HiFiNutNut said:
My mains do the same thing. I would count them as "LF sources" yes. With that situation I would use at least two more subs, maybe three - I highly doubt that you will need four. Put one in a corner behind the mains (actually depends on your room setup however). Another one at the other end of the room somewhere, but not in a corner. The third will be almost irrelavent where it goes. Then set them up per Markus posts.
Good God, I leave this thread for a day and look what happens!
Earl, if yur still there. I'd like to understand how your method generates consistant seat to seat responses. From what I've read you really only make one measurement. Yes, you move the mic around, but it's still onely one measurement. If I understand you, you are saying that in making the response in that area smooth, and the fact that you are using multiple subs, it will naturally tend to be consistant. In my investigations I have found some anecdotal evidence for this, though I did not find that it was always the case. But understand this: this is no substitute for making multiple measurements and optimizing them. Whether it is "good enough" is one thing, but it is not as good as can be had using multiple measurements. Your method is trial and error. I guess you could say that SFM is trial and error as well, but it uses multiple mic locations, multiple (potential) sub locations, and can easily analyze 20 million "trial and errors" in a few minutes.
FWIW, I think there's plenty of room for all these discussions on this thread.
Earl, if yur still there. I'd like to understand how your method generates consistant seat to seat responses. From what I've read you really only make one measurement. Yes, you move the mic around, but it's still onely one measurement. If I understand you, you are saying that in making the response in that area smooth, and the fact that you are using multiple subs, it will naturally tend to be consistant. In my investigations I have found some anecdotal evidence for this, though I did not find that it was always the case. But understand this: this is no substitute for making multiple measurements and optimizing them. Whether it is "good enough" is one thing, but it is not as good as can be had using multiple measurements. Your method is trial and error. I guess you could say that SFM is trial and error as well, but it uses multiple mic locations, multiple (potential) sub locations, and can easily analyze 20 million "trial and errors" in a few minutes.
FWIW, I think there's plenty of room for all these discussions on this thread.
gedlee said:
I also did a literature search for my original paper, also couldn't find much. Single sub optimization, sure (Groh was an early one). Multiple subs with matrix inversion type optimization, sure, it's all over the place. Floyd Toole had talked about the idea of cancelling modes in general in some of his powerpoints. But not much else. I'm not sure if the Genelec patent (DBA precurser) predates it.
cap'n todd said:
Todd
I think that you misunderstand. Taking several measurements and averaging them is the same thing as moving the mic over a long time window over the same spatial points. Both methods yield the same result. (See my 1991 AES paper on spatial averaging.)
I actually do take multiple measurements - six - because its easier, but make no mistake about it the results are exactly the same as if you had moved the mic over these same six locations.
As to your SFM doing mega calcs - well I think that you will find, down the road, that this is a waste of time, but I'll let you discover that.
You seem to be getting confused over the statistical properties of rooms - this is exactly the same problem that you and I have with your sample points NOT being statistically independent and the statistical bias in your results. I'm guessing that statistics is not you strong suite. (No offense intented, its just that statistics is where we always seem to disagree and it IS a specialty of mine.)
Pan said:Ah now you're harsh Wayne, let them have fun believing they are breaking new ground!
Here's a little test and article a guy did to get better understanding. It's in Swedish but the pics and graphs should be interesting for some of you.
In Sweden I know people that have been doing multiple subs for decades.
http://www.lts.a.se/Portals/0/artiklar/Rummet2.pdf
/Peter
I keep seeing references to "doing multiple subbwoofers", but there's a difference between "doing multiple subwoofers" and the methods being discussed here. Nobody here is claiming that they were the first to use multiple subs in a room. It's about the intelligent use and setup of multiple subs, i.e. optimization.
Also, there's a difference between saying "I did that 20 years ago" and having done a scientific investigation and published results. For the DBA, this has been done. For my subwoofer placement optimizatin and SFM, this has been done. For what Earl is talking about, I haven't really seen it. He does has a clearly written discription of it, and some sample results, but it's not really a scientific investigation, ast least not that I've seen. And I have seen some of his other research, so I know he is capable of doing a rigorous investigation.
---------
Todd Welti
Research Acoustician
Harman International
gedlee said:
Earl, I think you misunderstand. I do not average my measurements. I look at each individually, thus I can talk about seat to seat consistancy. In fact it is the cornerstone of my approach. You only have one measurement (i,e. one response curve), once you average the six. It does not take a PHD in statistics to figure out that you cannot say what the seat to seat consistancy is in your optimization, since you only have ONE measurement. If you are looking at the six measurements separately then thats another thing.
The other quesiton about how to sample the seating area (random versus grid) is interesting, and I thank you for bringing it up. For anyone not familiar with this issue, Earl suggested that a grid is an incorrect way to sample a seating area in a model, and that it shuld be random. My response was that
a. I sampled fairly densly at 2' intervals.
b. Real home theaters very often do have symmetric seating layouts.
cap'n todd said:
I agree with this, but there is probably some correlation between smooth average response and spatially smooth response. If you have some data on hand, could you check what kind of correlation we have between rankings of MSV and rankings of the other metric (the one for smooth average)? It could be interesting.
Another question regarding metrics, does it make more sense to use dB values or just raw magnitude values? dB would emphasize nulls more, with say 0 and -20 dB averaging to -10 dB while 1 and 0.01 would average to about 0.5 (-6 dB).
This is probably known my many here. But stating at the beginning might help. The thing about low frequency response in acoustically small rooms is that the modes are widely spaced. The result is that there are deviation form the sources free field transfer function which arise from two different mechanisms 1) peaks arise due to energizing under-damped modal resonances. 2) dip and/or nulls arise due to cancellation between out of phase modes. The key here is that the peaks are primarily single mode phenomena where as the dips are multi-mode. Eliminating single mode behavior is fairly easy form a theoretical point of vise, by locating sources at position where modes have nodes and by locating multiple sources which cancel modes (equal amplitude but inverted phase). When a mode is so canceled it will not contribute to the SPL at any location in a room, except when the position is very close to one source. Eliminating the dips is a little harder because of the multi mode dependence. To remove a dip at low frequency requires that all the offending modes be canceled. Simply moving sources around so that the response looks smooth in one position or another can not guarantee that the response will be smooth in another position because the phase relationships between modes, as well as the amplitude contribution from the different modes, is position dependent. SO, again, to guarantee smooth response over a wide area the requirement is the eliminate the contribution of all the modes which make significant contributions to the SPL. Todd's,s 4 woofer array is such an approach. In Todd's approach all odd number modes in the L and W directions theoretically cancel as does the first even order mode in the L and W directions. Along with this a number of tangential and oblique modes are also canceled. However, as Todd note in his paper, modes in the vertical direction are not addressed. If you look at simulation of Todd's approach, changing the vertical position, then there is significant change in the response as the vertical position changes. If listening height is constant this isn't a big deal, but there would certainly be a potential to hear different bass response when sitting or standing. The DBA basically take an array such as Todd's and places it of the front wall, canceling modes in the W and H directions. The second, identical array on the back wall, with inverted phase and delayed, is an attempt to remove low frequency reflections from the L direction, thus eliminating L direction modes. It goes the next step and yield, in theory, very uniform and smooth response independent of position, except when very close to a source. Both Todd's method and the DBA take advantage of symmetries in the modal behavior that exists in a rectangular room. With a 3 woofer array in a rectangular room there is a lack of symmetry. Once the first woofer is placed, say in a corner, when the second woofer is placed , if it is against a wall (doesn't matter if displace vertically or horizontally from a corner) the odd modes in the across room direction (source to source) will theoretically cancel. However, when the third woofer is place, It will re-excite the modes canceled by the first two woofers unless it is place at a node associated with the canceled modes. When simulation of the 3 woofer array is attempted it is possible to find position where the response is fairly consistent over a moderate listening area, but it is not very smooth. The only way to effective smooth the response is to add room damping and/or equalization. However, since many modes are still excited the damping or equalization can not correct for the position dependent variation in phase between contributing mode. Todd's approach and the DBA are not directly translatable to non rectangular rooms. Multi subs with arbitrary placement, whether 1, 2, 3, ..., n may certainly prove more effective in room of odd geometry but I see no scientific method here. It is ad hoc pushing source around until you get something acceptable. While an acceptable result is likely obtainable, there is little of a method here. It's like saying the way to grandma's house is through the woods. But which path to take is left as a mystery.
Breez said, "but there is probably some correlation between smooth average response and spatially smooth response. " Actually there is none at all. To have a correlation you need to know the number of measurement, the standard deviation and the mean, at least, before you can say much about the response. You also need to know something about the measurement. A long length MLS or impulse response may yield a different result than a swept or stepped sine and certainly a different result than a 1/3 octave pink or white noise measurement.
Breez said, "but there is probably some correlation between smooth average response and spatially smooth response. " Actually there is none at all. To have a correlation you need to know the number of measurement, the standard deviation and the mean, at least, before you can say much about the response. You also need to know something about the measurement. A long length MLS or impulse response may yield a different result than a swept or stepped sine and certainly a different result than a 1/3 octave pink or white noise measurement.
Any real room (with furnitures, open areas, leaks, boundaries with different properties, and with people in it...) will act to some degree as a non rectangular one.john k... said:
So what it the interest of using simulations if you cannot accurately model the room in the first place?
In such a situation Geddes' approch seems to be the most sensible one.
john k... said:
By correlation I meant between the rankings generated by Welti's SFM for different configurations.
http://en.wikipedia.org/wiki/Spearman's_rank_correlation_coefficient
High correlation would mean that the best configurations for spatially smooth result are also among the best for smooth average and vice versa.
How about this for a compromise :
- 2 large main floorstanders with 12" woofers, and good low frequency performance, i.e good enough to be considered 2 subs.
PLUS
- a single large subwoofer placed at other location
Effectively 3 subwoofers in "random" locations, but much simpler and a more efficient use of space & resources.
- 2 large main floorstanders with 12" woofers, and good low frequency performance, i.e good enough to be considered 2 subs.
PLUS
- a single large subwoofer placed at other location
Effectively 3 subwoofers in "random" locations, but much simpler and a more efficient use of space & resources.
cap'n todd said:
No Todd, I undersood completely, I always have, and again, I think that it is you who do not. It makes no difference WHEN one averages the signals in a linear system, first, or in the end. When one seeks the minimum spectral and spatial variance - seat to seat - this can be gotten with an average of the seats first and then fit the parameters or fit the parameters and then average the seats. Its all the same since in the edn its a single sound field that results and all paths to this result will be basically equivalent.
You are worrying about small details that end up making no difference. Its like "Do I average the dB or do I average the linear values?" Well, it turns out that in the limit of a number of measures this too makes little to no difference (been there done that). Maybe what you are missing is that spatial variance, when taken over a number of spatial points, becomes a spectral variance when the data is averaged. Hence I can now solve the problem as simply a spectral one.
What we are both seeking is the same thing, but we get there in different ways. Granted yours is precise, accurate and yields LOTS of data, its also very expensive and time consuming. Mine less accurate, not as precise, yields little data but its VERY inexpensive and easy. If they both end up with "good enough" ... well I think that choice is obvious.
You are right that I "could" do a scientific study of this technique, but thats not going to happen. I don't work in research any more and there is no time for academic studies. Lidia and I are planning (when we get the time) to do some psychoacoustic studies, but they will be on the perception of early reflections, which, I believe, is the more important aspect of the audio problem. I don't completely agree with Floyd on this topic. You and I agree on the goal, just not how to get there. Floyd and my differences are more fundamental.
I've been following this thread with great interest. I have no input on the technical aspect other than to say that when I used a sub, I found that running the mains full range and crossing the sub at 45 worked best, just the old trial-and-error method and for sub placement in the room as well.
I would like to make a few comments, however. I for one am interested in both approaches being presented here, the "good enough" approach of optimizing for limited subs and placement as well as the effort to analyze the room and optimization from the theoretical perspective. The former is how one eventually must finish a scheme to be used, but I see no reason to minimize or trivialize the latter approach of looking for a theoretical basis to find an optimum.
It seems to me that both should be part of the process, if one has access to software to further it. Were I to do multiple subs, I would much prefer to use the software for the initial phase, that being to select the most likely placement that is optimal. Then I would move on to the next step to optimize the driver responses with the necessary EQ.
My point is that the optimal is the goal, in all aspects, placement as well as the EQ afterwards, with whatever constraints on placement being set for whatever the reasons. This is no different than how I design mains. Each one has different requirements for driver placement with EQ applied afterwards. Using CAD for diffraction analysis for baffle dimensions and speaker placement is all done before building it, then optimizing the EQ of each driver after measuring. This is how I would prefer to address room subs and EQ as well. Software that could be used as a guide to optimal placement should come first if it's available. The followup EQ methodology is also indispensible. They should go hand in hand.
I really don't see why there's any animosity between the two camps here. The goal is the same. I'm learning from both sides and would not likely follow this if only one aspect were to be addressed. It would be rather boring, to be honest, and not nearly as informative.
Dave