ErinsAudioCorner

[napilopez]

Just imported your curves into REW, RMS Average gives me the same exact result as your vituixCAD one above. Can't seem to replicate your other results yet:

 

[hardisj]

I am currently at a loss. I’m out for the afternoon. I’m sure it’s something simple I am overlooking. Sometimes a break is needed to realize what it is. Normally, the revelations occur as I’m driving home from work rather than at work. LOL

 

[hardisj]

Well, after far too long troubleshooting my script, I have determined there is something going on with how Matlab is taking the average and what I'm doing with it so I emailed Matlab to see if they can help and for now I am using VCAD's output to generate the data for the below plot. Makes life easy. Import the table and plot what I want.


So, here I am... close to the finish line. I tried to stick to what is in Dr. Toole's book for the curves (dashes, dots, etc) but realized they likely did things the way they did because they didn't have colors. So, I built the graph the way I think it looks best. Matching sound power and SP DI curve colors (blue) and the same for early reflections (red). Dashed lines for only DI curves. Solid lines for everything else. I added the y-axis on the right to mimic what is in Dr. Toole's book. But I prefer the legend outside of the graphic because it's distracting otherwise so I shot for the way Klippel provides it at the top below the title (Matlab doesn't have a default for this which is stupid so I had to code that up specifically).

Anyway, here's my first full set of CEA 2034 curves. Took nearly 2 solid months of testing methods, understanding trade-offs, etc etc but I finally arrived at a data set I am proud of. And now that I've gotten the method for testing down as well as the scripts written to populate the graphs in a manner I like, I can hopefully start knocking more of these out. First up, though, I have to actually do a real review for the Buchardt S400 and get the data on my site.

Also, for those comparing with Amir's data one thing to note is that Amir used the tweeter as his reference angle. I used the midpoint between tweeter and woofer (per the manufacturer's direction).


A couple notes:
I provided two versions of the spectrogram: one is the full 360-degree view (from -180 to +180), with the frequency axis extending from 20 - 20kHz. The second version is a 'zoomed' version with a tighter window of ±90° and frequency axis from 200 - 20kHz. I will probably always do this. I also added some labels on the y-axis to indicate the direction the measurement was taken relative to the speaker. I think this helps gives newcomers an idea of what they are seeing.

I have also provided two different versions of the predicted in-room response: one with the target curve (1dB/octave) starting frequency at 100hz and the other starting at 200hz. In this speaker's case it doesn't much matter, though you can see a slight difference higher in frequency. However, if I were to have a speaker with a high-Q bass hump around 100hz (which I've seen before) then that would obviously effect how the predicted curve lines up with the target line. Plus, given the room is dominate below about 200/300Hz, I am more inclined to go with the version that starts the target line at 200hz. But your feedback and rationale is welcome.

I purposely provided the ±40° vertical response to mimic each other (same color for + & -). Also, adding all the curves to this graph creates a cluster in one's brain (same for the horizontal plot if I were to extend beyond 90 degrees). Really, to get an idea of directivity one should just look at the normalized spectrogram. I typically look at the on-axis and then the spectogram to see how the off-axis response corresponds to the on-axis. But I can make some changes or additions if enough people really see the need.

If you spot anything that you don't necessarily like feel free to make a suggestion. Just keep in mind I have been at this for days and may resent your feedback. LOL.

 

[hardisj]

I threw these together.... yay, nay?

I added the 'maximum resistance' label and dashed line to make it easy for people to see where the load dips to. Leave it, or remove it?

 

[HammerSandwich]

Leave it, or remove it?

Keep it and change "maximum resistance."

 

[hardisj]

Keep it and change "maximum resistance."

I guess it's a matter of how you perceive it. It's the highest load point... but the lowest value.

I initially had it as "minimum impedance" but thought that may confuse people. I can change it back to that.

 

[hardisj]

Here's some new ones for you guys to chew on... Compression.

The idea here is if a speaker is truly linear then there will be no fluctuation in output beyond what is expected (ie, if you double power you expect 3dB more output... that sort of thing). So, what Klippel's module allows me to do is sweep a frequency range using a range of input voltage and it will display a graph which *ideally* would have all results stacked on top of each other; that would imply the output is indeed linear... meaning the output SPL is what the system expects it to be based on the voltage differentials. When the lines do not stack on each other that is an indication of distortion or thermal compression. Here is a direct example:

Now, what I've done is export the above data to 'make it my own' and I have created some alternate ways of viewing the data.

Nothing big there. Same data. Just different layout. I can change colors. Not important right now. What is important is how it is presented. The above illustrates a speaker with practically no compression from 1 volt, or 77dB (mean @ 300-3khz), to 10v, or 97dB, (all at 1 meter). It's quite good!

I have made 2 additional plots based on the data above. This is simply the actual SPL output. I don't think this helps in any way because you have to understand the math and you need the data points to really see the point. But, just to show I've done it here we are:

However, for those who want to see the above but referenced to 1 volt input I have "normalized" the data to the 1v output. This version I believe is the most useful because it shows you what is going on referenced to 1 volt. And, yes, I *can* normalize to 2.5v (87dB)... but it's the same thing; just a different way of looking at it. The legend shows the input voltage and the THEORETICAL/calculated output; meaning the simple voltage ratio of 1v/x = ydBSPL; the viewer will have to do the math to say "okay, I'm losing approximately -- dB at this calculated output volume). I can do a bit more math to help you understand what the real number is but IMHO it's more useful to see what you're targeting and understand just what the compression is doing to that target. It gives you a feel for what the output really will be (and, it's hard to quantify a single SPL value across the whole band, thus the "mean" from 300-3khz).

 

[HammerSandwich]

People who don't have a clue about resistance, impedance or ohms will just skip it, right? And using the term backward won't help any confusion, IMO.

Not trying to be argumentative. I could settle for maximum conductance...

 

[hardisj]

People who don't have a clue about resistance, impedance or ohms will just skip it, right? And using the term backward won't help any confusion, IMO.

Not trying to be argumentative. I could settle for maximum conductance...

I don't disagree. I get tripped up on it many times because it's almost counter-intuitive. At least to my simple brain. I switched it to "minimum load = x.x". Let me know if that works.

 

[hardisj]

Step-Response. Automating this one takes a bit more thought (grab the first return to 0 which in this case is the bass but others would be the tweeter/mid transition swing so would have to add some smarts there... no time right now). So, for now I'm just gonna screencapture Klippel's interface. One not zoomed and one zoomed.

...... aaaaaaaaaaaaaannnnnnnnndddddddddddddddd .........

to round it out, here is the THD at varying levels.

I think at this point this is plenty of data for people to understand the performance of a loudspeaker (different set of metrics compared to a drive-unit). The only other thing I can provide that would useful is max SPL but that runs some risks of damaging the speaker and I just don't know if I feel comfortable doing that on speakers I can't even afford myself.



Until everything is published on my site, nothing is final. I may change some colors here or there. But mostly I'm satisfied with the presentation of each of these metrics as a whole. If you see anything that is glaringly wrong or absolutely needs to be changed let me know, though.

 

[HammerSandwich]

Works for me.

As does the normalized compression chart. My only comment is that you might want to determine a standard vertical scale to make comparisons easier. (Yeah, I'm the guy who opens 2 pics in different tabs & keeps switching to highlight the differences. Not everybody's gonna animate a la @BYRTT.) Something like +/-2dB? Tonal shifts won't be squished too much, and if it's off the chart, you know it's bad.

 

[hardisj]

Works for me.

As does the normalized compression chart. My only comment is that you might want to determine a standard vertical scale to make comparisons easier. (Yeah, I'm the guy who opens 2 pics in different tabs & keeps switching to highlight the differences. Not everybody's gonna animate a la @BYRTT.) Something like +/-2dB? Tonal shifts won't be squished too much, and if it's off the chart, you know it's bad.

So, I considered that. I try to do that for everything (another benefit of writing the scripts myself). But...
for the FR... the CEA2034 spec wants a 50dB window. Most speakers are in the 85-90dB sensitivity range so no issue there. But if I fix it to 50-100dB window and then I have a speaker that is outside the norm then that obviously shifts the scale.

The HD will always be a moving target because some speakers will have large distortion and others may not. If I bound it to a *guess* now that guess could be way off one way or the other. And then I've got to retroactively correct *everything* I did before the issue poked up its head.

Same thing for impedance/phase.

Same thing for compression. I've seen some speakers with pretty unhealthy levels of compression. So, let's say I set it for ±2dB; I could break that with a low sensitivity, small enclosure speaker. And then I'd be back to having to re-run everyone of my old datasets again.

So, this stuff kind of goes with experience and best-judgement.


that said, the alternative is to make a variable graph and then a 'fixed' one. For example: 3% THD is genuinely considered the breaking point for midrange/high frequency distortion. 10% is the target THD for a midwoofer. 20% for a subwoofer. I could lock a scale down to 10% THD and have that as a "zoomed" version. But, heck, then you get a speaker that is 5% (could happen with large, high sensitivity speakers) and you're like "blow it up!". So...

 

[hardisj]

I mean... it's just a copy and paste so why not. Here's the compression with a fixed ±2dB scale.

 

[hardisj]

And here's one for the distortion fixed at 10% THD:

 

[richard12511]

Here's some new ones for you guys to chew on... Compression.

The idea here is if a speaker is truly linear then there will be no fluctuation in output beyond what is expected (ie, if you double power you expect 3dB more output... that sort of thing). So, what Klippel's module allows me to do is sweep a frequency range using a range of input voltage and it will display a graph which *ideally* would have all results stacked on top of each other; that would imply the output is indeed linear... meaning the output SPL is what the system expects it to be based on the voltage differentials. When the lines do not stack on each other that is an indication of distortion or thermal compression. Here is a direct example:
View attachment 67733


Now, what I've done is export the above data to 'make it my own' and I have created some alternate ways of viewing the data.


View attachment 67740



Nothing big there. Same data. Just different layout. I can change colors. Not important right now. What is important is how it is presented. The above illustrates a speaker with practically no compression from 1 volt, or 77dB (mean @ 300-3khz), to 10v, or 97dB, (all at 1 meter). It's quite good!

I have made 2 additional plots based on the data above. This is simply the actual SPL output. I don't think this helps in any way because you have to understand the math and you need the data points to really see the point. But, just to show I've done it here we are:
View attachment 67735





However, for those who want to see the above but referenced to 1 volt input I have "normalized" the data to the 1v output. This version I believe is the most useful because it shows you what is going on referenced to 1 volt. And, yes, I *can* normalize to 2.5v (87dB)... but it's the same thing; just a different way of looking at it. The legend shows the input voltage and the THEORETICAL/calculated output; meaning the simple voltage ratio of 1v/x = ydBSPL; the viewer will have to do the math to say "okay, I'm losing approximately -- dB at this calculated output volume). I can do a bit more math to help you understand what the real number is but IMHO it's more useful to see what you're targeting and understand just what the compression is doing to that target. It gives you a feel for what the output really will be (and, it's hard to quantify a single SPL value across the whole band, thus the "mean" from 300-3khz).

View attachment 67741

97db isn't very loud for 1m, so I wouldn't expect many speakers to show tons of compression at that level. Can you go louder and see how it holds up?

 

[hardisj]

97db isn't very loud for 1m, so I wouldn't expect many speakers to show tons of compression at that level. Can you go louder and see how it holds up?

I can.

 

[richard12511]

I can.

That would be awesome. I've always wanted to see the point at which loudspeakers start compressing at various frequencies. It's a huge factor that greatly influences my purchasing decisions, but it's data that basically no one provides.

Don't break the speaker , but I'd love to see the point at which it starts showing compression. Nobody else is showing this.

 

[hardisj]

That would be awesome. I've always wanted to see the point at which loudspeakers start compressing at various frequencies. It's a huge factor that greatly influences my purchasing decisions, but it's data that basically no one provides.

Don't break the speaker , but I'd love to see the point at which it starts showing compression. Nobody else is showing this.

That's the tough part. If it's a cheapo speaker then, hey, I'm all for it. If it's an expensive speaker then I have to be a bit more cautious. So, I'll provide it when I can. But it'll be subject to my own risk of pucker factor. LOL

Even at 10v with the speaker in the garage and me running the tests from my PC inside the house via VNC, my dog was losing her mind. Kept running up to me and jumping on me like "dude, what is that!!!!?".

That said, I'll try to target 102dB @ 1 meter when I can. That's about 90dB @ 3 meters and 96dB @ 2 meters. Should be plenty adequate for even those who love to jam out in large listening rooms. But when I am skeered of doing that, it'll be about 96dB @ 1 meter target.

I'm also going to start at 2.83v because the 1v spec doesn't really give a good frame of reference. Since SPL is spec'd at 2.83v I'll use that as my reference and it may be easier for folks to "get" the data.

 

[hardisj]

I just ran the test again but from 2.83v to 14v. Remember, the values listed as "mean SPL" are based on the 0-degree mean SPL from 300-3kHz. So, there's some variance here in the single SPL values given in the legend. For example, 100dB is closer to 101dB at some points in the actual fundamental. But, it's close enough for this purpose.

Side note: It may have taken me a solid day but I am really glad I wrote a script for this. Was as simple as importing the new data .txt files and hitting "run".

 

[hardisj]

I added a note to (and fixed the header of) the normalized compression graph that explains what the graph is showing. This way if the graph is taken out of context it can at least be made sense of.