Hi Gerrit,
I have not made any comments recently because I have had difficulty in trying to figure out what you are doing.
I thought I'd come back to study those results when I had time to get my head around them - and then when I did there was more - and more - - -
As does Klaus, I too would appreciate more explanation, like a circuit diagram showing source, its waveform and the probe positions where the resullts come from.
Cheers .......... Graham.
I have not made any comments recently because I have had difficulty in trying to figure out what you are doing.
I thought I'd come back to study those results when I had time to get my head around them - and then when I did there was more - and more - - -
As does Klaus, I too would appreciate more explanation, like a circuit diagram showing source, its waveform and the probe positions where the resullts come from.
Cheers .......... Graham.
KSTR said:
But it has to be true that as far as linear performance is concerned it can't make a difference if the load is reactive or not, right?
You are talking about how an amp reacts to large signals where the large signal current and voltage are not handled the same resulting in a nonlinear effect which could not be quantified by a purely resistive load for other loads.
In any case, I typically use a loudspeaker load and then add a resistor. From these two measurements I can get the amps impedance, which, as long as the amp is linear, is valid for any load.
Do you agree?
Re: Red pill or blue pill?
Hmm. Could you, in a single post, summarize the important points in your posts? Don't get me wrong, but there are an awful lot of posts now and it is getting hard to get an overview. I might have missed the bottom line.
Gerrit Boers said:
Hmm. Could you, in a single post, summarize the important points in your posts? Don't get me wrong, but there are an awful lot of posts now and it is getting hard to get an overview. I might have missed the bottom line.
100% agreed. And perfectly feasible, too. One exception maybe open-loop outputs of some amps which really change in their (quite high) Zout signficantly (more than 20% or so) with output current level. Other exception maybe amps which have a deliberatly shaped Zout vs. frequency. The typical NFB amp will have most of its Zout in the wiring to the binding posts and in the protection relay, anyway.gedlee said:
- Klaus
Who's this feedback guy anyway?
Amice,
Please bear with me because I have one last bombshell to drop.
First I will offer you my explanation as to what these two plots represent.
The impedance response curve.
Because Qes is directly proportional to Re, or actually Re+Za (a for amp), what we have here is a plot of Qes versus frequency for a particular driver/load combination. This plot shall henceforth be known as the Gizmo plot.
I reversed the polarity of the mains plug of the NAD and I saw it in the Gizmo plot.
The impedance impulse response.
The timescale of the plots is 200µs. What you see here is the load as the driver sees it within this timeframe. The first point of the plot, t=0, represents the moment the shockwave of the noise hits the membrane of the driver.
I've thinking about why the numbers for the resistance vary for different drivers and I think I know the answer:
The impedance impulse response has the same properties as the acoustic impulse response. It must add up to zero. The value measured at t=0 for a purely resistive load can therefor not exceed 1/2 of the actual value. The measured value tells us something about the bandwith of the source i.e. driver. The sum of the absolute values should be equal to the actual value of the load.
Now for the new metric.
Name of the metric:
Gizmo factor
Essential properties of this metric:
For purely resistive loads the Gizmo factor is equal to the traditional damping factor.
The Gizmo factor is directly proportional to the imaging qualities of a particular loudspeaker amplifier combination. More Gizmo, more imaging.
Measuring the Gizmo factor:
The G-method is used to obtain the impedance impulse response for the amp under test and a precisely known 8Ω resistor. The Gizmo factor is calculated by dividing these two results just as with the traditional method for calculating damping factor.
Now I have to rest because I have slept very little since last thursday. I have found what I've been looking for and I've been able to reduce it to a single number, at least to myself that is.
How about this for symmetry, the audio interface I used for these measurements:
TC Studio Konnekt 48
Amice,
Please bear with me because I have one last bombshell to drop.
First I will offer you my explanation as to what these two plots represent.
The impedance response curve.
Because Qes is directly proportional to Re, or actually Re+Za (a for amp), what we have here is a plot of Qes versus frequency for a particular driver/load combination. This plot shall henceforth be known as the Gizmo plot.
I reversed the polarity of the mains plug of the NAD and I saw it in the Gizmo plot.
The impedance impulse response.
The timescale of the plots is 200µs. What you see here is the load as the driver sees it within this timeframe. The first point of the plot, t=0, represents the moment the shockwave of the noise hits the membrane of the driver.
I've thinking about why the numbers for the resistance vary for different drivers and I think I know the answer:
The impedance impulse response has the same properties as the acoustic impulse response. It must add up to zero. The value measured at t=0 for a purely resistive load can therefor not exceed 1/2 of the actual value. The measured value tells us something about the bandwith of the source i.e. driver. The sum of the absolute values should be equal to the actual value of the load.
Now for the new metric.
Name of the metric:
Gizmo factor
Essential properties of this metric:
For purely resistive loads the Gizmo factor is equal to the traditional damping factor.
The Gizmo factor is directly proportional to the imaging qualities of a particular loudspeaker amplifier combination. More Gizmo, more imaging.
Measuring the Gizmo factor:
The G-method is used to obtain the impedance impulse response for the amp under test and a precisely known 8Ω resistor. The Gizmo factor is calculated by dividing these two results just as with the traditional method for calculating damping factor.
Now I have to rest because I have slept very little since last thursday. I have found what I've been looking for and I've been able to reduce it to a single number, at least to myself that is.
How about this for symmetry, the audio interface I used for these measurements:
TC Studio Konnekt 48
KSTR said:
One more point of clarification. As long as the Zout vs. frequency is linear in Vout and Iout, then my technique will calculate the Zout vs. frequency with no problem. The only issue is that the technique assumes linearity in level, but not with frequency.
Ah, I see. I mistakingly assumed you meant a single point (wrt frequency) measurement. If we do a complete scan vs. frequency with a 2-Ch FFT analyser using noise or whatever stimulus we get the complete complex valued Zout-graph (mag/phase) from the complex transfer function voltage w.r.t. current. This, AFAIK, is actually the method programs like ARTA do impedance plots, usually with a high valued series resistor to approximate a current source.gedlee said:
- Klaus
.... which brings us back to Gerrit's mag(Zout). vs. freq graphs, which look like being created similarily. These, at the moment, seem to suffer from poor s/n-ratio (which is no wonder given the circumstances of the measurement as far as I understand those). This makes interpretation of them very difficult, I'd say.
- Klaus
- Klaus
KSTR said:
Yes, but the added resistor need only be a fraction of the Re of the speaker. The larger the value the better the SNR, but it deosn't have to be large. I use noise, I like noise, I understand noise.
Everyone's invited
There are a couple of things that need to be said. Let's start with the most important one.
All of this would not have been possible without the DIYaudio community. I do not claim the results for myself as I learned something else that is even more important to me: The breakthrough came the moment I put my ego and preconceived ideas (dogma) aside and starting looking at the facts.
I listened to what the driver had to say, it told me to take of my blindfolds. Turns out that there are two elephants in the room, and they are not called feedback, or solid-state. They are called ego and dogma.
To add some more symmetry to this thread I would like to ask Gary Pimm to design a Gizmo probe for the community, because that is what this is all about. The community, synergy.
Here are some specs I could think of:
I used used a 0.1Ω current sense resistor, as the precision of the G-method is also determined by the ratio of the CS sensor and Re, I would like to lower this to 0.01Ω. This means we need lots of gain, 80-100dB for current and 20-60dB for voltage. The amps must be differential i.e. balanced. The output should also be balanced.
Ground loops must be avoided at all cost so I would suggest battery power.
An 8Ω resistor should be also in there, maybe some more, that can be used as load.
Two sets of binding posts and some XLR or jack plugs, that's about it.
Any audio interface with balanced inputs can then be used for data acquisition. For the analysis a dual FFT analyzer used. The analysis software should also be able to provide the noise signal. Then you take the transfer of voltage over current and you get the Gizmo plot. The impedance impulse response can be used to calculate the Gizmo factor.
Let me present another way of looking at this. The formula for calculating Qes actually describes the situation where the driver is shorted, otherwise Re would be infinite. The Gizmo plot and the impedance impulse response are a representation of Qes in the frequency domain as well as the time domain for a specific driver/load combination.
The Gizmo plot is also where the infamous 'black background' is to be found. I hereby give you a metric for black background: The absence of PSU spikes from the Gizmo plot.
Some people may ask when there will be an academic publication, I submit to you that this thread actually is the publication. For me this thread is science in action.
In my previous post I presented a metric and I offer a hypothesis that can be tested:
The Gizmo factor is directly proportional to the quality of the imaging of a, and this is important, specific driver/load combination.
An abstract, more details, suggestions for further research, a dedication and acknowledgments will follow.
I have to get back to listening to music now, I still cannot believe what I'm hearing.
Here's how this work should be referenced:
Downside Up,
Impedance frequency response and impedance impulse response from the driver perspective.
Authors:
Gerrit Boers, Graham Maynard, Earl Geddes, Jean-Michel Le Cléac'h, Sreten, Swante, Gary Pimm, Klaus, Dave Dlugos, Forr, Pan, Eva and the immortal Harvey "Gizmo" Rosenberg.
This is also my thanks to you all, because if I'm not mistaken, your citation index will go through the roof.
Harvey, see you in Sto-Vo-Kor.
Sssh, I think I can hear the wings of a butterfly flapping.
There are a couple of things that need to be said. Let's start with the most important one.
All of this would not have been possible without the DIYaudio community. I do not claim the results for myself as I learned something else that is even more important to me: The breakthrough came the moment I put my ego and preconceived ideas (dogma) aside and starting looking at the facts.
I listened to what the driver had to say, it told me to take of my blindfolds. Turns out that there are two elephants in the room, and they are not called feedback, or solid-state. They are called ego and dogma.
To add some more symmetry to this thread I would like to ask Gary Pimm to design a Gizmo probe for the community, because that is what this is all about. The community, synergy.
Here are some specs I could think of:
I used used a 0.1Ω current sense resistor, as the precision of the G-method is also determined by the ratio of the CS sensor and Re, I would like to lower this to 0.01Ω. This means we need lots of gain, 80-100dB for current and 20-60dB for voltage. The amps must be differential i.e. balanced. The output should also be balanced.
Ground loops must be avoided at all cost so I would suggest battery power.
An 8Ω resistor should be also in there, maybe some more, that can be used as load.
Two sets of binding posts and some XLR or jack plugs, that's about it.
Any audio interface with balanced inputs can then be used for data acquisition. For the analysis a dual FFT analyzer used. The analysis software should also be able to provide the noise signal. Then you take the transfer of voltage over current and you get the Gizmo plot. The impedance impulse response can be used to calculate the Gizmo factor.
Let me present another way of looking at this. The formula for calculating Qes actually describes the situation where the driver is shorted, otherwise Re would be infinite. The Gizmo plot and the impedance impulse response are a representation of Qes in the frequency domain as well as the time domain for a specific driver/load combination.
The Gizmo plot is also where the infamous 'black background' is to be found. I hereby give you a metric for black background: The absence of PSU spikes from the Gizmo plot.
Some people may ask when there will be an academic publication, I submit to you that this thread actually is the publication. For me this thread is science in action.
In my previous post I presented a metric and I offer a hypothesis that can be tested:
The Gizmo factor is directly proportional to the quality of the imaging of a, and this is important, specific driver/load combination.
An abstract, more details, suggestions for further research, a dedication and acknowledgments will follow.
I have to get back to listening to music now, I still cannot believe what I'm hearing.
Here's how this work should be referenced:
Downside Up,
Impedance frequency response and impedance impulse response from the driver perspective.
Authors:
Gerrit Boers, Graham Maynard, Earl Geddes, Jean-Michel Le Cléac'h, Sreten, Swante, Gary Pimm, Klaus, Dave Dlugos, Forr, Pan, Eva and the immortal Harvey "Gizmo" Rosenberg.
This is also my thanks to you all, because if I'm not mistaken, your citation index will go through the roof.
Harvey, see you in Sto-Vo-Kor.
Sssh, I think I can hear the wings of a butterfly flapping.
Ok, so far I don't think I want to be a co-author of that article... 😀
An article needs
Abstract
Introduction
Method
Results
Conclusions
...and there is a good reason for this stereotypic look of most scientific publications, and that is clarity.
An article needs
Abstract
Introduction
Method
Results
Conclusions
...and there is a good reason for this stereotypic look of most scientific publications, and that is clarity.
Re: Who's this feedback guy anyway?
Please consider a change of terminology (or perhaps viewpoint). The "Q" of a resonance is defined, conventionally and for good reason, only at resonance. What you mean is more usually expressed by the loss angle (\phi, say), with \phi_o = 1/Q at resonance. Having loss angle as a function of frequency is the better approach.
Breaking the loss (as a function of frequency) into its various components is a useful technique. This is generally regarded as the standard approach.
Ken
(you probably remember that the Q of a system is defined as the energy stored / energy loss per cycle)
Gerrit Boers said:
Please consider a change of terminology (or perhaps viewpoint). The "Q" of a resonance is defined, conventionally and for good reason, only at resonance. What you mean is more usually expressed by the loss angle (\phi, say), with \phi_o = 1/Q at resonance. Having loss angle as a function of frequency is the better approach.
Breaking the loss (as a function of frequency) into its various components is a useful technique. This is generally regarded as the standard approach.
Ken
(you probably remember that the Q of a system is defined as the energy stored / energy loss per cycle)
Re: Re: Who's this feedback guy anyway?
I'd also like to point out that the impulse response is not defined for impedance. Impedance is a frequency domain concept and impedance is a time domain one. One can of cousre take a trasform of the impedance and call it an impedance impulse response, but such a thing has no recognition in the literature.
kstrain said:
I'd also like to point out that the impulse response is not defined for impedance. Impedance is a frequency domain concept and impedance is a time domain one. One can of cousre take a trasform of the impedance and call it an impedance impulse response, but such a thing has no recognition in the literature.
Graham Maynard said:
Hi Graham!
If the force (current) that drives the voice coil becomes more linear, then the displacement will be more linear as well.. and that can be illustrated by the lower harmonic and intermodulation distortion when a driver is fed via a high'ish impedance (in part of it's range) in contrast to a "zero impedance".
Must also say that my experience is the opposite compared to yours regarding the last part of your post. 🙂
/Peter
Gerrit,
Adding to the points made by Svante, Ken and Earl (which are -- more or less -- "only" semantic issues), I'd like put emphasis (again) on the noise problems I see in the plots, like this one.
I mean, those graphs, when taking them literally, would indicate sudden output resistance jumps of all test objects (including the plain old resistors) of several ohms all over the place in MF regions. If that really were the case one would hardly want to listen to such an amp. I doubt those graphs are truly repeatable, I suspect you'll get overall similar (on a coarse scale) but still very different (on a more detailed scale) "hash" with these graphs each time you repeat a measurement, don't you?
Given the very small signal levels involved and the gains needed for those, a synced averaging measurement approach (as described by Earl elsewhere in this forum) seems to be required to get i) the random noise level down by some 20dB and ii) to average out spuriae, hum, etc.
Further, I feel some sanity checks are in order, like measuring the output impedance basically in the same way, but using a voltage source directly and a resistor, and adjusted to levels/impedance comparable of what you get when using the speaker as a microphone (which is what I understand you did... still lacking a simple test setup drawing from your side, let alone exact details allowing other to perfectly recreate your findings -- I might try some own experiments if I find the time on the holiday this thursday).
And then using different values for the resistors to see if/how the amp reacts to the different loading (with the tube amp, mainly because of its output xformer, this coulb be an issue). This last point, that is that the load itself could make a difference for the actual drive impedance as seen by the load, is the only reason I can think of why one would get different Zout vs freq. graphs when using the speaker as the test signal source instead of a voltage source and resistor(s). One should keep in mind that this "backward-drive" type of measurements only test one point of the large signal conditions of the amp output, and that is voltage=zero, current=zero (unless we introduce bias conditions for those). Which is only part of the whole deal...
- Klaus
Adding to the points made by Svante, Ken and Earl (which are -- more or less -- "only" semantic issues), I'd like put emphasis (again) on the noise problems I see in the plots, like this one.
I mean, those graphs, when taking them literally, would indicate sudden output resistance jumps of all test objects (including the plain old resistors) of several ohms all over the place in MF regions. If that really were the case one would hardly want to listen to such an amp. I doubt those graphs are truly repeatable, I suspect you'll get overall similar (on a coarse scale) but still very different (on a more detailed scale) "hash" with these graphs each time you repeat a measurement, don't you?
Given the very small signal levels involved and the gains needed for those, a synced averaging measurement approach (as described by Earl elsewhere in this forum) seems to be required to get i) the random noise level down by some 20dB and ii) to average out spuriae, hum, etc.
Further, I feel some sanity checks are in order, like measuring the output impedance basically in the same way, but using a voltage source directly and a resistor, and adjusted to levels/impedance comparable of what you get when using the speaker as a microphone (which is what I understand you did... still lacking a simple test setup drawing from your side, let alone exact details allowing other to perfectly recreate your findings -- I might try some own experiments if I find the time on the holiday this thursday).
And then using different values for the resistors to see if/how the amp reacts to the different loading (with the tube amp, mainly because of its output xformer, this coulb be an issue). This last point, that is that the load itself could make a difference for the actual drive impedance as seen by the load, is the only reason I can think of why one would get different Zout vs freq. graphs when using the speaker as the test signal source instead of a voltage source and resistor(s). One should keep in mind that this "backward-drive" type of measurements only test one point of the large signal conditions of the amp output, and that is voltage=zero, current=zero (unless we introduce bias conditions for those). Which is only part of the whole deal...
- Klaus
Hi Pan,
The energies stored within LS cone suspensions and as moving mass momentum are subtracted from the transduced voice coil energy intended to displace air.
Where the drive is purely current it cannot displace air as linearly due to those cone mass and suspension subtractions.
However with voltage drive the current does not fall until back-EMF is generated due to voice coil movement, thus voltage drive better overcomes cone mass and suspension losses, BUT, traps that additional energy within the LS system - as Gerrit's early tests showed.
An optimum impedance can balance current/voltage drive and thus improve transient air displacement linearity versus the stored energies which additionally modify steady sine amplitude linearity.
As soon as you mention harmonic distortion you are talking about steady-sine excitation and examination, which is after the dynamic losses have been overcome, but before the stored energies which modify a frequency response are released !!!
The dynamic transduction errors arise before any response could stabilise sufficiently for quantifiable sine based measurements to be made; ie. they are missed and not reported, even though their effects are plainly audible.
If you do not agree then our experiences must be different.
Until effects are specifically demonstated in isolation, they cannot become recognisably identifiable.
Waveform linearity relates to transients as well as sines !
Cheers......... Graham.
The energies stored within LS cone suspensions and as moving mass momentum are subtracted from the transduced voice coil energy intended to displace air.
Where the drive is purely current it cannot displace air as linearly due to those cone mass and suspension subtractions.
However with voltage drive the current does not fall until back-EMF is generated due to voice coil movement, thus voltage drive better overcomes cone mass and suspension losses, BUT, traps that additional energy within the LS system - as Gerrit's early tests showed.
An optimum impedance can balance current/voltage drive and thus improve transient air displacement linearity versus the stored energies which additionally modify steady sine amplitude linearity.
As soon as you mention harmonic distortion you are talking about steady-sine excitation and examination, which is after the dynamic losses have been overcome, but before the stored energies which modify a frequency response are released !!!
The dynamic transduction errors arise before any response could stabilise sufficiently for quantifiable sine based measurements to be made; ie. they are missed and not reported, even though their effects are plainly audible.
If you do not agree then our experiences must be different.
Until effects are specifically demonstated in isolation, they cannot become recognisably identifiable.
Waveform linearity relates to transients as well as sines !
Cheers......... Graham.
- Status
- Not open for further replies.