hec != hoax ?
TTL is old school.
BTW, I had about the same discussion on this topic with Mike KS, 6 month ago. So if 'quite passionate' is synonymous with bored, then I am quite passionated indeed.
TTL is old school.
BTW, I had about the same discussion on this topic with Mike KS, 6 month ago. So if 'quite passionate' is synonymous with bored, then I am quite passionated indeed.
traderbam said:
It's OK. I have figured it out, there is no qualitative advantage in HEC wrt internal amp overdrive. What I said was that a classical fb amp, when clipping, can be overdriven because the feedback increases the effective input signal at the amp input terminal, and the amp of course always works with its high open loop gain.
My thinking was that if I would built a HEC amp with the same OL gain as the CL gain (not an output stage but a Vas stage for example), the overdrive at clipping would be much less than with the classical case, because the amp ol gain would be much less than with the classical case. But what happens is that the pos feedback loop in the HEC amp does continue to increase the effective amp input signal. So, even if the ol gain is much less, there is still internal overdrive because the effective input signal is much greater.
So, both in classical nfb and in HEC there is a mechanism that causes amp overdrive at clipping: either because of high ol gain, or because of pos fb. But I haven't yet looked at how the figures compare quantitatively.
Re: Re: Re: Re: Re: Re: Re: Re: Re: hec != hoax ?
Yes, you are right. But the Early Effect in the BJT cascodes is microscopic in comparison, since the signal swing is only that of the input signal, and is very small in comparison to Vcb of the cascodes.
It is also true that the first time I used this technique was with the JFET input differential pair. I always prefer JFET input pairs in my amps, but often when I get lazy I use BJT's in the front end for simulations 🙂.
I have a circuit that is functionally similar to the Hawksford cascode that I could use in the BJT cascodes if I wanted to. Maybe I'll try it and see if it makes any difference.
Cheers,
Bob
Edmond Stuart said:
Yes, you are right. But the Early Effect in the BJT cascodes is microscopic in comparison, since the signal swing is only that of the input signal, and is very small in comparison to Vcb of the cascodes.
It is also true that the first time I used this technique was with the JFET input differential pair. I always prefer JFET input pairs in my amps, but often when I get lazy I use BJT's in the front end for simulations 🙂.
I have a circuit that is functionally similar to the Hawksford cascode that I could use in the BJT cascodes if I wanted to. Maybe I'll try it and see if it makes any difference.
Cheers,
Bob
traderbam said:
Yes Jan.
Brian, you actually agree with me?? Where did I go wrong??😉
Jan Didden
On the contrary, where did I go wrong?
🙂What is it? It looks like a piece of furniture. Very pretty. Perfect for a cold winters evening.
traderbam said:
On the contrary, where did I go wrong?
What is it? It looks like a piece of furniture. Very pretty. Perfect for a cold winters evening.
Couple of 100 watts / channel, class A. Uses transmitter tubes. Supply voltage around 5kV. It's an audio power amp. More pictures at www.triodefestival.net
Jan Didden
janneman said:
I remain convinced the potentialy lethal consequences of working in and around these things is part of the appeal.
🙂
scott wurcer said:
Absolutely. As well as the nice white light from the thoriated tungsten heaters, and the soft blue shine of the mercury rectifiers. We're still carried away by beads and mirrors. 😉
Jan Didden
hec != hoax ?
It seems that some people have not clearly understood how my NFB output stages works. To let them gain more insight, I've put here the transfer functions.
There are two versions (see fig. 1 and 2), which differ only in the way of compensation, but perform identically.
It's also possible to give both versions some gain by adding R4 and R14 (see also R47 and R48 in post 2965)
Let:
a = gain of the source follower stage (Q9-10), normally ~0.96
C2 = effective input capacitance of Q9
t1 = R1 * C1
t2 = R2 * C2
t3 = R3 * C1
then the transfer function for figure fig. 1 without R4 and R14 is:
Vout = Vin*a*(1 + t1*s + t3*s ) / ( a + a *t1*s + t3*s + t2*t3*s*s)
and with R4 and R14:
Vout = Vin*a*( 1 + R3/R4 + t1*s + t3*s + R3/R4*t1*s ) / ( a + a*t1*s + t3*s + t2*t3*s*s)
The transfer function for figure fig. 2 without R4 and R14 is:
Vout = Vin*a*(1 + t1*s + 2*t3*s ) / ( a + a *t1*s + 2*t3*s + 2*t2*t3*s*s )
With R4 and R14:
Vout = Vin*a*( 1 + R3/R4 + t1*s + 2*t3*s + R3/R4*t1*s ) / ( a + a*t1*s + 2*t3*s + 2*t2*t3*s*s )
Clearly these functions represent a second order system, which also explains why under certain circumstances there exists a small gain bubble of 0.5...1dB at around 3MHz. But if a < ~0.85 or, more realistic, if the close loop gain is > ~1.2 (depending on the actual component values), this bubble has completely gone as well the accompanying overshoot. It should be noticed that the gain bubble in a complete amp is masked by other roll off mechanisms.
For best performance, set t1 = t2. This compensates for the pole created by the source resistors by a corresponding zero of the Miller compensation.
Cheers, Edmond.
It seems that some people have not clearly understood how my NFB output stages works. To let them gain more insight, I've put here the transfer functions.
There are two versions (see fig. 1 and 2), which differ only in the way of compensation, but perform identically.
It's also possible to give both versions some gain by adding R4 and R14 (see also R47 and R48 in post 2965)
Let:
a = gain of the source follower stage (Q9-10), normally ~0.96
C2 = effective input capacitance of Q9
t1 = R1 * C1
t2 = R2 * C2
t3 = R3 * C1
then the transfer function for figure fig. 1 without R4 and R14 is:
Vout = Vin*a*(1 + t1*s + t3*s ) / ( a + a *t1*s + t3*s + t2*t3*s*s)
and with R4 and R14:
Vout = Vin*a*( 1 + R3/R4 + t1*s + t3*s + R3/R4*t1*s ) / ( a + a*t1*s + t3*s + t2*t3*s*s)
The transfer function for figure fig. 2 without R4 and R14 is:
Vout = Vin*a*(1 + t1*s + 2*t3*s ) / ( a + a *t1*s + 2*t3*s + 2*t2*t3*s*s )
With R4 and R14:
Vout = Vin*a*( 1 + R3/R4 + t1*s + 2*t3*s + R3/R4*t1*s ) / ( a + a*t1*s + 2*t3*s + 2*t2*t3*s*s )
Clearly these functions represent a second order system, which also explains why under certain circumstances there exists a small gain bubble of 0.5...1dB at around 3MHz. But if a < ~0.85 or, more realistic, if the close loop gain is > ~1.2 (depending on the actual component values), this bubble has completely gone as well the accompanying overshoot. It should be noticed that the gain bubble in a complete amp is masked by other roll off mechanisms.
For best performance, set t1 = t2. This compensates for the pole created by the source resistors by a corresponding zero of the Miller compensation.
Cheers, Edmond.
Attachments
hec != obsolete kludge?
It’s getting quiet here, so it is time to stir the pot again with further reflections on HEC.
For a long time I considered HEC (and my own NFB variant too) as a hoax and an obsolete kludge. Meanwhile, I’ve changed my mind, but first let me explain why I think it’s a kludge and next why I think it is still a valuable an useful technique.
Looking at fig. 1 we have a typical amp, consisting of an input stage, VAS plus Miller compensation, EC unit and an output stage. The EC stage is supposed to correct the distortion of the output stage in such a way that the voltage at point Q, the VAS output, is exactly equal to the output voltage, V(out), right?
In other words, the right leg of the Miller cap (C1) sees the same voltage as V(out). So, why not tie C1 directly to the output (as Cherry did)? Shouldn’t such an arrangement have the same performance, because the only difference is that the VAS output is exposed to Vout plus a tiny error voltage (Ve), in stead of Vout alone.
If we ignore stability issues for the moment, and assume a sufficiently high gain of the VAS or a high enough output impedance, this error voltage has a minuscule effect on the VAS input currents, in particular I1, as this one is dominated by I2. So, theoretically, one might expect the same performance.
In real life the VAS has not only a limited current gain, but more importantly, also a limited voltage gain, mainly caused by the emitter degeneration resistor Re. Furthermore, as recommended by Cherry, the VAS output is loaded with additional ‘stopper’ caps between the base and the collector of the drivers. These are the main contributors to a less than perfect performance. But to what extent?
Let the impedance seen at the VAS output be Zo, at 20kHz, mainly determined by C3, as the output impedance of a decent VAS and input impedance of an OPS are considerably higher. Let’s further suppose that the current gain of the VAS is high enough so that we can ignore I3 and the current delivered by the input stage (I1) equals I2, which is Vout*2*pi*f*C1, btw.
Now, the error voltage at point Q create an error current Ve/Zo, which also appears as an error voltage across Re with an amplitude of Ve*Re/Zo, which on his turn is subtracted from voltage across C1, so I2 = (Vout-Ve*Re/Zo)*2*pi*f*C1. As a result, not all distortion has disappeared, but is reduced by a factor of Re/Zo.
Here are some numbers from an amp with vertical MOSFETs and an ideal VAS (current gain = 10000) and an ideal input stage (gm = 20mS). The conventional Miller compensated amp (fig. 1 without EC) gives a THD20 of 173ppm, with a 3rd harmonic as main component.
Fig.2 gives a THD20 of 0.55ppm, with 5th harmonic as main components and an overall THD reduction of 314. Is this reduction in accordance with the ratio Re/Zo as predicted? That’s a nasty job to calculate exactly, but let’s set the mean frequency of the harmonics between the 3rd and 5th, i.e 80kHz, then Zo is roughly equal to the impedance of C3 = 30kOhm, so Zo/Re = 300. Hmm… not bad compared to 314, and, compared to HEC, ten times better.
These figures are of course to good to be true. Mainly because this compensation scheme is highly prone to oscillations. However, the same applies to HEC, unless we use a kind of compensation, i.e. to let the correction roll off above a safe frequency, in practice at about 3MHz.
We can apply the same kind of roll off to Cherry’s scheme, called TMC, see fig. 3. Opposed to fig.2, this arrangement is stable, also in terms of driving a capacitive load, in the order of 20nF. According to my simulations, it is just as stable as HEC, or, compared to a conventional compensated amp, the stability is reduced to the same amount as with HEC.
In this example the transition frequency is set to 2.5MHz, giving an THD20 of 10ppm, but if C2 is connected to the emitter of the driver, the distortion drops to 5.3ppm, close to the performance of Bob’s HEC amp.
If this is the end of the story, then certainly one might conclude that HEC, and of course its my NFB nephew too, is a kludge, as about the same performance can be achieved by just one additional R and C.
BUT this isn’t the end of the story because we can apply both of these techniques at the same time to get an even lower distortion or if sub ppm figures are not the main target, we can lower the transition frequencies of HEC and TMC to get more stability.
The astute reader will notice that in the past I had a different opinion about combining these techniques, but that was based on the assumption that one cannot correct an output stage which is already corrected. While theoretically true, in practice not all distortion is removed by either HEC or TMC (due to the bandwidth limitation), so what is left over as distortion from HEC can be further reduced by TMC.
Cheers, Edmond.
It’s getting quiet here, so it is time to stir the pot again with further reflections on HEC.
For a long time I considered HEC (and my own NFB variant too) as a hoax and an obsolete kludge. Meanwhile, I’ve changed my mind, but first let me explain why I think it’s a kludge and next why I think it is still a valuable an useful technique.
Looking at fig. 1 we have a typical amp, consisting of an input stage, VAS plus Miller compensation, EC unit and an output stage. The EC stage is supposed to correct the distortion of the output stage in such a way that the voltage at point Q, the VAS output, is exactly equal to the output voltage, V(out), right?
In other words, the right leg of the Miller cap (C1) sees the same voltage as V(out). So, why not tie C1 directly to the output (as Cherry did)? Shouldn’t such an arrangement have the same performance, because the only difference is that the VAS output is exposed to Vout plus a tiny error voltage (Ve), in stead of Vout alone.
If we ignore stability issues for the moment, and assume a sufficiently high gain of the VAS or a high enough output impedance, this error voltage has a minuscule effect on the VAS input currents, in particular I1, as this one is dominated by I2. So, theoretically, one might expect the same performance.
In real life the VAS has not only a limited current gain, but more importantly, also a limited voltage gain, mainly caused by the emitter degeneration resistor Re. Furthermore, as recommended by Cherry, the VAS output is loaded with additional ‘stopper’ caps between the base and the collector of the drivers. These are the main contributors to a less than perfect performance. But to what extent?
Let the impedance seen at the VAS output be Zo, at 20kHz, mainly determined by C3, as the output impedance of a decent VAS and input impedance of an OPS are considerably higher. Let’s further suppose that the current gain of the VAS is high enough so that we can ignore I3 and the current delivered by the input stage (I1) equals I2, which is Vout*2*pi*f*C1, btw.
Now, the error voltage at point Q create an error current Ve/Zo, which also appears as an error voltage across Re with an amplitude of Ve*Re/Zo, which on his turn is subtracted from voltage across C1, so I2 = (Vout-Ve*Re/Zo)*2*pi*f*C1. As a result, not all distortion has disappeared, but is reduced by a factor of Re/Zo.
Here are some numbers from an amp with vertical MOSFETs and an ideal VAS (current gain = 10000) and an ideal input stage (gm = 20mS). The conventional Miller compensated amp (fig. 1 without EC) gives a THD20 of 173ppm, with a 3rd harmonic as main component.
Fig.2 gives a THD20 of 0.55ppm, with 5th harmonic as main components and an overall THD reduction of 314. Is this reduction in accordance with the ratio Re/Zo as predicted? That’s a nasty job to calculate exactly, but let’s set the mean frequency of the harmonics between the 3rd and 5th, i.e 80kHz, then Zo is roughly equal to the impedance of C3 = 30kOhm, so Zo/Re = 300. Hmm… not bad compared to 314, and, compared to HEC, ten times better.
These figures are of course to good to be true. Mainly because this compensation scheme is highly prone to oscillations. However, the same applies to HEC, unless we use a kind of compensation, i.e. to let the correction roll off above a safe frequency, in practice at about 3MHz.
We can apply the same kind of roll off to Cherry’s scheme, called TMC, see fig. 3. Opposed to fig.2, this arrangement is stable, also in terms of driving a capacitive load, in the order of 20nF. According to my simulations, it is just as stable as HEC, or, compared to a conventional compensated amp, the stability is reduced to the same amount as with HEC.
In this example the transition frequency is set to 2.5MHz, giving an THD20 of 10ppm, but if C2 is connected to the emitter of the driver, the distortion drops to 5.3ppm, close to the performance of Bob’s HEC amp.
If this is the end of the story, then certainly one might conclude that HEC, and of course its my NFB nephew too, is a kludge, as about the same performance can be achieved by just one additional R and C.
BUT this isn’t the end of the story because we can apply both of these techniques at the same time to get an even lower distortion or if sub ppm figures are not the main target, we can lower the transition frequencies of HEC and TMC to get more stability.
The astute reader will notice that in the past I had a different opinion about combining these techniques, but that was based on the assumption that one cannot correct an output stage which is already corrected. While theoretically true, in practice not all distortion is removed by either HEC or TMC (due to the bandwidth limitation), so what is left over as distortion from HEC can be further reduced by TMC.
Cheers, Edmond.
Attachments
To summarize, Edmond,
You started out wanting to show that deriving the VAS feedback directly from the o/p (rather than prior to the o/s stage in a conventional VAS) will produce as good as or better distortion performance than a conventional VAS combined with localized NFB around the o/p stage. This was what you meant by "HEC hoax" - that HEC is an inferior topology.
Now you have concluded that using a nested FB arrangement with localized NFB around the o/s stage contained within the VAS FB loop is even better.
Is that right?
You started out wanting to show that deriving the VAS feedback directly from the o/p (rather than prior to the o/s stage in a conventional VAS) will produce as good as or better distortion performance than a conventional VAS combined with localized NFB around the o/p stage. This was what you meant by "HEC hoax" - that HEC is an inferior topology.
Now you have concluded that using a nested FB arrangement with localized NFB around the o/s stage contained within the VAS FB loop is even better.
Is that right?
traderbam said:
Hi Brian,
Yes, except that (in a previous post) I meant by "HEC hoax" something different, i.e. HEC is equivalent to NFB.
This time, I was wondering in how far HEC and the NFB variant is a kludge as both can be replaced by TMC.
Indeed, I have concluded that using a nested FB arrangement with localized NFB (or HEC) around the o/s stage contained within the VAS FB loop is even better.
That's right.
Cheers, Edmond.
Edmond,
Ok, I concur with your original hoax opinion as you know. Looking at your nested FB idea, there may be a generalization to it that might or might not be useful. Your latest idea has the the o/p node as the NFB sense point for the localized o/s NFB, and the VAS stage NFB and the global NFB. So, might you be starting to test a hypothesis that all NFB should be derived from the o/p node? Just something for academic interest.
Brian
Ok, I concur with your original hoax opinion as you know. Looking at your nested FB idea, there may be a generalization to it that might or might not be useful. Your latest idea has the the o/p node as the NFB sense point for the localized o/s NFB, and the VAS stage NFB and the global NFB. So, might you be starting to test a hypothesis that all NFB should be derived from the o/p node? Just something for academic interest.
Brian
traderbam said:
Hi Brian,
That's right, that is, as long as AF is concerned. At HF it is a different story. Looking at TMC for example, above the transition frequency, the compensation reverts to the traditional Miller scheme.
The same principle applies to my NFB-OPS. At AF, the feedback is directly taken from the output, while at HF it is more or less overruled by feedback from the emitters of the drivers.
Cheers, Edmond.
EC amplifier designed, built, and working
See my latest post under "Power Amp Under Development" for photos and more details.
The amps utilize a single point EC topology in which the VAS stage drives a low power follower to provide a reference voltage that is then compared against the amp's output. The use of a 100+ MHz opamp as the EC amplifier stage guarantees sufficient BW to accurately compute the EC voltage. A simple resistive divider serves as the summing junction to feed the output stage. The output stage consists of a pair of VBE multipliers driven at their center point. The offset voltages formed thereby set the bias voltage and idle current for the succeeding pseudo darlington consisting of lateral FET devices.
With +/- 70V rails and a 650 va transformer, I have been able to obtain in excess of 400W/ch into a 4 ohm load.
I plan to write up the design in detail and publish it in the "Audio Express".
See my latest post under "Power Amp Under Development" for photos and more details.
The amps utilize a single point EC topology in which the VAS stage drives a low power follower to provide a reference voltage that is then compared against the amp's output. The use of a 100+ MHz opamp as the EC amplifier stage guarantees sufficient BW to accurately compute the EC voltage. A simple resistive divider serves as the summing junction to feed the output stage. The output stage consists of a pair of VBE multipliers driven at their center point. The offset voltages formed thereby set the bias voltage and idle current for the succeeding pseudo darlington consisting of lateral FET devices.
With +/- 70V rails and a 650 va transformer, I have been able to obtain in excess of 400W/ch into a 4 ohm load.
I plan to write up the design in detail and publish it in the "Audio Express".
Re: EC amplifier designed, built, and working
Sounds interesting! I did not see a schematic over on the other thread. Can you post one here, at least of the EC output stage, so we can discuss it?
Thanks,
Bob
analog_guy said:
Sounds interesting! I did not see a schematic over on the other thread. Can you post one here, at least of the EC output stage, so we can discuss it?
Thanks,
Bob
I think this is the simplied schematic:
http://www.diyaudio.com/forums/showthread.php?postid=1230139#post1230139
The discussion start a few pages earlier in this monster thread😀
Klaus
http://www.diyaudio.com/forums/showthread.php?postid=1230139#post1230139
The discussion start a few pages earlier in this monster thread😀
Klaus