Joshua_G said:
Hi Joshua,
I generally agree. Definitely more work can be done in this area to try to apply measurements to amplifiers in-situ with real loads playing real music.
For some of the interactions that are not dependent on high power levels, more work may be possible into real speaker loads without blowing them up.
For high-power issues, it is often possible to see outright amplifier mis-behavior when playing music at realistic levels into real loudspeakers if one looks for it. Under these conditions the amplifier and loudspeakers are only subjected to high power for brief intervals, which is OK.
I have often believed that a significant source of amplifiers sounding different is misbehavior of the amplifier under certain conditions, or differences in how the amplifier clips when it is briefly driven to clipping. The latter, of course, cannot explain differences in amplifier sound that are heard at low listening levels where clipping cannot possibly be happening.
One example of amplifier mis-behavior that can sometimes occur at higher levels is brief bursts of parasitic oscillation due to instability that arizes with larger signal swings as a result of device parameters changing under those larger signal swings. For this sort of thing I have been working on what I call a parasitic oscillation sniffer. It is a wideband amplifier with a steep high-pass rolloff above 20 kHz that will detect and amplify bursts of ultrasonic oscillation that may occur, without the output being masked by the program signal. The ac output can be looked at, or a rectified dc output representing the envelope of the oscillation burst can be looked at. The device has a bandwidth of over 100 MHz.
Cheers,
Bob
janneman said:You guys read the article by Dennis Colin on his noise measurements of Linear System's LSK389B low noise dual JFET:
http://www.audioxpress.com/magsdirx/ax/addenda/media/colin2993.pdf
Uses Scott's AD797 as the measurement amp, a nice touch I thought
Jan Didden
Anybody interested in this at all?
One conclusion was that two years ago those JFETs didn't meet spec, but now they seem to.
Jan Didden
PMA said:
I am especially looking forward Bob Cordell's reply to this
While there are exceptions to every rule, and almost any kind of amplifier can have an IIM issue, the point that I made in my work on IIM is that high open-loop output impedance made low by negative feedback is not, by itself, a cause of IIM or an exacerbator of IIM.
In my work, I actually built an amplifier and measured its IIM in accordance with Otala's test. This amplifier could be configured to have either a high open loop output impedance or a low open-loop output impedance. IIM was NOT higher in the amplifier with high open-loop output impedance. This was a fair apples-to-apples comparison; I'm not aware of anyone else who has done such a comparison in a properly controlled way. The details of what I did are in the AES paper on my website at www.cordellaudio.com. An article version was also published in Wireless World.
I also showed that the theory does not predict increased IIM when open-loop output impedance is high and is brought down to the same level with NFB.
IIM is generally reflective of changes in the closed loop output impedance as a function of output current. Designs that have very small changes in closed loop output impedance as a function of output current will generally have very low IIM, regardless of how the change in output impednace as a function of current was made very small.
Of course, for amplifiers with no global negative feedback, like Charles Hansen's, the change in output impedance is essentially the change in open-loop output impedance, since there is no loop. I am not aware of any IIM measurements that Charles has done on his amplifier, but they would be interesting to see. His MXR amplifier is an unusually good example of a no-NFB amplifier.
This all is NOT to say that there are no amplifiers with IIM problems. IIM is a useful test. It is only to say that, in the general sense, properly designed NFB is not the cause of IIM.
One of the best ways to avoid IIM is to use a Triple or the equivalent in the output stage and to have fairly high-current capability in the amplifier. Even if the Triple is driven by a VAS whose output impedance is extremely high at low frequencies, IIM will not usually be a problem in such a design.
Note that tube amplifiers are not immune to IIM, whether they use NFB or not. Similarly, any push-pull class AB tube amplifier also has crossover distortion caused by transconductance droop of the output tubes, not unlike the transconductance droop I described for MOSFET stages.
Cheers,
Bob
Bob Cordell said:
The problematic of testing is a large and difficult one.
May I suggest a similar idea.
As I explained in a previous thread, the linear swept sine wave impulse ( linear chirp) is a very handy signal that is never used in non linear testing.
It is an impulse made of a sine wave whose frequency is linearly swept from f1 to f2 during the duration of the pulse: Asin(at²+bt) during a time T . The beauty of this impulse is that it concentrates the energy in a band between f1 and f2. Of course you have energy outside because it is an impulse but the pulse is rather selective. By increasing the amplitude, you can generate in a transient way non linearities with the generating signal located mostly at low frequencies ( where the music is). What you will detect at high frequencies with a spectrum analyzer or with your method will be spuriuous signal. The pulse can be repeated with low duty cycle to ease detection. or visualisation.
This is real TIM test ( not the one from Otala
)I am planning to test the VI dual slope limiter ( FSOA). that I designed . Because the powers are rather high, do you have any advices/ suggestion to make a reactive dummy load.
Thanks
JPV
syn08 said:
I would generally agree, a single customer asking for a few 1000 screened parts will poison the inventory. But I have found the ones I bought were good.
Interesting would be an analysis of a random sampling for known JFET processing issues.
It is worth noting that RODE makes some of the quietest mics in the world with selected J305's (or so I was told). Selection is required, but starting at 3.5 cents each you don't need much yield.
The Scott Frankland Interview
I’ve had a chance to read the full interview with Scott Frankland. Excellent interview. I agree with nearly every word Scott said. Scott seems to be one of the most knowledgeable and level-headed tube designers around.
I urge all of you to read his comments in the interview – and read them carefully and in-context.
Scott is not an anti-NFB guy.
This, in spite of the fact that some of his tube amps do not use feedback. He does a good job of explaining why he does not use feedback with tube amps, and he is self-consistent in his reasoning. His reasons for not using NFB in tube amps do not necessarily apply to solid state amps.
Scott is correct in pointing out that NFB can make clipping sharper and that it is wise to avoid clipping. I believe that in poorly designed NFB amplifiers, the application of NFB can make clipping ugly as a result of poorly-managed internal overloads.
Scott is also correct in his reasoning about what he refers to as harmonic multiplication – the phenomenon that Baxandall investigated wherein the feeding back of distortion products to the input gives them a chance to multiply with the input signal and create new spectra. If feedback is applied to a forward path with only second and third harmonics, the result can be a spray of harmonics of decreasing amplitude out to very high frequencies.
Notably, Scott points out that the worst case for this phenomenon is a low-feedback situation and that the problem largely disappears as the amount of NFB is increased. This is a subtlety that escapes many well-known amplifier designers. Scott’s beef with using NFB in tube amplifiers is that in a tube amplifier he believes you cannot get enough NFB to get into the zone where harmonic multiplication is not a problem.
In this regard I think it is notable that the legendary McIntosh tube amplifiers succeeded in getting a lot of total negative feedback around the output stage with their unity-coupled output stage which provided quite a bit of local NFB.
Scott’s criteria for using NFB is that the amplifier with NFB does not have any larger higher-order products than there are with the open-loop design under the same conditions. This is not a bad criteria, although once the products get below, say, -130 dB I would not sweat it.
In my view, the key thing to keep in mind is that we rarely close the loop around a forward path that has just seconds and thirds. Just the existence of multiple stages alone gets us to higher-order products in the open loop. Moreover, crossover distortion virtually always produces a spray of high-order products. The higher-order products are there in the open loop in the first place, even if fairly low in amplitude.
Scott’s comments regarding TIM are also right on the money. He rightly points out that feedback will not cause more TIM as long as the stage receiving the feedback does not overload in a worst-case situation. This essentially translates to the amplifier having adequate slew rate. It is a slight over-simplification in regard to soft TIM, but it is close enough.
He is also correct in pointing out one of the paradoxes of negative feedback: NFB works best with a perfectly linear forward path, but then you probably don’t need negative feedback. Of course, many of us believe there is a huge middle ground where NFB is valuable, where forward paths are not perfect and can profit greatly from the proper application of NFB. Astute NFB designers do not try to fix up a highly non-linear forward path using NFB as a band-aid. Those who do give NFB a bad name.
Scott is also right on the money in regard to measurements and the importance of high-frequency distortion measurements. Notably, he pointed out that in his opinion tighter specs have yielded improvements and that ICs have gotten a lot better.
His comments on the interplay between the amplifier and loudspeaker are excellent, and are especially insightful when applied to tube amplifiers.
Overall a pretty solid guy.
Don’t take my word for it or my interpretation of what he said. Go read it yourself, it’s a good read.
Cheers,
Bob
I’ve had a chance to read the full interview with Scott Frankland. Excellent interview. I agree with nearly every word Scott said. Scott seems to be one of the most knowledgeable and level-headed tube designers around.
I urge all of you to read his comments in the interview – and read them carefully and in-context.
Scott is not an anti-NFB guy.
This, in spite of the fact that some of his tube amps do not use feedback. He does a good job of explaining why he does not use feedback with tube amps, and he is self-consistent in his reasoning. His reasons for not using NFB in tube amps do not necessarily apply to solid state amps.
Scott is correct in pointing out that NFB can make clipping sharper and that it is wise to avoid clipping. I believe that in poorly designed NFB amplifiers, the application of NFB can make clipping ugly as a result of poorly-managed internal overloads.
Scott is also correct in his reasoning about what he refers to as harmonic multiplication – the phenomenon that Baxandall investigated wherein the feeding back of distortion products to the input gives them a chance to multiply with the input signal and create new spectra. If feedback is applied to a forward path with only second and third harmonics, the result can be a spray of harmonics of decreasing amplitude out to very high frequencies.
Notably, Scott points out that the worst case for this phenomenon is a low-feedback situation and that the problem largely disappears as the amount of NFB is increased. This is a subtlety that escapes many well-known amplifier designers. Scott’s beef with using NFB in tube amplifiers is that in a tube amplifier he believes you cannot get enough NFB to get into the zone where harmonic multiplication is not a problem.
In this regard I think it is notable that the legendary McIntosh tube amplifiers succeeded in getting a lot of total negative feedback around the output stage with their unity-coupled output stage which provided quite a bit of local NFB.
Scott’s criteria for using NFB is that the amplifier with NFB does not have any larger higher-order products than there are with the open-loop design under the same conditions. This is not a bad criteria, although once the products get below, say, -130 dB I would not sweat it.
In my view, the key thing to keep in mind is that we rarely close the loop around a forward path that has just seconds and thirds. Just the existence of multiple stages alone gets us to higher-order products in the open loop. Moreover, crossover distortion virtually always produces a spray of high-order products. The higher-order products are there in the open loop in the first place, even if fairly low in amplitude.
Scott’s comments regarding TIM are also right on the money. He rightly points out that feedback will not cause more TIM as long as the stage receiving the feedback does not overload in a worst-case situation. This essentially translates to the amplifier having adequate slew rate. It is a slight over-simplification in regard to soft TIM, but it is close enough.
He is also correct in pointing out one of the paradoxes of negative feedback: NFB works best with a perfectly linear forward path, but then you probably don’t need negative feedback. Of course, many of us believe there is a huge middle ground where NFB is valuable, where forward paths are not perfect and can profit greatly from the proper application of NFB. Astute NFB designers do not try to fix up a highly non-linear forward path using NFB as a band-aid. Those who do give NFB a bad name.
Scott is also right on the money in regard to measurements and the importance of high-frequency distortion measurements. Notably, he pointed out that in his opinion tighter specs have yielded improvements and that ICs have gotten a lot better.
His comments on the interplay between the amplifier and loudspeaker are excellent, and are especially insightful when applied to tube amplifiers.
Overall a pretty solid guy.
Don’t take my word for it or my interpretation of what he said. Go read it yourself, it’s a good read.
Cheers,
Bob
For some here it is difficult to imagine the loudspeaker as a 2 way active transducer. It goes both ways. It emits sonic waves from the application of an input electrical signal, this everybody knows, but it also generates a LOW impedance electrical signal from stored energy in its inductive and capacitive components, stored energy in the resonant box, AND it actually picks up signal from other sources in the room, including its own early sound output.
Now, IF you use an output stage that has a very low output impedance without resorting to global feedback, then these extraneous signals will be suppressed at the output of the power amp, with or without global feedback. My designs do this naturally, and even though I use lots of global feedback, I don't get IIM.
However, IF you have an intrinsically high output impedance, then these extraneous signals coming back from the loudspeaker, have to go back to the input of the amp and will create an extra error signal injected into the feedback loop. This is the problem.
Now, IF you use an output stage that has a very low output impedance without resorting to global feedback, then these extraneous signals will be suppressed at the output of the power amp, with or without global feedback. My designs do this naturally, and even though I use lots of global feedback, I don't get IIM.
However, IF you have an intrinsically high output impedance, then these extraneous signals coming back from the loudspeaker, have to go back to the input of the amp and will create an extra error signal injected into the feedback loop. This is the problem.
- Status
- Not open for further replies.