??
Nelson Pass said:
Reiterating a point I made a long time ago, these curves imply
that you either want no negative feedback, or a lot of it.
Baxandall made the same point in his series. IIRC, I think he showed something up to 10db or so of feedback increased level of high order harmonics, and you had to go above that to start reducing them again.
http://www.analog.com/library/analogDialogue/Anniversary/10.htmlJohan Potgieter said:
See Question "Q. What about the inductance of wires?" at the end of the document. These are for straight "free air" wires/traces only (no other conductors etc nearby).
Rule-of-thumb value is 10nH/cm
http://www.ieee.org.uk/docs/emc1206a.pdf
(Page 14 has a graph/equation for traces above an GND plane)
Welcome to RF-Circuit design
Calculating proper trace impedances on a given PCB is a very well-paid art... but still useless if you don't know the actual parasitics of your parts (e.g. take care of the parasitics of isolating washers for power transistors on a grounded heatsink...)
Regards, Klaus
Hi John,
As a basement experimenter and ex-service guy (well, still kind of current), I even play with feedback levels and different circuit topologies. I have discovered that with different circuits, there is a happy spot were some feedback exists and the music still has life. Beyond this point the numbers get better still but the amplifier becomes dry and lifeless. I find this fascinating, same design but different character.
I guess I just contradicted Charles here, but what can I say? Possibly in a better design this does not occur. The difference is very plain though.
-Chris
I can't imagine where that came from. I expect all decent designers have actually tried and investigated feedback and how to apply it.
As a basement experimenter and ex-service guy (well, still kind of current), I even play with feedback levels and different circuit topologies. I have discovered that with different circuits, there is a happy spot were some feedback exists and the music still has life. Beyond this point the numbers get better still but the amplifier becomes dry and lifeless. I find this fascinating, same design but different character.
I guess I just contradicted Charles here, but what can I say? Possibly in a better design this does not occur. The difference is very plain though.
-Chris
I tend toward optimum feedback levels, myself. Charles is more extreme in his view. However, there is no arguing with his success. Personally, I think that open loop bandwidth is the most important factor. For example, the open loop bandwidth in my JC-1 power amp might be around 4KHz. I would prefer 40KHz, but I cannot afford the increase in measured distortion, because it would be about 10 times as much and the amp would not meet THX specs, which is important to us.
Fortunately, I don't have to hit arbitrary targets like THX, since I've only done things for myself and a handful of commissioned pieces. I cannot claim the level of performance that Charles quotes. Hopefully I'll get there at some point. For now, I still need a few dB of feedback to get the bandwidth up where I want it to be.
Grey
P.S.: Can't remember the distortion offhand...I seem to recall something on the order of .05-07% closed loop. I may have slipped a cog or two. It might be a little higher.
Sounds good, though, regardless of the numbers.
Grey
P.S.: Can't remember the distortion offhand...I seem to recall something on the order of .05-07% closed loop. I may have slipped a cog or two. It might be a little higher.
Sounds good, though, regardless of the numbers.
RF Design Practices
I agree that RF and microwave design is a very different kettle of fish. I have worked in the field for 10+ years, and am glad that I did not get rid of my old graduate E+M textbooks.
At audio frequencies (including ultrasonic harmonics) it is not necessary to invoke RF sophisticated practices. However, it is necessary to have a good grasp of parasitic capacitances and inductances as they apply on a per unit length basis for such things as twisted pairs and PCB traces. There are very good tools that are capable of calculating crosstalk, coupling, and loss effects for interconnect. Unfortunately, these tools are targeted for a professional user, and come with professional prices.
My suggestion is to use rules of thumb such as minimizing the proximity between input and output wiring and being prepared to add compensation capacitors when all else fails. I just completed an amplifier design on which I spent many weeks simulating and which showed no signs of instability. However, reality has the rude habit or intruding, and I was forced to apply some extra compensation to tame the oscillation.
I agree that RF and microwave design is a very different kettle of fish. I have worked in the field for 10+ years, and am glad that I did not get rid of my old graduate E+M textbooks.
At audio frequencies (including ultrasonic harmonics) it is not necessary to invoke RF sophisticated practices. However, it is necessary to have a good grasp of parasitic capacitances and inductances as they apply on a per unit length basis for such things as twisted pairs and PCB traces. There are very good tools that are capable of calculating crosstalk, coupling, and loss effects for interconnect. Unfortunately, these tools are targeted for a professional user, and come with professional prices.
My suggestion is to use rules of thumb such as minimizing the proximity between input and output wiring and being prepared to add compensation capacitors when all else fails. I just completed an amplifier design on which I spent many weeks simulating and which showed no signs of instability. However, reality has the rude habit or intruding, and I was forced to apply some extra compensation to tame the oscillation.
Johan Potgieter said:
Hi Johan,
Just like Klaus, I also use 10nH/cm as a rule of thumb.
As for the mutual inductance of twisted leads (or parallel traces) I just guess. Normally, I assume k=0.7 if the leads (to the O/P devices for example) are 5cm long. Maybe this value is to low, but we have also to take into account that at both ends the leads/traces are not any longer close together.
Klaus, do you have any suggestion regarding a first guess?
Regards, Edmond.
john curl said:
Hi John,
First of all, I really like your "Fellow designers" start to your post.
This is a really important discussion about amount of NFB and behavior of amplifiers as a function of NFB.
It is certainly true that there is a lot of disagreement, if not lack of understanding, about the use of high amounts of negative feedback versus low or no amount of negative feedback.
For one thing, I think it is important in our discussions to recognize that amount of NFB is often a function of frequency in the audio band, whether we agree that that is a good thing or not. For that reason, when talking about high versus low NFB, I think we need to be clear about where it is high and low.
In designs with high open-loop bandwidth equal to the audio band, the answer seems pretty simple, since the amount will be about the same everywhere in the audio band. For that we just need to say whether the feedback is 0 dB, 10 dB, 20 dB or 40 dB, for example.
It's not quite as straightforward for more typical designs, where we might have 20 dB of NFB at 20 kHz, but perhaps 60 dB of NFB at 200 Hz.
In other words, when loosely saying an amplifier is high NFB, do we just key off of the low-frequency value? That may not tell the whole story. Just a thought for some clarification.
Cheers,
Bob
Suggestion to the Moderator
This group of posts on the merits of negative feedback and the behavior of amplifiers as a function of amount of negative feedback is very important. However, they are not directly relevant to the BJT vs MOSFET subject.
The moderators might want to consider moving these posts to the feedback thread, where they would fit in perfectly.
Just a thought,
Bob
This group of posts on the merits of negative feedback and the behavior of amplifiers as a function of amount of negative feedback is very important. However, they are not directly relevant to the BJT vs MOSFET subject.
The moderators might want to consider moving these posts to the feedback thread, where they would fit in perfectly.
Just a thought,
Bob
Klaus, Analog_guy, Estuart,
Most informative - and sobering!!
While others will hopefully not choke on this irreverent intrusion of RF (yech) in the hallowed audio field, ... thank Heavens for resonance! (Now I've done it).
I mean that I have been active in the 150 MHz field in the design of radio animal tracking, down to nV sensitivities. But one usually works (very) narrow band, so there resonance took care!!
Back to palatable frequencies, I will use those handy hints. Thanks.
Most informative - and sobering!!
While others will hopefully not choke on this irreverent intrusion of RF (yech) in the hallowed audio field, ... thank Heavens for resonance! (Now I've done it).
I mean that I have been active in the 150 MHz field in the design of radio animal tracking, down to nV sensitivities. But one usually works (very) narrow band, so there resonance took care!!
Back to palatable frequencies, I will use those handy hints. Thanks.
First off let me say that I'm generally for the use of negative feedback, however there are a few negatives that I've noticed over the years. Some are fundamental, others involve poor/sloppy design.
I'm surprised the discussion has not considered overload and overshoot inside the the loop. Or did I miss it earlier in this thread?
I'd certainly think that an excellent amplifier with no global feedback, is going to sound better than one with high feedback when driven into overload/clipping. This might explain why I often read, and experience myself, that high power amps sound open and effortless even though many believe that they do not need the power. Perhaps they actually do.
I would not be surprised if the difference between amps with and without global feedback is more so, if not entirely, their clean overload recovery rather than higher order distortion issues.
Pete B.
I'm surprised the discussion has not considered overload and overshoot inside the the loop. Or did I miss it earlier in this thread?
I'd certainly think that an excellent amplifier with no global feedback, is going to sound better than one with high feedback when driven into overload/clipping. This might explain why I often read, and experience myself, that high power amps sound open and effortless even though many believe that they do not need the power. Perhaps they actually do.
I would not be surprised if the difference between amps with and without global feedback is more so, if not entirely, their clean overload recovery rather than higher order distortion issues.
Pete B.
PB2 said:
Pete,
I think you have certainly hit one of the nails on the head, namely overload behavior. First, it has been my experience by actual measurement in our workshops at RMAF and HE2007 that amps are in a lot of cases clipping more than we think. This is especially the case on well-recorded music that has not been compressed too much, and where perhaps a modest powered amplifier (say 100 wpc) is driving speakers of moderate to low sensitivity.
Tubes have a way of clipping more gracefully, and that may be a big part of the sound difference. I do not claim to be a Golden ears-equipped reviewer, but I have noticed that it is remarkable with at least one tube amp that it sounded as good as it did when it was surely clipping. But of course we are talking about brief clipping on transients in well-recorded music.
It is also fair to say that amplifiers with NFB will clip more sharply than those without NFB, although it can be dangerous to make generalizations. For example, some amps without NFB can clip poorly for reasons having nothing to do with NFB, such as saturation and sticking. By the same token, the way in which NFB amps clip can vary greatly as well. An NFB amp with proper Baker clamps or the like will clip very cleanly without sticking, but it will still clip quite sharply.
Ultimately, I am a big believer in well-designed passive-adaptive soft clipping circuits in front of solid-state amps that prevent the amplifier-proper from ever clipping (while not taking away the dynamic headroom). Of course, if one ever reviewed and measured such an amplifier with the soft clipping circuit engaged, they would (or the marketing department would) complain about the harmonic distortion of the amplifier rising well before clipping was reached.
Cheers,
Bob
Hi Edmond,estuart said:
I also use k=0.7 for twisted pair (never checked that, though. Read it -- in a good AppNote I think -- and believed it).
"The Scot" (Mr.Lesurf) has a nice Java applet for calculating some parameters of twin-feed wiring:
http://www.st-andrews.ac.uk/~jcgl/Scots_Guide/audio/Java/twin.html
As long as skin effect is not an issue the L/C values seem to be ok for RF-frequencies also.
Regards, Klaus
Clipping? The last time I tried to drive my amps into audible clipping, the police came and closed me down!
Also, everyone had left the listening room due to the racket. They would have probably cited me for disturbing the peace, if I was not holding an SPL meter.
Most clipping in early solid state power amps was due to V-I protection circuitry, and it came in early with REAL loudspeaker loads. Tubes can't drive very much, so they have to clip gently.
Now let's use an example: My JC-1 and the WATT 1 loudspeaker (my personal speaker)
First the 1W sensitivity is 91dB, without a subwoofer. So at 10W, I should have 101 dB spl, and at 100W maybe 111 spl, and at 400W 117 spl.
Darn, a lease breaker if there ever was one! And that is for one channel only, without a subwoofer.
The lesson is: 'Real man' amps don't easily clip. That includes Nelson and Charles in this design group. It is only the wimpy solid state power amps of 50W or less that are going to have any significant problems, and I recommmend making a bigger amp, instead of putting a 'band-aid' on a small one.
Also, everyone had left the listening room due to the racket. They would have probably cited me for disturbing the peace, if I was not holding an SPL meter.
Most clipping in early solid state power amps was due to V-I protection circuitry, and it came in early with REAL loudspeaker loads. Tubes can't drive very much, so they have to clip gently.
Now let's use an example: My JC-1 and the WATT 1 loudspeaker (my personal speaker)
First the 1W sensitivity is 91dB, without a subwoofer. So at 10W, I should have 101 dB spl, and at 100W maybe 111 spl, and at 400W 117 spl.
Darn, a lease breaker if there ever was one! And that is for one channel only, without a subwoofer.
The lesson is: 'Real man' amps don't easily clip. That includes Nelson and Charles in this design group. It is only the wimpy solid state power amps of 50W or less that are going to have any significant problems, and I recommmend making a bigger amp, instead of putting a 'band-aid' on a small one.
