gpdb
Meter Reader 1st Class
Posts: 66
Likes: 5
|
Post by gpdb on Aug 9, 2022 13:19:05 GMT -5
I would assume there's a simple answer to this, but I haven't found a consistent one on the internet. Is there a specific way to calculate Q-factor as a single number? It would make it much easier to see the differences in Q in comparison bode plots. Any help would be appreciated.
|
|
|
Post by antigua on Aug 9, 2022 20:24:59 GMT -5
It's a bit of a pain, but you do as described here www.learningaboutelectronics.com/Articles/Quality-factor-calculator.php#answer , You start at the peak, go downwards -3dB (assuming that the peak is at least 3dB in height, which is often not even the case), and there will be two frequency points that intersect the X-axis there (this is the bandwidth at -3dB of the peak), you plug them into the calculator, and select "bandpass filter", and "calculate" . For example, the unloaded peak of this Firebird bridge is at 10.4kHz, at -3dB down from there, the left frequency is about 8500Hz and the right side is about 12000Hz , so if I put in those values the resulting Q factor is 2.8. Unfortunately the Velleman bode plot isn't super accurate, so the Q factor calculation is pretty crude. The most accurate method would be to look at the actual data file to get the recorded amplitudes and frequencies at the resonance, but that's a lot of work for not a lot of benefit. On the other hand the Velleman makes it easy to measure the amplitude of the peak with the double Y markers, you set one marker at the baseline of inductive output, and the other at the top of the peak. In this case, it says 9.4dB for the unloaded bridge. That can be more useful than the Q factor, because it means that the resonance, if you could somehow hear it, is a whole 9dB louder than everything preceding the resonant peak, and in the loaded context, 6dB louder at 4.3kHz, which is good practical information to have. It tells me I'm going to want to work with the tone knob or use 250k pots with these pickups. The eddy currents throw a wrench in the Q factor aspect, though. With a pickup like a Filter'tron, both the Q factor and the amplitude at resonance don't reflect the low-pass attenuation profile of the pickup. This causes Filter'trons to have a higher proportion of low end to high end, despite the low inductance and high resonant peak. In a case like this, the fact that it has any Q factor can be misleading because it will sound more like a pickup without any Q factor. For this measurement, I noted that the amplitude at resonance is actually below that of the inductive response, -1.7dB, and when you look at that along with the 4.46kHz cut off, it tells you the pickup has a lot of bass as well as treble response, because 4.46kH is a rather high peak frequency for a guitar pickup, thanks to the fact that Filter'trons have a very low inductance. The negative amplitude at resonance tells me that 500k or 1meg pots are probably needed to prevent the pickup from having too much bass and sounding too muddy. And if the negative amplitude at resonance is really low, like this Chinese neck pickup that had a brass cover and steel pole pieces, with a peak amplitude of -4.4dB, and a resonant peak of only about 3.7kHz, then I know this pickup will sound muddy no matter what, even with 1meg pots it will sound muddy. So for a simple Fender style single coil, you can know almost all there is to know just by looking at the measurement numbers, but if a pickup has lots of steel parts or conductive covers, like the Filter'tron, you have to have a bode plot to reveal what attenuation is caused by the eddy currents. For pickups with brass covers, there's usually no Q factor, the eddy currents load the resonance down to nothing, and you just have a low-pass filter with no resonance. The Filter'tron seems to have a interesting profile because the "H" hole cover is designed to reduce eddy currents, but the large filister screws in the pickup still cause substantial eddy currents, but with a different geometry than a cover. The geometry of the conductive metal parts around the coil, as well as their permeability and resistivity all factor into the degree of attenuation and the frequencies where they're more prominent, it's a complexity that can only be seen in a bode plot.
|
|
|
Post by aquin43 on Aug 10, 2022 6:13:43 GMT -5
I would assume there's a simple answer to this, but I haven't found a consistent one on the internet. Is there a specific way to calculate Q-factor as a single number? It would make it much easier to see the differences in Q in comparison bode plots. Any help would be appreciated. The Q of a pickup is not at all well defined. A useful reference is The peak in the response will colour the sound of the pickup, even when the peak is below the low frequency response as in a pickup with eddy current losses or a metal cover. The pickup can be reasonably closely modelled as three filters in cascade; a low pass shelving filter from the cover, another low pass shelving filter from the eddy current losses and a peaked second order filter from the coil and capacitive load.
None of the "loaded Q" stuff seems to be of much use either, unless you are content to use the guitar with the volume always at full. Turning down the volume by even a few dB completely alters the pickup loading and usually moves the resonant peak considerably while adding a further low pass filter of the pot and the cable capacitance.
|
|
timtam
Meter Reader 1st Class
Posts: 53
Likes: 24
|
Post by timtam on Aug 10, 2022 6:33:58 GMT -5
It's a bit of a pain, but you do as described here www.learningaboutelectronics.com/Articles/Quality-factor-calculator.php#answer , You start at the peak, go downwards -3dB (assuming that the peak is at least 3dB in height, which is often not even the case), and there will be two frequency points that intersect the X-axis there (this is the bandwidth at -3dB of the peak), you plug them into the calculator, and select "bandpass filter", and "calculate" . For example, the unloaded peak of this Firebird bridge is at 10.4kHz, at -3dB down from there, the left frequency is about 8500Hz and the right side is about 12000Hz , so if I put in those values the resulting Q factor is 2.8. Unfortunately the Velleman bode plot isn't super accurate, so the Q factor calculation is pretty crude. The most accurate method would be to look at the actual data file to get the recorded amplitudes and frequencies at the resonance, but that's a lot of work for not a lot of benefit. If you were to add links to the those data files to the website I am sure one of us could whip up an Excel sheet to calculate the Q factor. I've done lots of curve feature location/extraction jobs in Excel (all with functions rather than VBA code). Finding the peak and the two -3dB frequencies, and then doing the Q factor calculation, is easy enough. And the checks you described - to ensure it's a valid number - would not be too hard to build in too. When I did the quartile-based 'box-and-whisker' plots of your data to work out the typical low/medium/high ranges for DCR, inductance, resonant frequency and capacitance in each pickup class, I really felt like the Q factor was the missing piece.
|
|
|
Post by aquin43 on Aug 10, 2022 9:26:19 GMT -5
The 3dB down points measurement is for a bandpass filter. The pickup is a low pass filter, not a bandpass. If you are determined to consider it as a simple damped second order filter, which is approximately true for a strat pickup but for few others, the Q is pretty much equal to the ratio of the peak response to the low frequency response, converted back from dB. Thus a peak of 10dB implies a Q of just over 3 and 20dB a Q of 10. A Q of 0.707 or below will give no peak. A big problem is that the plotter doesn't have direct access to the coil but is pre-filtered by the cover and the eddy current loss path so the measurement of low values of Q is pretty meaningless.
|
|
|
Post by ms on Aug 10, 2022 10:29:14 GMT -5
Given the complexities as shown in Zollner's paper, here is reasonable method to measure the Q of a pickup for a relative guide to its sound.
First pick a frequency at which to measure the Q. I choose 3 KHz because it is near the middle of the range of resonant frequencies achieved in practice and near the maximum sensitivity of human hearing.
Next measure the complex impedance as a function of frequency, and look at the resulting curves to make sure that you have a good measurement. This uses the pickup alone, no loading, and the measurement device should not place a significant load across the pickup. The method Zollner shows is good. I prefer a method requiring no active electronics except the recording interface and computer.
Now compute 2*pi*f*L/R; this highlights the ratio of energy storage to energy dissipation. L is the the effective inductance at 3 KHz, derived from the imaginary part of the complex impedance at 3 KHz, L = 2*pi*f*Zi. It can be different from the low frequency inductance because of eddy current losses. R is the real part of Z at 3KHz, and it also includes eddy current losses. The resulting Q can be as high as about 8 for very low loss ferrite cores and no metal parts in pickup. Of course it is much lower when there is lots of conductive metal in the pickup, especially steel cores and a thick high conductivity cover.
|
|
|
Post by aquin43 on Aug 10, 2022 11:19:02 GMT -5
Given the complexities as shown in Zollner's paper, here is reasonable method to measure the Q of a pickup for a relative guide to its sound. First pick a frequency at which to measure the Q. I choose 3 KHz because it is near the middle of the range of resonant frequencies achieved in practice and near the maximum sensitivity of human hearing. Next measure the complex impedance as a function of frequency, and look at the resulting curves to make sure that you have a good measurement. This uses the pickup alone, no loading, and the measurement device should not place a significant load across the pickup. The method Zollner shows is good. I prefer a method requiring no active electronics except the recording interface and computer. Now compute 2*pi*f*L/R; this highlights the ratio of energy storage to energy dissipation. L is the the effective inductance at 3 KHz, derived from the imaginary part of the complex impedance at 3 KHz, L = 2*pi*f*Zi. It can be different from the low frequency inductance because of eddy current losses. R is the real part of Z at 3KHz, and it also includes eddy current losses. The resulting Q can be as high as about 8 for very low loss ferrite cores and no metal parts in pickup. Of course it is much lower when there is lots of conductive metal in the pickup, especially steel cores and a thick high conductivity cover. That is a good way of measuring the intrinsic Q of the inductance and should provide a useful figure of merit. Very few here seem to measure the pickup impedance even though it can give the most direct insight to the possible behaviour of the pickup as a filter.
For example, the Gretsch filtertron shown above doesn't have a quantifiable Q when measured via the exciter but it is obvious from the shape of the curve that the core of the pickup remains pretty lively up to 5kHz and the sound of its interaction with any capacitive load will be there and will be heard even through the shielding and eddy current low pass filtering that is evident in the Bode plot. Such filtering might easily be offset by the amplifier tone controls.
|
|
|
Post by antigua on Aug 10, 2022 11:31:41 GMT -5
None of the "loaded Q" stuff seems to be of much use either, unless you are content to use the guitar with the volume always at full. Turning down the volume by even a few dB completely alters the pickup loading and usually moves the resonant peak considerably while adding a further low pass filter of the pot and the cable capacitance. I think a lot of players keep the volume at "10", which is why a big deal is made of 250k vs 300k vs 500k pots. Ideally everyone might have 1 meg pots and always keep the tone control down a hair, but few players do that in practice. If you were to add links to the those data files to the website I am sure one of us could whip up an Excel sheet to calculate the Q factor. I've done lots of curve feature location/extraction jobs in Excel (all with functions rather than VBA code). Finding the peak and the two -3dB frequencies, and then doing the Q factor calculation, is easy enough. And the checks you described - to ensure it's a valid number - would not be too hard to build in too. When I did the quartile-based 'box-and-whisker' plots of your data to work out the typical low/medium/high ranges for DCR, inductance, resonant frequency and capacitance in each pickup class, I really felt like the Q factor was the missing piece.
I have the data for the majority of them most likely. I usually save the data and the plot image. It would be an all-day project to upload them and link them, though, and there's no distinction between neck, bridge loaded and unloaded in the data file, because Velleman doesn't have a utility to allow that, so it would be labor intensive to process the data. Most of these pickups have rather predictable Q factor based on the construction, and the DC resistance to a lesser extent. So many pickups on the market are just a PAF based humbucker with all the same parts, and the Q factor is almost always the same, only differing depending on whether there is a cover, so I think it would probably be better to give an approximation, like the Q factor is approximately ~1 or ~2.5, or in many cases, zero. I know there is a pickup maker out there who talks a lot about Q factor as being an important metric of a pickup on his website, like a defining characteristic of the pickup, but you find that most all Strat pickups, or pickups of a given type, have the same Q factor because they use all the same metal parts and types of metals. And as said above, the inherent unloaded Q factor of the pickups is changed by the voltage division tone and volume pots, and it's only at a maximum when both volume and tone pots are fully up.
|
|
|
Post by Yogi B on Aug 10, 2022 17:21:05 GMT -5
I haven't found a consistent one on the internet. Seems like that applies equally here. First pick a frequency at which to measure the Q. I choose 3 KHz because it is near the middle of the range of resonant frequencies achieved in practice The trouble with picking a fixed frequency is that you don't want to be too near the resonant frequency, with typical unloaded pickups that's quite unlikely but if 3kHz were very close to the resonant frequency you could (incorrectly) calculate a small, possibly even zero, Q from the method you describe. I think it would probably be better to give an approximation, like the Q factor is approximately ~1 or ~2.5, or in many cases, zero. Following on from the above, explaining why a Q of zero is an obviously incorrect result — a Q of zero would imply the transition between a filter's passband & stopband would span an infinite frequency range — though that's probably not the best description. Basically, if we had a first order filter (a rolloff of 6dB/octave in the stopband) with zero Q theoretically it wouldn't have either a passband or stopband but a constant rolloff of 3dB/octave at all frequencies — that's obviously problematic since there's an infinite number of octaves between 0Hz and any other frequency. Or another way of thinking about it borrows ms's definition of Q for a series inductor & resistor: for zero Q either the inductance must be zero or the resistance must be infinite — neither of which result in a frequency dependent filter.
It still has issues, but a method to determine Q that also works for lower values is based upon group delay (negative of the derivative of phase with respect to angular frequency): \begin{aligned} Q &= \tau_{g0} \times {\omega_0 \over 2} \\[1.5em] &= \left.-{d\phi \over d\omega}\right|_{\omega=\omega_0} \times {\omega_0 \over 2} % \\[1.5em] % &= \left.-{d\phi \over df}\right|_{f=f_0} \times \left.{df \over d\omega}\right|_{\omega=\omega_0} \times {2 \pi f_0 \over 2} % \\[1.5em] % &= \left.-{d\phi \over df}\right|_{f=f_0} \times \left.{d \over d\omega}{\left(\omega\over2\pi\right)}\right|_{\omega=\omega_0} \times {2 \pi f_0 \over 2} % \\[1.5em] % &= \left.-{d\phi \over df}\right|_{f=f_0} \times {1 \over \bcancel{2 \pi}} \times {\bcancel{2 \pi} f_0 \over 2} \\[1.5em] &= \left.-{d\phi \over df}\right|_{f=f_0} \times {f_0 \over 2} \end{aligned} Where f 0 can be approximated by the frequency where half the maximum phase change occurs, i.e. − 3/ 4 pi (−135°) assuming a third order lowpass.
|
|
|
Post by ms on Aug 10, 2022 18:39:39 GMT -5
I haven't found a consistent one on the internet. Seems like that applies equally here. First pick a frequency at which to measure the Q. I choose 3 KHz because it is near the middle of the range of resonant frequencies achieved in practice The trouble with picking a fixed frequency is that you don't want to be too near the resonant frequency, with typical unloaded pickups that's quite unlikely but if 3kHz were very close to the resonant frequency you could (incorrectly) calculate a small, possibly even zero, Q from the method you describe. I forgot to mention that I compute the parallel C and then "unparallel" its impedance from the measured impedance. Then there is no problem with the resonance. Edit: A couple of examples of Q measured in this way (Q3K) are shown here: guitarnuts2.proboards.com/thread/9877/measuring-high-performance-strat-pickups
|
|
timtam
Meter Reader 1st Class
Posts: 53
Likes: 24
|
Post by timtam on Aug 10, 2022 20:55:07 GMT -5
Reading the Zollner and Zwicker paper (linked above) on Q again, it seems that after they canvas all the complex issues around trying to arrive at a valid measure of Q (which I don't pretend to fully understand), they do hit on a pragmatic approach. And indeed one that they have used elsewhere (eg in the WRHB analysis series). That is, the simple 'peakedness' of the impedance frequency response curve - the peak height divided by an estimate of its width (from the two -3dB points). Their preference for that impedance curve (over the more common bode plot ?) seems based on their assessment that a simple OP amp circuit is 'easier' than an exciter setup; and perhaps also with it being a more 'hill-like' curve, with less ambiguous -3dB points (than an integrated or unintegrated bode plot ?).
But it doesn't seem that any of that really precludes making one's own pragmatic choice to estimate 'Q' as the peakedness of a bode plot's resonant peak with -3dB points, with the practical limitations antigua outlined.
Coincidentally, I was reminded recently that Gibson used to quote Q factors on their pickups, at a brief moment in time (now passed) when they provided copious technical detail on them. Although curiously inductance was omitted, and AFAIK no detail on how the resonant frequency was measured was provided (loading etc ?). And the quoting figures of like DCR to multiple decimal places, and resonant frequency down to 1Hz, suggests they were perhaps just one-off measurements of a single pickup. For the LP Standard 2015:
|
|
|
Post by antigua on Aug 10, 2022 21:44:12 GMT -5
Coincidentally, I was reminded recently that Gibson used to quote Q factors on their pickups, at a brief moment in time (now passed) when they provided copious technical detail on them. Although curiously inductance was omitted, and AFAIK no detail on how the resonant frequency was measured was provided (loading etc ?). And the quoting figures of like DCR to multiple decimal places, and resonant frequency down to 1Hz, suggests they were perhaps just one-off measurements of a single pickup. For the LP Standard 2015: With that resonant peak, I'd guess there was a guitar cable involved, something in the area of 470pF capacitance results in that peak frequency if the inductance is between 4 and 5 henries, which can be inferred from the DC resistance, turn count and wire type. The Q factors, being 5.4 and 5.35, I can't even imagine how they came up with that. Even without a cover, the Q factor of most PAF types hovers around 1 using the -3dB corner freq. method, as well as when I measure the Q at 1kHz with the DE-5000 LCR meter. I think it would probably be better to give an approximation, like the Q factor is approximately ~1 or ~2.5, or in many cases, zero. Following on from the above, explaining why a Q of zero is an obviously incorrect result — a Q of zero would imply the transition between a filter's passband & stopband would span an infinite frequency range — though that's probably not the best description. Basically, if we had a first order filter (a rolloff of 6dB/octave in the stopband) with zero Q theoretically it wouldn't have either a passband or stopband but a constant rolloff of 3dB/octave at all frequencies — that's obviously problematic since there's an infinite number of octaves between 0Hz and any other frequency. So if you have a constant roll off, would you rather put a Q of "N/A" or like <0.1 ?
|
|
|
Post by ms on Aug 11, 2022 7:11:39 GMT -5
Reading the Zollner and Zwicker paper (linked above) on Q again, it seems that after they canvas all the complex issues around trying to arrive at a valid measure of Q (which I don't pretend to fully understand), they do hit on a pragmatic approach. And indeed one that they have used elsewhere (eg in the WRHB analysis series). That is, the simple 'peakedness' of the impedance frequency response curve - the peak height divided by an estimate of its width (from the two -3dB points). Their preference for that impedance curve (over the more common bode plot ?) seems based on their assessment that a simple OP amp circuit is 'easier' than an exciter setup; and perhaps also with it being a more 'hill-like' curve, with less ambiguous -3dB points (than an integrated or unintegrated bode plot ?). I think the more "hill-like" curve is the deciding factor in favor of the impedance curve. Even if you take out the 6 db per octave rise of the bode plot, you still are left with a low pass response, and there is no simple way to derive the Q accurately unless it is very high. The impedance curve is better; it is much less asymmetrical, falling to near zero on the high end and to the dc resistance on the low end. In other words, not perfect when the Q is low, but better.
|
|
|
Post by Yogi B on Aug 13, 2022 18:55:44 GMT -5
The trouble with picking a fixed frequency is that you don't want to be too near the resonant frequency, with typical unloaded pickups that's quite unlikely but if 3kHz were very close to the resonant frequency you could (incorrectly) calculate a small, possibly even zero, Q from the method you describe. I forgot to mention that I compute the parallel C and then "unparallel" its impedance from the measured impedance. Then there is no problem with the resonance. That'll certainly help. Something I didn't think about is that a fixed frequency is necessary because the assumption is that it's an ideal inductor & ideal resistor thus theoretically the numerator, Im(Z), would be directly proportional to frequency whereas the denominator, Re(Z), would be constant (and equal to DCR) — thus Q would be directly proportional to frequency. For real pickups this isn't true, Re(Z) isn't constant (even after removing the affect of the capacitance), but certainly neither is it directly proportional to frequency. ...as well as when I measure the Q at 1kHz with the DE-5000 LCR meter I think I've mentioned this before, and you probably already know anyway, the DE-5000 appears to use Q = Im(Z) / Re(Z) — plus I don't know if even knowing that fact is really all that helpful. Edit: A couple of examples of Q measured in this way (Q3K) are shown hereThat's also interesting because it includes the phase responses, which is pretty rare. It also shows why my approach could be problematic in practice: I have no idea what's causing the 360° phase shift between 20kHz & 60kHz on the regular Strat pickup (I assume the vertical line is due to phase wrapping -270° = +90°). (The greater than zero phase at low frequencies I'm putting down to influence from noise.) I did think of an alternative approximation of f 0 being where the group delay was at a maximum, that would be immune to changes in the order of the filter (e.g. your ferrite pickup with virtually no eddy losses is only a 2 nd order filter thus the group delay measurement would be at -90°, and I'd still say same with the regular Strat pickup though the aforementioned peculiarities don't help). However, much like using peak magnitude, this frequency could be much too low with low Q — or (again, like peak magnitude) non-existent. A problem common to any approach, is that there isn't really a value for Q — there's multiple overlapping filters, each with their own Q. When they're close enough we can treat them as a single number, but that breaks down when they're vastly different. This is made painfully obvious by pickups such as Filtertrons where it would make more sense to say that they have, for example, a Q of about 2 and a Q of about 0.25. Following on from the above, explaining why a Q of zero is an obviously incorrect result — a Q of zero would imply the transition between a filter's passband & stopband would span an infinite frequency range — though that's probably not the best description. Basically, if we had a first order filter (a rolloff of 6dB/octave in the stopband) with zero Q theoretically it wouldn't have either a passband or stopband but a constant rolloff of 3dB/octave at all frequencies — that's obviously problematic since there's an infinite number of octaves between 0Hz and any other frequency. So if you have a constant roll off, would you rather put a Q of "N/A" or like <0.1 ? Probably the latter, but my last remark of that quote should've covered why constant rolloff isn't actually something you'd see in reality. Sure, losing a fixed number of decibels per octave off to infinity is fine, but traverse the frequency spectrum in the other direction and it's equivalent to gain per octave — so at some point that'd have to stop.
|
|
|
Post by aquin43 on Aug 14, 2022 11:09:39 GMT -5
What are we measuring the the Q for?. Take the Filtertron above. If it is merely the frequency response in isolation that is required then there is no Q because Q is not defined for such frequency responses. If, however, the interaction of this pickup with other components is desired, then it is clear that the frequency response can be split into two parts in cascade. There is a non-resonant low pass filter followed by a low-pass resonator. The resonator definitely has a fairly high Q and will ring when excited by transients. The way that this behaviour comes about through the mediation of the eddy current loss path has been well aired on this forum. Most pickups will have a reasonably high Q (be underdamped) and will ring when excited by an impulse applied to their output terminals even when their Bode plot seems to be very flat and lossy. I see this as further justification for measuring the Q from the impedance curve at the output terminals, which is the only way of getting direct acces to the LCR resonator. Since the essential parameters L and R that determine the Q vary with frequency, a proper characterisation of the pickup would either model the lossy inductance in sufficient detail and abandon Q altogether or measure Q at several frequencies in the expected working range.
|
|
|
Post by antigua on Aug 14, 2022 14:10:04 GMT -5
...as well as when I measure the Q at 1kHz with the DE-5000 LCR meter I think I've mentioned this before, and you probably already know anyway, the DE-5000 appears to use Q = Im(Z) / Re(Z) — plus I don't know if even knowing that fact is really all that helpful. So if you have a constant roll off, would you rather put a Q of "N/A" or like <0.1 ? Probably the latter, but my last remark of that quote should've covered why constant rolloff isn't actually something you'd see in reality. Sure, losing a fixed number of decibels per octave off to infinity is fine, but traverse the frequency spectrum in the other direction and it's equivalent to gain per octave — so at some point that'd have to stop. The Q @ 1kHz measurement has been effective at telling apart brass and nickel silver covers. Q factor in a pickup is useful for determining whether to pair it with 250k, 500k or 1meg pots, lest you get a peak in the guitar sound that is audible, but AFAIK, not a whole lot else. I'm pragmatic, I don't have use for more esoteric measurement values and methods, and I don't imagine gpdb does either if he's making a public facing data set. If there is a practical use for the Q factor of a pickup modelled with pure inductivity for example, I don't know what it is, I don't know how it informs me as a consumer of guitar pickups.
|
|