rdf said:
Interesting you should mention that. I recently finished up an 807 amp project. For this, I added 6.9dbv of local feedback (807 plates-to-driver grid) as this was the recommended LFB I got from a tech report. When adding the gNFB, I had a miscalculation
An externally hosted image should be here but it was not working when we last tested it.
and wired in an Rf about five times smaller than it should have been, for some 43dbv of gNFB. Completely one-dimensional; no "ambience". An externally hosted image should be here but it was not working when we last tested it.
Getting the right value for some 10dbv of gNFB was much better, and with that, I'm not seeing any lack of ambience at all. The feedback got the bass under control, so that it's not dominated by speaker resonance, and it cleaned up the mids and highs very nicely. Tried it for some time with no feedback of any sort, and that had poor bass, and overly bright mids and highs that could be annoying with some program material. The feedback doesn't seem to have done the ambience any harm.
Later auditioned a solid state amp for comparison purposes to see how the home-built stacked up against a commercial product that got some pretty good reviews. Did that sound just plain bad. It uses some sort of MOSFET topology -- not sure what it is exactly, but I can take a good guess. Here's the specs:
Again, no ambience and completely one dimensional. In addition, the bass sounded overly loud, but lacking detail. The mids had no brightness at all, and vocals simply receeded into the same bland, homogenized background, making the lyrics harder to understand.
Needless to say, the folks who loaned it didn't have to worry about getting it back.
Hi Miles, thanks for the sanity check. Your description matches my experience closely. Again, absolutely no idea why but the reduction in ambience is most audible at high frequencies and moves down the spectrum with increased feedback. At the end of the road the result is "the bass sounded overly loud, but lacking detail..mids had no brightness at all, and vocals simply receeded into the same bland, homogenized background.." To be fair the source material on which it's most apparent is movies with all that 'head related transfer function' DSP jiggery pokery happening.
I have a ss amp in for a friend, doing some minor repairs. It is labelled a reference amplifier, early 1990's? Brand starts with 'P' (whats the policy on naming please). It weighs a tonne and probably cost one too.
After much time listening to no feedback, I could spot it straight away. Now this is a luxurious amp, quiet, capable and well appointed. But the feedback stands out like the proverbial.
Images are defocused. They live under a plexiglass dome around my speakers. There seems to be a harshness (best guess could be from the beer can caps), and it sound like it is trapped and complaining about it (doncha love trying to put intangible things into words 😉 )
The bass is well weighted and powerful, but it takes my horns and makes them sound like a regular closed box. Transient response is noticeably affected at the impedance extreme.
In my earlier days I'd have walked through fire for this amp. Now I find myself tearing it down. Sigh 🙁
After much time listening to no feedback, I could spot it straight away. Now this is a luxurious amp, quiet, capable and well appointed. But the feedback stands out like the proverbial.
Images are defocused. They live under a plexiglass dome around my speakers. There seems to be a harshness (best guess could be from the beer can caps), and it sound like it is trapped and complaining about it (doncha love trying to put intangible things into words 😉 )
The bass is well weighted and powerful, but it takes my horns and makes them sound like a regular closed box. Transient response is noticeably affected at the impedance extreme.
In my earlier days I'd have walked through fire for this amp. Now I find myself tearing it down. Sigh 🙁
rdf said:
Actually, adding feedback did cause some loss of ambience and spatiality. I did play the 807 amp for a month with no feedback at all to see how that worked before adding the feedback loops. However, that's an acceptable trade-off considering how much better the bass is, and that the excessive brightness to the highs doesn't have much the same effect as fingernails on a blackboard with some material (especially Ozzy Osbourne). Plenty of ambience remains since the feedback factor is not so high.
There is a theory that, since feedback tends to fill the noise floor with random, subaudible noise, that this removes subtle clues that help to identify the direction and distance of the sounds we hear, both in RL and from amps and speeks. Same theory as to why some folks favor vinyl over CDs: A-to-D conversion removes all the subaudible noise that may have that vital information. Who knows if that's right or not? Psychoacoustics is full of unknowns.
That's fascinating on a couple of levels. First is again how independent observations appear to correlate so well. I agree competely with the perceived change, the benefits of the trade-off (depending on circuit) and the description of the way a low or no feedback circuit can sound up top. There's a common perception tubes can do no wrong at high frequencies.
The second is that as I applied feedback to my most recent circuit, a spectral analysis showed the PS noise rectification peaks went down but the 'flat' noise floor beneath came up at low frequencies. I had/have no explanation for the behaviour and put it down to measurement artifacts.
The second is that as I applied feedback to my most recent circuit, a spectral analysis showed the PS noise rectification peaks went down but the 'flat' noise floor beneath came up at low frequencies. I had/have no explanation for the behaviour and put it down to measurement artifacts.
Miles Prower said:
If this was a court case the poor feedback would probably have been hanged (shot, choked, whatever). It is experiences like these that has given feedback a bad name.
Now I don't doubt these testimonies for a moment, but I would make a "sanity" plea here. I am repeating somewhat, but please keep in mind that feedback is the single most powerful tool to linearise an amplifier's rersponse, and therefore also the most abused one. After all, no feedback must mean no ill effects from misusing it - so what can go wrong (except for audible distortion).
But having said that ....🙁 🙁 Can all the amplifiers in the above mentioned experiences have had a feedback abuse problem? I can of course not say - it would seem unlikely. Only, there is simply no reason for properly used feedback (iow a more linear amplifier) to cause that. There are enough excellent stereo set-ups out there using NFB amplifiers (both tube and ss) to support this. I did say earlier that the phase behaviour of amplifiers (very different tolerance components) can contribute to lack of spacial image. I would really hesitate to vilify feedback per se for these shortcomings, but I am at a disadvantage not having ever been given such a "dead" amplifier to analyse.
Then, it is probably known that ambience and such is mainly a function of the different reflection patterns reaching the (basically) 2 microphones, and that this information exists largely from 2 KHz upward. I have difficulty in visualising that subaudible "noise" (define noise - aperiodic?) can have an effect here. Again ...🙁
I hate to relate the following, at my own expense,,, but my (simple) car sound setup produces the best ambience I have heard in a long time, and I am talking about FM only! (better than my exalted (!) home system). One day I am going to take that car system out, connect it up in the living room and work this out once and for all. (No, the car amplifiers also use feedback!)
Regards.
When going through the feedback dilemma personally, I drew what seemed a logical conclusion from what I had experienced. Feedback will do little damage to a signal that doesn't need it.
The more the fed back signal resembles the input it's fed to, the less damage, it would seem.
High frequencies are often subject to phase shift. I have attached a drawing. It might be conceptually illustrative if not typical.
The more the fed back signal resembles the input it's fed to, the less damage, it would seem.
High frequencies are often subject to phase shift. I have attached a drawing. It might be conceptually illustrative if not typical.
Pioneer A-757planet10 said:
Attachments
Hi Indm,
There also seems to be a sweet spot where the benefits begin to be outweighed by the problems of feedback.
-Chris
Yes!
There also seems to be a sweet spot where the benefits begin to be outweighed by the problems of feedback.
-Chris
Indm,
You didn’t say exactly what you had concluded from your sketch (post #108), but the implication of sketch D is that you’d get some sort of non-linear output or something “kinky” with higher frequency content than the original, simply by adding a phase-shifted and inverted sine wave (or a time slice thereof) to the original sine wave. This simply cannot be true because:
sin(wt) + sin(wt+p) = 2cos(p/2)sin(wt+p/2)
Where w = 2*pi*f, t = time, p = phase shift in radians (any arbitrary value)
The sum is simply a sine wave at the SAME frequency as the original, but phase shifted by p/2 radians, and the peak amplitude is now 2cos(p/2), a constant (this term is not a sinusoidal wave itself since it is not a function of the variable t).
I’m not jumping into the “feedback is good” or “feedback is bad” argument right now, but at least I wanted to suggest that your sketch might be misleading you. If you very carefully redraw the sum of two sine wave sections, maybe using Excel to calculate points along the way, you’ll see another clean piece of a sine wave which crosses through zero without the flat spot in sketch D.
You didn’t say exactly what you had concluded from your sketch (post #108), but the implication of sketch D is that you’d get some sort of non-linear output or something “kinky” with higher frequency content than the original, simply by adding a phase-shifted and inverted sine wave (or a time slice thereof) to the original sine wave. This simply cannot be true because:
sin(wt) + sin(wt+p) = 2cos(p/2)sin(wt+p/2)
Where w = 2*pi*f, t = time, p = phase shift in radians (any arbitrary value)
The sum is simply a sine wave at the SAME frequency as the original, but phase shifted by p/2 radians, and the peak amplitude is now 2cos(p/2), a constant (this term is not a sinusoidal wave itself since it is not a function of the variable t).
I’m not jumping into the “feedback is good” or “feedback is bad” argument right now, but at least I wanted to suggest that your sketch might be misleading you. If you very carefully redraw the sum of two sine wave sections, maybe using Excel to calculate points along the way, you’ll see another clean piece of a sine wave which crosses through zero without the flat spot in sketch D.
Ok Brian, good call. I remapped using calculated values and cad. I should learn not to draw sinewaves by hand (at least without some kind of feedback 😀 ).
It reminds me of one of my teachers trying to draw envelopes. We used to laugh at them behind his back. Karma hey?
Anatech, I still live in a land of confusion, but what you said is starting to look good. Maybe its time to look at fixing that OPT rolloff.
It reminds me of one of my teachers trying to draw envelopes. We used to laugh at them behind his back. Karma hey?
Anatech, I still live in a land of confusion, but what you said is starting to look good. Maybe its time to look at fixing that OPT rolloff.
Attachments
Hi lndm,
Sometimes the best way to understand something like this is to build a discrete amplifier and play with the feedback ratio. This will change your damping factor a bit, and you should pad the input to keep your levels equal.
You have to build it and listen to figure this out. There is no way to sim it either. 😉
-Chris
Sometimes the best way to understand something like this is to build a discrete amplifier and play with the feedback ratio. This will change your damping factor a bit, and you should pad the input to keep your levels equal.
You have to build it and listen to figure this out. There is no way to sim it either. 😉
-Chris
I have listened to my current amp (my own design, 6sn7/kt66 PSE) with 1.5, 3, 6, 12, 18dB of feedback. Beyond 6-12dB there were diminishing returns. 1.5dB seemed to have hardly a noticeable effect, except for something hard to describe that all fb levels had. I was cautious about stability and RF introduction. I did like the subjective improvements of the better specs, but currently choose to live without the FB.
Before that, I ran a williamson with 6550's. Couldn't make up my mind between 7dB of feedback and none, kept changing.
Before that I ran a hign power mosfet AB (and class A at one point) / valve hybrid. The AB mosfets sounded like they measured well, but sound was irritating. Feedback only mangled this. With 20dB feedback, the amp behaved like it had plenty of control, but I could hear that the problem was just buried. Besides, it sounded lifeless.
So Anatech, I am no authority but I can speak from experience. I avoid feedback because I haven't been able to make it work for me.
To be truthful, I wish I could.
Before that, I ran a williamson with 6550's. Couldn't make up my mind between 7dB of feedback and none, kept changing.
Before that I ran a hign power mosfet AB (and class A at one point) / valve hybrid. The AB mosfets sounded like they measured well, but sound was irritating. Feedback only mangled this. With 20dB feedback, the amp behaved like it had plenty of control, but I could hear that the problem was just buried. Besides, it sounded lifeless.
So Anatech, I am no authority but I can speak from experience. I avoid feedback because I haven't been able to make it work for me.
To be truthful, I wish I could.
I built a KT-88 P-P (pentode wired) amp once with a vairiable global feedback control. I used this amp for a while before passing it around to several of my friends. One of them bought it. Everyone agreed that it sounded best with the feedback knob turned "almost all of the way down" but not "all the way down". It was also agreed that the best feedback position was dependent on the type of music being played, and how loud it was being played.
As the solid state revolution played out in the early 70's many amp makers kept designing amps with ever increasing open loop gain and applying larger levels of feedback to linearize them. This made for great specs on paper, but a generic sound. Remember from engineering school, as the open loop gain rises, the closed loop amplifier's characteristics are determined entirely by the feedback loop. The amplifier itself drops out of the equation.
I find that local feedback (around the output stage only) can be applied in pretty large doses with only beneficial results. Global feedback (from speaker terminals to amp input stage) must be applied sparingly if at all. Here again different users will have different opinions as to how much is right for them. The "best sound" rarely coincides with the lowers measured distortion. This is especially true with a SE tube amp.
As the solid state revolution played out in the early 70's many amp makers kept designing amps with ever increasing open loop gain and applying larger levels of feedback to linearize them. This made for great specs on paper, but a generic sound. Remember from engineering school, as the open loop gain rises, the closed loop amplifier's characteristics are determined entirely by the feedback loop. The amplifier itself drops out of the equation.
I find that local feedback (around the output stage only) can be applied in pretty large doses with only beneficial results. Global feedback (from speaker terminals to amp input stage) must be applied sparingly if at all. Here again different users will have different opinions as to how much is right for them. The "best sound" rarely coincides with the lowers measured distortion. This is especially true with a SE tube amp.
Nigel Goodwin said:
I try to avoid getting involved in this ancient debate, but I have to comment on this. I think I know what Indm was trying to say and it really is about objective measurements.
It comes down to selecting a metric. Historically, a simple unweighted THD measurement has been used to compare amps. But, there are MANY MANY people that have come to the conclusion that it's not a good metric; it just doesn't correlate to perceived quality of music reproduction. The whole point of choosing a metric is to help us objectively rank amps according to how well they perform their intended job. Unweighted THD doesn't help.
What I think Indm was suggesting is that some sort of weighting on the distortion spectrum might result in a more useful metric. It might not be perfect; only a fool would take it as the final arbiter of which amp is best, but it might actually be useful while unweighted THD seems to be nearly useless.
There is the Gedlee metric developed by Earl R. Geddes and Lidia W. Lee. It was presented at the 115th convention of the Audio Engineering Society (2002) as Convention Paper 5890. For a while you could download a preprint from http://www.gedlee.com but it doesn't seem to be available anymore. At least I couldn't find it. The metric involves more than simple weighting of harmonics, but according to the authors it is a reliably good tool. The basics might be in the online book by Geddes (available at gedlee website.) I don't know, I have not read through all of the chapters.
My little CCS-6c45/EL84SE gets regularly abused with different feedback techniques, loading, bias points, etc.. It's a test bed and way to play with many of the ideas discussed here. I thought the design was finally nailed, 15 ma through the 6C45, 150 ohm cathode, 2.2K or about ~10dB global negative feedback. The distortion spectrum was as good as any I've achieved with this circuit, ~0.5% 2nd with 3rd and 4th each another ~20dB down from each other and everything else below ~-110dB at full power. Nothing visible above the 6th (-125!) Save for about twice the 2nd harmonic it left the Mullard 3-3 which once inhabited this chassis for dead.
Yesterday it was rejigged again to confirm some of the impressions I'ld posted. Feedback was switched to cathode on the output only, the front end returned to LED bias. It's a much better sounding amp, with three times the measured distorion. Images break completely free from the speakers again, to the point where the anti-phase and DSP'd effects underlying most all contemporary movies are almost disorienting. The interchannel phase differences in good, two mic'd classical recordings generate a huge wall of sound. (Pre-emptive comment: no, it's not an inter-channel phase or balance issue, I measured. Mono is dead centre. No visible gremlins to the limit of a 60 MHz scope, grid stoppers everywhere.) Granted, the bass isn't quite as solid and sibilance has shifted from high and too sharp to softer but just slightly 'slurry', but the overall tradeoffs for me weight heavily to the no-feedback circuit.
Tubelab nails it. I have no explanation but cathode feedback does less harm than global.
Yesterday it was rejigged again to confirm some of the impressions I'ld posted. Feedback was switched to cathode on the output only, the front end returned to LED bias. It's a much better sounding amp, with three times the measured distorion. Images break completely free from the speakers again, to the point where the anti-phase and DSP'd effects underlying most all contemporary movies are almost disorienting. The interchannel phase differences in good, two mic'd classical recordings generate a huge wall of sound. (Pre-emptive comment: no, it's not an inter-channel phase or balance issue, I measured. Mono is dead centre. No visible gremlins to the limit of a 60 MHz scope, grid stoppers everywhere.) Granted, the bass isn't quite as solid and sibilance has shifted from high and too sharp to softer but just slightly 'slurry', but the overall tradeoffs for me weight heavily to the no-feedback circuit.
Tubelab nails it. I have no explanation but cathode feedback does less harm than global.
Hi Indm,
I am now experimenting with two feedback networks. One around the output stage only, the other around the voltage amp stage only. The two are not shared at all except at the node between the two systems.
We shall see.
-Chris
I have found that same range for global feedback. Higher and it seems to make music sterile.
I am now experimenting with two feedback networks. One around the output stage only, the other around the voltage amp stage only. The two are not shared at all except at the node between the two systems.
We shall see.
-Chris
anatech said:
Indeed.
I have yet to play with PP, but in SE, 6-8dB seems the absolute top for gNFB with a very "rich" tube before it sounds.... stale.
2-4dB is more an avg.
lndm said:
An interesting statement, which I pondered for a while...
It cannot refer to the input signal, for we assume that to be the standard that we want to hear unchanged. "Signal" must therefore refer to the output, but if that does not need feedback, it implies that it is perfect, and so is the amplifier - thus no need for feedback in the first place!
We are then back to the point where feedback is applied solely to reduce amplifier induced abnormalities, thus affecting the (output) signal. That applies to all amplifiers because I think it is common cause that all basic amplifiers generate audible distortion. Ergo, we are back to the statement that all feedback is bad. Again, this appears to conflict with measured and practical experience.
Regarding the effect of various degrees of feedback, we cannot disregard the fact that not all harmonics are objectionable. As said before, 2nd (all an octave higher) will enrich most music, while 3rd appears to make brass etc. sound more crisp. The danger here must be that if feedbackless reproduction sounds "nice", then applying feedback will indeed make it more dull. To me that would pose the question that one must first decide whether one wants the music to sound "better", or whether one wants an exact reproduction of the input. (And we have no way of telling whether the latter was in fact attractive; we were mostly not there to hear the original.)
It would appear to me that saying that an optimal amount of feedback is best is an anomaly logicwise. If one starts off with a flawed output, then somewhere in the middle (with some feedback) it is best, and improving amplifier linearity beyond that with more feedback makes it worse again, until it suddenly becomes perfect again in the limit where the amplifier is perfectly linear. I would suggest that in this case hearing tests are not a good measure of amplifier quality per se, for the reasons stated in the previous paragraph. They are a good measure if one's taste is the criterion, but then optimal design techniques do not necessarily count. (In all of this I am obviously excluding abused feedback, where poor design generates more problems than it solves.)
Regards.
Geek said:
With PP 807s set up for the "audiophile" Q-Point (1.8% THD before applying feedback), 6.9dbv local (recommended in tech report) and 10dbv gNFB worked out just great. Just enough to get the bass under control, kill the high end harshness, and not to completely flatten the sound.
From my findings, solid state amps with fast, BJT finals can tolerate quite a bit of gNFB (Helpful hint: quasi-complementary sounds significantly better than straight complementary.) MOSFET amps can't tolerate so much, and VTs even less, before the sound becomes completely homogenized, dull, lifeless.
- Status
- Not open for further replies.