 of my original paper. The Q22-23 error amplifier transistors set the bias voltage across the emitters of the Q26-27 drivers of the MOSFET gates, so no other junctions are involved here. This is accomplished by feedback action.
In practice, for thermal feedback, I mounted Q22 on the heatsink. This gave just about the right coefficient of thermal compensation. See Figure 4. In doing so, I used an MPS-U06 transistor with a tab for convenient mounting.
There are, admittedly, two things that are sub-optimal about this approach. The first is that often transistors with mounting tabs are medium power transistors with a lower ft that can be gotten otherwise. Secondly, this mounting arrangement on the heatsink tends to increase the lead lengths on Q22 in the error amplifier. Ideally, other means can be used to introduce the desired heat sink temperature dependence, but they will tend to add some complexity.
Cheers,
Bob
hec != hoax ?
Hi Bob,
Vanderkooy and Lipshitz published their paper bofere your presentation: January 1980.
You still haven't answered my question: what does the other point of view reveals in terms of stability, distortion reduction etc.
It didn't take me a couple of iterations to get it to be competitive, rather comparative. Besides, may I remind you that I had to improve your output stage before it came close to the performance of my design.
Are you teasing me? First, it didn't take me 25 years, just one day.
Second, the only reason I've designed it was because someone (Pete) challenged my to come up with competitive NFB variant.
Third, I was never interested in either HEC or a NFB contender, as there are simpler ways to get the same performance.
Actually, there isn't any need to simulate or even build a complete amp. Comparing the two output stages in every relevant detail is enough. Nevertheless and only to satisfy my own curiosity, I've simmed a complete amp (without NDFL). It was just as stable as other high performance amps I've simmed and the THD20 was below 1ppm (BW=100kHz)
I don't recognize any elegance in HEC.
1. I steals about 22V from the VAS
2. The need for a boosted front end power supply
3. A lot of power dissipation in the drivers an EC circuitry
4. The need for precision resistors and maybe even a trim pot.
Thank you so much.
That's right, but it was never my intention to do so.
Although I agree with Vanderkooy and Lipshitz and their view on error feedback, I'm not upset about HEC, I'm just not charmed by HEC and that has little to do with my 'personal laws of semantics'.
Cheers, Edmond.
Bob Cordell said:
Hi Bob,
Vanderkooy and Lipshitz published their paper bofere your presentation: January 1980.
You still haven't answered my question: what does the other point of view reveals in terms of stability, distortion reduction etc.
It didn't take me a couple of iterations to get it to be competitive, rather comparative. Besides, may I remind you that I had to improve your output stage before it came close to the performance of my design.
Are you teasing me? First, it didn't take me 25 years, just one day.
Second, the only reason I've designed it was because someone (Pete) challenged my to come up with competitive NFB variant.
Third, I was never interested in either HEC or a NFB contender, as there are simpler ways to get the same performance.
Actually, there isn't any need to simulate or even build a complete amp. Comparing the two output stages in every relevant detail is enough. Nevertheless and only to satisfy my own curiosity, I've simmed a complete amp (without NDFL). It was just as stable as other high performance amps I've simmed and the THD20 was below 1ppm (BW=100kHz)
I don't recognize any elegance in HEC.
1. I steals about 22V from the VAS
2. The need for a boosted front end power supply
3. A lot of power dissipation in the drivers an EC circuitry
4. The need for precision resistors and maybe even a trim pot.
Thank you so much.
That's right, but it was never my intention to do so.
Although I agree with Vanderkooy and Lipshitz and their view on error feedback, I'm not upset about HEC, I'm just not charmed by HEC and that has little to do with my 'personal laws of semantics'.
Cheers, Edmond.
Why the EC trimpot?
Here is perhaps a way of looking at EC that puts it into perspective in regard to its negative feedback nature.
Although HEC certainly can be seen as a negative feedback system, I think that a useful distinguishing feature of it from more conventional NFB implementations is the fact that the forward gain is achieved with positive feedback, rather than with straight gain stages. I'm not saying this is better or worse, just different.
In the past, when a circuit has been presented as being a version of HEC, I have always looked for the presence of a positive feedback path in the topology.
This also explains the need for precision resistors or a trimpot. Basically what you are doing when you twiddle the pot is that you are adjusting the feedback factor in the positive feedback loop to exacly unity positive feedback, which of course corresponds to causing the forward gain to tend to infinity at low frequencies.
Because a more traditional approach using NFB relies on lots of forward gain (the more the better), it does not need precision or an adjustment to get that high gain.
Cheers,
Bob
Here is perhaps a way of looking at EC that puts it into perspective in regard to its negative feedback nature.
Although HEC certainly can be seen as a negative feedback system, I think that a useful distinguishing feature of it from more conventional NFB implementations is the fact that the forward gain is achieved with positive feedback, rather than with straight gain stages. I'm not saying this is better or worse, just different.
In the past, when a circuit has been presented as being a version of HEC, I have always looked for the presence of a positive feedback path in the topology.
This also explains the need for precision resistors or a trimpot. Basically what you are doing when you twiddle the pot is that you are adjusting the feedback factor in the positive feedback loop to exacly unity positive feedback, which of course corresponds to causing the forward gain to tend to infinity at low frequencies.
Because a more traditional approach using NFB relies on lots of forward gain (the more the better), it does not need precision or an adjustment to get that high gain.
Cheers,
Bob
Re: hec != hoax ?
Hawksford published "Distortion Correction in Audio Power Amplifiers" Presented at the 65th Convention 1980 February 25-28, London
The paper you cited is actually from 1979 "Error correction in power amplifiers", John Vanderkooy and Stanley p. Lipshitz, Presented at the 63rd Convention May 15-18
One before and one after:
"Is Zero Distortion Possible with Feedback?", Lipshitz, Stanley P.; Vanderkooy, John, Presented at the 76th Convention 1984 October 8-11, New York
And here is my favorite "feed forward" approach of EC:
"Design and construction of a feedforward error correction amplifier", Susumu Takahashi, Susumu Tanaka, Sansui Electric Company Limited, Tokyo, Japan, Presented at, the 65th Convention, 1980 February 25-28, London
It would be very interesting if somebody could put together a paper to prove the equivalence of these three approaches: HEC, NFB, FF.
Edmond Stuart said:
Hawksford published "Distortion Correction in Audio Power Amplifiers" Presented at the 65th Convention 1980 February 25-28, London
The paper you cited is actually from 1979 "Error correction in power amplifiers", John Vanderkooy and Stanley p. Lipshitz, Presented at the 63rd Convention May 15-18
One before and one after:
"Is Zero Distortion Possible with Feedback?", Lipshitz, Stanley P.; Vanderkooy, John, Presented at the 76th Convention 1984 October 8-11, New York
And here is my favorite "feed forward" approach of EC:
"Design and construction of a feedforward error correction amplifier", Susumu Takahashi, Susumu Tanaka, Sansui Electric Company Limited, Tokyo, Japan, Presented at, the 65th Convention, 1980 February 25-28, London
It would be very interesting if somebody could put together a paper to prove the equivalence of these three approaches: HEC, NFB, FF.
Re: Why the EC trimpot?
Hello Bob,
In my HEC amp, I had a trimpot to tune the pos feedback for min THD. The resistance value for optimum setting is around 2.7k. I just did a quick check, and changes of this resistor value of up to 80 ohms (which is almost 3% deviation) in either direction increases the THD by about 50% in RMS value, higher harmonic peak values a bit more.
I have built 3 implementations with 1% resistors and they all are very close to each other in THD. Of course this may be to some extend implementation dependent, and/or I got lucky, but although trimpots can surely give you the best value, my experience is that in practise it isn't really required.
Jan Didden
Bob Cordell said:
Hello Bob,
In my HEC amp, I had a trimpot to tune the pos feedback for min THD. The resistance value for optimum setting is around 2.7k. I just did a quick check, and changes of this resistor value of up to 80 ohms (which is almost 3% deviation) in either direction increases the THD by about 50% in RMS value, higher harmonic peak values a bit more.
I have built 3 implementations with 1% resistors and they all are very close to each other in THD. Of course this may be to some extend implementation dependent, and/or I got lucky, but although trimpots can surely give you the best value, my experience is that in practise it isn't really required.
Jan Didden
Re: hec != hoax ?
Hi Edmond,
I must admit I am not familiar with that one. I surely would have referenced it in my papers if I was aware of it. Was it published in the JAES? Can you please provide me with the exact reference? Or, better yet, please provide us with a link to the actual paper.
Thanks,
Bob
Edmond Stuart said:
Hi Edmond,
I must admit I am not familiar with that one. I surely would have referenced it in my papers if I was aware of it. Was it published in the JAES? Can you please provide me with the exact reference? Or, better yet, please provide us with a link to the actual paper.
Thanks,
Bob
Uvidiu wrote:
In my opinion, the "hoax of HEC" (the non-feedforward version which this thread has investigated) is the ignorant claim that it is "error cancellation" or "reverse feedforward". Hawksford nearly made this mistake in his first paper (I assume by the standard of his paper that he was a bit of a novice at this time) and Bob made this mistake in his. Subsequent Hawksford papers avoided this suggestion and stuck to the NFB description.
Oh, I would be very interested to see that! Mainly because it would require a new form of mathmatics that has hitherto not existed on this planet.
In my opinion, the "hoax of HEC" (the non-feedforward version which this thread has investigated) is the ignorant claim that it is "error cancellation" or "reverse feedforward". Hawksford nearly made this mistake in his first paper (I assume by the standard of his paper that he was a bit of a novice at this time) and Bob made this mistake in his. Subsequent Hawksford papers avoided this suggestion and stuck to the NFB description.
Re: Re: hec != hoax ?
Hi Bob,
Here is the exact bibliography:
J. Vanderkooy and S.P. Liphsitz, "Feedforward Error Correction in Power Amplifiers", JAES, Vol. 28, pp. 2-16, Jan./Feb. 1980.
Regrettably, I've only a paper copy, but I suppose you can download it from the JAES website.
Cheers, Edmond.
Bob Cordell said:
Hi Bob,
Here is the exact bibliography:
J. Vanderkooy and S.P. Liphsitz, "Feedforward Error Correction in Power Amplifiers", JAES, Vol. 28, pp. 2-16, Jan./Feb. 1980.
Regrettably, I've only a paper copy, but I suppose you can download it from the JAES website.
Cheers, Edmond.
traderbam said:
Brian,
This is an interesting statement and I would appreciate if you could support it.
I have some time ago developed a generic HEC non-linear model based on first order non singular perturbation theory, which essentially avoids the fundamental issue that affects the NFB/PFB approach: the system is not linear, hence the superposition theorem does not strictly apply. Fortunately, the results show that a quasilinear approach is valid, so the NFB/PFB approach renders results close to a nonlinear analysis.
Anyways, I haven't encountered any major mathematical difficulties, so I was wondering what is your experience in such?
Set aside it is obviously impolite, in all truth, based on your contribution around, I don't find you qualified to make strong blank statements like "ignorant claim" on other people work.