Erin's Audio Corner (my new review YouTube channel/website)

[ErinH]

FWIW, here's a spectrogram I coded in Matlab using my data I posted yesterday. The two white lines show the -3 and -6dB values, as requested by another individual. Per the title, the response is normalized to the on-axis response, so you can see some areas where the off-axis response has a slightly higher SPL value than the on-axis response (3-4kHz, for example). I prefer this way of viewing spectrogram data because it tells me how the response varies relative to the on-axis response. But I can easily change the program to deliver absolute SPL levels as well. I prefer to write my own scripts because it puts me in control of how I present the data as opposed to using bits and pieces of others' software (such as aRTA or VirtuixCAD; both great programs, just not how I exactly want to present the data).

 

[ErinH]

(Edit: Sorry about the extra large photos: I'll re-size if I have the time)


About a month ago I purchased a pair of Elac DBR62 speakers and sent it to Amir at AudioScienceReview so he could measure it with his Klippel Near Field Scanner. Results here:
Elac Debut Reference DBR-62 Speaker Review | Audio Science Review (ASR) Forum


The reason I was interested is because I'm starting back up doing reviews and thought having the NFS 'anechoic' measurements would be a big help in determining my measurement method for reviews. My initial plan was to measure outdoors for far-field mid/high frequency and merge with a ground-plane measurement for low frequency. So, here's the first measurement...





Test # 1) Speaker on platform ~ 8.5 feet off the ground; window to about 13ms with accuracy down to approximately 100hz.

Speaker flush with front of platform (picture was taken before I began tests). Mic at 1m; signal 2.83v. Note: I don't care that the mic stand can cause a reflection for this measurement; this is just a test of a test.

Pros: This method has the benefit of having very high resolution as you go above 200hz.
Cons: being subject to the elements (cold, wind, rain; external noise not really an issue in my area) and having to hoist a speaker up 8.5 feet in the air and risk throwing out my shoulder should I ever test a floorstanding speaker.



Here's a photo of the setup:

Here's the result compared to Amir's. You can see some decent differences namely above 1khz. Notably the 1-2khz region.

Now, I wouldn't say that test was 'final'; it was just a "proof of concept" and I suspect that I could probably get better accuracy with an additional test and higher # samples that feed the average. Given the hassle of setting the speaker up high I thought I would try a few other methods out to see how they compare.


Ok. That brings you up to speed to today's testing. Where I tried a number of things. I'll list the method, the pros/cons and provide a picture of the setup as well as my result vs Amir's result. Note: The stand was thrown together from scrap wood; if I were to use this it would change. But I didn't want to waste too much time because storms are setting in this afternoon and I needed to get to work. Also, ignore levels for now. I used a mic calibration file for FR but haven't checked the level with the pistonphone; I only care about the general trends for this process.



Test # 2) Measurement on a 5 foot stand in my garage; gating was about 4 ms.

Same stand used below but inside my garage with a ceiling height of about 10 feet. The speaker was positioned about 5 feet above the ground with the tweeter at about 10" off the platform. Mic at 1m; signal 2.83v.

Pros: Don't have to worry about the elements nor do I have to hoist the speaker 8 feet in the air.
Cons: Poor resolution below 1khz. I don't think I would even bother posting this data. For a single drive unit it's fine. But, speaking from experience, I need good resolution to determine where to merge LF response data of ported loudspeakers.

I didn't take a picture of this setup.

Here's the result compared to Amir's:

Test # 3) Outdoors on stand. Same stand as before. Window ~ 7ms.

Mic at 1m; signal 2.83v. Note: I don't care that the mic stand can cause a reflection for this measurement; this is just a test of a test.

Pro: Better resolution than garage measurement since ceiling reflection isn't there and thus gating is extended an additional 3ms or so. Don't have to hoist the speaker up 8 feet in the air.
Con: Outoors; weather/environment. Still not as reflection-free as the 8.5 foot platform or ground plane measurement (see below).

Here's the result compared to Amir's:

Test # 4) Ground Plane Measurement: Speaker Angled ~ 8 degrees; window out to 40ms before first reflection.

Speaker was angled at about 8 degrees so that the tweeter would be pointing on-axis with the microphone (well, as on-axis as it can be without burying the mic and pointing the mic at the DUT). Mic at 2m; signal 2.83v. Note: Though the response was gated here, if I were to actually use this method I would move to the back yard where I could get the nearest reflection about 40 feet away which should get me down to around 20hz.

Pro: Invulnerability to reflection; I can get ever further out in my backyard. High resolution in to the very low frequency region.
Con: Outdoors. Diffraction effect of speaker changes due to the baffle doubling (via the mirror image effect). HF > 10khz seems to be a pain. From my research that's generally accepted as questionable. Though, I had decent results.

Phone placed at tweeter on baffle and used in 'selfie' mode to make sure the microphone was at the center of the image (trick I learned on another website).

Here's the result compared to Amir's:

Thoughts:

At face value, each of my methods look pretty reasonable. And all are pretty much "industry standard". You can go to a website or magazine and see any number of reviewers using any of the above methods. But I wanted to do a comparison myself and see if there was any one method that is better than the other. Unfortunately, there isn't. I was hopeful the Ground-Plane method would yield the utmost accuracy (aka: match what Amir has). But it's missing the 600 hz - 1khz bump Amir's data shows and there's a slight difference in the LF region. Potentially diffraction effects; but playing around in EDGE and VirtuixCAD I didn't see the influence. Otherwise, it looks pretty dang good.



There's two possible reasons I'm seeing differences:
1) none of the methods I use are fool-proof
or
2) I shouldn't put all my trust in to the NFS (not saying it's wrong; just saying it may also have a few things that make 97% accurate instead of 100% accurate; keeping in mind that an anechoic chamber has its own issues).

I think the answer may actually be a large heaping of #1 and a small portion of #2. All of that said, once you combine these possibilities with the level of detail I'm viewing this in... really, none of my results are extremely different from the NFS results. They're just different in some areas and alike in others. I mean, I'm usually within 2dB difference. Which, yea, it's not minimal but in the grand scheme of things, I'd say that's not too bad. I want to be closer, though.



Ultimately, this is where I am at at this point:
I still want to re-test on the 8.5 foot platform. Maybe I can do that tomorrow.
Measuring in the garage is convenient. But measuring outdoors provides better resolution, especially in the area where I would need to perform low-frequency response stitching and that is useful in determining how best to align the two sets of measurements.


I would prefer to find a method that yields the best accuracy without the need for additional post-processing. Though, when I started this venture my plan was to use the outdoor farfield measurement for > 200hz response and merge that with a ground-plane measurement for low frequency.

But if I could figure out how to calculate the diffraction impact then I could just remove it from the result. I could also use the near-field technique but if I were to get an oddly shaped port (flared) or something that's harder to easily measure and quantify then that would impact accuracy as well. There's also the "mic in the box" method but it has the issue of being able to fit a mic; some speakers (like this Elac) won't allow my mic inside the slot port.


If you've read this far I assume you have some vested interest in these results. So, I welcome your opinions, as either or both fellow testers and reviewers. Would you say the ground plane measurement is "close enough"? Is there a formula or set of formulas I can incorporate in to my Matlab script (which post-processes the data) to account for diffraction in the ground plane measurement? Would you stick with the original plan, combining "speaker in the sky" (Test #1) for mid/high frequency with ground plane for LF?

One thing is for sure: No one can tell me I didn't commit due diligence before I started posting data. It is the nature of the internet for someone to try to find something wrong but I am putting in the legwork up front to make sure I get this as good as I can. I have searched high and low and literally could not find one source where someone took the time to do multiple types of measurements and compare the results to anechoic.

 

[shadyJ]

I don't understand how method #1 isn't getting a good match to Audio Science Review. That is assuming Audio Science Review is getting true anechoic results. The mic and the speaker have to be in the exact same positions in each test, and maybe a problem was the positioning was a bit different. Like I said before, you should get a better boom pole for the mic. It's not hard to do, and your current mic holder is always going to be a problem at high frequencies.

I would iron out whatever problems you are having with method 1, and use that combined with ground plane for low frequencies. Method 1 will not work for tower speakers. Use something like method 3 for towers. Even then, that will only be safe for smaller floor-standing speakers. It will be very hard to hoist towers over 60 lbs on that stand without help. I hope you have a strong friend handy for that.

As for full range ground plane measurements, sometimes I have acquired results that were reasonable approximations to the free-air measurements, but sometimes they could be way off. I would say you should practice with more speakers before you place your faith in full-range ground plane.

 

[ErinH]

I don't understand how method #1 isn't getting a good match to Audio Science Review. That is assuming Audio Science Review is getting true anechoic results. The mic and the speaker have to be in the exact same positions in each test, and maybe a problem was the positioning was a bit different. Like I said before, you should get a better boom pole for the mic. It's not hard to do, and your current mic holder is always going to be a problem at high frequencies.

I would iron out whatever problems you are having with method 1, and use that combined with ground plane for low frequencies. Method 1 will not work for tower speakers. Use something like method 3 for towers. Even then, that will only be safe for smaller floor-standing speakers. It will be very hard to hoist towers over 60 lbs on that stand without help. I hope you have a strong friend handy for that.

As for full range ground plane measurements, sometimes I have acquired results that were reasonable approximations to the free-air measurements, but sometimes they could be way off. I would say you should practice with more speakers before you place your faith in full-range ground plane.

Just to be clear, this isn't my first rodeo. I am just simply providing all the various methods out in the public; something I wish more people were willing to do. Seems a lot of people just pick a method and go with it and don't discuss what led them to their decision. (not you; just in general)

I understand your concern for the boom pole but again, this is all proof of concept stuff. I understand the impact and it's negligible relative to the other things I'm looking at at this very moment. I will use a properly extended boom when I decide on which method I want to go with.

I have also seen NFS data that doesn't match true anechoic data so there's some ambiguity here.

I agree the best method is likely the first one I went with and that's been the plan. But I was more curious to see what the other options results were/are. If nothing else I can tell people I did my due diligence.

 

[ErinH]

Also worth noting the NFS itself has a microphone holder that is exposed to the mic:

I would hope that is somehow part of the calibration, however, the holder is not symmetrical and at varying axes the influence would be different. If I had to guess, I would say it is not accounted for (calibrated) because that would require a calibration for every single measurement point. But I don't know for sure.

 

[ErinH]

I quit posting here for a bit but continued to post on DIYA, having discussion with some others. Therefore, some of what you see below may seem odd as it may be worded as a reply to others. But, ultimately what I did was I ran down all the different ways to test with a focus on the ground plane method for a few reasons. The ultimate discovery is that if you're measuring in an open grass field you can discount the data above 300hz or so (see further measurements/discussion).

I haven't seen anyone do this kind of multi-test approach to compare and contrast the different methods. So, I'm simply sharing to help you guys get a better understanding of the pros and cons of the different methods.


Alright... here goes a lot of posts...

 

[ErinH]

Alright, guys. I slept on this a bit and thought, "what if I take the merging approach and apply it to a ground plane measurement both indoors and outdoors" with the notion being that indoor GP measurement would be a way to help shore up any concern over HF response when measuring outdoors due to wind/noise/etc.

Below is a picture of the setup for the MF/HF GP portion:

(that tandem bike rocks!)



The nearest reflecting point was about 6 feet away. Here is the impulse response gated:

And here is the result compared against the NFS as well as my driveway GP measurement conducted yesterday where I was concerned some wind could be causing issues. Note: The SPL isn't an exact match because the distance may have been off a couple inches in either measurement; I'm just looking for trends.

And this is the result of the speaker on an 8.5 foot stand vs my GP measurement today and the NFS measurement:

All of these results of mine look pretty close to the NFS results. However, the striking difference is in the 600-1kHz region where the NFS shows approximately a +2dB difference over both my GP and outdoor 8.5 foot stand measurements.


So, do I quit obsessing over this difference? Do I call the GP method adequate and just carry on; potentially using the indoor GP measurement for mid/high frequency merging when conditions are a bit more windy/noisy? I am willing to do (2) sets of ground plane measurements if it saves me from breaking my back and/or wallet worrying over hoisting large speakers in to the air. And based on what I'm seeing in my results, the GP measurement matches as well as my high-stand mounted speaker measurements do to the NFS, other than the diffraction effect due to the mirror-image baffle. The only concern I currently have with the GP measurement is off-axis measurements. Does setting the speaker on a turntable that is lifted off the ground matter? I suppose I could just make a long 'table' to align with the turntable that runs the length of the path between speaker --> mic and place the mic on it at the other end. The height would alter the response some but if I'm only using the indoor measurement to avoid any concerns over very high frequency (>8kHz) then that is not of concern, either. I'll test this out later.

 

[ErinH]

Now that Amir has tested the Buchardt S400, which I just received yesterday as well (not the exact same speaker he tested; just the same model), I took the opportunity to do a little bit more testing. It has been raining off and on this morning so I proceeded with the garage-based ground plane measurements.

Picture (the garage door was opened just to get some light for the photo; was shut during testing).

Nearest reflection point from either the mic or the DUT was about 7 feet. Window of IR is about 8 ms wide.


I ran a series of tests but I'm providing what matters most. Note, I smoothed all data to 1/24 octave resolution.

Here is the FR comparison:

And to better see the difference, here is the NFS ÷ Ground Plane.

The delta is really quite good. The resolution of my measurement isn't as high due to the 8ms window so that explains some of the low-mid frequency deviance. I was specifically looking for differences in the 600-2kHz region because that's where my Elac DBR62 differences showed up. I am seeing a similar trend here; my 600-1kHz region is a little bit lower, while my 1-2kHz region is a little bit higher. I mean, these are nitpicking levels. But assuming this is a trend caused by the NFS hardware itself then I'm OK with that conclusion and am happy to continue on. Assuming, of course, my off-axis measurements reveal similarly good matches with the NFS results.



**Of course, my results do not show the same 520Hz high-Q peak the NFS shows but that's simply because this, once again, is an indoors measurement with a relatively low window. Assuming my speaker has the same quality it will/should show up once I perform the GP test outdoors and can set my window to 40 ms. **

There could also be some very minor differences in HF due to imprecise aiming on my part; I didn't get the laser level out. I just eyeballed it to be on-axis horizontally. But the vertical reference axis was the tweeter itself.

Overall, however, this additional data set gives me a pretty good feeling about using the ground plane measurement. It's a really close match to the NFS result. I'll test outdoors when the weather permits and also conduct polars to see if there are any glaring issues.

Have to thank Amir again for posting his data. It's a really good way for me to sanity check my own methods and provide me even further confidence that what I am getting is quite good.

 

[ErinH]

Here you go! I made this post on ASR so I'll just copy/paste it from there. I specifically bolded a portion of it that I think you'll want to keep in mind...



I performed my own spin on a DIY'd turntable. Now, this is, once again, not something I would consider *FINAL*. It was an indoors ground plane measurement and therefore the window of reflection free response is only about 6ms long which means accuracy only down to ~166hz, and the DUT was lifted about 2 inches off the ground as it was placed on a turntable and therefore, comparing back to back against the "direct to floor" method, there is some variation, notably below 1kHz. The plans for future tests is to conduct ground plane measurements outside so I get a longer gate and bury the turntable so it sits flush with the ground so there is no misleading data due to it being on a 2" platform. But I believe the response >1kHz is at least useful as a comparison within itself (IOW, comparing on/off axis response within my own dataset).



The center of rotation is the front baffle. Some reviewers place the DUT at the center of the turntable, thus making the center of the speaker itself the rotation point. My rotation point is the baffle. You'll see what I mean in these pictures:



This is the 0 degrees (on-axis) setup:

This is the 90-degrees off-axis setup:

And here are the results of my horizontal. Again, keeping in mind the notes I mentioned above.

 

[ErinH]

Okay, back to the ground plane measurements. Ground/grass surfaces are not ideal. Everything I've read about measuring in the ground plane says you need a hard reflective surface. I didn't realize how important this was. I'll explain.


I re-tested the Buchardt S400 in my driveway. Here's a picture of the setup:

Here is the result compared to the results Amir posted and the factory curve posted on Buchardt's website:

That's pretty good correlation within those two sets (and notice those two both used the NFS and don't even match which is being discussed in the S400 thread).




Back to my opening statement that the the grass is not ideal...

When the 'rules' of ground plane measurements say you need a "hard" surface they mean it. Grass is NOT a good place to conduct ground plane measurements. And I understand why now. Rather than keep this information to myself I wanted to share in case it might be useful for someone here in the future. If you already knew this, congrats. If you didn't, let me be your example of why not to bother.

First, here's a picture of the backyard. Where the black chunk is where the speaker and mic were placed; very, very low grass. You can see it's obstruction free for a good bit; about 40 feet from the center is the fence and my house. Nothing else in the way.

In the below graphic you can see my driveway measurement mentioned above compared to (2) different backyard measurements (BOTH were taken with a 2x3 foot mirror on the ground at the mic): The blue graph is with the speaker and mic on the ground with nothing between. Just bare ground/grass. The red graph is with a long piece of plywood running the stretch of space between the microphone and the speaker. You can see the ground is responsible for a LOT of absorption. I don't know if it's the grass because my backyard is kept cut VERY low and it is pretty bare as you can see in the photo above. But, regardless of the exact influence of grass vs ground the results are eye-opening. The plywood also is not reflective enough alone.

I am going to buy concrete pavers and see if lining them up in a 2x8 foot section between the speaker and mic will help get me results that mimic the driveway. I am hopeful it will. I don't think it matters that the area next to the speaker and mic are not hard reflective surfaces but I don't know for sure until I measure. If it doesn't then I will either have to test in the driveway or consider pouring a small concrete slab in the backyard. I like measuring in the backyard because it's a traffic-free area. My kid has a lot of friends in the neighborhood and they are always in and out of my garage space and in the driveway in the summer. Plus I want to avoid weird stares from my neighbors.





Anyway, like I said, I didn't necessarily have to share this and I know some don't give a rip. But if it helps at least one of you then I've done my job.

 

[ErinH]

I measured in the backyard again but this time with pavers. I made a run about 1.5 x 8 feet long of 1.7" thick pavers purchased from Lowe's this morning. Here's a photo of the setup:

Here's the pavers result (blue) overlaid with the driveway measurement conducted a couple days ago (black) and the plywood measurement (red):

I then decided to try a few things...

So, first I laid a strip of plywood ON TOP of the pavers, running the length of the pavers (blue). I didn't take a picture of this but it is literally the same as above just with plywood laying on top of all the pavers.
I also tried laying the plywood crossways in front of the microphone (green).

I think this combination of data is showing me that I need more than just a length of pavers between the mic and the DUT. I need a large area. Something like a driveway or patio where is more concrete area.

Since I was already in my backyard I decided to do a quick sanity check to make sure nothing was wrong with the hardware or software by measuring the speaker on my back patio like you see below. This measurement was gated due to the wall you see in the photo, so it does't have the high resolution of the other measurements.

Conclusion:
From the above you can see the patio sanity check works out; it matches the driveway response fairly well (with less resolution due to the ~ 4ms gating). So, it is confirmed, the pavers/plywood/grass backyard method cannot be used. I (you/we?) need a much larger slab of concrete than a 16 inch wide set of pavers and/or possibly multiple layers and much more area of plywood.

I honestly thought the pavers would make up the difference. But they did not. I don't know if the result is because the pavers are not dense enough to fully reflect sound, if it's the fact that there is grass around, that the pavers are not 100% flat (but lined up well enough I would have thought), or that my backyard has a gradual slope. I'm just lost on the explanation here but maybe it's the culmination of different things. Not only does the top end still experience a large dip around 6kHz but there's also a dip in the midrange compared to the driveway measurement.

So, I will be using my driveway, patio or an empty parking lot for any ground plane measurements. Based on the data, I encourage you guys to do some testing of any non-paved space you'd like to use and compare it against a measurement in a parking lot or driveway.

And, yes, I realize this may seem like I am obsessing. I am. I simply want to get the most accurate measurement I can but I also want to understand what makes other methods inaccurate.

 

[ErinH]

Ok, now that the decision on where to measure has been made, I want to investigate how to measure:
1) Mirror vs no mirror
2) Mic angled vs mic flat


Note: Any photos I show were taken today in the sunlight to show you how I measured LAST NIGHT. They are just for reference; the measurements were all made at the same time as the datasets above.


All of the following measurements were done at 1 meter to help me get a little bit more reflection-free time but it did not effect the relative trends that I measured at 2 meters. Here is the comparison (note: the mic was flat on the ground as you can see in section #1 below):

Okay, so the mic will be placed at 1 meter for all following test results. Moving on...



#1) Let's look at the first concern, with and without a mirror:

For setup reference, here are the two setups:

And when the mic is flat, there is just a small gap between the mic and the surface:

Here are the results:

As you can see, the comparison with and without mirror laying flat on the ground/mirror are practically the same. So, assuming the surface you are measuring on is a hard surface like concrete I would say you can go without using a mirror.

Now, just for the heck of it, let's see if adding a mirror and angling the mic changed anything...

Pictures of setup. The mic is angled toward the ground and the shell is touching the actual ground.

Results:

Not a real surprise that the difference here is essentially null. We saw in the above when the mic was flat the mirror made no difference. Same thing here with the mic angled.

Now, that means we can assume that the results With Mirror == Without Mirror. Regardless of mic position.




#2) Mic angled vs mic flat:

Comparing "without mirror", referencing the above photos for setup, here is the overlaid results:

These results show higher frequency combing. I can't say how much of this is the result of the mic holder causing reflections or if this is literally all caused by the mic being angled. I would need to remeasure with foam or something that wouldn't cause a reflection to hold the mic at an angle to know for sure. I will need to consider this a bit more. What I've read is that aiming the mic down is best to get it on the same plane but I worry about the reflection from the holder so I will try to come up with a way to do this. Maybe something as simple as a thin block wedge where there is no empty space and therefore no chance for a reflection. If I re-test this aspect I will post back the results.











***** Extra Info! Wind matters, duuuuhhhhhhh.... but I still wanted to see how much. *****

And for those who may wonder about wind, I measured twice yesterday. All the data I showed for the pavers results were done in the morning and then again at night. It was quite windy yesterday morning with gusts up to about 15mph. So I scrapped that and decided to wait until the winds subsided. But I saved those results and have plotted them vs the nighttime measurement below. "Wind" is red. "No wind" is black. Again, these are with pavers and intended only to show the effect the 10-15mph winds had on my mesurements (2 meter distance).

 

[ErinH]

I guess I need to point out that everything that I’ve read in the books indicates what you need is a large flat reflective area. I basically just did the roundabout way of proving that to be true. On one hand I feel rather silly and dumb For not having understood and trusted what was meant. But I know me well enough to know that if I didn’t see this through on my own that I would not have benefited from the experience. So, ultimately I guess I just proved to myself that what I’ve read is accurate. The issue I had was that I saw so little talk about testing speakers full range; everything seem to be dedicated toward subwoofer and low frequency response only. Now I feel more confident in using this method for full range testing. You live and learn.

 

[ErinH]

Please, allow me to vent...

Yesterday the weather was nice. No wind. The neighborhood was quiet. So, I set out to measure the S400 and get it done. I ran horizontal and vertical measurements. About 30 total measurements.


But, I made a stupid mistake. Given that it seemed the "mic angled down" measurements were a close match to what was expected, I decided to go this route but I made a foam "shroud" to prevent any reflections from the mic holder altering the result. Like so:

My mistake was: I didn't test the 'shroud' first. So, I completed all of these measurements, went and looked at the results and saw this:

Man, that is a LOVELY sight to see. Look at that high resolution! Oh, but wait.... Above 3kHz the data is junk!!!! Sonuva!!!!...

So, today I measured in the garage (ground plane, 2 meters, gated to about 8ms): 1) with shroud and 2) without shroud. The "with shroud" reduces output above 3kHz by about 1-1.5dB.

I am about to lose my ever-loving mind here.

I really want accurate data out of a single measurement (ie, no 'splicing' or 'blending'). My issue with splicing is it requires nearfield measurements of each woofer/vent and results in a nearfield bump. My issue with blending is it requires additional modeling that would also increase lead time. Plus, there's the concern that you really can't just assume a NF mic measurement of the different sources will behave the same off-axis; especially when you don't have high resolution due to the window size. And then imagine splicing a single NF set of measurements with multiple off-axis measurements. But from this little bookshelf alone I can already tell you the polars wouldn't be correct and the resolution/accuracy in the midrange would also suffer. Proof? Look at my above set of horizontal measurements; see the split begins about 100-200hz? On-axis NF + FF stitching won't show you that. I want that data.

A single ground plane measurement would give me data down to < 30Hz; that's resolution better than some anechoic chambers. It also means no need to blend or splice. That also means there is less chance for error in complicated enclosures where there are different sound sources on different sides of a large enclosure (ie, woofer on front, vent on back). Cons of the ground plane measurement are simply that it is outside and I have to contend with weather and neighborhood kids. Not to mention having to walk 40 feet to go turn the speaker (and I have played with different automated designs but they all elevate the speaker on a platform which only alters the data).


I have one more thing I want to try. If that is no bueno I'm gonna have to really think this over and consider the pros/cons of the above. I keep falling back to the driveway ground plane measurement but again, that has its own set of cons. I think my only solace in all of this is that "if it were easy, everyone would do it". I just didn't realize how much legwork I was going to be putting in to get confidence in my measurements. I almost wish I didn't know what the data should look like. I tell ya', ignorance is bliss, folks.


If any of you think I am being too complicated about this: I am. I know it. But that's because I want this data to be accurate. I could easily do some 5ms windows in my garage, do some NF measurements, slap 'em together and call it a day. And that's not a knock against others who don't have a choice but to do that. It is simply that I know I can do better... but how badly do I want to do better is the real question. Hey, if nothing else, maybe I can help you guys avoid the same mistakes I am making.

/rant

 

[ErinH]

At this point I am pretty sure most who were paying attention have probably "checked out" at this point in the whole "how am I going to measure speakers" discussion but I'm going to make one final post on the topic because I had to feed my own curiosity...



To recap where my mental state is:

Driveway ground plane measurements look quite good. They make life a bit easier because there is no need to stitch. But they do come with cons (I'll cover this later). The only issue with GP in the driveway is neighborhood kids and other random things causing distractions when I am measuring. The backyard would be ideal but as I discovered in earlier tests, the grass is too absorbent and even a 2x8 foot section of plywood wasn't enough. I either need a large concrete pad or some other area of reflective surface. As a "last ditch" effort of trying to make the backyard GP measurements work, yesterday I went to Lowe's and purchased 3 large sheets of plywood. The idea was I would try them laying in the backyard as a "large, flat, reflective surface". I could have purchased more but I thought if 3 isn't sufficient to do what I need I might as well just stick with the driveway if I am going to do GP measurements.



First I had to wet the sheets down because they were bent. I learned this trick from my skatepark/ramp building days. Wet the concave side, lay it on the concave side. I let it set overnight. This morning they were all mostly flat.

Then it was on to testing. I tried ALL sorts of configurations between measuring longways and sideways. Here's some random photos...

But, I'll spare you the results of EVERYTHING and get to the point more quickly:

First up, let's look at the data from placing the mic on a single sheet of plywood vs the "bare ground" I took a couple weeks ago vs the (windowed) ground plane measurement I conducted in my garage:

As you can see, there is some improvement in the high frequencies when using the single plywood sheet compared to the bare ground measurements. Notably, with the plywood the HF response no longer has a sharp dip it did in bare grass. But, the plywood measurement shows a dip between 100-300hz and compared to the garage GP mesurement the HF response is still not what it should be; down by about 2-3dB above 1kHz.





What happens when I add a small piece of OSB behind the mic (Purple) and then as another test I add a small sheet behind the microphone as well (orange)?

Practically zero difference when placing the small section of OSB behind the speaker and then both the speaker and the mic.



No point wasting time doing that. But, further results will all have both these scrap OSB pieces in the measurement.

Okay, so what happens when I add a second sheet of plywood to the side (brown)?
What about adding a 3rd piece of plywood to the side, with the speaker/mic in the middle (green)?
What about shifting the speaker/mic off to one piece of plywood (purple)?

Moral of this story:

Don't bother wasting money on plywood thinking if you create enough surface area you'll get accurate results. Maybe if you buy a LOT of plywood. Maybe. But then you have to take it up and put it down for test. If you want to do ground plane measurements you need a LARGE, FLAT, concrete area. I wasted $60 on this experiment but at least I know now. So, let me save you money: Don't waste your time doing what I did with the plywood. As you can see it doesn't fix the issue entirely.

(now, maybe I could 'calibrate' the measurement for this plywood but I don't trust that.)


I considered buying the OSB with the radiant barrier to see if it would help with reflection and maybe fix the HF. I also considered buying a large sheet (4x8 foot) of the dry-erase board sold at Lowe's. Then there's the trough between 100-300Hz that *I think* may be fixed by more surface area. But I am so incredibly broke with having spent all this money on "this vs that"... I am literally selling old, beloved video games (NES nostalgia rules!) and anything else I don't need at this point to help replenish my bank account after all of this stuff I've bought the last month.

 

[ErinH]

Since I was in my backyard I thought ... might as well go ahead and throw the speaker back up on the platform and test there to see how the ground plane garage measurement compares to the 4pi "free space" measurement at 36 inches. After adjusting level differences (36 inch vs 2 meter/GP), you can see the two are practically identical!

And then I took another measurement from 12 inches:

The nearfield gives a bit better resolution on the lower end at the *potential sacrifice of the accuracy in the 1-2kHz region.

*This could be aiming; I wasn't being critical about being "dead on" to the tweeter in these tests; just getting close enough to see what difference the trend showed, if any.





Moral of this story:
Ground Plane measurements can, indeed, yield highly accurate results identical to their 4-pi counterpart. And it is a whole lot easier to place a speaker on the ground than it is to hoist it 8+ feet in the air.

 

[ErinH]

At this point I have the following options if I want high resolution and accuracy:
If I want to avoid merging NF with FF measurements then I go with the ground plane measurements in the driveway and just deal with the heat, the neighbors and other environmental factors.

If I want to avoid the environmental factors, I can use the NF/FF merging using ground-plane measurements in my garage for farfield and for nearfield, experiment with outdoor ground plane for low frequency if I am concerned about external noise influencing the result or use the standard close-mic method. I prefer ground plane to 4-pi because I've already shown it to be as accurate and raising the speaker up to 8+ feet doesn't seem to make much sense anymore when I know I can get just as good results by using the ground plane measurement in my garage. I still want to experiment with how to build a rotating platform that doesn't corrupt the data (the same way baffle step effects a speaker design) but I may have to wait until I sell a few things before I can make another trip to the hardware store.


Now, that said, I don't have to conduct ALL my tests in the same fashion. I could conduct some outdoors but if, for example, I wanted to test a speaker when it was freezing cold outside then I would just test in the garage and navigate the merging challenges.


That's where my head is at at this moment. Let's see how long this sticks. What's the over/under for a single night?

Of course, as overwhelmed as I feel doing this stuff, sometimes I feel like I'm just gonna quit.

 

[ErinH]

After sleeping on this I have decided that merging Low Frequency with High Frequency data (similar to NF/FF merging) is the only way I will be able to do this and provide the data to the accuracy I desire in most cases. That means I will have to measure ground plane both indoors and outdoors. (If I happen to have really nice weather conditions where the outdoors measurement results in the same HF response as indoors then I won't bother at all with splicing).

Now, there are options to merge the data: using the various software available (virtuixCAD, DIY'rs excel sheets) or I can use Klippel's script. However, this all requires me to merge one at a time. To get vertical and horizontal data, that's merging about 60 files down to about 30. That would be agonizing (and most importantly, prone to accidental error). So, I wrote a script in Matlab to do a "hard splice" in a batch. I provide it all the files, tell it where I want the splice to be and it calculates the delta and makes the low frequency graph merge to the upper frequency; same premise as NF/FF merging. I am ignoring phase for this aspect and focusing solely on the resultant FR. The plan is to measure all angles... I'll have to do it twice: in the garage for mid/upper and outdoors for lower. But once I do that, I can load the FR files as txt and my script will take care of the rest and throw them all on a plot. I have another version for a "soft splice" which will try to blend data points but for now the hard splice method works well as long as I choose a frequency where there is good consistency between trends (and I have the data resolution to support it). Anyway, here's an example for a single on-axis response, combining the response in the midrange area.

For the time being I have provided the output of my combined LF/HF merged data with that of ASR's as well as Buchardt's NFS data. My result is right in between theirs. And, therefore, IMHO, certainly acceptable. Take note that I have provided two of Buchardt's NFS results: one is "original" which was measured before they implemented a slight crossover modification and the other is the "updated" which was measured more recently with the modification in place.

I still need to resolve the turntable aspect to save me time of turning the DUT manually. Making the center of rotation at the baffle creates some challenges when you have a turntable that is built with the platform an inch or two above the ground; you get a baffle step effect from the lifted platform in addition to reflection from the platform itself. I've got a couple ideas to try. I'll post on that later. If all else fails, I'll just use a piece of cardboard and turn the DUT manually as I did earlier this week.



At any rate, I'm making forward progress. Yay!

 

[ErinH]

Okay. I used my script to run the horizontal polars. Looks pretty darn good.



I'll have to clean up the script but I am happy with how it works so far. I have it set up to where I point Matlab at the directory and it prompts me for the stitch frequency, I tell it where, then it dumps out the following plots. I'll have to add info for verticals. Then there's the impedance/phase, IMD, max SPL, and other tests. But I think I finally have decided on how I am going to test and the scripts will help me automate the presentation of the data so I can focus on the analysis aspect.

 

[ErinH]

... and that's it. All caught up.


I am going to try to recap everything I have learned thus far.



As we know by now, there are many ways to measure a speaker's response. All with their own pros and cons. My efforts reveal that a 2-pi ground plane measurement can give the same accuracy as a 4-pi "speaker in the sky" measurement. And it's a lot easier. So I will measure in the garage for mid/high frequency accuracy and outdoors as needed for low frequency stitching. As for rotating the speaker, I built a turntable using a NEMA-23 motor and a USB stepper motor controller but the platform height is currently of concern. I'll work through this. If all else fails I'll manually rotate the speaker. It's easier to do this than it is to wait for the perfect weather and rig up a heavy speaker high in the sky. That's the method I choose. If others choose to use the 4-pi speaker on a stand method that's fine. Again, both are fine methods.

However, it is worth noting that in my garage using the Ground Plane method I can place my DUT and mic in the center of the garage, clear items out from the sidewalls and get about 11ms free of reflection window as you can see at the bottom of this post. This is significantly better than if I were to place the speaker on a stand in my garage and measure; at which point I would get about 4ms of reflection-free time (due to the floor and ceiling in my 10 foot tall garage space). That's the difference in having a data point approximately every 90hz (11ms) vs every 250hz (4ms). And for no other reason encourages one to at least consider a proper ground plane measurement in lieu of the standard 4-pi method.


The thing to note about ground plane measurements are they need to be in an area that is obviously far, far away from the nearest boundary (wall, vehicle, etc) if you want low frequency accuracy (same as any other method). And it also needs to be in an area that has a large reflective surface. You don't need a mirror (as I have shown). But concrete is necessary and grass will not cut it (ha! no pun intended). If you are looking to perform LF response only measurements and all you have is an open field to measure in, then you can probably get away with 300hz as the maximum range (my data shows 400hz in my backyard with very low cut grass). I would not trust the data higher than that, however. Plywood is NOT a good substitution.


Any measurements performed outdoors need to be performed when there is practically no wind. A little wind is tolerable but more than a couple mph and the HF data is corrupted and considerably moreso if the wind is higher (as I have shown).


And, finally, these are all just my results. I encourage you to do your own tests if you have the desire to measure your own loudspeakers. Let my results guide you but understand that different surface areas and conditions will effect the results and you need to quantify those differences before you go willy-nilly with measurements. I have seen a lot of measurements since I started this quest that make me question the accuracy and reliability because of the knowledge I have gained from my own testing of the different methods. But, such is life. As long as you understand what you're doing and you can back it up with data then you can better provide analysis.


Hopefully some of you gained some useful information from this. I know it has been enlightening to myself as well as many others in the other forums.


- Erin






Ground Plane Impulse Reflection-Free Window ( ~ 11ms wide; 90hz data point intervals; mic 2 meters from DUT)

Stand Mounted Impulse Reflection-Free Window ( ~ 4ms wide; 250hz data point intervals; mic 1 meter from DUT)