DIYMobileAudio.com Car Stereo Forum banner

rePhase,a loudspeaker phase linearization,EQ and FIR tool v.2 for car

28K views 113 replies 21 participants last post by  oabeieo 
#1 · (Edited)
Hello everyone ,

In this thread we will discuss rePhase and it’s uses on many different platforms to be used in a car. Let’s not talk about running jriver or cpu convolution unless it’s going to be used in a car for the most part this will be discussing FFT convolution and relatively lower tap filter making techniques.
General fir filtering , we will go over the concepts, and the uses for corrections to be made in a car with fir capabilities. We will go over how rePhase and REW are interchangeable and how to make good measurements to be applied to your speakers using REW and how to make a measurement that would normally not mean much matter.

I have invited Thomas Drugeon (a.k.a Pos) from DIY to join us., Pos wrote rePhase and I hope he chimes in from time to time to help keep my info accurate :p I urge anyone who downloads rePhase and uses it to please go to surge force and contribute to Thomas for such an excellent software, same with John from REW this is a professional gesture we all should do to show our appreciation;)

I would rather not let this thread be crashed by noobs , if anyone doesn’t know what fir filtering does or is , please educate yourself first on what the general terminologies are. In no way am I trying to drive anyone away however fir filters are complicated and it takes a lot of time to explain the basics and a simple search will give you all the answers you could possibly want. Once familiar with the basic terminology than please join in discussion. :)


So, I’ll get some of you up to speed on what I personally have done. In my car I use 4 dsps, 1 Dirac Live ddrc22di and 3 minidsp 2x4HDs. My system consists on a clairion headunit NX-706 running 96khz toslink with volume control embedded so volume is done completely upstream. That feeds into the Dirac Live 2ch Dirac room correction box that I use basically as my global EQ and the first stage of my correction. Toslink out of that into an active 3way toslink splitter that divides the toslink into three that goes into each of the 3 minidsp 2x4HD minisharcs and than analog single ended RCA out to 4 amplifiers in a 4way system.

The 2x4HDs is where I use the minisharcs and make 8 seperte fir for each driver (channel) to have , eq, crossovers, phase corrections , and magnitude corrections, spots where Dirac needs small adjustments all in each fir. That is where I will be talking about rePhase and how it is setup in my car.. however as far as rePhase , my system is unique and I could get the same results with just rePhase alone, I chose to use the Dirac box so when I change equipment or drivers out setup time is minimized. (I’m a freak and change speakers a lot) in a normal persons system no Dirac box would be needed. RePhase can get everything done right the first time.

So now I would like to talk about what it is exactly in a car that makes fir filtering really make a lot of sense. Mostly a linearized mid-bass HPF and the subwoofer LPF is a must have. But beyond that basic everyone should have in a car the principal of how “phase” works in a car, how to read the measurements and how to apply a measurement using rePhase. And lastly after all the corrections are done for each side (left and right respectively) how to apply another correction on top of all the separate corrections to offset not just the timing differences between Left and Right but to correct for the comb-filters caused by interaction of the Left and Right channels caused by the time of flight delay. :kid:

I will share this article from Dirac Research first to get everyone up to speed on exactly what that correction looks like and what has to be done to achieve it realistically in a car with the use of fir and rePhase! :devil:


:drummer: without further ado the article from Dirac Research: This is a must read for anyone trying to use fir in car. As I can’t go into everything in one post this thread will be evolving over the next weeks and months.
I’ve dug myself into every rabbit hole possible but once I was able to start to apply these type of filters is when I fell into the dam glory hole, and 2ch audio in car becomes an entirely different animal and takes on things never thought possible. Once one has begun to get this type of filtering, it becomes silly to think about all the ppl bickering about trying to get the “time alignment” controls set perfectly, when actually “time alignment “ controls alone seem to collapse the stage and compress the recording so it’s right biased. As inducing a delay on the signal causes a serious comb filtering issue between left and right , than the actual time of flight distance compounds the problem even worse.

Here’s the article for everyone to read. I’ll stop here for the night and continue tomorrow or the next day.

https://static1.squarespace.com/sta...+Obtain+a+Good+Stereo+Sound+Stage+in+Cars.pdf


:)
 
See less See more
#2 ·
Here is a couple links we should all read to save me time in explaining some basics that REW and rePhase do.

https://www.computeraudiophile.com/...nerate-amplitude-and-time-domain-corrections/



And the link to the multi parent rePhase thread over at DIY for anyone who’s interested
rePhase, a loudspeaker phase linearization, EQ and FIR filtering tool - diyAudio


I will start gathering screenshots tonight to get into building the basics on how to make a lovely 2 seat listening position car or a down right deadly 1 seat listening position car, with standard door / dash speaker placement without minimized PLDs (path length differences) and start to jump right into it ;)
 
#4 · (Edited)
I will be speaking now in regards to minidsp 2x4HDs , or minidsp mini-sharks tied to a conventional DSP like a helix or with any range of added DACs or with any minidsp platform dsp like the 6x12 with added mini-sharks.


The analog devices shark processors are used in minidsp products and in other products like the najda boards. The ADSPxxxx is basically a 32bit floating DSP most are 8ch output and use a AKM codec as the DA converter which is quite ample for a run-of-the-mill chipsets.

Hopefully everyone has download a copy of rePhase and started to play with its controls.

You open rePhase to a blank setting. You can import your REW auto eq right into rePhase and maximize the low filter tap at 2042taps per 2 outputs (4096 on 4ch) at 96k (24/96) input or get 4084taps per 2ch at 44/48k inputs. So if your using a mini-shark ( a raw ADSP chip on a board connected through the I2s link to a compatible DA or input bus from minidsp) instead of the 2x4HD you could probably get away doing the entire correction in the fir by itself without the need to deploy any iir based biquad banks for PEQ or APF functionality. A purely fir based correction would be sure the have no recursion. Meaning, the silence of filter ringing caused by iir infinite amount of ringing in the attenuation cycles as samples are fed back into the input endlessly.
This “noise” is almost impossible to hear unless you’ve been making filters an obsession (like me ;) ) otherwise using iir peq banks for minimum phase adjustments or even as all pass functions go as well. Using as much iir controls when running 96khz on a mini-shark (2x4HD) will be a must to be able to get the taps needed to make usable change to the system in the fir banks.

In that instance the user can start in rePhase and do the Minimum phase portion first and export raw coefficients into the iir peq slots so there’s no need to jump platforms, the entire correction can be worked from rePhase alone. After your done working the minimum phase parts of the correction you can simply move forward and generate some fir!

If running a multiway and using multiple fir, make sure to give file names that include the tap setting, bit rate, general description on target import bank....that way when your sending 8 files you can manage what is actually loaded in the bank and can help when needing to use the delay setting in dsp to offset the fir fft length. For example a file name could look like ;
Left6G40ND-8ms-LR2LPF150hz-LR4HP1600hz-linphxo-.bin

That would tell me I’m working with a 6.5” Beyma 61/2 midrange with an fir that has 8ms of delay , a band pass linear-phase crossover and there crossover points and alignment in the file name. So when it’s a live fir Loaded in dsp I can have enough info about the functionality of the fir when bouncing around channels and not forgetting. Think of the long file name as leaving yourself a breadcrumb trail as you go WAY down deep in the rabbit hole. That file name will be your only way back to reality after going all night at it :p

Think of it as a guiding fairy


Oh come on andy! Where did that come from :D ”more fairy fun than ever!”!Lmao!! :D:D:D
Making sure to manage and save a file or configuration to your platforms is vital to working an fir. Spitting out a 10min fir will get you in the ballpark , but to hit home plate going back and making small adjustments to your dsp and fir is in the order of things so getting good habits will only make it much easier to see into the depths of it when working a correction.


Filter output delay can be managed and integrated as part of each fir if there’s no outboard delay controls available, you just have to move the impulse x amount of ms forward in the impulse centering controls on rePhase. I find it easier to use the delay of the fir ? And whatever speaker (most likely the sub) has the longest fft in “ms” will have no delay added , examples;

If sub fir is 10.63ms because you used a 2042tap fir and use a rectangular window and center the impulse on the peak energy if the ir and use a 1024t fir on midbass with rectangular windows And center on peak centering with a delay of 5.35ms you would subtract 10.63-5.35 and plug the answer (5.28ms) into the
Midbass delay setting to get both speakers to play at the same time. If you add the 5.28ms before the ir peak they will both be 10.63ms delayed fir. I like to use delay , it’s faster for me and easier to manage. So bring a calculator to the car with you. You’ll need it for this and much more to come
That’s enough for tonight, I’ll gather screenshots and go more tomorrow

Chao
 
#9 · (Edited)
So we’ve talked about different convolution engines
And talked about file naming and general organizational aspects of of filter management .
We’ve read the basic principles on the time domain and the near side bias problem in cars.

So it comes to a fork in the road where one has to decide which approach to take. Solving the near side bias problem for one seat or two. It will take a completely different approach on each method. This is where we need to start running some numbers to see what’s achievable.

We need to measure distances of the car dimensions, and the speaker positions relative to the car dimensions and where you or two will be sitting.
So let’s make an example car. Let’s say the car is

5’ wide (door panel to door panel)
4’ tall (Floor to ceiling )
7’ long (corner of windshield to back glass corner)

(You can take some 1/2space measurements and 1/4space measurements as well to help make sence of some odd reflections that may appear in measurements)
For example we’ll just stick the the basic room mode problems.

So if we had a door mounted midbass/midrange combo driver we could say there 5’ apart.
Sound will cancel when they are 180° Out of phase. Let’s run the math and get those basic numbers and determine what frequency we will start to see a combfilter start to emerge.

So 5’ wife car and we sit let’s say 36 inches from the left speaker and 60 inches from the right speaker. Wait that’s more than 5’.......exactly, if the speakers are mounted in front of you it will add a few inches to the total measurement, so that will change the “math” a bit and will change the way the combfilter appears on measurement, but we can still see where the direct on axis energy will collide and get in the ballpark.


Keep in mind REW has a room sim feature that does all this for you, it’s excellent in fact but someone needs to ask John at REW to make the room size variations smaller because I can’t get the smalles setting to match the size of my car. If you have a big truck you might get lucky.

So let’s take the distance the speakers are offset from center to center now , about in front of the lower part of the driver seat directly in front of the speakers , let’s say we measure 36’ driver and 48’ passenger , okay now the numbers add up to our total 5’. Let’s run that and see what we get. Any frequencies that wavelength in ms or feet whatever you use when the ratio is 1:1/2 it will cancel, if you sat exactly double the distance away from side to side it would be easy to calculate the frequencies. Find the frequency at the 1st comb and work your way up to the 3rd comb that’s where we will focus an fir correction, at higher frequencies we will use IID principles and HRTF in general to get it right. Because both ears matter more for each side up high and down low even SPL through the magnitude is dominant between left and right respectively relative to where your seated

So let’s get the basics on what we’re doing
Speed of sound = 1125.33ft~sec

Knowing that we can use any online calculator and determine just about anything or use your calculator manually.
Let’s determine how many ms the car is wide first
At 1125.33ft sec that means there 0.89ms in one foot (approx 13.5” is 1ms) some ppl say in car a ms is equal to a foot, which is almost right, and definitely works good but once you get farther than 1foot it drifts off so keep the math tight and do it the right way.

At 5’ wide that 4.45ms wide car. Let’s determine what frequency is 4.45ms long now. That roughly 252hz (4.465’ 4.99ms) it’s close enough to 250 to sadly say you could calm it the 250hz band. So what happens with a car 5’ wide at 250hz? Each speaker will be able to play one full cycle without any obstruction, this will be the beginning on the “Schroeder” or room mode frequency on the X axis. Now what happens in between that 250hz time path is where things get dicey. It’s good to know the math in “ms” how far you are from each speaker, and how many ms the car is in dimensions so we can calculate where to start and stop out room correction, once the frequency is longer than the room than we will start to be more and more certain the speakers will behave more in a minimum phase way. Also the pressure in the car will create nodes of high pressure and low pressure at your position in a different manner so. So any changes we make in those lower frequencies we make to amplitude will also affect phase directly and proportionally. At frequencies above that we will have to start to look at the combfilter issue caused by the two speakers interaction mostly in the x axis which we determined to be 4.45ms So 250hz and up.

Andy W. Has stated many times that 250hz is a fundamental frequency in most cars and almost all cars it’s safe to say 250hz is a good spot to start any all pass functions. And he’s right. I’ve discovered this as well and 250hz is a great place to start the first all pass to combat the 1st combfilter. We know that the car is 4.45ms so the first cancellations that are at frequencies shorter than 4.45ms we could call it the 1st combfilter. This combfilter will appear in different spots around the car and should be the inverse of itself on the opposite side of the car. This first combfilter is where we need to really focus a lot of what we do in the fir on and all eq work needs to be worked around this 1st big visible notch in the magnitude.

This leads us to get to know our cars. We can’t just cut a measured frequency in half and say arbitrary that’s where it will cancel , we need to take measurements of each Left and Right separately than measure Left and Right together and watch the notches appear. Because of funky 1/4space loading and the transfer function of each Left and Right respectively we might see the 1st notch show up +/- 1/2 octave Or so.

So one has to look at the data collection of measurements of frequency response and look at the interaction between left and right using balance controls, and the tape measure measurements and the physical spaces and put together a picture of the transfer function of the car itself as a whole. Once you can see what frequencies cancel and what frequencies are reinforced because of the room and the numbers are pretty close to the hard math you can start to generate a correction filter that will work. The less PLD there is the more achievable a two seat car is the the HF. As frequency rises the need for less PLD becomes a serious issue. But on the same token the higher in frequency we go the more HRTF takes over and some IID properties of hearing take over where amplitude can make or break a two seat car before PLD will. So determine how many ms. That is past 1khz. 1khz is mostly a frequency where depending on PLD but most cars you can start to begin to slowly just get an even mag response between channels and get a clear image for both sides. The worse PLD is it might get into the 1.2-1.4khz before we can start to apply that. Anything under about 2ms is good to use 1khz as a great reference to start making amplitude the driving force for localizations. So basically between 250hz and 1khz is where I call the “diffuse” region and good minimum phase behavior is the main driving force To get a good transfer function.

Obviously speaker placement should be considered first, if the speakers can be placed in a way the will radiate there energy or sound power evenly through its passband entirely evenly on both sides for both passengers or at least the driver that will be the single most important aspect to getting a good fir to be easy to manage and not fight you every step of the way through every peak and null.


This may seem like a big ramble but it’s important on how we approach the fir, we will see in further posts how this all is key to getting a correction filter to do what you ask of it

Still getting screenshots prepped. More tomorrow after turkey


Happy thanksgiving to everyone!
 
#10 ·
Sorry guys , Black Friday at the shop is tearing the arse outa me.

Haven’t had time, got to get through this weekend.

One thing I also wanted to talk about is the 70-80hz botch we all have. This notch has a cause but also an illusion, but truly no different than the notch you have between 250-600 with the big fat peak behind it.

Being we mount subs in the back of our cars mostly I’ll be talking about a rear mounted sub , although omnidirectional yes, but placement will lend to the transfer function on the loudspeaker. The sub frequencies are much longer than any dimensions of the car, so as the sub starts to compress the air in the back of the car the air will build momentum as the air molecules get tighter together, they start to look for a path out and transfer towards the other possible escape paths with low air resistance, the size of the car that is longest will is the last escape path for the high pressure air to move to, once the long part of the car starts to compress it will be proportionate to the amount of air the speaker pistoniclly can produce at one fourth cycle. (We can hear in 1/4waves under pressure) as the pressure itself will cause our eardrums to make the rest of the motions to complete the cycle (but at 1/4 the volume as well) that’s why we need large subwoofs. Need to move lots of air to hear the bass proportionately. So when it’s at half cycle at that frequency the air that had had already started to build pressure is rebounding back towards the opposite node (pressure zone in cabin depending on transfer function cant say where that will be) but can say as it rebounds it will interfere with the 1/2 cycle and cancel itself out. This is why there’s a huge dip in the impulse response in a car. The car body pushes back the air and acts like a diaphragm and causes a big ripple in the IR.

This also means where it’s roughly 1/8th we will get some reinforcements. Say hello to cabin gain. So cabin gain makes the 70-80hz dip “seem” a lot worse than it actually is. Even though the dip looks severe and everything you to to try to EQ it doesn’t work we will go over some ways to make the transfer function more even with fir so you can get it to sound smooth without the need for gobs of time smearing PEQ notches or boosts to make things worse.

Once the cabin gain has been either damped or EQed down , you’ll be surprised how much more power you can give the sub without blowing it up. The big dip can be salvaged by properly addressing cabin gain , gain structure, and using hass techniques to “move” the bass forward using the midbass while breaking up the cancellation node a little bit using the midbass fir.

The notched can fairly easily be made to go from -15db to -3db without adding any other drivers or using bigger drivers. The one thing I’ve noticed about when using hass technique, is that the leading speaker needs to sound very good and have no breakup modes or fs issues even if it is down -10db compared to the speaker on the trailing side of timing it will make all the nastys from a speaker come out with laser precision. Also don’t go for absolute max hass effect , yes you can get the bass to move up front but push timing back .2-1ms , keep the nastys hidden by the cancellation. Amplifying fs breakup by 10db isn’t pleasant
 
#11 ·
WELL !

I got a bunch of screenshots and started writing a post in PhP and discovering that was not the direction I want to go with this thread ,

I was basically demo-ing REW rePhase filters in auto EQ and decided I’m going to skip magnitude correction from 20-20k and go through it in a multiway.
There’s that link I posted that goes through it anyway, so I sorta wasted tonight but still got a couple screenshots I’ll share ,

Sorry for the setback, i want the screenshots to not be confusing and to not go way off on. Tangent and have to spend weeks clarifying details because I went in wrong order.

Back to the drawing board to capture screenshots. This time I know a way to explain technique and reasoning that will be much more compatible with any magnitude correction approach or any mag correction using various DSPs. So I’ll build on this thread and just do a standard magnitude correction using EQ from 80hz to 20k flat using whatever method as long as it’s averaged multiple times and separate left and right EQ work starts at 250hz and up. Under 250hz just averaged Left and Right simultaneously and eq -ed together.

Weather auto eq or manual a flat mag is a good baseline to start.

This is what I would do as a suggestion but obviously you know how your car behaves and sounds with your gear and install , so I want to leave out my personal preferences on things that have different routines.


So , multiway eq work goes as follows

1 measure each speaker with the crossover turned either off or at least 1 oactave past its desired crossover point. close mic with RTA and moving mic method in a slow averaging window and circle the mic around the edge of speaker , to capture its off axis output and all around the speaker. In the area where the crossover will slope knock down any peaks with PEQ. Whatever your method, this will ensure the linear phase crossovers that you will apply in rePhase will do for one , but also keep the phase of the system from making a change that would be difficult to track down and isolate later. Don’t worry about making the crossover slope area super flat, just no peaks, the rest leave alone unless it’s drastic and more than 6db, maybe consider a steeper filter or different crossover if speaker can’t play to full power an oactave below its crossover point. This will also ensure that any radical eq changes we make later in the pass bands won’t move the crossover slope shape to a frequency that is unsafe of too low for the speaker.

If your wondering how that can happen , think about it, if you cut most of the pass band and than gain it up to compensate for the lost signal the crossover area that may not have eq would be untouched thus making the amplitude in the crossover higher in relation to the passband that has most of its energy cut. That would extend the crossover acoustically way past it’s crossover point.

We going to sorta try to avoid having to stack a bunch of IIR crossover on top of fir crossovers to acquire an acoustical slope. But sometimes just adding the extra iir crossover on top of a fir crossover might be needed to get the gain structure required to fill some nulls. If combing is severe where we will be using more than 6db of eq you may want to start with a little more gain on that driver compared to other drivers when eq work is done.
I’ve had as much as 10db difference in gain between drivers in crossovers pre eq just to get them to work right at crossover. So do what you do with your gear, get a flattened mag and leave the nulls alone. I always EQ everything down and never go past -12db and I get the mag down within 3-6db of the bottom of null. Figure if you eq the mag down to the bottom of each null , at the bottom the speaker will get spitty if driven to hard and a soggy - not snappy non transient sound will be the result , although more spectral and better sounding at low volumes , leave the nulls down just a wee bit, but definitely eq everything else down a bit as much as well.

I’ll put more up soon, I’m organizing!
 
#12 ·
So oabeieo you have 3 hds with an optical splitter and that works no signal degradation am i correct? I was thinking running 4 hds with an optical splitter in my bmw using toslink on the optical ring bus. Ill just use a mobridge converter. I have 14 speakers now if i dont add a seperate sub if i did that will bump me up to 16 channels so 4 hds. Maybe i can run dirac on each hd later i wonder if that's overkill lol
 
#13 ·
yes 3 HDs and a dirac 2ch dig in/out in front of them

its toslink! cant degradate light, Its just optical light. No degradation what so ever and no noise what so ever.

Only think with toslink is jitter, and that has nothing to do with a spliter.
or any other digital source. Its a problem with clock syncing. and power supplies for digital clocks etc etc ...


no it works marvelous. The active split-er allows syncing. The passive spliters work without sync but may be a sample off from one another.

The HDs are excellent. At first I though they were a bit janky howver after really getting into them , they are quite good. I just thought a bigger box with nicer case meant better engineering. not so. for the size and the fact they are 12v and car work with zero mods and have zero noise. I love them
 
#14 · (Edited)
Okay here is a screen shot of rePhase Start up screen.

You can select a output file from just about any format you could think of.
CPU convolution, stereo convolution, mono, 24bit fixed, floating, 32bit,,,etc etc ..


you can see many different tabs for the controls, I wont post screen shots now of all of them as the thread develops I'm sure we will see them all

you can do phase EQ, normal EQ, linearize a existing crossover, or make a linearized crossover in fir, add delay, get rid of that annoying box rise and GD caused by a port with the box tool. So many thinks to do.

you can invert a measurement (or anything on the screen) so you can manually draw out a correction, you can invert a correction, it can compensate for any existing passive crossover alignment with the compensate function (which I have found many more uses for than just that)

Once you hit the "generate" button it makes the fir to the directory listed and named what you name it. Once generated the result and prediction will be overlaid. As long as the result is close (meaning a lot of things by close we will talk more about later) it will sound good. You can simply change windowing, centering values, tap amounts toill you can get a result that is acceptable. At first I though it had to be "perfect" to what the graph shows or ringing will happen. That's not entirely true, and again we will talk more later about that. Generally with small tap high sample rate filters a mixed phase filter is ideal. A combination of either iir and fir filters or a combination of minimum phase filters and linear phase filters, (one in the same, just outside of small tap boxes the ability to everything in fir becomes a issue so iir will be used in combo to get rid of ringing, per-ringing, and get the acoustical shape to match the fir.

You can also have REW do the eq work for you. I will not go deep into that, as I stated earlier but will include a screen shot to see it..

Here's a few screen shots to get started. I need to do some trial measurements now so I can import some very basic loudspeakers responses so the phase isn't a mess so as I explain things, no one will get lost when It comes to the more complicated things like summing two driver responses and how to identify problems and what to work and not work on...





 
#16 ·
Sound a lot like myself :D A little bit freaky about DSP

In all honesty Dirac is a excellent tool hands down ,

But for the guy that can settle on a set of speakers and settle on not changing out his or her equipment every month and spent the time with rePhase one could really nail down a correction that performance is better than a Dirac tune with no mods, a tweaked Dirac tune is very good , but for those with the manual fir rePhase can do wonders .


Can’t wait to use Dirac car platforms, not available as of yet but from what I hear is in the works, I have very high hopes for it.
 
#20 · (Edited)
Okay before I move forward , I need to make sure EVERYONE understands this. There is a 100% chance you will get completely lost and the rabbit will eat you for lunch.

So , take 15min out of your day and read this entire page start to finish. Don’t be a ***** and get board 1/8th the way through or tell yourself you can do this without understanding this principle. . Read the whole thing.

https://www.roomeqwizard.com/help/help_en-GB/html/minimumphase.html#top
 
#21 ·
:worried: Okay so (that was a joke BTW he he)


More coming soon.

Once I get going and get all my info organized so it makes sense we will see how things need to be invertable

Meaning , you have to be able to reverse the polarity and have the opposite happen. If there’s a reflection that has diffraction attributes to it , we will see that diffraction can not be un-diffracted.

We will be discussing how to pull phase around and get a reflection to behave a bit better so that it’s high and low air pressure spots are not in a offensive spot.

Keep in mind moving phase can act like moving sound forward or delay but it all comes out at the same time. Nothing is delayed , it’s just structured a bit different that’s all ;)

More soon
 
#23 ·
YES with minidsp 2x4HDs, I go over a few minidsp products in the first few posts that are compatible. But with focus on the minisharc (2048t at 96k or 4096t at 48k)

APL1 , almost had the trial up and running , still want to try it. Got massively sidetracked and had a revelation with rePhase. The APL1 trail is on back burner.

It sounds stellar (I hit the glory hole by gosh) (I’m such a nerd)


My car is definitely that car that sounds nothing like it should relative to where the speakers are placed. It definitely sounds like a ton of DSP is applied , but not in a bad way at all, quite the contrary in fact. Can tell lots of dsp is in use, but dsp done well. Very dynamic , no harsh spots, goes very loud with stage staying exactly the same at all volumes.


Once I went to a concert (smallish venue about 2500ppl) and it had a very similar effect. Except in the car it’s reproduction and has all the artifacts of a recording engineer, solid center , etc.... but anyhoo my car reminds me of that venue. I remember the concrete building it was in seemed to have no reverberations and seemed like i couldn’t hear any echo.




———-


Sorry for slow to post, been very busy at work, the holiday season in car audio is slammin , we’re doing a lot of installs working long hours,

But after January I should start having more time, I stepped down from management to move to a store closer to home, (5hr a day drive too hard on the Fam) so I’ll finish the year in my store than transfer. Makes me pretty sad, I love that position and worked pretty hard to attain it, but need to open new chapter. So, I’ll have more time for everything else. (Should be a good thing for now)



I saw this pic and stole it from a site and wanted to share it , it shows the different zones as far as acoustic behavior is concerned in a room. Each “zone” needs to be addressed separately as they all react differently to the overall manipulation of the audio signal.

I have a lot of screen shots made and am making posts in HTML and PHP , as soon as I feel the info is as accurate as I can explain I’ll post. I sometimes get ideas and thoughts mixed so I want to be careful to not have misinformation or convey a point that is right but comes across incorrectly. I assure everyone it will be through enough to make good use of it


 
#24 · (Edited)
so, here are a few pics of some measurements at the listening position. I did a left and right , no-EQ just crossovers (mixed phase, some iir some fir and some half and half) you can see the phase wrap wheres there's iir crossovers or a time delay.

I added a 5cy IR window to everything. Than I rand Dirac Live , No pre-tune, with a flat response (cant remember if I had tilt, either way just for conversation. Lets look at what is going on before and after.

Ill post about it after so I can write from my phone instead of this bulky computer..Also a view of the rePhase files and minidsp configs
















 
#25 · (Edited)
So ,
I used Dirac as a fast example of what one of the big goals is just for speed


If you look at the left and right no correction (can see what measurement I have open)
And left and right with correction it’s pretty easy to see what’s going on.

Phase is not pancake flat. There are wraps still , and that’s OK! You can not hear a phase change on a single speaker, and that’s the point of this post to get you going on what to do.
It’s when you introduce the other speaker/channel (depending on what your doing) the goal is to have left and right phase and magnitude to be as close as symmetrical as possible (at least for now, later we will go apeshit on APFs and shape the stage and completely disregard the symmetry, but that’s way way way down the road. For now, just get them looking symmetrical)

(Sorry for not making my limits the same but you get the drift)

As long as the Left channel vs right channel are making the same shapes in magnitude and phase it will be “phase coherent “ (not a real word but can be used)

Keep in mind if your spl meter is calibrated on your measurement software and you take measurements from the same point in space on the driver side , the passenger side will look like it’s phase is more “positive “ throughout the magnitude. That’s what you want. You don’t want to try to “correct” the magnitude of the phase to match the phase on the left by correction. I moved my scroll bar and so you can’t really see it because I added windows because I took these measurements a while back. The reason the right side will be positive is it has more delay, (time of flight, unless you window it out which you will) but at first the right will look a few degrees positive in the linear view and simply more positive in log.


If you were to try to correct for that the center image would be directly in front of you instead of between the speakers, you want to let the magnitude in “db” move phase for you throughout the FR magnitude. By simply reducing the output on the left side will move the phase with it in the opposite direction. After all your signal delays are done and FR is made to your liking and is symmetrical on both sides, let the minimum phase behavior move the phase magnitude. We will only be focusing on the shape of the phase vs Left and Right and how the shape relates to the FR magnitude.

If we look the Dirac Software made Left and Right look dam near exactly the same on Left and right in FR and phase (BTW FR means frequency response, but you knew that)

This view is in the linear axis just for viewing. Even in the linear axis you can see some issues where phase isn’t exactly the inverse of FR. It could be just the view , the measurements, the window, but it’s always a good idea to start to pay attention to those anomalies with each view window your in. Start to get a feel for the room.

But even with this , one could set the magnitude so that FR has the most symmetry as possible between channels, asymmetrical crossovers can be very useful and asymmetrical level setting between channels in a multiway can be very valuable before running to the eq.
Your crossovers can make a lot of good changes for you and save a lot of backtracking later down the road.

After EQ Left and Right you should have a good baseline on what to do next. There will be back tracking, be ready, trust your ears also, Ive skipped over hundreds of validation measurements by just using my ears, once you get a hang of what a phase change will sound like when played against its other stereo half you can get a knack for what’s off and needs fixed to get left and right to be more symmetrical.

Once you get some real symmetry happening with crossovers and levels and eq , than its time for a few more sets of measurements. The phase changes you make you will want to compare what individual drivers are doing as far as phase behavior vs when it’s paired with its fellow drivers on like channel and that will let you get an idea of what’s happening when they sum.
When you take your 2nd set of measurements, and have a decent flat (or reference target magnitude) between left and right its time to do a different viewpoint. You’ll want to get rid of the time of flight delay and generate the minimum phase response. Getting rid of time of flight delay I like to just remove samples from the IR, the acoustic timing method will use the HF driver as a dominant reference point. I like to loop back and move the IR manually, I like to see what happens as I move the delay out of the IR window. If you center before the peak, in middle of peak, or where ever , find out how long your foriour is, make sure Left and Right have the same windows applied and same settings in your analysis. That way if you remove 2 samples or 200 samples.

Once you generate a minimum phase you can now see inverting that’s happening as amplitudes change and how the phase should look a lot like the mag inverse.
In any spots that do not look the same between left and right and that do not follow the general inverse shape of the FR those are areas we will want to look deeper into.

For a two seat tune simply (ha simply it’s no easy thing) make measurements of the passenger side as well, the the driver and passenger to have the similar phase response that is a mirror of itself. Does not have to be perfect (it can’t!) just not swing to -180deg to +180deg between combfilters. Get it so they both +/-90deg on both sides to mirror each other. Some spots will be achieved some won’t be so much but just make it best you can. This will get into more later, let’s start with a one seat tune before we whip out our shlongs.


Any phase change will affect FR as well, and any phase change in one area will move phase all the way through the spectrum in some areas. After a correction is made its pretty important to listen and RTA with some noise and re adjust eq or levels. Once you change phase the way the speaker behaves in its space will change the FR as well. Could make some areas louder with reinforcement or quieteter with interference. A phase change will change the interferences all over the place and a new measurement must be made after levels are re balanced. Once you get few the first few you’ll notice some cancellations actually sound better and should be kept as it may add to the entire room character better it’s sound envelopment or spacial impression.

Next posts will be covering those types of views and how to import the measurements into rePhase and start making corrections!
 
#28 ·
Yes in fact that’s a sweet way to do it (if your ok with car pc lag) but yes with
Jriver as a convolution engine and rePhase you get get a very high tap count.

Keep in mind CPU convolution and FFt convolution are a bit different, but a lot of ppl use a pc as a DSP and run half million tap filters with good success. I tend to like fft better with shorter impulses (the smearing stays manageable)
 
#32 ·
I was hoping through some awesome Dirac magic that it could help alleviate the mono center channel that most DSP users are stuck with due to companies not investing the pricey licenses for center channel processing.

Something along the lines of:

Since the center is R+L...

Only the R signal would come out of the Center in conjunction with the right speakers when performing the right side tune. And only the L signal would come out of the Center in conjunction with the left speakers when performing the left side tune..

So when played together through some magic that oabeieo could explain we would in a way derive the Dirac Virtual Center

Sighs.. patiently waiting for that Audiofrog unit.
 
#34 ·
Audiofrog is working on a unit that can upmix stereo to surround sound including proper Center channel by signal steering via Penteo. No details on whether it will have fir filter capabilities.

There are a bunch of differences. First, in the matrix upmixers, the center is a summed L and R and the steering angle computer turns the levels of the channels up and down according to the calculated vector to maintain stage width. Second, the rears are a L-R signal.

In our Upmixer, the sound field in the stereo recording is separated into mono (same in L and R), Intermediate information (similar but not exactly the same in left and right--the sounds between the center and the left or the right) and differential information (only right and only left). The mono info is sent to the center. The intermediate and differential info is spread over the left and right, the sides and the rears--kind of like a horseshoe if you were to turn all of the dials to 11.

This provides great stage width, a stable center and no artifacts in the sides and rears.
Only Andy has the knowledge of when it will be out for retail. So until then I guess let’s not stray too far since this thread is for RePhase and Dirac Live.

As far as the DDRC-88A, it does not have an up mixer to my knowledge. It can perform Dirac Live calibration to 8 seperate channels. I’m curious to know how many taps per channel? Same as DDRC22 divided amongst 8 channels or is itvmore like 4 DDRC22’s in one?
 
#35 · (Edited)
Elgrosso is right , Dirac isn’t release anything for car as of yet

As far as a center ch. I would simply buy a logic 7 up mixer and insert that in the chain if you want a true center that has the proper decoding for what you want to do.

If you have a ddrc 22d before the helix i would run the correction without the center ch turned on (rears ok) so it doesn’t mess with the 1st measurement that sets delays, but also would leave the eq work improper. I would just use rew after your Dirac run and eq the center in the helix to match the response of the left and right summed response . Meaning play both left and right together, use PN and do a moving mic average and just manually eq the center to match that response down to 250hz.or so if tiny center at least to 400.

For your center you will want to insert a toslink splitter and get a signal split before the ddrc so you can add a 2x4HD or DDRC24 to not be a part of your Dirac run , you won’t want to try To upmix a Dirac correction as left and right will have different electrical responses respectively to the original source and up mixing is dependent on phase of left and right so that can’t be altered to get a proper upmix.

With that dsp you could get the center to shape the way you want, obviously your upmixer would have to go between the source and any Dirac


The ddrc88A or BM is ideal for a updstream upmixer as it’s seperate corrections tied to a single target. That how I would do it

Ditch the helix , use a ddrc88xx and a logic7 upmixer before it. Pretty easy and would be one hell of a sweet setup, and the BM add on you can get Dirac to use all the channels to make a dedicated sub output and also mix portions of each correction to any channel,
You could mix 25%of left correction and 25% of right correction on top of the center correction if there’s massive differences in the drivers FR pre correction and mix parts of each correction and send it to other drivers to blend the sound if the locations are radically different as far as sound characteristics go.

The same exact acoustic response can sound completely different if one loudspeaker is playing let’s say into glass and another is playing into carpet or a soft padded seat.....
So the 8&BM allows you to mix each Dirac correction onto other drivers to balance that in 1% increments. Pretty sweet. The combo of mixed corrections and seperate Dirac targets on each ch but tied to a single target would give you the ability to make it do whatever you want as far as how it sounds.

But again as far as upmixing , again just needs be done upstream than run corrections on each respective channel.
 
#36 ·
Thanks!!

I couldn’t discern what effect would occur when performing the Dirac Live calibration while a mono sum center speaker was turned on.

It’s interesting you say to ditch the Helix. I’ve actually been toying with the idea of fitting a surround upmixer unit in my glove compartment. I snagged an Outlaw Audio 975 from eBay that I was thinking of powering off a mini pure sine wave inverter upstream of the DDRC-88A with the BM plugin.

If you can fit and power the Outlaw unit it opens the door to:

-Analog and digital inputs including Optical,coax, and hdmi (perfect for iPad or tablet source).

-192 kHz 24-bit DAC's for each of its 8 output channels that can be processed to a surround sound matrix galore such as:

Dolby TrueHD, Dolby Digital Plus, and Dolby Digital decoding

DTS-HD Master Audio, DTS-HD High-Resolution Audio and DTS decoding

Dolby Pro Logic IIz, Dolby Pro Logic IIx, Dolby Pro Logic II, and DTS NEO:6 processing/upmixing from Stereo in either Movie or Music modes

It can even set crossovers for each channel. Very tempting. As is your ms8 Logic 7 route. I’ve only heard BMW’s modified implementation of logic 7 for their vehicles. Not the true Lexicon or Harmon Logic 7 upmixing that’s often touted as the best.

Do you know how many taps are available to each channel on the DDRC-88A?

In regards to the DDRC-22D + Helix Route, could I perform the calibration while Helix sends R-L and L-R signals to the rears? Or is it best to perform the calibration with a stereo rear and then modify the matrix post calibration?
 
This is an older thread, you may not receive a response, and could be reviving an old thread. Please consider creating a new thread.
Top