How to get impedance measurement estimate from speakerbench.com with error <1 #4
Comments
|
Please explain your exact measurement setup. Ideally attach a simple diagram sketching the wiring and where the impedance sense resistor is located. Do you have a real voltage source (e.g. a power amplifier) in the circuit? |
|
And I wonder what the measurement voltage is. The driver is tiny so perhaps it is being driven nonlinearly. |
|
I agreed to take the same driver for repeated measurements. Today he will be with me. The picture shows the calculated REW parameters based on my measurements and the factory ones. |
|
Thanks for sharing the datasheet. It seems the fit by REW is quite good, with a just OK match on DCR (the datasheet could also not be perfect with the driver at hand), an Mms within 0,5% of factory specs and Bl within 1.5%. That's not bad. Beta is the viscoelastic property and it is found to be almost zero (?). In this respect Speakerbench is nice in that it verifies the quality of the data. The input data might fail for several reasons. Adding and removing the masses should be done with great caution in this case, avoiding cone motion in the process (in practice it can be difficult for such a small driver). The impedance sweep should be long enough that the driver stabilizes, if a10 second sweep doesn't cut it, try 30 seconds. Keep the signal low enough to avoid high excursion. For woofers we say 0.25 Volt, I'm not sure what voltage level can be accepted by this little driver. The viscoelasticity is visible when adding/removing masses and you see a drop in the impedance peak, but it is also visible down the slopes of each impedance peak. What the analysis in Speakerbench complains about, is that something is 'off' and it can't all be right. This 'off' can be caused by motion (e.g. when adding or removing masses, or too high a measurement voltage), but it could also be that the driver isn't driven by a true voltage source, as in for example if you have a large DC resistor in series with the driver (anything above 0.5 ohm is considered large). In spite of beta not being detected correctly, you can probably go ahead and do box simulation with the found parameters, just accepting that without beta being detected correctly (and maybe also Re), box-simulations will be less spot-on. If you attempt to find a way for Speakerbench to accept your measurements, probably in the process you will learn how sensitive the added-mass method is to various stimuli and in the process learn to become excellent at performing these measurements, not just for Speakerbench, but in general. It can be a bit painful to walk down this road for a while, until one day you will succeed. |
|
Thanks for the clarification.
I looked at my very first measurements. In some Beta is zero. In some 0.01 0.05 0.005. Is this parameter related to some other simple dependency? To roughly evaluate Beta based on this second one.
There is only test lead resistance in series with the driver, which is currently 0.25 ohms. I have a task to make it less than 0.1 ohm. Only the Audient iD14 sound card is involved in the measurements. There is no additional amplifier. |
|
I have doubts regarding the driver mount. It is attached in a vertical position to the mounting holes of the basket with screws to the measuring stand. It's wooden. But the screws are metal. And they are magnetic. Although they are not involved in movement, can they change the magnetic field of the driver magnet? |
|
Regarding beta, it is not unusual to find it around 0.1, if the surround is low-loss rubber. 0.25 ohm and test leads is not a problem. No need to reach for 0.1 ohm. I think the screws holding the chassis must be acceptable, it is a normal use-case. It is (much) worse if you have screws around the magnet system (shorting the backplate and topplate), then you would immediately loose Bl ... but your measurements indicates, this is not the case (Bl is close to manufacturer specs). We warn against magnetic weights on the cone, since they are moving and the magnetic stray field is always the strongest on the front side of the driver. Your screws holding the basket are static and should not cause a headache. |
|
I calibrated the output voltage of the sound card in REW. The picture shows impedance peaks at high magnification. I just took a series of free air measurements. 120 mV - 1170 mV. In steps of 50 mV, 14 measurements, and then 100 mV. The difference between the first and last peak was 5 ohms. The resonant frequency changed from 143 hertz to 138 hertz. Now it’s clear why the height of the peaks was not enough. Because the measuring voltage was too high. The question then becomes how to determine the 'correct' or range of 'correct' measurement voltages for any driver? |
|
I got measurements from 10 mV to 2270 mV. Below 30 mV there is a lot of noise, but with smoothing the measurements are normal. |
|
Interesting analysis at various levels. I guess you may conclude it is worth trying to measure at or around ca. 100 mV signal level. |
|
Audient iD14 has two modifications. mkI and mkII. The output impedance of the analog outputs and the input impedance of the analog inputs should be the same here and there. The picture shows the specifications of both. As I see, the output impedance is specified <100 ohms. Input >8 kOhm. |
|
The output impedance of 100 ohm is much too high. I hope you can find an amplifier somehow and try with it as a voltage source. |
|
There is an amplifier. hegel h90. |
|
Damping factor 2000. That should work. :-) |
|
It seems to me that any external sound card for musicians, like mine, has an output impedance of about 100 ohms. RME, Motu, Presonus, and others. Many people use them for measurements. So I did the same. I did impedance measurements (not TS parameters) with only the sound card, and then with it and the amplifier earlier. The results were no different. 1 to 1. Therefore, I decided that the amplifier was an extra link. By the way, measuring the values of capacitors and inductors by measuring impedance only with a sound card was always accurate. |
|
Impedances: free air, 83 g, 166 g. Voltage 50-700 mV. The voltage step is signed. Sound card only. |
|
From 700 mV to 300 mV the resistance increases smoothly, after 200 mV there is a sharp increase. Which zone can be called a zone of linearity, which zone can be called a zone of nonlinearity? |
|
These are very nice scans. 50mV-100mV looks linear in the sense that the resonant frequency isn't changing much. What is changing with voltage is the suspension damping -- and it depends on voltage even at very low voltage. So, we could say the driver is quasilinear here. I suspect you will get a good rating to the fit for V < 100mV. What was the voltage used in your original dataset? |
I didn't even think about it. But it was 1.4 volts. |
|
Can you try to run the 70mV scan through Speakerbench? What is impressive here is that you have all the data to do a Speakerbench voltage scan, uncovering the power-dependence of the parameters! |
|
I checked how REW treats different data sets depending on voltage. His model fits the data well from 100 mV to 50 mV. 90 mV and 60 mV are best. |
|
But with such data there is a strong discrepancy with the factory parameters of the TS. |
|
Don't worry about the factory parameters for now. That's a separate can of worms. Typically the factories test at "realistic" voltages that are quite high. Claus can comment further here. |
|
I see everything, but I still can’t wrap my head around the idea that data obtained at such low excitation levels has more weight than data obtained with real, 1 V to 3 V, input signals. Nobody listens to music at such low levels. Even if music listening levels are not the argument here, looking at the driver when the excitation signal arrives, no movement is visible. The sound is audible, but very quiet. The voice coil is located in an air gap, nothing prevents it from moving even with very small signals. But the bending of the rubber suspension should become an insurmountable obstacle to movement if the excitation signal is too small. That is, the driver membrane will not be able to move. Then you will get imperfect TS parameters. |
|
Of course the data obtained at realistic voltages is critically important. What we are doing now is something more theoretical; namely, getting Speakerbench to give you a good rating. When that happens, we know that everything is internally consistent and accurate. There are many reasons unrelated to voltage level that can cause the rating to be poor. Once we have that you can carry out fitting at higher voltage to see which parameters change and which do not. The hope is that, for example, Re, Bl, Mms, etc do not change but C0 and beta will change, giving you a quantitative idea of the driver behaviour versus voltage. |
|
About the driver not moving at very low voltage, I do not think this is an issue. We can prove this if you can get a good fit at low voltage. |
|
This is the Speakerbench data evaluation for 90.2 mV. |
|
Excellent! How do the advanced parameters compare to your original 1.4V fit? |
|
I'll enter the data for 50-100 mV. And for 1.4 V. Then I’ll make an archive and upload it here. |
|
I would like to know if it makes sense to take measurements with an amplifier as Claus recommended? |
|
If you have reached 'Excellent' rating without a power amplifier, then no need to go there. My concern with the output impedance is that this is not a voltage source, but if your setup with REW somehow compensates for this (I don't know), then you're good. Regarding previous comments about voltage vs. listening level. The measurement is for linear parameters. The driver at hand does not support linear behavior at normal listening levels. If you wish for this to be different, then you must select a different driver... The original datasheet is created with the Klippel measurement system, I believe using LPM module. Klippel handles Large Signal (with the LSI module) and possibly LPM is a derived subset. Although Klippel might use a larger signal, the output parameters are still small-signal parameters (Klippel also offers something in-between). |
|
I've read your attached ADV file, and beta = 0.065 - 0.067 sounds plausible. It looks like this parameter is now identified correctly with high accuracy. Re = 6.28 ohm is probably highly accurate too. Bl = 3.75 Tm. Mms = 1.98 gram. It could be interesting to compare the dataset with the manufacturer specs, but I have no doubt the found parameters are accurate. I can't read the attached datasheet (above graphics image) anymore (??). [Later] - Hey, now I can read the above graphics image. We have: The box simulation in Speakerbench was meant to facilitate the use of our advanced parameters for classical (linear) box simulation. I am not aware of that many box simulators that can handle non-linear parameters (LEAP comes to mind), and we haven't really taken that step either ... not yet. Before moving into the nonlinear domain, all parameter identification and box modeling should rest on a foundation of linear parameters, then one can expand modeling from there, at least that's my opinion (I don't see how a non-linear model which gets the simplified linear parameters wrong could ever become correct). |
|
I downloaded the ADV file for myself. How can I use it?
This means that a voltage of 60,70,80 mV creates linear conditions in the driver. Although the 700mV-400mV section also looks linear.
It indicates 1.93 grams, 0.07 ounces in parentheses. There is also a discrepancy in the resonance frequency at high voltage, at 1,5 V 138 hertz. |
The ADV file is for re-upload into the Datasheet Creator. Here you create the final Speakerbench Datasheet (SBD-file).
Yes, your driver at hand is clearly level-dependent 12- 20% higher in resonance frequency than the official datasheet. This is not a measurement error, but due to the driver suspension being stiffer than the driver that was used for datasheet measurement. |
|
Claus, I have two thoughts regarding the discrepancy between the received data and the manufacturer's data. The output resistance of the measuring setup affects the obtained parameters. Then I have to measure again with an amplifier. Second, one part of the parameters should be measured at low voltages, the other at high voltages. |
|
|
For this driver, the linear section was found to be closer to 50 mV. As I understand it, linearity for a driver is when only the impedance peak resistance changes with a change in the input voltage. The peak frequency does not change (almost). Did I understand correctly? |
|
Speakerbench does not only look for the impedance peak, but analyzes the character in a range down the sides of the impedance peaks. The change in resonance frequency depending on the applied voltage is normal in electro-dynamic drivers. For Speakerbench to determine the linear parameters, we just need the measurements to show linear properties across three measurements. Jeff wrote in an earlier message you can 'scan' for higher voltages, but at some point Speakerbench cannot detect linearity in and between the measurements, so you get at 'Failed' rating. |
|
Here is a plot of the variation of fit parameters versus voltage. The curves are normalized to the lowest voltage. Only the suspension resistance (r0) and beta parameter (beta) are varying. These two parameters effectively determine the frequency-dependent damping. This is a really excellent dataset. You could almost publish it.
|
|
Jeff, Claus, thank you for your help in understanding what Speakerbench is and how it evaluates data. |
|
Speakerbench rated the data obtained at a voltage of 50 mV as excellent, and I went to the boxing simulator, determined the required box volume and obtained the estimated frequency response. After making a real box and measuring at higher levels (1.5-3 V), will I get the same frequency response as predicted by Speakerbench? |
|
Speakerbench does not hold any data about the nonlinearity of the driver, sorry. Increasing eg = 2.83 Volt to a higher voltage in the simulation tool simply results in linear scaling. Nonlinearity in the port output is also not part of Speakerbench. |
|
Thank you. |
I did a lot of impedance measurements to get the TS parameters, but didn't get the accuracy I needed.
The text was updated successfully, but these errors were encountered: