Thiele-Small Parameters measurement using dual added mass method.

sm52

Member
Joined
Mar 14, 2019
Messages
462
the three files M0, M1 and M2
There are no error messages. Only the technical data file at the time of the error from firefox. Three files. Free air, 2.26g, 4.5g.
 

Attachments

sm52

Member
Joined
Mar 14, 2019
Messages
462
Bernard, thank you. I now have data from REW, Speakerbench and the driver manufacturer. In short, the REW data is much closer to the manufacturer's data.
 

sm52

Member
Joined
Mar 14, 2019
Messages
462
John Mulcahy,
found that after measuring the impedance, the component model is calculated differently for different cases. If the model only includes a resistor and an inductance, in my case, the REW shows an inductance of 2.011 mkH. If there are more elements in the model, the inductance is 188 or 185 mkH. The real one should be a little more than 200 mkH. That is, 2.011 mkH should probably be multiplied by 100.
 

Attachments

Last edited:

sm52

Member
Joined
Mar 14, 2019
Messages
462
Based on my measurements in post #54, REW calculated a Qts of 1,196. The manufacturer gives 0.78. Bl REW 2.726, the manufacturer gives 3.04. Can such a large discrepancy between these parameters be caused by different conditions before measurements? The manufacturer pre-applied a 20W stepped sine wave for 20 hours, and I didn't.
 

jcandy

Registered
Joined
Dec 3, 2017
Messages
11
Location
San Diego
@sm52

I'm sorry you're having instability issues with Speakerbench. As John noted, Speakerbench doesn't require javascript. We use the Dash/Plotly framework to serve the apps via your browser, so it should work on your phone, tablet or PC without any special libraries or apps. One thing I can suggest is (carefully!) clearing your browser history and trying again. If you still have issues I'll do my best to troubleshoot.
 

jcandy

Registered
Joined
Dec 3, 2017
Messages
11
Location
San Diego
@sm52

Your ZMA files appear to be using 122K frequency points! That is on the order of 100X larger than what we are used to dealing with, so it may be stretching your browser memory.
 

sm52

Member
Joined
Mar 14, 2019
Messages
462
jcandy
I did smoothing 1/24, 96 PPO. The number of lines as and before. This time I got it. But. In this case, the error Sorry. Then I got everything without smoothing. In this case, the Fair error. That is, smoothing reduces accuracy and increases error.
What I noticed. Previously, I entered the values of two weights, then loaded the zma files. This time I loaded the zma files, then entered 2 weights. Loading now and before is fine. The browser thinks for half a minute, a json file appears. But when you hover over it, there is no way to get a download link. At the same time, the processor is overloaded, it considers something, it overheats. The reason is overheating during incomprehensible calculations. This time I did nothing, I waited for about three minutes. The link appeared, the json file downloaded in half a second. Further, when you click on the FIT tab, the browser hangs for 1-2 minutes. then everything works.
Now a question. Why need to download the json file to computer and then upload it to your site? If your site made a json file, let it move it to the FIT tab. Without download or upload.
 

jcandy

Registered
Joined
Dec 3, 2017
Messages
11
Location
San Diego
@sm52 I don't know that there's much we can do to make this process easier or less painful for such large ZMA files. The dataset is considerably reduced in size before the fit, so one possibility would be for us to add the option to apply the data reduction before offering the Z-file for download. If I see other users regularly using large ZMA files I might consider this. But as I mentioned before, keep in mind that your ZMA file is about 100X larger than it needs to be.

About your question regarding carrying the Z-file forward automatically to the fit: the web programming framework makes this more complicated than you might imagine. To do so I would need to define a new internal browser memory space and make it shared across callbacks. We had initially thought to do this, but discovered that forcing the user to download the Z-file enforces a useful workflow such that one is left with a data record after fitting session is complete.
 

Bernard

Member
Thread Starter
Joined
May 22, 2017
Messages
108
Hello Jeff, sm52,
The other day I uploaded sm52's files, Speakerbench worked properly, and I posted the results.
My pc is running Windows 11 Pro, version 21H2, build 22000.527. MS Edge is my browser and I also use NordVPN when I want to watch F1 for free on RTS, the French speaking Swiss TV. ;-)
Upload worked fine with and without VPN.

Yesterday, I installed the latest Firefox 64 bits version and I redid the test with sm52's files.
Everything worked fine until I wanted to download the Json file.
The screen froze and even with Alt+Ctrl+Del I couldn't stop the Firefox program.
I had to shut down the pc wildly by holding down the power button.
When I rebooted, I uninstalled Firefox and returned to Edge as my default browser.
I didn't try to figure out why Firefox was freezing.
 
Last edited:

sm52

Member
Joined
Mar 14, 2019
Messages
462
jcandy
about the size of the zma file. Yes, he's big. But notepad opens it quickly. So it's not a problem.
I have 8 GB of RAM. When firefox hangs, 36-38% is busy. Before working with REW and Speakerbench, 18% is occupied. So the lack of memory is not a problem for me.
I think that when trying to generate a json file and then load it, there is some kind of unexpected loop that is erroneous.
 

jcandy

Registered
Joined
Dec 3, 2017
Messages
11
Location
San Diego
@sm52

I added a downsample option that will reduce the size of the Z-file significantly and make the download less painful. The upload will still be slow but hopefully this will work better for you since the real killer was trying to serve such a large JSON file for download. Let me know if that helps.

Thanks for your feedback!
 

Attachments

jcandy

Registered
Joined
Dec 3, 2017
Messages
11
Location
San Diego
Hi Bernard,

Thanks alot for your help here.

Bernard said:
The other day I uploaded sm52's files, Speakerbench worked properly, and I posted the results.
My pc is running Windows 11 Pro, version 21H2, build 22000.527. MS Edge is my browser and I also use NordVPN when I want to watch F1 for free on RTS, the French speaking Swiss TV. ;-)
Upload worked fine with and without VPN.
This is good to know. I appreciate the feedback. IMO the uploads with these files are slow but should complete on most systems. Claus and I have tested on Windows, Linus and OSX and see slowness but never a crash. However ...

Bernard said:
Yesterday, I installed the latest Firefox 64 bits version and I redid the test with sm52's files. Everything worked fine until I wanted to download the Json file. The screen froze and even with Alt+Ctrl+Del I couldn't stop the Firefox program. I had to shut down the pc wildly by holding down the power button. When I rebooted, I uninstalled Firefox and returned to Edge as my default browser. I didn't try to figure out why Firefox was freezing.
I think the killer is the large JSON file (the Z-file) that the browser serves for download. This causes awful slowdown on all my computers and makes the browser sluggish even after closing the tab. As you can see in my post to sm52, the downsample option makes the JSON file smaller and hopefully makes the issue less severe.
 

sm52

Member
Joined
Mar 14, 2019
Messages
462
jcandy
I checked how everything works if you use a downsample of data in a zma file. There are no freezes, the browser thinks, but not for too long. The result is slightly different from the full data. For example, the total error is 0.83(downsample) and 0.82(full data).
 

Bernard

Member
Thread Starter
Joined
May 22, 2017
Messages
108
Hello Jeff,

Any difference during uploads and between the non-activated downsample :
Fit quality : GOOD Zero-mass error : 0.18 Full-mass error : 0.29 Electrical error : 0.50 Error (fail > 1) : 0.49 Datafile (JSON) : ADV20220318_132951.json
and the activated downsample :
Fit quality : GOOD Zero-mass error : 0.18 Full-mass error : 0.29 Electrical error : 0.50 Error (fail > 1) : 0.49 Datafile (JSON) : ADV20220318_133314.json

Problem related to the machine used?
I just got one problem at the very beginning of Speakerbench with too large files.
Cheers,
Bernard
 

sm52

Member
Joined
Mar 14, 2019
Messages
462
I used metal washers for fasteners of different sizes as weights. Today I checked how they react to approaching the speaker membrane. A speaker with a membrane diameter of 8.7 sm at a distance of 3 sm begins to attract a puck suspended on a thread. Then she sticks to the membrane. This means that washers are not suitable for measuring Thiel-Small parameters. As well as any weights interacting with a magnet. Non-magnetic weights must be used.
 

jcandy

Registered
Joined
Dec 3, 2017
Messages
11
Location
San Diego
I used metal washers for fasteners of different sizes as weights. Today I checked how they react to approaching the speaker membrane. A speaker with a membrane diameter of 8.7 sm at a distance of 3 sm begins to attract a puck suspended on a thread. Then she sticks to the membrane. This means that washers are not suitable for measuring Thiel-Small parameters. As well as any weights interacting with a magnet. Non-magnetic weights must be used.
Yes, that is correct. Non-magnetic weights only should be used. We have added some text in the manual to reflect this (see attached photo).
 

Attachments

sm52

Member
Joined
Mar 14, 2019
Messages
462
jcandy
How to understand what is the cause of a large electrical error? The mechanical error is small, and the electrical error is large:
(wfit): Mechanical error [%] = 0.065
(wfit): Electrical error [%] = 0.581 0.678
 

jcandy

Registered
Joined
Dec 3, 2017
Messages
11
Location
San Diego
How to understand what is the cause of a large electrical error? The mechanical error is small, and the electrical error is large:
(wfit): Mechanical error [%] = 0.065
(wfit): Electrical error [%] = 0.581 0.678
Those particular numbers are not normalized and have only a relative meaning. For a better normalized electrical error you should use the electrical error at the top of the results, not the lower gray area.
 

sm52

Member
Joined
Mar 14, 2019
Messages
462
you should use the electrical error at the top of the results
The electrical error at the top is about the same as at the bottom, my question is how to understand it? I take several measurements, do not change anything in the wires and devices, the electrical error was small, about 0.2, and suddenly it becomes about 0.6-0.7. And I changed only the weight of the weights.
This is one. Secondly, your site works fine once, the second time too, and the third time you visit, it does not display the 'Downsample' window. I click 'refresh', the 'Downsample' window appears, the text in the file download windows disappears. I press refresh again, everything changes in a circle. One will disappear, then another. Further. I have uploaded the measurements. All rectangles are green, the json file has appeared. I want to see 'Data Check'. I click, one time there is information, another time it is not. Same with the tabs in the 'Fit' window.
 

sm52

Member
Joined
Mar 14, 2019
Messages
462
John Mulcahy
I measured the driver impedances in free air and in a box. REW calculated the TS parameters. I entered the volume of the box 10l. Vas turned out to be 10.7 liters. It seemed strange to me. Then I entered the volume of the box 20l. REW calculated Vas 21l. That is, with any volume of the box, Vas will turn out a little more. For measuring TS parameters with the added volume method, is this correct?
 

John Mulcahy

REW Author
Joined
Apr 3, 2017
Messages
5,436
VAS is proportional to the sealed box volume used in the measurement, but it also depends on the peak frequency from the sealed box measurement. If you measure in a larger box you will get a different peak frequency and end up with the same VAS (within the bounds of experimental error).
 

sm52

Member
Joined
Mar 14, 2019
Messages
462
I understand. Thank you. When measured by the method of additional volume, Vas turned out to be 10.7 liters. And when measured by the method of dual added mass 13.4l.
1. When there is a driver with unknown TS parameters from the manufacturer, how can you be sure that you get the correct measurements?
2. Is the dual added mass method actually more accurate than the added volume method? Indeed, in real conditions, no mass is ever glued to the driver diffuser.
 
Top Bottom