Matt
All quite correct observations and the kind of thing that a seasoned experimenter tries. The novice tends to take "a" measurement and call it "data", but it can often be quite variant. Notice that I did not say "wrong", because its not, its a valid "single sample", its just that the data is variable and no one "sample" can be called "the answer".
All quite correct observations and the kind of thing that a seasoned experimenter tries. The novice tends to take "a" measurement and call it "data", but it can often be quite variant. Notice that I did not say "wrong", because its not, its a valid "single sample", its just that the data is variable and no one "sample" can be called "the answer".
I've also read in the posts of some people measuring room improvements who argue that they took multiple measurements, since they all were "consistent" by their standard, they only reported one. Makes me wonder if they picked the two measurements which showed the greatest change and then touted that as the improvement their bass absorber made.
I often can see a lot of changes in the results of the phase response too. I've mentioned before that I was unable to get clean phase response of your speakers (or any speakers for that matter). It seems like averaging is a much better way of getting results worth trusting.
I think a lot of people don't realize that measurements of sound response are not directly measured in the same concrete sense as say measuring temperature. It's based on an algorithm which itself is based on certain assumptions of the nature of sound in a given space. Well ok, Algorithm might be the wrong word, mathematical formula. A related issue is the idea of a statistical space, and what that entails. For instance, you can't average say 100 measurements of the "A" situation, make a change and take another 100 measurements of the "B" situation, and then compare directly and assume that the averaged differences indicate significant changes. You need to calculate an error term, since the reason for the difference may basically be a fluke drive by the amount of variance measurement to measurement.
It probably gets far too technical for the average reader, but I could post the method I used to figure out if the multi-sub approach was making a significant improvement in my room. I estimate that I took around 30 or 40 measurements per scenario, using a weighted averaging, and ran a significance test to take into account the amount error variance found measurement to measurement. I also made sure that no one location was measured only once, and weighting accomplished via more measurements at priority locations.
Using the above mentioned method I was able to calculate that 3-4 subwoofers located as you mentioned showed a statistically significant improvement in response smoothness above and beyond that of other multi-sub methods, single sub methods, 3-band PEQ, and 48 square feet of corner bass traps. Of note was that the best method, combining PEQ, 4 subs, and the bass traps was not actually significantly different than just the 3 subs method (with proper setup), when I corrected the significance for the increased number of variables in the model. The biggest issue I found, anecdotally, was that adding in the eq, bass traps, and 4th sub seemed to make setup variation a little less of an issue. I think the only reason it didn't turn out to be significantly better was that it couldn't get much better than what I achieved with the 3 subs.
I often can see a lot of changes in the results of the phase response too. I've mentioned before that I was unable to get clean phase response of your speakers (or any speakers for that matter). It seems like averaging is a much better way of getting results worth trusting.
I think a lot of people don't realize that measurements of sound response are not directly measured in the same concrete sense as say measuring temperature. It's based on an algorithm which itself is based on certain assumptions of the nature of sound in a given space. Well ok, Algorithm might be the wrong word, mathematical formula. A related issue is the idea of a statistical space, and what that entails. For instance, you can't average say 100 measurements of the "A" situation, make a change and take another 100 measurements of the "B" situation, and then compare directly and assume that the averaged differences indicate significant changes. You need to calculate an error term, since the reason for the difference may basically be a fluke drive by the amount of variance measurement to measurement.
It probably gets far too technical for the average reader, but I could post the method I used to figure out if the multi-sub approach was making a significant improvement in my room. I estimate that I took around 30 or 40 measurements per scenario, using a weighted averaging, and ran a significance test to take into account the amount error variance found measurement to measurement. I also made sure that no one location was measured only once, and weighting accomplished via more measurements at priority locations.
Using the above mentioned method I was able to calculate that 3-4 subwoofers located as you mentioned showed a statistically significant improvement in response smoothness above and beyond that of other multi-sub methods, single sub methods, 3-band PEQ, and 48 square feet of corner bass traps. Of note was that the best method, combining PEQ, 4 subs, and the bass traps was not actually significantly different than just the 3 subs method (with proper setup), when I corrected the significance for the increased number of variables in the model. The biggest issue I found, anecdotally, was that adding in the eq, bass traps, and 4th sub seemed to make setup variation a little less of an issue. I think the only reason it didn't turn out to be significantly better was that it couldn't get much better than what I achieved with the 3 subs.
The variance in measurements is indeed not well understood and this is true even in engineering and sometimes among loudspeaker designers. They want to believe that a measurement that they take is "good" if its what they expect or wwhat they would like to see - thats actually quite natural and human. But the facts are quite different. There are so many ways that measurements can get messed up. Even a temperature change or the heating system coming on can cause changes from the breeze. Watch a RTA as it fluctuates - it the measurement weren't variable this wouldn't happen. Things are a LOT better these days, I know my resolution and data stability jumped when I started to use HOLM impulse - the techniques are know for their robustness. But there is still a lot of uncertainty in any measurement and the finer the detail you try and look at the more uncertain it is. Makes refinement down to a high degre very difficult.
For sub placement and optimisation a lot of people use REW. I've found when you take the same measurement twice without changing anything, you get exactly the same line - it makes the previous line disappear because they are so close. It doesn't seem necessary to keep measuring the same thing multiple times when there is no change in the measurement. I'll have to try it again to check, but last time I tried this, that was the result. Is this what you are talking about? Or do you mean that the measurement will be different with small changes in the mic position, air movements, temp etc?
Hey Paul,
Ok so two points here. First, I find that REW is pretty course compared with better softwares I've used. REW is commonly used, its a nice free toy, but when you consider what it was designed for, the level of precision we are talking about wasn't as crucial. Conversely, software like HOLMimpulse was designed around the most current thinking in measurement software design, and is as accurate as it gets. This really is less of an issue at these low frequencies than at higher frequencies, but it may explain your experience.
However, I also have REW on this laptop, and I too have used it. I have also compared the results between it and various other softwares I have, some I paid for, some were free but more accurate. It's close, but never identical. I too have found that REW doesn't change much when I keep measuring over and over again, but I don't trust its accurate given my other results. However, even with REW, take a measurement, but don't move the mic or anything else. Turn the computer off, walk away, do something else for a few hours, walk back, turn the computer back on, and take the measurements again, and tell me if they are identical? Its been my experience that even with REW this isn't the case. Now look at that amount of variation (its small) and compare it to the improvement you might see with some bass traps for instance (its often also small at these very low frequencies) and it brings the results into question. That is what I am talking about.
Add to that the proper way to setup the bass response in a room, given that a home theater has multiple seating positions, and not just one money seat, and you have to take multiple measurements at multiple locations. Even for the money seat you need to take more than one. The bass response varies drastically with even small movements of the mic, and consider this relative to your head, you realize rather quickly why multiple measurements should be taken at each listener position about the area you intend to listen in. You then need to average these measurements to get the best estimate of the response for each listening position, and all of them need to be averaged to get your best estimate of your room response.
So again, the variation you get, which complicates finding small improvements in a room come from, error in the software itself at such low frequencies, and second, the need to measure around multiple locations due to actual variations in response over small changes in distance.
Ok so two points here. First, I find that REW is pretty course compared with better softwares I've used. REW is commonly used, its a nice free toy, but when you consider what it was designed for, the level of precision we are talking about wasn't as crucial. Conversely, software like HOLMimpulse was designed around the most current thinking in measurement software design, and is as accurate as it gets. This really is less of an issue at these low frequencies than at higher frequencies, but it may explain your experience.
However, I also have REW on this laptop, and I too have used it. I have also compared the results between it and various other softwares I have, some I paid for, some were free but more accurate. It's close, but never identical. I too have found that REW doesn't change much when I keep measuring over and over again, but I don't trust its accurate given my other results. However, even with REW, take a measurement, but don't move the mic or anything else. Turn the computer off, walk away, do something else for a few hours, walk back, turn the computer back on, and take the measurements again, and tell me if they are identical? Its been my experience that even with REW this isn't the case. Now look at that amount of variation (its small) and compare it to the improvement you might see with some bass traps for instance (its often also small at these very low frequencies) and it brings the results into question. That is what I am talking about.
Add to that the proper way to setup the bass response in a room, given that a home theater has multiple seating positions, and not just one money seat, and you have to take multiple measurements at multiple locations. Even for the money seat you need to take more than one. The bass response varies drastically with even small movements of the mic, and consider this relative to your head, you realize rather quickly why multiple measurements should be taken at each listener position about the area you intend to listen in. You then need to average these measurements to get the best estimate of the response for each listening position, and all of them need to be averaged to get your best estimate of your room response.
So again, the variation you get, which complicates finding small improvements in a room come from, error in the software itself at such low frequencies, and second, the need to measure around multiple locations due to actual variations in response over small changes in distance.
I want to thank you guys for this latest round of posts as I am new to the room measurement arena. I am about to start taking measurements and it seems kind of daunting as I am no computer wiz and have never done any measurements before so this will be a first. I am planning on using the rew software to start with. One question first should I wait until I get the Abbeys I have ordered into the system or go ahead now and do measurements as I have the subs that I intend on using already in the system? And 2 can someone give me an idea of the cost of the HOLMimpuse software and where to get it. Also the ideas on averaging sure make a lot of sense even to a guy who has'nt even measured anything yet. Thanks again fellas.
Paul
At LF you can get two measurements taken within a short time intervale to match sure, but HFs, system teardown and re-setup, different days - your unlikely to get the same measurements under those conditions.
DWR - How the Abbeys integrate into the subs is a critical factor and so until you have those in place you can't do much./ The subs are "fit" to the Abbeys, so they have to be in place.
At LF you can get two measurements taken within a short time intervale to match sure, but HFs, system teardown and re-setup, different days - your unlikely to get the same measurements under those conditions.
DWR - How the Abbeys integrate into the subs is a critical factor and so until you have those in place you can't do much./ The subs are "fit" to the Abbeys, so they have to be in place.
Hi Mike.
Can one use HOLMImpulse to determine the optimal gain/phase/crossover settings for three subs? Are you aware of any step-by-step directions for doing this set-up that exists out on the web? Setting up my subs with my Abbey's is something I still need to do as I move into my new house/listening room.
Thanks.
Can one use HOLMImpulse to determine the optimal gain/phase/crossover settings for three subs? Are you aware of any step-by-step directions for doing this set-up that exists out on the web? Setting up my subs with my Abbey's is something I still need to do as I move into my new house/listening room.
Thanks.
Anand
Holm is somewhat difficult to use in this situation because it does not like to do steady state LF measurements. But its possible. I'm going to try and develop a technique and post it since somewhere someone has to do this.
Set the sample rate as low as possible and there is a setting for "steady state" although its not called that and I don't remember what its called. Averaging is not easy either. Basically what I want to do is have HOLM write out the data and write a windows program to analyze it and fit the curves. HOLM can't do everything, but it takles raw data very well. A better link between HOLM and Windows would be great, but for now writing and reading text files works. The Polar Plot software that I just posted does this.
Holm is somewhat difficult to use in this situation because it does not like to do steady state LF measurements. But its possible. I'm going to try and develop a technique and post it since somewhere someone has to do this.
Set the sample rate as low as possible and there is a setting for "steady state" although its not called that and I don't remember what its called. Averaging is not easy either. Basically what I want to do is have HOLM write out the data and write a windows program to analyze it and fit the curves. HOLM can't do everything, but it takles raw data very well. A better link between HOLM and Windows would be great, but for now writing and reading text files works. The Polar Plot software that I just posted does this.
and I might note, since I'm the one who talked so much about this averaging technique, I don't use HOLMimpulse for the averaging. In fact, I've only started using this software in the last 3-6 months. In the past I used a software from Germany which allowed me to output the individual measurements as data files, and then I would import them into as raw data into SAS, and averaged the data in that software. SAS is a statistical package for stat experts, not average joe, so this really isn't an option for most people. Matlab is an easier and cheaper option if people really want to try this, but you still need to have somewhat of an understanding of statistics (A) and the concept of a statistical acoustic space (B). I'm quite savvy with acoustics, I'm quite novice when it comes to acoustics, but I'm learning.
SAS is an excellent statistics software and can do anything you could possibly want it to, as long as you know how to do it. However, its not very user friendly, and it's graphing package is horrible. For my own research work I far prefer SPSS for most of what I do, as I find, especially in the newest versions, that I can do everything I could do with SAS, and if I don't know the syntax code, I can always go back to the point and click method.
SPSS is specifically designed for social science research and is really the standard software in my field. When I applied for jobs, knowing SPSS was preferable to most employers than SAS, as even if I was equally savvy with either, they primarily knew SPSS. I see no reason why this same kind of analysis couldn't be done with SPSS as well, but for most people I think MATLAB is better. It has a lot better interface and graphing abilities than SPSS or SAS, and you can apply a broader range of mathematical algorithms to the data unrelated to statistics.
SPSS is specifically designed for social science research and is really the standard software in my field. When I applied for jobs, knowing SPSS was preferable to most employers than SAS, as even if I was equally savvy with either, they primarily knew SPSS. I see no reason why this same kind of analysis couldn't be done with SPSS as well, but for most people I think MATLAB is better. It has a lot better interface and graphing abilities than SPSS or SAS, and you can apply a broader range of mathematical algorithms to the data unrelated to statistics.
Just a thought. Why not talk to the people at HOLM Acoustics and see if they like your ideas?
Just a thought. Why not talk to the people at HOLM Acoustics and see if they like your ideas?
I don't know anyone there, but its a good idea. Do you know, are they going to continue support of the software? HAd Ask still been on the forum I was going to approach him with exactly that thought, but he's not arround anymore.
Have no connection to HOLM Acoustics. Try with the contact info at their homepage? (Read about the lack of presence of Mr Ask after posting to you).
What is it that makes HOLM Impulse stand out from the rest of the measurement programs?
I have not used it myself yet. Using Arta at the moment. Arta has a flawed user interface if you ask me. Arta is a good program, but would have been an excellent program if these flaws were fixed.
What is it that makes HOLM Impulse stand out from the rest of the measurement programs?
I have not used it myself yet. Using Arta at the moment. Arta has a flawed user interface if you ask me. Arta is a good program, but would have been an excellent program if these flaws were fixed.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.