But before you make the boards for this amp, you need to design 2 brand new amps, CKI and CKII first

[Edited by Dave_M on 11-16-2006 at 01:48 PM.]

---

[Quoting cetoole]

JFET inputs look nice, but will you be stocking 2SK170/2SJ74, or potentially the Linear equivalents when both are out, AMB?

---

---

[Quote]

Also, I would think it might make sense to add protection diodes to the 7815/7915 regs.

---

---

[Quote]

Finally, is there any specific reason you chose the triangular pin configuration for the TO-92 parts? I think the straight tape configuration would make it a bit less tight in many areas,

---

---

[Quote]

Why dont you feel adding a CRC or CLC filter would be useful here AMB?

---

---

[Quoting Dave_M]

haha Cavalli Kan III worked out well

---

[Edited by amb on 11-16-2006 at 04:06 PM.]

The usual caveats that sims are all perfect and that all devices of the same type are identical, the PS is perfect, etc, etc.

For 40mW continous power into 32R at 1KHz (50mA peak current) I get:

2kHz - .015%
3kHz - .001%
>=4kHz - negligible

Zo ~= 0R4

These results are for 60mA idle current in the output transistors.

These are pretty good sim results for an amp with no feedback.

I see that amb is already leaving me out of the name (clever as it is).

[Edited by runeight on 11-17-2006 at 12:24 AM.]

---

[Quoting runeight]

I see that amb is already leaving me out of the name (clever as it is).

---

Also, going back to input impedance for BJT version.. if you consider the input impedance of just the amp and don't include the pot, because from the amp's point of view, the pot is part of the source (I think that is the case). And say you had a 100k pot, then it would be possible for the output impedance of the source (including pot) to be higher than the input impedance of the amp? What I'm trying to say is, for the BJT version, it is better to use a low value pot like 10k? because in audio you usually want the output impedance to be lower than the input impedance of the next stage.

The introduction of the vol pot is actually a little more complicated than we've discussed and what it does depends on which point of view you take, the source or the sink (amp).

Most vol pots are drawn vertically with the wiper shown somewhere along the resistor. Let me call the part above the wiper connected to the source the "top" and the part below the wiper connected to ground the "bottom."

What we really have is the source's output inpedance in series with the top part of the vol pot. These two series resistors are then in series with a third resistor made from the bottom part of the vol pot in parallel with the input impedance of the sink (the amp). As in this diagram:

From the amp's point of view it has a Zi of the paralleled resistor. It then sees a source whose Zo is the sum of the actual Zo plus the top part of the vol pot.

From the source's point of view it sees a load resistor consisting of the top part of the vol pot in series with the paralleled resistor made from the bottom part in parallel with the actual Zi of the amp.

Thus, what each part (source or amp) sees depends on where the vol pot is set. This also means that the freq response of the front end of the amp can depend on where the vol pot is set since the top portion of the vol pot is in series with the capacitances of the front end. This may be noticeable if there is a large Miller capacitance, but most of the time is not.

All this to say that, yes, when the vol pot is set low, from the amp's point of view, there can be a much larger source impedance than the effective input impedance of the amp.

But, what really matters is what does the source see?? Because it's the loading effects on the source that we are generally concerned with. And the source sees at best the full resistance of the pot (volume set to zero) to at worst the parallel resistance of the pot and Zi (volume set to max).

If the Zi of the amp is 100k and the pot is 100k then the source sees at best 100k (min volume) and at worst 50k (max volume).

MHO is that it is generally better for the vol pot to be bigger to increase the value of the load that the source sees. But this must be balanced against the series resistance that the pot puts at the input and the attendant increase in noise and the affect on freq response. Otherwise we would just make the pots 10M to have a very high input impedance to the amp. So for this reason and others that probably go back to early tube designs, we tend to stick with 50-100k for the volume controls.

[Edited by runeight on 11-17-2006 at 05:26 PM.]

The few times I worked with that topology on power amps those
resistors were either 100 phms or 68 ohms. So, I'd welcome
some insight to the higher values.

Thanks

But I first saw CFB in the Amber Series 70 amps that used
National's LM391, thus have only been exposed to it there.

In order to achieve 30mA quiescent current, I had to reduce the Vbe multiplier resistor R18's value slightly. To provide a wider adjustment range, I conferred with runeight and we have decided to change R18's value to 2.4KΩ and the R19 trimpot to 1KΩ. Runeight will post an updated schematic reflecting this change.

The following is the frequency response and phase response graph. The response is flat within the audio band. In the subsonic region, a peak is observed, reaching a maximum of 1.3dB at 1.17Hz, below which the response falls sharply. This is due to DC servo action. At the ultrasonic end, the -3dB point is 1.96MHz. Since the input is 1V, the graph shows that the gain of the amp is almost exactly 9x (19dB).

The phase response is also flat within the audio band, +20° at 1Hz, -4.4° at 100KHz, and -43° at 1MHz.

Here is a harmonic distortion spectrum, with a 1KHz fundamental, and the input level adjusted to produce 1Vrms (1.4142V peak) at the output. Only the 2nd and 3rd harmonics are significant, at 0.0085% and 0.013%, respectively. None of the higher order harmonics exceed 0.0013%.

Below is the output impedance vs. frequency plot. The Y-axis is labeled as millivolts but is actually in milliohms. The output impedance is about 0.46Ω over the entire audio band, with a peak and rolloff in the subsonics due to the DC servo.

The 100KHz square wave response is shown below. It is characterized by fast slewing and gentle settling with no overshoot and ringing.

This amp performs exceedingly well for a no NFB amplifier. I think that you guys are going to like it.

However, there are those who hate NFB in all its guises and those who don't mind using lots of it (as in the b22 and SS power amps) and those who use some amount of it (as in many tube power amps). There are many, many opinions on the sound of NFB.

The main features of global NFB are that it theoretically 1) reduces harmonic distortion, 2) lowers the output impedance and 3) increases the bandwith.

These are all good things, but like many other things in the world of audio, you can't make bad amp into a good one by simply applying NFB. You can't make a bad amp into a good one by using boutique components.

Over the years, lots of designs, particularly SS amps, have simply used massive amounts of NFB to cover up a bad design and/or crappy components. Consequently NFB is often associated with really bad sounding SS amplifiers.

OTOH, a well designed amp, such as the b22, can be made better with NFB. How much NFB depends on the particulars of the design.

Even though NFB does those three great things listed above, it is not without its own inherent problems. There are three of these that I know about.

One is that after a while all amps with lots of NFB tend to sound the same. At least I have heard this from others who have listened to a wide variety of amps. And this is believable to me.

The second problem is that when you send the output of an amp back into its input you don't get perfect NFB because there is a phase lag as the signal traverses the amp to the output and back into the input. Because of the phase lag you don't get perfect NFB type cancellation of non-linearities introduced by the amplifier. Instead, because of the phase differences, you can get even worse intermodulation components that diminish or destroy the good effect of the NFB.

The third problem is that there is always a phase shift in the ultrasonic realm where, eventually, the phase shift goes to 180d. At this point the amp will oscillate and so the NFB loop has to be compensated to hammer this oscilation, generally with a cap in parallel with the feedback resistor and, in tube amps, a forced rolling off of the gain of the input stage below the critical frequency. I called this a problem. It is not really a problem, but just a design detail that has to be handled.

So, you can't make a bad amp better by just slinging NFB into the circuit. But you can apply NFB to a properly designed amp to get better results.

If you look at amb's b22 graphs, you'll see that the amp has almost no phase shift in the audio band and that it is very fast. This means that you can apply a fair amount of NFB without suffering from the bad effects of NFB. There is about 40db of NFB in the b22 and the actual performance measurements indicate that the amp is doing better because of this. This is because amb first designed a very good, low distortion, high speed, medium gain amplifier before applying global NFB.

Some would argue that even with this good behavior in the audio band, there is increasing phase shift in the ultrasonic region that will create intermodulation components. I don't know what effect this may or may not have.

I guess this was a long answer to the question. The short answer is that NFB is neither good or bad, it's only the particular application of it makes it one or the other or somewhere in between.

I am pretty sure of one thing. Almost every commercial power amp that you've ever heard has NFB and you have, therefore, heard what is "sounds" like.

Of course, no one would ever use NFB if the active components in the amp did not generate inherent distortion because there would be no reason TO use it.

Most, if not all of the opamp-based amplifiers that are popular use some amount of NFB. In general this happens even when a discrete buffer is used because the buffer is inserted into the global feedback loop of the opamp so that the opamp can zero the offset. We don't even think about using NFB with opamps because it's the only way to control their gain, but many would consider opamps to be poor audio amps because of this NFB.

With regard to the CKIII and no NFB. It can be a real challenge to design a good amplifier with no NFB and low THD and low Zo. The CKIII has these qualities. Its THD is quite low and its Zo is not bad. Since we don't want to use NFB unless we MUST use it, the CKIII falls into a category of an excellent amp that does not use NFB.

There are members of this forum who are far more knowledgable than I am about the details of NFB. Hopefully they will also take time to answer and correct any errors in my explanation.

[Edited by runeight on 11-25-2006 at 05:12 PM.]

Both local and global negative feedback are utilized in the β22. Local feedback is used to control the gain of each individual stage of the amp (so that a moderate target open loop gain is achieved), while global feedback is applied to set the overall closed loop gain (and to achieve the performance enhancements that runeight has mentioned). This is quite different than the approach taken in some designs where the open loop gain is made as high as possible, so that a large amount of global feedback is applied (the amount of global feedback is the difference between the open and closed loop gains). Pretty much any opamp-based design will fall into the latter category, where the opamp's open loop gain could be as high as 120dB or more, forcing a large amount of global feedback.

[Edited by amb on 11-25-2006 at 07:08 PM.]

So now you guys have two new amazing amps to choose from. Each from a completely different design perspective.

The b22 from the perspective of an extremely well done basic amp expressly designed to have substantial global NFB (with a new, fast PS to go with it).

The CKIII from the perspective of an amp expressly designed to have no global NFB (with PS all on a single board).

Or you can build both of them and compare.

[Edited by runeight on 11-25-2006 at 11:12 PM.]

Bottom line is - now I want to build both