Zero Feedback Impedance Amplifiers

[Susan_Parker]

Hi Sheldon,

Thank you.

Originally posted by Sheldon
Even more options. This thread is now over 100 pages. It's nice to see this clever design get some attention.

Thanks to everyone for their interest and taking the time to post questions, observations, comments and results from their own versions.

Much appreciated.

Best wishes,
Susan.

[suiraMB]

Originally posted by Sheldon
Even more options. This thread is now over 100 pages. It's nice to see this clever design get some attention.

Definitely. It is a very interesting approach, pared down to the basics. The next step being to optimize it, and determine which enhancements add to its performance, and which detract from it.

Relieved. I was worried I had missed something. Wouldn't have been the first time.

No, this time I missed something. Yet again, not for the first time

Ok, you're gonna have to slow down. Too many ideas now. I have single ended outputs, but if I had balanced, I'd try the coax trick.

Yeah. A universal problem. My TODO-list always grows faster than I can process it, regardless of how quickly I work

Originally posted by Susan-Parker
Watch this version carefully. The separated windings will enable small mismatches between the mosfets to take off and the whole thing will oscillate.

These are JFETs, not MOSFETs. If I read Mr. Curl's comments on another forum correctly, JFET followers are significantly less prone to oscillation than MOSFET followers.

If oscillation turns out to be a problem, gate resistors, ferrites and gate-source capacitors are all viable tools to eliminate it.

This can be cured by putting low value resistors across the paralleled sources, something like 1 or 2 ohms, on each side.

Won't 1-2 ohms significantly affect the idle current?

The voice coil will overheat unless you use very low biasing and the inductance will not be enough for full range.

I was actually thinking about lowering the voice coil DC resistance (e.g. by paralelling sections) to counteract this problem. 5A at 0,2, for example, would only be 5W on average. I seem to recall this design did well also with low bias currents, so that part shouldn't be a problem.

The inductance is more difficult, I'd imagine. What kind of inductance would one need to get full range operation?

What would work is turning the speaker motor around and using a bifilar coil instead of a permanent magnet and then magnetizing the voice coil with a low(ish) level of DC.

Thanks, that's a brilliant idea.

Come to think of it, wouldn't it be possible to drive the voice coil with the same signal as the "magnet", or would this be yet another case where the inductance cannot be made to reach the desired level?

[Sheldon]

Originally posted by suiraMB
Since the circuit operates in class AB, you should be able to get 25W or so from a single pair with regular heatsinking, or 50W from a single pair with the modified heatsinking. This at high bias.

I'm pretty new to all this, but this issue seems central to the difference between this design and a regular pp amp. And not understanding this has made me want to pursue it even more (what issues that speaks to, I'll leave to the observer). You cite class AB operation. Susan cites class A operation. So what's going on here? As the light in my noggin slowly goes from black to dull red (or is that light at all), it occurs to me that the crux of the matter is the fact that the rail voltage actually swings with the drive voltage. Unlike the usual situation in which one rail is fixed positive, the other negative. So it seems like class AB amps in that the power output can exceed the idle dissipation, but like class A in that significant current flows in both devices throughout the ouput cycle. Or?

Sheldon

[suiraMB]

Originally posted by Sheldon
You cite class AB operation. Susan cites class A operation. So what's going on here?

Sorry. I guess it's a case of semantics.

As I understand the design, both transistors are in conduction at all times, which fits some definitions of class A, as neither transistor will cut off at any point, entirely eliminating crossover distortion.

However, I believe the conduction angle (at least in terms of linear operation) is less than 360 degrees, which excludes the class A designation in its strictest sense.

Please correct me if I'm wrong, and also note that I do not mean to imply inferior sonics or design by my statement.

As the light in my noggin slowly goes from black to dull red (or is that light at all), it occurs to me that the crux of the matter is the fact that the rail voltage actually swings with the drive voltage. Unlike the usual situation in which one rail is fixed positive, the other negative. So it seems like class AB amps in that the power output can exceed the idle dissipation, but like class A in that significant current flows in both devices throughout the ouput cycle. Or?

This has been my understanding of the circuit as well.

In the strictest sense, the circuit is not class A, as its conduction angle is not 360 degrees at maximum output.

However, herein lies one of the more elegant aspects of this design, in my opinion:

Unlike a conventional class AB amplifier, Susan's design has a continously variable conduction angle, in that the conduction angle is proportional to the signal, and the conduction angle always exceeds 180 degrees, meaning you get no cross-over distortion.

In fact, I believe you can select the minimum conduction angle by adjusting the bias, though I haven't tried to derive a formula for it.

[suiraMB]

Regarding additional series resistance to lower the bias on the JFETs, I'd suggest putting a single, shared resistor between the primaries and the rail, not one per leg. I believe Sheldon was the one thinking about this.

[Susan_Parker]

Originally posted by suiraMB

These are JFETs, not MOSFETs. If I read Mr. Curl's comments on another forum correctly, JFET followers are significantly less prone to oscillation than MOSFET followers.

If oscillation turns out to be a problem, gate resistors, ferrites and gate-source capacitors are all viable tools to eliminate it.

Won't 1-2 ohms significantly affect the idle current?

Sorry, I haven't explained myself clearly.

Using the method of paralleling devices but having each device on its own winding is an attractive method of ensuring that each part doesn't current hog as each one has it's own separate load to look after.

However there is a problem with oscillation as each device tries to optimally do its follower 100% degenerative feedback thing. Note that this is completely separate to the oscillation some devices are prone to from having too low a value of gate resistor.

The resistors are to couple adjacent device sources i.e. between the devices, they are not in series with the windings or across the transformer to ground.

Therefor there will be no current flow (of any significance) as all the sources are at the same voltage point (within a few mV anyway).

However what this does is provide a low resistance path between the sources and ensures that they all stay in step with each other since the windings above DC actually have significant impedance.

So in the simplest case of having two pairs of devices, one pair per side, each on its own winding one would use two resistors. One is across the first pair's sources, the other is across the second pair's sources.

(Will draw up a schematic at the weekend.)

Hope this is a little clearer.

Best wishes,
Susan.

[suiraMB]

Originally posted by Susan-Parker
Sorry, I haven't explained myself clearly.

Actually, you've been pretty lucid. I'm sure I've been a lot less clear about things.

Using the method of paralleling devices but having each device on its own winding is an attractive method of ensuring that each part doesn't current hog as each one has it's own separate load to look after.

Actually, current hogging isn't as much of a problem as it would be with bipolars, or even MOS fets. However, the bias voltage can get a bit high when you have several devices on the same winding, due to the autobias arrangement.

Plus, seperate windings should give better high-frequency performance, according to some pages I read. I could be wrong here, though.

However there is a problem with oscillation as each device tries to optimally do its follower 100% degenerative feedback thing. Note that this is completely separate to the oscillation some devices are prone to from having too low a value of gate resistor.

Ah. Yes. I misunderstood.

The resistors are to couple adjacent device sources i.e. between the devices, they are not in series with the windings or across the transformer to ground.

That makes a lot more sense, yes.

Hope this is a little clearer.

Yes. Perfectly clear.

Although I'm no good with transformers, so if you're feeling awfully bored, you could explain the details of the hows and whys of this kind of resonance.

[Sheldon]

Originally posted by suiraMB
Regarding additional series resistance to lower the bias on the JFETs, I'd suggest putting a single, shared resistor between the primaries and the rail, not one per leg. I believe Sheldon was the one thinking about this.

I was, thanks. I think Susan recommended the same, so that's 2 to 0 in favor. Sounds like a mandate to me.

Sheldon

[Sheldon]

Originally posted by suiraMB
In order to cool them properly, I'd recommend the suggestion form someone in another thread, namely putting indium solder on a heatsink, then heating that until the indium solder melts, then using a vice to press the devices against the heatsink until the indium solder cools again. This will give very good heat transfer characteristics.

In chasing up materials to do this, I come across two possibilities with wildly different pricing. Indium Corportion has a variety of solders for electronics that they list on their web store. For a 1 meter, .75mm piece, the prices are from $180-350. Ouch.

The other type is a brand used by mostly by hobbyists and jewelry makers. The brand is Tix. It seems to sell for about $15 for 20 8cm pieces. The only information I could find seems to indicate that it comprises indium, lead, and tin. It has a melting point of 135 degrees C. Unless someone knows some reason why this wouldn't work, I'll give this type a try.

Sheldon

[suiraMB]

Originally posted by Sheldon
In chasing up materials to do this, I come across two possibilities with wildly different pricing. Indium Corportion has a variety of solders for electronics that they list on their web store. For a 1 meter, .75mm piece, the prices are from $180-350. Ouch.

Indeed it is somewhat expensive. You don't need very much per device, though.

The other type is a brand used by mostly by hobbyists and jewelry makers. The brand is Tix. It seems to sell for about $15 for 20 8cm pieces. The only information I could find seems to indicate that it comprises indium, lead, and tin. It has a melting point of 135 degrees C. Unless someone knows some reason why this wouldn't work, I'll give this type a try.

Well, it's cutting it fine, but it should work, provided you have a thermometer that is reasonably accurate in that range. The Lovoltechs can handle 150 degrees storage temperature, which leaves you a headroom of 15 degrees over the melting point of the solder. Remember that the device will be at this temperature for some time, since you'll need to heat the entire heatsink in order for this to work. You'll also want to start low on the bias, I think, since you don't know the thermal conductivity of this solder.

Since this topology generally employs a sub-360-degree conduction angle, you may want to incorporate some advice from another forum member (again, I forget who). Basically, solder the devices to a large copper bar, then solder that to the heatsink. While this may slightly reduce constant dissipation, it will improve transient thermal performance, as the mass (and higher heat conductivity) of the copper bar will absorb larger heat transients.

Whether this trick works for you is, I guess, dependent on the bias setting you choose. Of course, a 12dB fan would also help a lot, possibly enabling you to go as high as 30W idle dissipation (operating temperature of 80 degrees).