Finding peak of two measurements ??

[mamba76]

Hi all. Iv attached two impulse filter wav responses. They are cuts I need to make. Is there a way or arithmetic process that can combined these responses. Like if you were to draw a line underneath these two to get one response or just sticking to the low points of both. Iv attached an image to help. My photoshop free hand skills suck but the thin red line is what Im after. just the lower peaks of both responses.
Thanks

 

[Matthew J Poes]

Hi all. Iv attached two impulse filter wav responses. They are cuts I need to make. Is there a way or arithmetic process that can combined these responses. Like if you were to draw a line underneath these two to get one response or just sticking to the low points of both. Iv attached an image to help. My photoshop free hand skills suck but the thin red line is what Im after. just the lower peaks of both responses.
Thanks

Can you explain more of what you are trying to do?

If you multiply two response files you get a response file that is equal to file A modified by the transfer function of file B. For example, if you have a speaker response file and a filter response file (transfer function of the filters) and you multiply them, the resultant file is the speaker response file modified by the filters. Is that what you need to do?

If you took two filter files and multiplied them, it would be the same as you would get if you applied those two filters at the same time, they would essentially modify each other to equal a new third filter transfer function equal to what would happen if the two independent filters were active on the same signal at the same time.

adding and subtracting can be thought of as adding and substracting sources. It isn't so intuitive. Most people think of adding as something like 2-1=1. If the average response of a file is 100dB and a second file with the same exact response was subtracted from it, you would think you get 0, but you actually get 97dB. That is because you simply removed a source. If you wanted to know the difference (what someone might intuitively think of as subtracting) then you need to actually divide. That creates the difference of two responses. This is one way to get the transfer function from a modified signal. Take the original signal, the modified signal, and divide them. The resultant signal is the difference between the two, or in this case, the transfer function of the filters (which could, for example, be the effect of a speaker system and room on a measurement sweep signal).

 

[ddude003]

Hi Mamba, I think the function you are looking for is a pair wise vector minimum... In other words you want the lesser values of the two vectors that would create/be a 3rd vector... I see that REW has several functions that perform the math Matt suggests above and no MIN or MAX function... Maybe a REW feature request???

By the way there is an existing Vector Average...

 

[mamba76]

Hi, thx for the help guys. Yeah ddude, to get the peak of normal FR measurements or lower peak of these impulses is what Im after. Why Im trying to do it is a little more complicated to explain! Its to balance and optimise the responses of the two main car speakers at drivers right and left ears. This includes the reflections that add to opposing responses (UK driver so mostly right side) Been trying various ways for a couple of years but have recently gained some new tools and knowledge from this site. Its just putting it all together to get my own process nailed so its quicker the next time.
Like Matt says arithmetic with responses is not intuitive and still not easy to grasp so I appreciate Matt's explanations. This stuff just ain't around on google.
Nearest Iv got is taking an average of 2 FRMs adding that measurement to itself then bring the offset down 4db. This don't work well with these filter responses tho. I was thinking of making a target house curve the exact shape of what I want but can only load one FRM at a time.
Its basically a merge but different from REWs (don't know what that is for either)
A min/max feature would be great, so would target curve designing software! )
whats the vector average?

 

[Matthew J Poes]

Hi Mamba, I think the function you are looking for is a pair wise vector minimum... In other words you want the lesser values of the two vectors that would create/be a 3rd vector... I see that REW has several functions that perform the math Matt suggests above and no MIN or MAX function... Maybe a REW feature request???

By the way there is an existing Vector Average...

If that is what he wants to do, then I misunderstood his question completely. That would imply calculating the minimum amplitude point for a pair or multiples of a response file (i.e. draw the vector of the minimum of the set of response files). What purpose would that serve?

 

[Matthew J Poes]

Hi, thx for the help guys. Yeah ddude, to get the peak or lower peak of these or normal FR measurements is what Im after. you could have 3 or 4 FRMs and want the peaks known, Why Im trying to do it is a little more complicated to explain! Its to balance and optimise the response of the two main car speakers, including all the reflections, cancellations and all the other strange phenomenons to improve the image/soundstage from the worst possible place to put speakers - a car door! Been trying various ways for about 2 years now and have some good tools and knowledge. Its just putting it all together to get my own process nailed so its quicker the next time.
Like Matt says arithmetic with responses is not intuitive but Im starting to understand it more. I learnt this just through trial and error.
Nearest Iv got is taking an average of 2 FRMs adding that measurement to itself again then bring the offset down 4db. This don't work as well with these filter responses. I was thinking of making a target house curve the exact shape of what I want but can only load one FRM at a time.
It basically a merge but in the true sense of the word.

I'm afraid it is still unclear to me why you would want to know the minimum of the pair of filters? What purpose would that serve?

I optimize sound systems all the time, I was even an IASCA champion at one point, I've never heard of or had to do what you are describing. I think we can help, but I need to know more about what you are trying to accomplish.

Maybe explain in more tangible terms what you are doing, that might help.

Try to remember that there is no amount of DSP processing that can overcome certain placement and dispersion related limitations. You can never make a speaker in a car door perform like a speaker in an enclosure in a room 12 feet in front of you. That is impossible. What you can do is adjust a set of speakers placed in disparate locations so that their phase tracking and time alignment is correct, potentially even ensure they are linear phase (i.e. with FIR or mixed phase filters). You can fix the time alignment problem between the left and right speakers for a single listener (but you can never fix it for all listeners in a car). As such, you don't want to go too nuts with the filters because there is only so much you can fix.

 

[ddude003]

Hi Matt and Mamba, I agree with you Matt that what Mamba was, at first, asking for was difficult to discern... However, it became obvious to me upon seeing his "mock-up" photoshopped attachment that he was looking for a MIN and MAX function to go along with the Average function... In many applications MIN, MAX and Average math functions are used... However, I am not sure if these will have the expected value Mamba is looking for... I do encourage Mamba's pursuit and exploration into the unorthodox areas of signal processing and acoustic research...

"Good math and small teams win..." -Alan Kay-

 

[Matthew J Poes]

Hi Matt and Mamba, I agree with you Matt that what Mamba was, at first, asking for was difficult to discern... However, it became obvious to me upon seeing his "mock-up" photoshopped attachment that he was looking for a MIN and MAX function to go along with the Average function... In many applications MIN, MAX and Average math functions are used... However, I am not sure if these will have the expected value Mamba is looking for... I do encourage Mamba's pursuit and exploration into the unorthodox areas of signal processing and acoustic research...

"Good math and small teams win..." -Alan Kay-

That is fine as long as they serve a meaningful acoustic purpose. It is unclear to me what purpose it serves and I fear it may be an endeavor already well studied and understood.

There is certainly good math to be had in acoustical physics, but we still need to understand the purpose of a min.max function to make sense of it.

 

[mamba76]

Hi Matt and Mamba, I agree with you Matt that what Mamba was, at first, asking for was difficult to discern... However, it became obvious to me upon seeing his "mock-up" photoshopped attachment that he was looking for a MIN and MAX function to go along with the Average function... In many applications MIN, MAX and Average math functions are used... However, I am not sure if these will have the expected value Mamba is looking for... I do encourage Mamba's pursuit and exploration into the unorthodox areas of signal processing and acoustic research...

"Good math and small teams win..." -Alan Kay-

Matt respect for winning IASCA! Thats what I have in mind maybe - one day. I really do understand all that you've said. but don't think I can explain what Im doing easily, not until I can prove it works. I only care about the drivers listening position at the moment.
I have a target curve and identical DIY response (all 0deg phase) for all my drivers including an overall target that when I want to tweak what I hear I go and redo these targets. measurements are always done with mic at drivers left ear and drivers right. The ones i use to judge what I hear to what I see is with all, or in this case just the mid range pair, of speakers playing, take the information received by the left ear and same with right ear then average. For me the better the info matches on both sides the better the image with regards to intensity. Timing these below 400hz is easy as there is only one center with regards to bass and this can't be EQd. So I use the mid target as a guide to cut the peaks from measurement made at Rear Rmid, Rear Lmid, Lear Lmid and Lear Rmid so they are just under the guide this way there is no peaks / resonances when they are playing together.
When you measure one speaker at one ear then measure the same speaker at the other ear obviously theres a difference if you apply EQ to it there will always be that proportionate amount of info received by the other, like a signature or footprint. I want that footprint to be a little less than the direct sound so you hear the from the intended direction and not the reflection. Sorry this is the best explanation I can give. I think maybe I have gone slightly mad but well done if you read this! These filters are what I want applied to one speaker and are from the direct sound and the reflected sound.

 

[Matthew J Poes]

Matt respect for winning IASCA! Thats what I have in mind maybe - one day. I really do understand all that you've said. but don't think I can explain what Im doing easily, not until I can prove it works. I only care about the drivers listening position at the moment.
I have a target curve and identical DIY response (all 0deg phase) for all my drivers including an overall target that when I want to tweak what I hear I go and redo these targets. measurements are always done with mic at drivers left ear and drivers right. The ones i use to judge what I hear to what I see is with all, or in this case just the mid range pair, of speakers playing, take the information received by the left ear and same with right ear then average. For me the better the info matches on both sides the better the image with regards to intensity. Timing these below 400hz is easy as there is only one center with regards to bass and this can't be EQd. So I use the mid target as a guide to cut the peaks from measurement made at Rear Rmid, Rear Lmid, Lear Lmid and Lear Rmid so they are just under the guide this way there is no peaks / resonances when they are playing together.
When you measure one speaker at one ear then measure the same speaker at the other ear obviously theres a difference if you apply EQ to it there will always be that proportionate amount of info received by the other, like a signature or footprint. I want that footprint to be a little less than the direct sound so you hear the from the intended direction and not the reflection. Sorry this is the best explanation I can give. I think maybe I have gone slightly mad but well done if you read this! These filters are what I want applied to one speaker and are from the direct sound and the reflected sound.

Thanks for the props, but...I should say I won a state championship. While I made it to Nationals, I didn't compete in enough total competitions to be competitive for the final national championship. If I recall correctly I placed 7th overall, but I never came in less than 2nd place in any single competition, I just didn't have enough points. I got out of it because I learned something. obviously, everyone who was competing at this level had sponsorships. You would need to be very wealthy otherwise. All of my equipment was cost or free and much of it was prototype odd-ball stuff. Officially I was part of Kicker competition, but I had Eclipse, Nakamichi, Dynaudio, and Veritas audio (I think they are now defunct). My last year of competition I gave up Kicker and took on Memphis audio instead. What I realized was that winning wasn't about anything more than having the most-best equipment and going to a lot of shows. It really wasn't about sound quality (which is what I cared about). By the end I was using DSP equalization using Dolby Lake processing and I could easily score perfect in most areas. I felt like I was cheating. I always got a perfect frequency response score because....I knew were the mic had to always go and I made a specific mode that measured perfect flat in that mic position.. It wasn't even close to the mode I listened in.

I'm afraid I still don't understand what you are doing though. How does a filter come from direct and reflected sound? Do you mean that is the shape of the filter you generated based on a measurement of both? In a car you always have direct and reflected and separating the two would be exceedingly difficult (given that the car is so small that the reflections are all generally under 5ms (other than those coming from the back of the car or opposite side).

When you say you are trying to increase the direct sound relative to reflected, are you trying to actively cancel the reflected sound somehow? Li,ke use the rear speakers to cancel reflections?

Could you post your response measurement MDAT files so I can maybe see what is going on? Maybe I can better understand the current issues.

I would need to see the total response of the left and right side. If it's just individual drivers it won't do me a lot of good.

 

[mamba76]

More money more points, that figures, maybe Ill give IASCA a miss. So you did what volkswagon did with emmisions lol.
What Im doing is just another experimental session so I cant explain it but when you eliminate certain ways you end up with somethings that works. That's the hobby I suppose. A peak would be handy but Ill find another way.

 

[ddude003]

Matt, Vector Average, MIN and MAX are in just about every math library and toolbox I can think of... R, Octave, Matlab and Wolfram Mathematica, as well as many math and science applications...

I would propose that if Vector Average (already included in REW) is meaningful then I would suggest that MIN and MAX are too...

I respect your knowledge and experience... And in this case Mamba is the arbiter of meaningful acoustic purpose...

 

[Matthew J Poes]

More money more points, that figures, maybe Ill give IASCA a miss. So you did what volkswagon did with emmisions lol.
What Im doing is just another experimental session so I cant explain it but when you eliminate certain ways you end up with somethings that works. That's the hobby I suppose. A peak would be handy but Ill find another way.

Sorry, I can't be more helpful!

Yeah I got really disenchanted with IASCA and left the hobby. I also had to focus on school. The year I quit was my last year of undergrad, I was working on a Cornell Research Internship, and applying for PhD programs. It didn't make sense to continue in a hobby that made me very little money and was really seriously time consuming.

The entire car audio industry also nearly collapsed around that time. It was on it's way up due to movies like Fast and Furious and then that huge boom died. I was involved before and after the bust and I was shocked at the change. We all were riding that rise like crazy. The money in the industry was so great that those of us with sponsorships were having unheard amounts of gear sent to us. The teams could afford to truck our cars all over the country as part of the deal. By the end only top competitors could do that, the rest had to pay. I wouldn't be shocked if nobody is having that level of service anymore.

 

[Matthew J Poes]

Matt, Vector Average, MIN and MAX are in just about every math library and toolbox I can think of... R, Octave, Matlab and Wolfram Mathematica, as well as many math and science applications...

I would propose that if Vector Average (already included in REW) is meaningful then I would suggest that MIN and MAX are too...

I respect your knowledge and experience... And in this case Mamba is the arbiter of meaningful acoustic purpose...

Vector Average's acoustic purpose is that it allow frequency averaging of spatial measurements that maintains the timing information.

I know that math programs have this capability. I am a statistician, I use R, SPSS, SAS, and you are right, these abilities are in them because that flexibility is needed, it serves a purpose. I'm sure there are filter analysis or algorithms that would need this, but REW is more of a basic measurement software. When designing filters for sound systems, you often don't need that level of sophistication unless using it in some kind of algorithm that is trying to use that information, say the vector min or vector max, to calculate something for a filter. In setting up a car or room, you wouldn't need that, but you might need vector average for the timing piece. That is because it's nice to still be able to look at the impulse, waterfall, wavelet, etc. of the average response.

There very well could be a use for min and max, I just can't think of one that makes a lot of sense. John is really the person to ask on that front. He knows far more than I do. My training in this was an unfinished degree in electrical engineering and later taking courses in acoustical physics and audio engineering (neither earning a degree, just to enhance my knowledge). I'm certainly not all knowing.

 

[mamba76]

I think IASCA is in the states, in the UK closest thing is EMMA awards. They do have certain criteria for each entry level and rules attached but it sounds like a head ache anyhow.

I should add that if I had one peak response I would create or import a flat line target then run REW EQ or my own and apply that to the speaker. You may need to reverse every filter to get a mirror image then apply. Played about with it but no holy grail yet.

if I merge A (Lear Rmid) and B (Rear Rmid) at 500hz it sort of works for these 2.

 

[Matthew J Poes]

I think IASCA is in the states, in the UK closest thing is EMMA awards. They do have certain criteria for each entry level and rules attached but it sounds like a head ache anyhow.

I should add that if I had one peak response I would create or import a flat line target then run REW EQ or my own and apply that to the speaker. You may need to reverse every filter to get a mirror image then apply. Played about with it but no holy grail yet.

if I merge A (Lear Rmid) and B (Rear Rmid) at 500hz it sort of works for these 2.

What does lear and rear refer to specifically here?

 

[ddude003]

What does lear and rear refer to specifically here?

Somewhere in the above thread Mamba was talking about taking measurements at one ear then the other... Guessing left ear and right ear...

 

[Matthew J Poes]

Somewhere in the above thread Mamba was talking about taking measurements at one ear then the other... Guessing left ear and right ear...

Oh right. That was kind of obvious.

 

[mamba76]

Done it. Used a series of 'merge A to B' function at the intersection points and accumulated both responses into one.

whether this is any help overall I don't know but its good to know ))

 

[ddude003]

Nice work Mamba... Good luck with your project...

"We keep moving forward, opening new doors, and doing new things, because we're curious and curiosity keeps leading us down new paths." -Walt Disney-

 

[ddude003]

Oh right. That was kind of obvious.

Puffed corn doodles are not a breakfast cereal... ;^)

And this thread brings up some important areas for REW... The ability to script set of actions that we use repeatedly... And the ability to script our own maths to apply to the various data sets...

 

[mamba76]

Thanks all. Will let you know if any success..

 

[ddude003]

Now that I have thought about this a little bit maybe Mamba is approaching car eq correction in a manor similar to headphone eq dual channel cross-feed algorithms. As in a separate eq for each channel and/or driver...

 

[Eric SVL]

And this thread brings up some important areas for REW... The ability to script set of actions that we use repeatedly... And the ability to script our own maths to apply to the various data sets...

If the actions involve interacting with items that are always in the same place on the screen, you could just use a macro recorder.

 

[Matthew J Poes]

Now that I have thought about this a little bit maybe Mamba is approaching car eq correction in a manor similar to headphone eq dual channel cross-feed algorithms. As in a separate eq for each channel and/or driver...

So as I alluded to before, these things sometimes have solid research behind them already. Crossfeed is about sending sound from one channel to the opposite side and eqed to mimic in room sound. You wouldn’t and shouldn’t do this in an open environment. Headphones have no acoustic environment outside the ear. No delayed reflections. Since that’s unnatural we want to add that back in. In a room it isn’t desirable to attempt to add that on top of an existing environment. It’s one reason those fake Hall sounds are so unpopular and unrealistic.