Subject is a CFB amp, two pairs of output devices. I've built CFB amps before and subdued the oscillations and made them work. This one does something new. I've always been careful to be sure the heatsink is grounded: A- for saftey and B- to avoid it being an antenna.
On this amp, if I leave the heatsinks floating I see no oscillation on my scope. If I connect them to ground I see signs of oscillation on the scope and the output transistors start heating up even with no input or load. I noticed that when I touch a heatsink the oscillations go away (I'm on a insulated pad and have one hand in my pocket - I'm nut that nuts!) It's hard to figure out the frequency of the oscillation as it seems to be across a band not too far above the audible range. The zobel resistor stays completely cool which indicates this is not a truly HF oscilation.
With only one set of output devices (and the rails lowered with a variac) there was no sign of this - I couldn't even induce oscillations by putting caps across the dummy load. This of course suggests I am dealing with oscillations stemming from the CFB topology.
Any ideas would be appreciated.
On this amp, if I leave the heatsinks floating I see no oscillation on my scope. If I connect them to ground I see signs of oscillation on the scope and the output transistors start heating up even with no input or load. I noticed that when I touch a heatsink the oscillations go away (I'm on a insulated pad and have one hand in my pocket - I'm nut that nuts!) It's hard to figure out the frequency of the oscillation as it seems to be across a band not too far above the audible range. The zobel resistor stays completely cool which indicates this is not a truly HF oscilation.
With only one set of output devices (and the rails lowered with a variac) there was no sign of this - I couldn't even induce oscillations by putting caps across the dummy load. This of course suggests I am dealing with oscillations stemming from the CFB topology.
Any ideas would be appreciated.
My guess is that there is a low capacitance from unearthed heatsink to ground and a somewhat higher capacitance from collectors (i.e. signal) to heatsink. With the heatsink floating there is not enough capacitance for much effect but when it is grounded the larger capacitance comes into play.
Have a careful look just exactly where you are earthing it to. Is it injecting currents into a low signal level earth area?
Have a careful look just exactly where you are earthing it to. Is it injecting currents into a low signal level earth area?
I agree with Circlotron. Assuming the collectors are the signal output, you may try to gnd the heatsink and connect the two collectors by a smallish (10-20pF ) cap. If the collectors are not the signal outputs but the supply connections, try to put an extra supply decoupling (.1uF) from the collectors to the central gnd, using as short leads as possible. You then effectively bypass the supply wiring inductance.
Jan Didden
Jan Didden
Sam9,
I dont know much about you, your posts and what you have built in the past since I'm quite new to this board, but I have some questions before I would try on giving any clue about your problem.
Could you tell if you are using MOSFET's or BJT's in the output stage?
Where does this ocsillation occur, output or upper/lower side of the transistor?
If upper/lower side of the transistor, does it occur att both sides?
What kind of resistor (if any) do you use between the transistor and output?
How does it look around the VAS?
Schematic would be most appreciated, could you post anything or draw up something simple?
I dont know much about you, your posts and what you have built in the past since I'm quite new to this board, but I have some questions before I would try on giving any clue about your problem.
Could you tell if you are using MOSFET's or BJT's in the output stage?
Where does this ocsillation occur, output or upper/lower side of the transistor?
If upper/lower side of the transistor, does it occur att both sides?
What kind of resistor (if any) do you use between the transistor and output?
How does it look around the VAS?
Schematic would be most appreciated, could you post anything or draw up something simple?
Thanks for the feedback. When I woke this morning after sleeping on it, it ocurred to me that the structure, heatsink-silpad-transitorback, consitutes a sort of crude capacitor. This dovetails with the advice from Circlotron. I'm going to check over the grounding again and make sure it is solid and low impedance. This at an expiremental stage so its not in it ultimate permanent enclosure; instead it's set up more for quick removal/replacement so I may have gotten a bit casual about grounding.
Yes, there is a cap in parralel with the feed back resistor. That has been both helpful and rather touchy on past ocassions. Another item that came to mind when I woke up. Also reducing the gain had ocurred to me. I think these are more stop gap measures though since the amps was rock stable with only one output pair installed. I still suspect something specific to the output section.
On the the other hand putting a small cap between the output collectors is a new idea. Since there are two parrallel output pairs, do you suggest a cap across each pair or just one across the two output resitors (.22R)? The later is physically easier to arrange.
My schematic capture software doesn't export graphic images that would be of suffient resolution to be helpful. However, there topology is not innovative. IS a single long tailed pair, CC biased by two diodes which also bias the CC in the VAS, VAS is beta enhanced configuration. and output is as described. Its not an exact copy of a specific textbook or cookbook amp but there is nothing in it you wouldn't have seen before. Output devices are 2sa1943/2sc5200, drivers 2sd669/2sb649 and TO-92s are 1n5551/1n5401.
Yes, there is a cap in parralel with the feed back resistor. That has been both helpful and rather touchy on past ocassions. Another item that came to mind when I woke up. Also reducing the gain had ocurred to me. I think these are more stop gap measures though since the amps was rock stable with only one output pair installed. I still suspect something specific to the output section.
On the the other hand putting a small cap between the output collectors is a new idea. Since there are two parrallel output pairs, do you suggest a cap across each pair or just one across the two output resitors (.22R)? The later is physically easier to arrange.
My schematic capture software doesn't export graphic images that would be of suffient resolution to be helpful. However, there topology is not innovative. IS a single long tailed pair, CC biased by two diodes which also bias the CC in the VAS, VAS is beta enhanced configuration. and output is as described. Its not an exact copy of a specific textbook or cookbook amp but there is nothing in it you wouldn't have seen before. Output devices are 2sa1943/2sc5200, drivers 2sd669/2sb649 and TO-92s are 1n5551/1n5401.
Hi sam9!
What you mean by CFB Amp?
I have thinked that you talk about a Current Feedback Amp...but now i'm not so sure...as you talk about:
If it's not a current feedback design the sugestions that i gave to you in my previus post don't apply!!!
Are you talking about a Counpound feedback Pair Output Stage??(CFP).
What you mean by CFB Amp?
I have thinked that you talk about a Current Feedback Amp...but now i'm not so sure...as you talk about:
If it's not a current feedback design the sugestions that i gave to you in my previus post don't apply!!!
Are you talking about a Counpound feedback Pair Output Stage??(CFP).
CFB = collector feedback. Output is taken off the collectors, emitters collect to rails and driver emitters commect to output collectors. It has certain theoretical advantages with respect to thermal tracking, tight local feed back (low distortion) etc, but often exhibits local oscillations there are very hard to pin down and cure. Appearently, adding parralel output pairs adds to all of the above.
Here's the results of using everyone's suggestions:
A-(Ciclotron)Found a way to bolt the heatsinks to the tempory enclosure but still be able to quickly take the board out and replace it. This got rid of the intermittent behavior and I was left with more "normal" oscillating behavior.
B-(Tube_Dude)Removed the small cap parrallel to the feedback resistor. This greatly reduced but did not eliminate the oscillations. Odd, when I was running this with only one pair of output devices, adding this cap cleaned up the oscillations so well that even with a variety of caps across the dummy load, I was getting a beautiful, crisp output signal on the scope. Now with two pairs, removing it helps!
C- (Tube_Dude) I checked the values of the feedback resistors. Due to a typo using my schrematic capture software, I had set the gain about 20% higher than intended! Correcting this eliminated the rest of the oscillations, at least under no-load condations.
However, I still get oscillations with an 8 ohm dummy load. A big bloom of "fuzz" at the the + and - peaks when the input signal in increased to just a few mA. Increasing further and it eventually just flat-lines until the signal is brought back down. Since this seems to indicate something wrong at the peaks, I'll next try reducing the resistance in the [driver base]-[rail]-[output device emitter] path per Tube_Dude. After that comes the cruder approach of increasing the compensation cap and/or reducing gain another notch.
After this project, I think I've had enough CFB for a while the next amp is going to be an EF output. What good is squeezing out a few percent more distortion if the sucker insists on oscillating if you look at it cross-eyed?
A-(Ciclotron)Found a way to bolt the heatsinks to the tempory enclosure but still be able to quickly take the board out and replace it. This got rid of the intermittent behavior and I was left with more "normal" oscillating behavior.
B-(Tube_Dude)Removed the small cap parrallel to the feedback resistor. This greatly reduced but did not eliminate the oscillations. Odd, when I was running this with only one pair of output devices, adding this cap cleaned up the oscillations so well that even with a variety of caps across the dummy load, I was getting a beautiful, crisp output signal on the scope. Now with two pairs, removing it helps!
C- (Tube_Dude) I checked the values of the feedback resistors. Due to a typo using my schrematic capture software, I had set the gain about 20% higher than intended! Correcting this eliminated the rest of the oscillations, at least under no-load condations.
However, I still get oscillations with an 8 ohm dummy load. A big bloom of "fuzz" at the the + and - peaks when the input signal in increased to just a few mA. Increasing further and it eventually just flat-lines until the signal is brought back down. Since this seems to indicate something wrong at the peaks, I'll next try reducing the resistance in the [driver base]-[rail]-[output device emitter] path per Tube_Dude. After that comes the cruder approach of increasing the compensation cap and/or reducing gain another notch.
After this project, I think I've had enough CFB for a while the next amp is going to be an EF output. What good is squeezing out a few percent more distortion if the sucker insists on oscillating if you look at it cross-eyed?
"my experience has been that CFBs are tough to get to work stablely, this is particularly true with mosfets."
At least two authors list the benefits of CFB, and then as a kind of afterthought mention that they are subject to oscillations for which a sound understanding and remedy remains elusive. It's easy to miss this caveat after reading how wonderful they otherwise are. I suppose one additional benefit is that if I ever get this thing to run the way I want, it will be a big ego boost.
At least two authors list the benefits of CFB, and then as a kind of afterthought mention that they are subject to oscillations for which a sound understanding and remedy remains elusive. It's easy to miss this caveat after reading how wonderful they otherwise are. I suppose one additional benefit is that if I ever get this thing to run the way I want, it will be a big ego boost.
The problem is that your output stage is just too slow, that why you decreased compensation and the oscillation problem got better....
I can see four solutions:
1-use only one output device...
2-increase driver current...
3-put a folower between the driver and the output device...(this creates other problem because of the big increase in the local open loop gain)
4-scrap that crapy output stage and build a folower that gives a much better results wen conected to a reactive load...
millwood
if you realy understand the problem you will see why mosfets(gate capacitence is multtiplied by the gm of the mosfet in commun source mode) are a big no no in a conpound folower
cheers
I can see four solutions:
1-use only one output device...
2-increase driver current...
3-put a folower between the driver and the output device...(this creates other problem because of the big increase in the local open loop gain)
4-scrap that crapy output stage and build a folower that gives a much better results wen conected to a reactive load...
millwood
if you realy understand the problem you will see why mosfets(gate capacitence is multtiplied by the gm of the mosfet in commun source mode) are a big no no in a conpound folower
cheers
"The problem is that your output stage is just too slow, that why you decreased compensation and the oscillation problem got better...."
Pehaps, but at least as far as the actual devices go 2sa1943/2sc5200 anf 2sd669/2sb649 are about as "fast" as one is going to find from the usual sources. But see "2-", below.
"I can see four solutions:
1-use only one output device..."
Not an option given tne intended application. Also future applications, I anticipate would need even more.
"2-increase driver current..."
I'm trying out some mods today to see if this this can be done. Plus another idea or two.
"3-put a folower between the driver and the output device...(this creates other problem because of the big increase in the local open loop gain)"
Most likely I would be substituting one agravation for another -- not progress
"4-scrap that crapy output stage and build a folower that gives a much better results wen conected to a reactive load..."
This is the most likely resolution. However, I don't agree with the adjective, "crapy". More like "tempermental", such as when young I had an experience with a French girl friend and a very fast Italian car. Frustrating, expensive and ultimately a waste of time - but educational.
Pehaps, but at least as far as the actual devices go 2sa1943/2sc5200 anf 2sd669/2sb649 are about as "fast" as one is going to find from the usual sources. But see "2-", below.
"I can see four solutions:
1-use only one output device..."
Not an option given tne intended application. Also future applications, I anticipate would need even more.
"2-increase driver current..."
I'm trying out some mods today to see if this this can be done. Plus another idea or two.
"3-put a folower between the driver and the output device...(this creates other problem because of the big increase in the local open loop gain)"
Most likely I would be substituting one agravation for another -- not progress
"4-scrap that crapy output stage and build a folower that gives a much better results wen conected to a reactive load..."
This is the most likely resolution. However, I don't agree with the adjective, "crapy". More like "tempermental", such as when young I had an experience with a French girl friend and a very fast Italian car. Frustrating, expensive and ultimately a waste of time - but educational.
Sam,
I suspect you have oscillation of the Complementary Feedback Pair.
Place a 100pF (for starters, might need more) between the base and collector of each driver.
This will reduce gain of the driver at high frequencies, improving Nyquist stability.
Furthermore, you can sometimes stop oscillation without sonically destructive base/collector caps on the driver by introducing 0R22 to 0R47 emitter resistors into the output devices, between rail and emitter. This reduces gain within the CFP also, but does not introduce sonic penalty. It does however cost rail efficiency.
I would tactfully agree with the comments about 'crappy CFPs'. I'd strongly suggest a couple of emitter followers in cascade.
Cheers,
Hugh
I suspect you have oscillation of the Complementary Feedback Pair.
Place a 100pF (for starters, might need more) between the base and collector of each driver.
This will reduce gain of the driver at high frequencies, improving Nyquist stability.
Furthermore, you can sometimes stop oscillation without sonically destructive base/collector caps on the driver by introducing 0R22 to 0R47 emitter resistors into the output devices, between rail and emitter. This reduces gain within the CFP also, but does not introduce sonic penalty. It does however cost rail efficiency.
I would tactfully agree with the comments about 'crappy CFPs'. I'd strongly suggest a couple of emitter followers in cascade.
Cheers,
Hugh
sam9 said:
Sam, you only get the ego boost after your ego has been busted working out a solution in vein,
.I think many of the problems associated with CFB is in wiring / layout. I can count the number of my working breadboarded CFB amps I have done in one hand (a couple at best). One such beast took a couple years and I still failed,
. I guess the stray capacitance / inductance of the wires on a breadboard does wonders in getting the circuitry to oscillate.Your chance of getting it to work on a pcb is likely better but for small time hobbist, we really cannot afford to revise PCBs endlessly. so it may be one of those areas where it pays to get a proven layout for a CFB amp - I think P3A uses CFB.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.