Target curves and headroom

[Mikkel]

Hi Bernt and others who read this,

 

I have a question regarding target curves. When drawing a B&K house curve and generating the filters I loose approx 20-25 db of headroom. My speaker response for front left, front right and surrounds shows approx 15 db roll-off from 9.000 hz to 20.000 hz. Part of the reason is that these speakers (compared to the center speaker) are off-axis, another reason when it comes to the front left and right speakers are a loss of energy when the last octave travels the 4,5 meters from the speakers to my seat. Measuring the speakers at a distance of 1 meter shows a somewhat flat response even in this last octave.

 

If I let the target curve follow the natural roll-off of the speaker only a loss of 10 db is required.

 

Given this description is it normal to experience this huge a cut in headroom? (or perhaps the tweeters are just poorly powered/build)

 

 

Best regards,

Mikkel

[OlavK]

+1 What headroom is actually lsot when the filtered response is lowered >15dB?

Regarding target curve I see the same behaviour as you, and I am curious how the best target curve should look like. Currently I use something very close to B&K, and I am trying to convince myself that is sounds great even if I find it a bit dull. It seems to suppress some crispiness in the high frequencies. It may be habit and it may be sub-optimal. I do not know.

[Mitch]

From the Audiolense help file:

----------------

"We often get questions about why the filters attenuate the output so much. The short answer is that in order to avoid digital clipping, the average sound pressure level has to be attenuated by, typically 6-10 dB to give enough headroom for the single frequency that needs the most amplification. The frequency response of the correction filters – as they appear in the main form will give a 100% accurate picture of how the filters will attenuate the output. The sound pressure level of the evaluation charts doesn’t give the right picture in this regard.

If you see excessive attenuation in the correction filter, you should look for unnecessary boosting of single frequencies. Typically this can occur in the top treble when the treble boost limiting is disabled, and the measured speaker response drops before the target response. Such cases can easily be negotiated by activating the treble boost limiter or by making a target that doesn’t give as much boost"

-----------------

Depending on which Convolver you are using, you can make up the gain.  For example, if using convolverVST, it has a gain slider that you make the gain back up by adding 10 dB for example. Protip - press the shift key when moving the slider.  Or if you are using JRiver's Convolution engine, you can enable "normalize filter volume".  If that is not enough gain, then a nice free gain VST plugin is: http://www.bluecataudio.com/Products/Product_GainSuite/ 

With respect to the B&K curve, if it sounds dull to you, then change the target.  Do you have "prevent treble boost" checked on or off?  If on, try the same procedure with the treble boost off.  Some folks prefer flat out till the natural roll-off of the tweeter (i.e. no -3 dB at 2KHz like in the B&K curve).  Other folks use the "Sean Olive" target at the top of slide 24 here: https://docs.google.com/file/d/0B97zTRsdcJTfY2U4ODhiZmUtNDEyNC00ZDcyLWEzZTAtMGJiODQ1ZTUxMGQ4/edit?pli=1&hl=en    This curve is more of a slope starting at 0 dB at 20 Hz and then sloping to -10 dB at 20 KHz.  I am listening to and liking this "slope"  right now. Finally, depending on how dead or live your room is will also have an impact on the target used.  For example, my really live room liked the B&K curve.  But when I added acoustical treatments, I found it sounded a bit dull and started to experiment.

The best peice of advice I got from Bernt is to experiment.

Hope that helps.

Cheers,

Mitch

[Mikkel Gylling Hangaard]

Hi Mitch,

 

Is it preferable to let the target curve follow the natural roll-off of the tweeter? In my case I have a 13db roll-off fom approx 10khz. It follows that the filter will correct for this, but is this a mistake to do so? A lot of transients are lost if I don't but if there is a good reason not to I'd listen to it, indeed.

 

 

Best regards,

Mikkel

2012/8/31 Mitch <mitch...@gmail.com>

--
Audiolense User Forum.
http://groups.google.com/group/audiolense?hl=en?hl=en
To post to this group, send email to audio...@googlegroups.com
To unsubscribe, send email to audiolense+...@googlegroups.com

--

Venlig hilsen,

 

Mikkel Gylling Hangaard

______________________________________

Ph.d.-studerende

Cand.mag. medievidenskab
BA medievidenskab og politisk teori & filosofi

Tlf.:      3121 0418

E-mail: mikkel.gyll...@gmail.com

[Bill Street]

Hi Mikkel,

 

"Is it preferable to let the target curve follow the natural roll-off of the tweeter?" I'm not as experienced with Audiolense as Mitch (it was his posts that got me interested in Audiolense in the first place), but I would say it is preferable. I would also say it's just as important to follow the natural roll off at the low end as well. My thinking is the natural roll offs at extremes are there for a reason. Maybe it's a function of your room, your speakers, not sure. If we try and extend the low and high frequency roll offs with our targets, it's like we're trying to make our equipment/room do something it's not able or willing to do.

 

It seems the main goal of software like Audiolense should be to smooth the peaks and dips in the response. Trying to use it to extend the natural high or low frequency response of our rooms and equipment does not seem like a good idea to me. When I'm experimenting with a new filter, I try and get the roll offs of the target at extremes to match the measured response roll offs as closely as possible. Using the Impulse Response graph to check the simulated response after each tweak of the target really helps me to zero in on the "best " target. You can compare the amplitude of the simulated impulse after each tweak and you may be suprised how much difference even tiny adlustments to the target can make. Also notice how "smooth" the simulated response looks after each tweak. After each tweak of the target, you can also note how different the target looks in the Impulse Response window )more or less jagged looking). In my experience, the goal of tweaking the target is to get the smoothest looking target as shown in the Impulse Response window, which will also mean a smooth simulated impulse response (by "smooth" I'm mainly referring to the lack of jaggies/spikes after the main impulse). I also pay close attention to the amplitude of the simulated impulse. I've found the higher the amplitude, the better sounding the filter, all other things being relatively equal.

 

I'd be very interested in hearing other thoughts on my comments. Hopefully Bernt will add some comments as well.

 

Good luck,

 

Bill

[Mitch]

Hi Mikkel, what Bill says has been my experience as well.  I could not have written it better, nice job Bill!

 

This is a passband filter design following the natural roll-off of the speakers at the frequency extremes.  The wider the bandwidth of the passband filter, the better tonal balance, but not so much that you are boosting the extremes (by much ;-)

 

I have some screen shots I will try and post later.  Btw to give you an example of tweaking, I am on my 28th target :-)  It really does pay to spend the time fiddling with the frequency extremes, especially the top end, and looking at the impulse response.  As Bill says, you can tune it in and the reward is pretty amazing sound quality.  But it takes a while to correlate what is being displayed in the graphs and what is being heard.

 

Cheers, Mitch

[Mitch]

A quick walkthrough.   While the fine tuning is very specific to my speaker/room combo, I think the general process would apply to most.

My speakers/listening position is set up as an equilateral triangle.  I use a distance laser measurer to line up the mic to the speakers.  I then run a measurement and have a quick look at the impulse response to see where I am at.  I move the mic slightly, re-measure and repeat until I get it spot on like so:

http://i1217.photobucket.com/albums/dd381/mitchatola/impluseequalateral.jpg

I am zoomed way in on the time scale (20 nanoseconds per horizontal division).   I put a piece of tape on the floor directly over the tip of the mic so I can usually get the mic into position pretty quickly when it is time to re-measure.  The mic is pointing down the center line and is both at the height of my ears and my speakers tweeters.

Once I have everything lined up, I take a keeper measurement.  I check the impulse response to see if it is still spot on.

There are two variables  to solve for.  One is the Correction Procedure Designer to create a custom filter procedure and the other is to create a target frequency response.  I am going to skip over the custom filter procedure as that is a subject of another topic.  I suggest reading Bernt’s 4.4 release notes.

As a side note, pay close attention in the Correction Procedure Designer to whether you engage the prevent treble and/or bass boost checked on.  This may roll off the frequency extremes too early for your particular speaker/room combo.  I think the HF rolloff kicks in at 15KHz, which may be too early.

In my case, I have unchecked both boxes.  That means I am responsible for tailoring the rolloff of the frequency extremes in the target.  As Bill has pointed out, we want to tailor the frequency extremes to the natural rolloff of the drivers.

From a target perspective, for my speaker/room combo, I find Sean Olive’s slope of 0 dB at 20 Hz to -10 dB at 20 KHz (see slide 24) to suit my reasonably live room.  Btw, I listen to mostly rock, so may be too dull for classical.  Or for others that have a really dampened room, this also may sound too dull.  If so, move the 20 KHz data point up (or down) until it sounds in the ball park.  Now comes the tailoring.

In my case, my target looks like this:

http://i1217.photobucket.com/albums/dd381/mitchatola/target-1.jpg

How did I arrive at the roll-offs?

Here is what my target looks like against the raw frequency response of my system

http://i1217.photobucket.com/albums/dd381/mitchatola/targetmatching.jpg

I have played with so many targets that I am xperimenting with extending the frequency extremes a bit to give my system a fuller sound.  I can get away with it to a certain degree as I have large efficient floor standing speakers with a 4 ohm 15” driver and mid and high frequency compression drivers/horns being driven by a wideband Class A amp.

If you look at the high frequency close-up, you can see that there appears to be a dip at 21 KHz and 23 KHz.  I have adjusted my target to smooth that out a bit, but follow the natural roll off after that.  Given that the wavelength of 21 KHz is around 0.6 of an inch, it could be a phase cancellation in the 1” throat of the HF compression driver or mating horn.   Then again the mic cal file only goes to 20 KHz...  Regardless, sounds good to me and took a while to dial in.

http://i1217.photobucket.com/albums/dd381/mitchatola/HFmatch.jpg

Here is what the low frequency roll-off looks like.  I am experimenting with extending the low end a bit.  So far, it does not seem to overload the amp or 15” driver.  Still fiddling.

http://i1217.photobucket.com/albums/dd381/mitchatola/LFmatch.jpg

Here is what it looks like all up from an all up frequency response perspective.  Note the filter is boosting the very extreme ranges of my speaker system, as an experiment.  It is almost like a very narrow band parametric eq., but does follow the natural roll-off right after the boost.  This fine tuning is very specific to my speaker/room combo:

http://i1217.photobucket.com/albums/dd381/mitchatola/Everything.jpg

The trick seems to be balancing the target with the impulse response.  For example, here is one of my target’s that rolloffs the high end frequency response early

http://i1217.photobucket.com/albums/dd381/mitchatola/target17.jpg

Here is the corresponding impulse response:

http://i1217.photobucket.com/albums/dd381/mitchatola/target17impulseresponse.jpg

And here is the impulse response from my latest target from above:

http://i1217.photobucket.com/albums/dd381/mitchatola/simulatedimpulse.jpg

If you compare the two impulse responses, note how much higher peak in my my last target’s impulse response, compared to the other one.  The first one measures 0.011 and my latest one measures 0.016.   Sounds better too.  The small amount of ringing before and after the peak is not audible to me.  I am not using preringing control in my Correction Procedure filter.

So as you keep tweaking the matching of the target high frequency roll-off to the natural roll off of your tweeters frequency response, keep checking the impulse response for maximum peak and as little ringing as possible.

I do AB listening compares by using JRiver’s native Convolver.  I can play a test song I am familiar with and switching filters in real time as the Convolver just inserts a short mute when the filters switch.  Once I decide which one I like better, I listen for a number of days and take inventory of what I am hearing and look for opportunities to fine tune the filters even more.

Hope that helps.   Would love to hear others experiences too, as I am sure others have tricks and tips too.

Cheers, Mitch

--------------------------

[Brad]

Why bother with the "dead nuts" mic positioning? Audiolense is going to set delays in the filters to make all the drivers "hit" that spot at the same time. I would think a multi-position measurement average would be best.

[JB]

Hi Mitch,

 

Thanks for sharing your walkthrouh.

 

Here is a pic of my the impulse response of my system:

 

http://imageshack.us/photo/my-images/209/impulse.jpg/

 

The initial spike reaches 0.034 while the consequent fluctuatins is kept below 0.001. (factor of 34!)

When zoomed out it almost looks like a perfect impulse response. However when I zoom in I see fluctuations going on for 25ms.

But I guess no matter how close to perfect a real system is, you can always find fluctuations if you just zoom in enough.

 

The question is, is the zoomed out state the correct state on which to base the verdict of my system?

Would you say that it is a good or a bad response?

 

Thanks!

 

Best regards,

Jarle

[Bill Street]

Hi All,

It's great to see screenshots from others filters. I hope this isn't derailing the original topic too much.

Mitch I'd like to make some comments on your walk through post. It looks like you're using 32k filter length? Any special reason? It appears your measurement sample rate is 96k. I have no understanding at all of how filter rate/sample frequency relate, but to me it would seem longer filter rate would go with higher sample rate? Any explanation on how this works would be appreciated.

Looking at your target, I don't think your little dip above 20k is a good idea. Wouldn't it be preferable to be smoothing that measured dip out instead of preserving it (assuming it's even audible)? I would also expect eliminating that from your target would improve the "look" of your simulated impulse response.

A couple ideas to try as well, maybe try adding a new anchor point to your target around 15k (can experiment with exact position). This will make the smoothed response follow the target rolloff more closely. You will also see a noticeable change in the simulated impulse response as you experiment with this. Another thing I've found makes a difference is moving the target vertically in the target designer. For example, your 20 Hz anchor is at 0 dB. Try shifting it down to -5 or even -10B. You'd shft the other points on the slope the same amount so you keep the same slope. The anchors on the rolloff may or may not need to be adjusted. The idea with these is to maintain the identical rolloff even though the target iself has been shifted relative to the measured response. I'll be posting some pics at the end of this post and one shows my target in designer window. You'll see mine's shifted down about 8 dB.. This shifting has a definite effect on the simulated impulse response result. It can change the actual horizontal position of the impulse, the amplitude. You'll notice if you try this.

I notice on your simulated impulse response, it looks like right and left are no longer perfectly aligned, like you original measurement. That could be a result of your target designer settings. Maybe experiment with those a little and see if it can bring them back to alignment. The above suggestion about shifting the target horizontally may also have some effect, although I think it would more shift both channels together instead of individually.

I also think your overall simulated impulse response can be improved on, besides the things I've already mentioned. I remember from another post of yours, seeing the amplitude up around 0.050, while now it's only around 0.016. I'm assuming your equipment setup hasn't changed that drasically, so we'd have have to think that difference is all a result of your filter design. Whether it's a difference in the maximum dB correction you're using now, the correction design parameters, the target itself, it's hard to say. But all of these will have an effect on that amplitude. The negative excursion of the impulse is almost as big as the positive. I'm sure that can be improved as well.

Mitch, I hope you don't take these comments as criticism. They're not meant that way at all. They're all only suggestions and comments based on my own experiences. I have no idea how any of them will work in another setup.

Another thing I've started paying attention to is the Group Delay simulated response. I have no idea what it represents but I have noticed the more smooth/flat the simulated group delay response, the "better" the resulting filter sounds. Everything adjustment, parameter setting seems to make a difference to this result. One of my screenshots shows my designer window settings, which may look a little strange (especially the before peak settings), but these were mainly made based on the resulting group delay response, trying to achieve the smoothest response possible.

Brad mentioned multiseat correction. My setup is 2 channel only but I've also started using mutliseat with this latest beta release. I take 3 measurements. A left, center and right measurement, with the left and right no more than about 10" or 12" from the center. I can't really say I notice any difference between multiseat and single measurements soundwise but I just thought I would mention it. I do notice the the multiseat frequency response can look quite a bit different from the single measurement. I've attached pics of both mine, all other settings identical.

I use none of the betatesting modes, mainly because as you can see in my impulse response, I have no visual pre-ringing. I did experiment with it, mainly the "A" method and found it could actually introduce audible pre-ringing, even though it wouldn't be noticeable visually in the simulated impulse response. As part of my test track listening to compare results, I have a couple that are just male voice, talking, for channel identification. He says things like "right channel, left channel" and a few other similar things. For whatever reason, that is my torture test for pre-ringing. I may never hear it during music, but I can hear even the tiniest amount of pre-ringing using those tracks. I used 2ms/100 Hz and had no audible pre-ringing. As soon as I went away from either of those settings, I would hear it. I initially thought using "A", even though I had no pre-ringing to start with, sounded better. I ended up disabling it since I generally think the less going on filter-wise, the better.

I'm also using the 64 bit float filter setting from the latest beta. No idea what it does, but it did seem to make some kind of overall improvement. Probably all in my mind, but who knows.

Here's some screenshots from my current filters:

This is my impulse response:  http://screencast.com/t/LjOVtxKegv

Corresponding frequency response:  http://screencast.com/t/4FOv0AKsy

Single Measurement frequency response ( notice the differences in the high end response in this one and the previous pic, which was multiseat correction, everything else the same):

http://screencast.com/t/ama0auL8A9D

My target and filtered measurement:  http://screencast.com/t/RdqE1sEhcMD

Target in target designer. Note the vertical shift:  http://screencast.com/t/Uq9YuhaM2Glo

Group delay:  http://screencast.com/t/8joTvA5CmwGR

Again, I'm very interested to hear any comments as well as seeing and hearing about others experiences. Even if you think what i'm doing is totally useless, at least you'll know what not to do.

Bill

 

[Mikkel Gylling Hangaard]

Wow, great to see all this activity on the subject. Due to time constraints only one comment: it was mentioned not to boost the upper frequencies but to follow the natural roll off. Wouldn't it be an idea to measure the speaker, on axis and at 1 meter and then at the listening position. Comparing the two would give a hint on the changes of the room.

 

My experience with this is that the room resonances causes a lot of gain on frequencies below approx 8khz while the highest octave response remained the same (more or less). If I then build a filter that should be flat from 20-20.000hz wouldn't audiolense just reduce the room gain to match the high frequency response? The result in Bill's and Mitch's cases would be a loss of 10 db headroom - which is somewhat insignificant isn't it?

 

 

Best regards,

Mikkel

2012/9/2 Bill Street <stree...@gmail.com>

 

--

Audiolense User Forum.
http://groups.google.com/group/audiolense?hl=en?hl=en
To post to this group, send email to audio...@googlegroups.com
To unsubscribe, send email to audiolense+...@googlegroups.com

[Allan Hansen]

I do not follow the natural rolloff at either end.

My speakers are open baffle, so they absolutely need a low end boost
to overcome the baffel drop.

I used natural high frequency rolloff in the beginning. But since I
switched to XO 2x3 stereo I prefer a fairly flat response.

[Brad]

You need to zoom in on the time axis so you can see what is goin on. Can't tell how good it is like this.

[Mitch Global]

Great to see folk sharing pics and experiences.  Would love to see more graphs!

Brad, understood.  But my best sounding filter is when I have the equilateral triangle set up dead on to begin with.  If the mic position is off, and I take a measure/correction, it sound close, but the tone quality is slightly different as is the timing/imaging in my AB listening tests.  Go figure.  Btw, got any graphs to share?

Hey Jarle, thanks for sharing. As Brad mentioned, it would be cool to see the zoomed in view of your simluated impulse response.  In case you have not seen it, at the end of the Help file is a section on “Hidden functions and details you might like to know” has all the ways to zoom in plus more.

HI Bill, thanks for the feedback, all good!  With respect to the sample rate/filter rate, I remember Bernt saying something about this.  You can see in my filter designer that I am using a 65K filter length:

http://i1217.photobucket.com/albums/dd381/mitchatola/TTDfilter.jpg

 

There are a few other items of note.  I am using a reasonably large amplitude correction (18 dB), but a fairly short time domain correction (40 ms at 10 Hz).  You mentioned that you saw one of my simulated impulse response peaks  at .045, I am pretty sure it is this one: 

http://i1217.photobucket.com/albums/dd381/mitchatola/LinearPhase.jpg

And here is the correction procedure used for that one:

http://i1217.photobucket.com/albums/dd381/mitchatola/CorrrProcDesigner.jpg  

Note the time domain subwindow is at 500 ms @ 10 Hz.  That’s considerably more time domain correction than my other filter procedure of 40 ms.  My understanding is that is why the amplitude difference between the two simluated impulse response peaks.

If that is the case, maybe Bernt can confirm,  then it makes sense to work on getting a good correction procedure designer filter first and then work on the target response for best impulse amplitude peak for a specific filter design.

In my room, I find a 40 ms @ 10 Hz time domain correction, using a linear phase target, gives me the best tone quality and timing/imaging for me.

Bill, I appreciate your advice.  I have already done most of what you are suggesting.  I have put anchors every KHz after 10 KHz and tried many variations of targets.  The one I have now, is the first time that the very top on my compression horn tweeters sound “right” to me.  Now the top is as transparent sounding as I can get it.  I am trying not to buy one of these: http://www.madisoundspeakerstore.com/bullet-tweeters/fostex-t500amkii-super-tweeter-new-version/   :-) 

However, I have not tried your shifting down on the target.  I will give that a try.  With respect to the two curves not quite matching, I am pretty zoomed in on the time scale… 

Looks like you are using a min phase target?  Also looks like you are also using 800 ms of correction at 10 Hz and down to 80 ms at 100 Hz.  How did you arrive at that?  Have you tried a much shorter time window at the low frequencies?  If so, what was the sound like?

Hey Allan, it would be cool to see some graphs if you want to share.  I am interested  (and I am sure others on the forum) in open baffles speakers. 

Cheers,

Mitch 

--

[Bernt Ronningsbakk]

This is a great thread that goes to the core of H2 produce the best possible correction filter.

 

It seems like some people prefer - or some systems works better with – a target that has a significant downslope, while others have good reasons for preferring a close to flat response through most of the passband. I think this is an area where each has to decide on his own. As a general rule, if it sounds dull, go flatter. If it sounds too lean and lacking of meat, go steeper. There are a couple of target tilt buttons in the target designer and they are there for a good reason.

 

+ 1 on Mitch’s “It really does pay to spend the time fiddling with the frequency extremes, especially the top end”. That is my experience as well.  I suspect that the optimal tweeter target is highly tweeter specific.

 

The formula for frequency resolution  is  {sample rate} / {correction taps}. The math goes like this: 44100 taps @ 44100 Hz will give 1 Hz / tap. 65536 taps @ 44100Hz will give approx. 0.7 Hz per tap. This resolution is the exact spacing between each correction point in the frequency plot. My very conservative rule of thumb is a factor of 10: 1 Hz resolution will give good correction down to 10Hz with 100% certainty.  10 Hz resolution will work well down to 100 Hz etc. I any case, the simulated response will tell how well it works, and shorter filters with less resolution will usually work as well or almost as well down to a factor of 2-3.

 

There will always be a risk of introducing digital clipping if digital gain is added in the convolver. Automatic gain adjustment should be turned off and manual gain set to zero for the best tradeoff between gain and the risk of digital clipping. Adding 3 dB gain or less is not likely to give audible clipping. Adding 6 dB gain is likely to give some audible clipping on compressed recordings (recordings that play loud to begin with).

 

The height of the peak in the simulation corresponds directly with high frequency level and extension.

 

Kind regards,

 

Bernt

[Bernt Ronningsbakk]

Hi Mitch,

 

When I compare TTD with frequency correction here, the impulse response peaks are very close, but slightly higher on the TTD correction. 0.047 vs 0.042. I did that with 15dB of max correction boost, treble & bass boost limiters turned off and with a target that was conservative enough to ensure that there were no boost limiting in operation anywhere.

 

Note that you have to compare p-p here. Not from zero to max, but from max to next point which is usually the min point.

 

Another thing that may complicate the comparison between fcorr and TTD corr is that the true peak typically will hit a sample with TTD correction, while with a frequency correction, the true peak can be anywhere between two samples. This has no consequences for the sound quality, but may influence what we see here.

 

In theory it should be the same or perhaps slightly higher for a TTD due to a better synchronized speaker behavior with more of the total SPL being direct sound.

 

Kind regards,

 

Bernt

[Mikkel Gylling Hangaard]

Hello Bernt,

 

What amplitude is desirable to reach when designing the target?

 

I have linked to pictures of my setup for people to comment on. Any input is very much appreciated :-)

https://picasaweb.google.com/mikkel.gylling.hangaard/Audiolense?authuser=0&authkey=Gv1sRgCNrT3-DAqOms1AE&feat=directlink

 

Best regards

Mikkel

 

2012/9/3 Bernt Ronningsbakk <bernt.ron...@lyse.net>

--

Audiolense User Forum.
http://groups.google.com/group/audiolense?hl=en?hl=en
To post to this group, send email to audio...@googlegroups.com
To unsubscribe, send email to audiolense+...@googlegroups.com

[Bernt Ronningsbakk]

Hi Mikkel,

 

I have never paid much attention to the amplitude. The target you are using is on the flat side of the preferred spectrum, but I’ve seen and heard equally flat targets that sounds great, so you should go by ear. The round-off in the top makes sense because it corresponds well with how your speaker measures. I think you can afford a bit more low frequency extension and perhaps you should spend that on a more gradual round-off in the bass. It is quite steep as it is right now. And steep may lead to more ringing.

 

The screen shots are somewhat blurred, but if the labels on the x axis on the impulse response are 1 ms apart, you have visible pre-ringing around 700 Hz. If you hear it I recommend that you activate the “mid frequency” in the correction procedure designer and apply a short time window at 500 Hz.

 

Kind regards,

 

Bernt

[Mikkel Gylling Hangaard]

Hi Bernt,

 

Thanks for your response.

Yes, it is at times harsh. It depends very much on how the albums or movies are mixed, though... and unfortunately.

I have made a filter with a B&K house curve which on average is a lot better. I'll upload it when I get time later.

 

Do you advice for or against partial correction where e.g. I turn off correction for the back left and right speakers (7.1-setup) above 10khz? I'm asking since they shoot over my head instead of facing the listening position. The sound is therefore quite a lot off-axis meaning the last octave from 10khz-20khz have a drastic roll-off which audiolense tries to correct for - quite unnecessarily.

 

 

Best regards,

[Bernt Ronningsbakk]

I bet that most of the roll-off is caused by the directionality of the microphone…

 

You have two, maybe 3 alternatives to get the back speakers right.

 

Alternative 1 is to add targets to your target “family”. You can make separate targets for the rear speakers and shape them any way you wish.

 

Alternative 2 is to do a measurement where you group the speakers. Or a speaker by speaker measurement. And reorient the microphone for each group or each speaker. These measurements have pause between speakers and thus, the timing difference between speakers can be wrong. If you do a short “all in one” measurement by the start of the session, the correct timing will be carried over to the speaker by speaker measurement. Or you can enter whatever delay you want to use manually.

 

Alternative 3 is to aim the microphone towards the ceiling. I don’t have any experience with that myself, but I believe there  are Audiolense users who have tried this.

[Mitch Global]

Hi Bernt,

 

Thanks for setting me straight.  I went back and redid my target again carefully following the roll-offs and got a much better result. 

 

http://i1217.photobucket.com/albums/dd381/mitchatola/simimpres.jpg 

 

If measuring peak to peak, would that be roughly about 0.067?

 

re: The height of the peak in the simulation corresponds directly with high frequency level and extension.

 

Thanks for that.  I seem to notice that the low frequency roll-off also has some impact on the peak as well...?

 

Mikkel, re measurement mic pointing straight up.  I have done that before, but you need the corresponding mic calibration file.

 

Cheers,

 

Mitch

--

[Mikkel Gylling Hangaard]

Hello Bernt,

 

Thanks again for your valuable input.

You are right that directionality is the biggest reason. The other reason is the room affects the frequency band up to aroudn 10khz by gaining that band.

 

Anyway, about alternative 2. I tried doing a short all-in-one, then a speaker by speaker. But the timing doesn't seem to be carried over to the speaker-by-speaker measurement. New values are entered when doing that measurement.

 

Should I write the timings of the all-in-one down and then enter them manually in the speaker-by-speaker afterwards?

 

 

Best regards,

 

Mikkel

2012/9/4 Bernt Ronningsbakk <bernt.ron...@lyse.net>

[Peter Hsieh]

Hi Bernt,

Can you elaborate how you could tell  700Hz pre-ringing from the impulse response?  Thanks.

Peter

On 9/3/2012 4:34 PM, Bernt Ronningsbakk wrote:

[Bernt Ronningsbakk]

What you suggest will work

 

Perhaps I need to take a second look at the carry-over functionality here…

[Bernt Ronningsbakk]

On the IR view there is a sinus like pre-ringing oscillation. Peak to peak (1 period) is perhaps 1.3 ms give or take, corresponds to a few hundred below 1 kHz. In the frequency response I can see something going on around 700 Hz, so I think that’s probably it.

 

Kind regards,

 

Bernt

 

From: audio...@googlegroups.com [mailto:audio...@googlegroups.com] On Behalf Of Peter Hsieh
Sent: Monday, September 03, 2012 10:46 AM
To: audio...@googlegroups.com
Subject: Re: [audiolense] Re: Target curves and headroom

 

Hi Bernt,

--

[markus_kr]

Calculation of preringing frequency:

>>>>>>>>>>>>>>>

On the IR view there is a sinus like pre-ringing oscillation. Peak to peak (1 period) is perhaps 1.3 ms give or take, corresponds to a few hundred below 1 kHz. In the frequency response I can see something going on around 700 Hz, so I think that’s probably it.
<<<<<<<<<<<<<<

That means:
(pre-ringing oscillation length = 1,3 ms)


pre-ringing frequency = 1 / pre-ringing oscillation length (in seconds)
                                     = 1/0,013 s
                                     = 769 1/s     = 769 Hz

best regards
Markus

[markus_kr]

Correction... (1/0,0013 s)

Am Dienstag, 4. September 2012 17:20:23 UTC+2 schrieb markus_kr:

Calculation of preringing frequency:

>>>>>>>>>>>>>>>
On the IR view there is a sinus like pre-ringing oscillation. Peak to peak (1 period) is perhaps 1.3 ms give or take, corresponds to a few hundred below 1 kHz. In the frequency response I can see something going on around 700 Hz, so I think that’s probably it.
<<<<<<<<<<<<<<

That means:
(pre-ringing oscillation length = 1,3 ms)


pre-ringing frequency = 1 /

pre-ringing oscillation length (in seconds)
                                     = 1/0,0013 s

[Bernt Ronningsbakk]

Yes,

 

A bit of math gets you in the ball park. Then you can examine the frequency response of the measurement. There is usually a magnitude peak or dip in play where pre-ringing occurs.

 

Kind regards,

 

Bernt

[Jarle Bergene]

Then there must be something strange with my target....

When I zoom in it becomes evident that it takes as much as 20ms just for for the impulse response of the target to reach its first positive maximum.. (ignoring those very cloes to the spike)

My target doesn't differ significantly from the other targets that has been uploaded here. And those targets have impulse responses that are more or less quiet after just 1ms,

my target impulse response:

http://imageshack.us/photo/my-images/196/targetimp.png/

target:

http://imageshack.us/photo/my-images/811/targetb.png/

Can someone please tell me what's going on here?

Regards,

Jarle

--

[Bernt Ronningsbakk]

There’s nothing wrong with your target, Jarle. It is dead silent until it rises straight up and reach first max within 1 or 2 samples. This is as it should be. About 1.3 ms after the first peak, your target is dominated by the low frequency roundoff. That is normal too. The high frequency activity that goes on for a little more than a millisecond is due to the roundoff in the treble. If you draw a target that ends straight out to the right, the target will look cleaner, but not the corrected speaker. What you see here is a strictly physical & mathematical relationship between a certain frequency response (your target) and the corresponding minimum phase time domain behavior. Also if you make a low frequency roundoff that has a lower Q alignment, the target will have less “decay” than what we see here.

 

Kind regards,

 

Bernt

 

From: audio...@googlegroups.com [mailto:audio...@googlegroups.com] On Behalf Of Jarle Bergene

Sent: Monday, September 10, 2012 9:30 PM
To: audio...@googlegroups.com

[Jarle Bergene]

Thanks!

 

Will try to obtain a more flat response in the upper end. I think I can gain a little bit by more optimal speaker toe in and listening height.

I would like to see a more clean impulse response! The impulse response of Brad's system looks very clean.

Brad, would you like to share a screenshot of your target in the frequency domain? Awesome setup by the way!

[Brad]

Hi Jarle,

If you look at the plot I attached to my earlier post labeled "impulse response" that shows my target as a thin brown line.

Brad

[Jarle Bergene]

Thanks Brad,

 

Found it! (picture called frequency response, shows the target and simulated response in frequency domain)

 

Our targets don't differ that much.

 

I have a 10db slope from 20Hz to 20kHz, while you have a 10dB slope from 20Hz to 10kHz. Regarding the tweeter roll off we are both following the response of our tweeters closely.

One should think that I would benefit from having a tweeter that has a more extended frequency response? (only drops 5db between 10kHz and 20kHz) Instead my target has awfully many oscillations during the first millisecond.

 

I realize that drivers with extended frequency response have the tendancy to roll off rather sharply when the knee finally comes. Is this the problem in my case? If so, is there any way to come around this? Other than making a more shallow roll off after the knee and thereby overruling the actual response? (suppose that solution would have some negative effects...)

 

Bernt, could you shed some light on this please?

 

Regards,

Jarle

[Bernt Ronningsbakk]

A steeper tweeter roll off will lead to more activity in the time domain. It is a physical necessity and has nothing to do with the quality of the gear involved etc. Very often the steepness that you see at the very end comes from the reconstruction filter of the dac and not from the tweeter itself. There is a fair chance that you will see a more gradual roll off if you measure to, say, 40kHz at 96 kHz sample rate. But there is also a chance that other high frequency noise in the system will surface… I don’t have your measurement fresh in mind, but this is the case in general.

 

Various tests, and the countless efforts from DAC manufacturers to optimize the reconstruction filter indicates that it is audible, at least when you play back  44,1 kHz material. Upsampling or resampling will make a difference, but it doesn’t eliminate the fundamental  problems.

 

A lot of the focus in DAC development the for the few last years seems to have gone into minimum phase vs linear phase vs a blend of the two, but also the steepness and the cut off frequency matters. Some of the most sophisticated reconstruction filters are only down a few dB at the nyquist frequency, which means that they will produce high frequency artifacts – mirror images - above the nyquist frequency. According to this paper :  http://www.mlssa.com/pdf/Upsampling-theory-rev-2.pdf such artifacts may actually do something good to the sound quality. This paper is a good read as it covers a lot of what happens from the recording studio to the domestic listener.

 

The target response used in  Audiolense, with the slope and the phase behavior will have an influence here. But don’t ask me what the best solution is. As can be seen from the above paper, there is a combination of many “imperfect world factors” that determines the final sound quality.

[Mitch Global]

+1 Bernt

 

I measure to 48Khz with a 96Khz sample rate.  Attached is a pic of the unfiltered hi frequency response and target.  Also attached is the corresponding target impulse response.  Note no "jaggies" and it is a linear phase target.

 

The reward for me has been the top end sounds extended and natural.

 

With respect to linear phase versus minimum phase, according to this description: http://www.dspguru.com/dsp/faqs/fir/properties  a linear phase filter "time aligns" the frequency response so all frequencies arrive at the same time.  Whereas minimum phase has the possibility of introducing phase distortion.  How that is implemented and/or affects the filters produced by Audiolense, and whether it is audible or not, I can't say.

 

With respect to targets.  I came across another approach from Bob Katz's excellent book called, "Mastering Audio, Second Edition: The Art and the Science"    Bob's suggestion is for a flat frequency response from at least 30 Hz to 2Khz and then begin a slow rolloff (which follows the natural roll-off of the tweeter).  This approach is working for me.

 

Cheers,

 

Mitch

[Peter Hsieh]

Hi Mitch,

You mentioned in the last paragraph that Bob Katz's book suggest flat frequency response from 30Hz to 2KHz, but the target curve you attached show a slope from 900Hz to 2KHz.� Was the point you tried to make related to high frequency roll off, not on the <2KHz flat?� Thanks in advance.

Best,

Peter




On 9/12/2012 9:50 PM, Mitch Global wrote:

+1 Bernt

�

I measure to 48Khz with a 96Khz sample rate.� Attached is a pic of the unfiltered hi frequency response and target.� Also attached is the corresponding target impulse response.� Note no "jaggies" and it is a linear phase target.

�

The reward for me�has been�the top end sounds extended and natural.

�

With respect to linear phase versus minimum phase, according to this description: http://www.dspguru.com/dsp/faqs/fir/properties� a linear phase filter "time aligns" the frequency response so all frequencies arrive at the same time.� Whereas minimum phase has the possibility of introducing phase distortion.��How that is implemented and/or affects the filters produced by Audiolense, and whether it is audible or not, I can't say.

�

With respect to targets.� I came across another approach from Bob Katz's excellent book called, "Mastering Audio, Second Edition: The Art and the Science"����Bob's suggestion is for a flat frequency response from at least 30 Hz to 2Khz and then begin a slow rolloff (which follows the natural roll-off of the tweeter).� This approach is working for me.

�

Cheers,

�

Mitch

�

On Wed, Sep 12, 2012 at 2:36 AM, Bernt Ronningsbakk <bernt.ron...@lyse.net> wrote:

A steeper tweeter roll off will lead to more activity in the time domain. It is a physical necessity and has nothing to do with the quality of the gear involved etc. Very often the steepness that you see at the very end comes from the reconstruction filter of the dac and not from the tweeter itself. There is a fair chance that you will see a more gradual roll off if you measure to, say, 40kHz at 96 kHz sample rate. But there is also a chance that other high frequency noise in the system will surface� I don�t have your measurement fresh in mind, but this is the case in general.

�

Various tests, and the countless efforts from DAC manufacturers to optimize the reconstruction filter indicates that it is audible, at least when you play back �44,1 kHz material. Upsampling or resampling will make a difference, but it doesn�t eliminate the fundamental� problems.

�

A lot of the focus in DAC development the for the few last years seems to have gone into minimum phase vs linear phase vs a blend of the two, but also the steepness and the cut off frequency matters. Some of the most sophisticated reconstruction filters are only down a few dB at the nyquist frequency, which means that they will produce high frequency artifacts � mirror images - above the nyquist frequency. According to this paper : �http://www.mlssa.com/pdf/Upsampling-theory-rev-2.pdf such artifacts may actually do something good to the sound quality. This paper is a good read as it covers a lot of what happens from the recording studio to the domestic listener.

�

The target response used in� Audiolense, with the slope and the phase behavior will have an influence here. But don�t ask me what the best solution is. As can be seen from the above paper, there is a combination of many �imperfect world factors� that determines the final sound quality.

�

Kind regards,

�

Bernt

�

From: audio...@googlegroups.com [mailto:audio...@googlegroups.com] On Behalf Of Jarle Bergene
Sent: Wednesday, September 12, 2012 9:38 AM
To: audio...@googlegroups.com

Subject: Re: [audiolense] Re: Target curves and headroom

�

Thanks Brad,

�

Found it! (picture called frequency response, shows the target and simulated response in frequency domain)

�

Our targets don't differ that much.

�

I have a 10db slope from 20Hz to 20kHz, while you have a 10dB slope from 20Hz to 10kHz. Regarding the tweeter roll off we are both following the response of our tweeters closely.

One should think that I would benefit from having a tweeter that has a�more extended frequency response?�(only drops 5db between 10kHz and 20kHz) Instead my target has awfully many oscillations during the first millisecond.

�

I realize that drivers with extended frequency response have the tendancy to roll off rather sharply when the knee finally comes. Is this the problem in my case? If so, is there any way to come around this? Other than making a more shallow roll off after the knee and thereby overruling the actual response? (suppose that solution would have some negative effects...)

�

Bernt, could you shed some light on this please?

�

Regards,

Jarle

�

�

�

On Wed, Sep 12, 2012 at 2:30 AM, Brad <hul...@mac.com> wrote:

Hi Jarle,

If you look at the plot I attached to my earlier post labeled "impulse response" that shows my target as a thin brown line.

Brad

On Tuesday, September 11, 2012 3:24:41 AM UTC-5, JB wrote:

Thanks!

�

Will try to�obtain a more flat response in the upper end. I think I can gain a little bit by more optimal speaker toe in and listening height.

I would like to see a more clean impulse response! The impulse response of Brad's system looks very clean.

Brad, would you like to share a screenshot of your target in the frequency domain? Awesome setup by the way!

�

Regards,

Jarle

�

�

�

On Mon, Sep 10, 2012 at 11:07 PM, Bernt Ronningsbakk <bernt.ron...@lyse.net> wrote:

There�s nothing wrong with your target, Jarle. It is dead silent until it rises straight up and reach first max within 1 or 2 samples. This is as it should be. About 1.3 ms after the first peak, your target is dominated by the low frequency roundoff. That is normal too. The high frequency activity that goes on for a little more than a millisecond is due to the roundoff in the treble. If you draw a target that ends straight out to the right, the target will look cleaner, but not the corrected speaker. What you see here is a strictly physical & mathematical relationship between a certain frequency response (your target) and the corresponding minimum phase time domain behavior. Also if you make a low frequency roundoff that has a lower Q alignment, the target will have less �decay� than what we see here.

�

Kind regards,

�

Bernt

�

From: audio...@googlegroups.com [mailto:audio...@googlegroups.com] On Behalf Of Jarle Bergene

Sent: Monday, September 10, 2012 9:30 PM

To: audio...@googlegroups.com

Subject: Re: [audiolense] Re: Target curves and headroom

�

Then there must be something strange with my target....

When I zoom in it becomes evident that it takes as much as 20ms just for for the impulse response of the target to reach its first positive maximum.. (ignoring those very cloes to the spike)

My target doesn't differ significantly from the other targets that has been uploaded here. And those targets have impulse responses that are more or less quiet after just 1ms,

�

�

my target impulse response:

�

http://imageshack.us/photo/my-images/196/targetimp.png/

�

target:

�

http://imageshack.us/photo/my-images/811/targetb.png/

�

Can someone please tell me what's going on here?

�

Regards,

Jarle

�

�

�

�

�

�

�

On Mon, Sep 3, 2012 at 5:13 AM, Brad <hul...@mac.com> wrote:

You need to zoom in on the time axis so you can see what is goin on. Can't tell how good it is like this.

On Sunday, September 2, 2012 4:27:59 AM UTC-5, JB wrote:

Hi Mitch,

�

Thanks for sharing your walkthrouh.

�

Here is a pic of my the impulse response of my system:

�

http://imageshack.us/photo/my-images/209/impulse.jpg/

�

The initial spike reaches 0.034 while the consequent fluctuatins is kept below 0.001. (factor of 34!)

When zoomed out it almost�looks like a perfect impulse response. However when�I zoom in I see fluctuations going on for�25ms.

But I guess no matter how close to perfect a real system is, you can always find fluctuations�if you just zoom in enough.

�

The question is, is the zoomed out state the correct state on which to base the verdict of my system?

Would you say that�it is�a good or a bad response?

�

Thanks!

�

Best regards,

Jarle

�

��

�

�

�

kl. 06:16:52 UTC+2 s�ndag 2. september 2012 skrev Brad f�lgende:

Why bother with the "dead nuts" mic positioning? Audiolense is going to set delays in the filters�to make all the drivers "hit" that spot at the same time. I would think a multi-position measurement average would be best.

�

On Saturday, September 1, 2012 6:41:54 PM UTC-5, Mitch wrote:

A quick walkthrough.�� While the fine tuning is very specific to my speaker/room combo, I think the general process would apply to most.

My speakers/listening position is set up as an equilateral triangle.� I use a distance laser measurer to line up the mic to the speakers.� I then run a measurement and have a quick look at the impulse response to see where I am at.� I move the mic slightly, re-measure and repeat until I get it spot on like so:

http://i1217.photobucket.com/albums/dd381/mitchatola/impluseequalateral.jpg

I am zoomed way in on the time scale (20 nanoseconds per horizontal division).�� I put a piece of tape on the floor directly over the tip of the mic so I can usually get the mic into position pretty quickly when it is time to re-measure.� The mic is pointing down the center line and is both at the height of my ears and my speakers tweeters.

Once I have everything lined up, I take a keeper measurement.� I check the impulse response to see if it is still spot on.

�

--
Audiolense User Forum.
http://groups.google.com/group/audiolense?hl=en?hl=en

To post to this group, send email to audio...@googlegroups.com
To unsubscribe, send email to audiolense+...@googlegroups.com

�

--
Audiolense User Forum.
http://groups.google.com/group/audiolense?hl=en?hl=en

To post to this group, send email to audio...@googlegroups.com
To unsubscribe, send email to audiolense+...@googlegroups.com

--
Audiolense User Forum.
http://groups.google.com/group/audiolense?hl=en?hl=en

To post to this group, send email to audio...@googlegroups.com
To unsubscribe, send email to audiolense+...@googlegroups.com

�

--
Audiolense User Forum.
http://groups.google.com/group/audiolense?hl=en?hl=en
To post to this group, send email to audio...@googlegroups.com
To unsubscribe, send email to audiolense+...@googlegroups.com

�

--
Audiolense User Forum.
http://groups.google.com/group/audiolense?hl=en?hl=en
To post to this group, send email to audio...@googlegroups.com
To unsubscribe, send email to audiolense+...@googlegroups.com

--
Audiolense User Forum.
http://groups.google.com/group/audiolense?hl=en?hl=en
To post to this group, send email to audio...@googlegroups.com
To unsubscribe, send email to audiolense+...@googlegroups.com

[Bernt Ronningsbakk]

Yes it is correct that min phase introduces phase distortion while lin phase preserves phase.

 

But a frequency region in a linear phase pulse can and will still produce time domain distortion. The total quantity of ringing in the time domain is not a lot different between the two; the difference is how it is distributed. In a min phase filter all ringing happens after the peak. In a linear phase filter there is symmetry before and after the peak.

 

Kind regards,

 

Bernt

[Mitch Global]

Hi Peter,

 

Attached is a pic of my target designer.  You can see that in the designer the entry for 2 KHz is 0.  However, when "smooth targets" is clicked on, you can see the 900 Hz to 2 KHz slope, which basically puts the response -1 dB at 2 KHz.

 

I acquired a new DAC on Monday and been making adjustments since.  Basically down to the last dB or so of tweaking. 

 

If you look at the frequency response again, I am just cutting off the tops of the peaks between 4.5 KHz to 14 KHz.  I may decide to lessen the amount of target attenuation in this range by a dB or so, which when smooth targets is checked on, should bring the 2KHz point back up to 0 dB.  Make sense? 

 

Quick question for Bernt.  When matching the target to the HF roll-off, does one match it to the very top peaks of the measured response or an average?

 

Cheers,

 

Mitch

On Wed, Sep 12, 2012 at 8:33 AM, Peter Hsieh <chun...@gmail.com> wrote:

Hi Mitch,

You mentioned in the last paragraph that Bob Katz's book suggest flat frequency response from 30Hz to 2KHz, but the target curve you attached show a slope from 900Hz to 2KHz.  Was the point you tried to make related to high frequency roll off, not on the <2KHz flat?  Thanks in advance.

Best,

Peter

On 9/12/2012 9:50 PM, Mitch Global wrote:

+1 Bernt

 

I measure to 48Khz with a 96Khz sample rate.  Attached is a pic of the unfiltered hi frequency response and target.  Also attached is the corresponding target impulse response.  Note no "jaggies" and it is a linear phase target.

 

The reward for me has been the top end sounds extended and natural.

 

With respect to linear phase versus minimum phase, according to this description: http://www.dspguru.com/dsp/faqs/fir/properties  a linear phase filter "time aligns" the frequency response so all frequencies arrive at the same time.  Whereas minimum phase has the possibility of introducing phase distortion.  How that is implemented and/or affects the filters produced by Audiolense, and whether it is audible or not, I can't say.

 

With respect to targets.  I came across another approach from Bob Katz's excellent book called, "Mastering Audio, Second Edition: The Art and the Science"    Bob's suggestion is for a flat frequency response from at least 30 Hz to 2Khz and then begin a slow rolloff (which follows the natural roll-off of the tweeter).  This approach is working for me.

 

Cheers,

 

Mitch

 

On Wed, Sep 12, 2012 at 2:36 AM, Bernt Ronningsbakk <bernt.ron...@lyse.net> wrote:

A steeper tweeter roll off will lead to more activity in the time domain. It is a physical necessity and has nothing to do with the quality of the gear involved etc. Very often the steepness that you see at the very end comes from the reconstruction filter of the dac and not from the tweeter itself. There is a fair chance that you will see a more gradual roll off if you measure to, say, 40kHz at 96 kHz sample rate. But there is also a chance that other high frequency noise in the system will surface… I don’t have your measurement fresh in mind, but this is the case in general.

 

Various tests, and the countless efforts from DAC manufacturers to optimize the reconstruction filter indicates that it is audible, at least when you play back  44,1 kHz material. Upsampling or resampling will make a difference, but it doesn’t eliminate the fundamental  problems.

 

A lot of the focus in DAC development the for the few last years seems to have gone into minimum phase vs linear phase vs a blend of the two, but also the steepness and the cut off frequency matters. Some of the most sophisticated reconstruction filters are only down a few dB at the nyquist frequency, which means that they will produce high frequency artifacts – mirror images - above the nyquist frequency. According to this paper :  http://www.mlssa.com/pdf/Upsampling-theory-rev-2.pdf such artifacts may actually do something good to the sound quality. This paper is a good read as it covers a lot of what happens from the recording studio to the domestic listener.

 

The target response used in  Audiolense, with the slope and the phase behavior will have an influence here. But don’t ask me what the best solution is. As can be seen from the above paper, there is a combination of many “imperfect world factors” that determines the final sound quality.

 

Kind regards,

 

Bernt

 

From: audio...@googlegroups.com [mailto:audio...@googlegroups.com] On Behalf Of Jarle Bergene
Sent: Wednesday, September 12, 2012 9:38 AM
To: audio...@googlegroups.com

Subject: Re: [audiolense] Re: Target curves and headroom

 

Thanks Brad,

 

Found it! (picture called frequency response, shows the target and simulated response in frequency domain)

 

Our targets don't differ that much.

 

I have a 10db slope from 20Hz to 20kHz, while you have a 10dB slope from 20Hz to 10kHz. Regarding the tweeter roll off we are both following the response of our tweeters closely.

One should think that I would benefit from having a tweeter that has a more extended frequency response? (only drops 5db between 10kHz and 20kHz) Instead my target has awfully many oscillations during the first millisecond.

 

I realize that drivers with extended frequency response have the tendancy to roll off rather sharply when the knee finally comes. Is this the problem in my case? If so, is there any way to come around this? Other than making a more shallow roll off after the knee and thereby overruling the actual response? (suppose that solution would have some negative effects...)

 

Bernt, could you shed some light on this please?

 

Regards,

Jarle

 

 

 

On Wed, Sep 12, 2012 at 2:30 AM, Brad <hul...@mac.com> wrote:

Hi Jarle,

If you look at the plot I attached to my earlier post labeled "impulse response" that shows my target as a thin brown line.

Brad

On Tuesday, September 11, 2012 3:24:41 AM UTC-5, JB wrote:

Thanks!

 

Will try to obtain a more flat response in the upper end. I think I can gain a little bit by more optimal speaker toe in and listening height.

I would like to see a more clean impulse response! The impulse response of Brad's system looks very clean.

Brad, would you like to share a screenshot of your target in the frequency domain? Awesome setup by the way!

 

Regards,

Jarle

 

 

 

On Mon, Sep 10, 2012 at 11:07 PM, Bernt Ronningsbakk <bernt.ron...@lyse.net> wrote:

There’s nothing wrong with your target, Jarle. It is dead silent until it rises straight up and reach first max within 1 or 2 samples. This is as it should be. About 1.3 ms after the first peak, your target is dominated by the low frequency roundoff. That is normal too. The high frequency activity that goes on for a little more than a millisecond is due to the roundoff in the treble. If you draw a target that ends straight out to the right, the target will look cleaner, but not the corrected speaker. What you see here is a strictly physical & mathematical relationship between a certain frequency response (your target) and the corresponding minimum phase time domain behavior. Also if you make a low frequency roundoff that has a lower Q alignment, the target will have less “decay” than what we see here.

 

Kind regards,

 

Bernt

 

From: audio...@googlegroups.com [mailto:audio...@googlegroups.com] On Behalf Of Jarle Bergene

Sent: Monday, September 10, 2012 9:30 PM

To: audio...@googlegroups.com

Subject: Re: [audiolense] Re: Target curves and headroom

 

Then there must be something strange with my target....

When I zoom in it becomes evident that it takes as much as 20ms just for for the impulse response of the target to reach its first positive maximum.. (ignoring those very cloes to the spike)

My target doesn't differ significantly from the other targets that has been uploaded here. And those targets have impulse responses that are more or less quiet after just 1ms,

 

 

my target impulse response:

 

http://imageshack.us/photo/my-images/196/targetimp.png/

 

target:

 

http://imageshack.us/photo/my-images/811/targetb.png/

 

Can someone please tell me what's going on here?

 

Regards,

Jarle

 

 

 

 

 

 

 

On Mon, Sep 3, 2012 at 5:13 AM, Brad <hul...@mac.com> wrote:

You need to zoom in on the time axis so you can see what is goin on. Can't tell how good it is like this.

On Sunday, September 2, 2012 4:27:59 AM UTC-5, JB wrote:

Hi Mitch,

 

Thanks for sharing your walkthrouh.

 

Here is a pic of my the impulse response of my system:

 

http://imageshack.us/photo/my-images/209/impulse.jpg/

 

The initial spike reaches 0.034 while the consequent fluctuatins is kept below 0.001. (factor of 34!)

When zoomed out it almost looks like a perfect impulse response. However when I zoom in I see fluctuations going on for 25ms.

But I guess no matter how close to perfect a real system is, you can always find fluctuations if you just zoom in enough.

 

The question is, is the zoomed out state the correct state on which to base the verdict of my system?

Would you say that it is a good or a bad response?

 

Thanks!

 

Best regards,

Jarle

 

  

 

 

 

kl. 06:16:52 UTC+2 søndag 2. september 2012 skrev Brad følgende:

Why bother with the "dead nuts" mic positioning? Audiolense is going to set delays in the filters to make all the drivers "hit" that spot at the same time. I would think a multi-position measurement average would be best.

 

On Saturday, September 1, 2012 6:41:54 PM UTC-5, Mitch wrote:

A quick walkthrough.   While the fine tuning is very specific to my speaker/room combo, I think the general process would apply to most.

My speakers/listening position is set up as an equilateral triangle.  I use a distance laser measurer to line up the mic to the speakers.  I then run a measurement and have a quick look at the impulse response to see where I am at.  I move the mic slightly, re-measure and repeat until I get it spot on like so:

http://i1217.photobucket.com/albums/dd381/mitchatola/impluseequalateral.jpg

I am zoomed way in on the time scale (20 nanoseconds per horizontal division).   I put a piece of tape on the floor directly over the tip of the mic so I can usually get the mic into position pretty quickly when it is time to re-measure.  The mic is pointing down the center line and is both at the height of my ears and my speakers tweeters.

Once I have everything lined up, I take a keeper measurement.  I check the impulse response to see if it is still spot on.

 

--
Audiolense User Forum.
http://groups.google.com/group/audiolense?hl=en?hl=en
To post to this group, send email to audio...@googlegroups.com
To unsubscribe, send email to audiolense+...@googlegroups.com

[Jarle Bergene]

Hi Mitch,

Why measure all the way up to 48kHz, wouldn't it suffice to measure up to say 30kHz to get a picture of the real tweeter roll-off?

Also, is it considered safe to measure all tweeters up to 48kHz? Before I try it's better to be safe than sorry ;)

Regards,

Jarle

[Mitch Global]

Hi Jarle,

On my frequency response graph, there is signal out to 45 KHz (-56 dB).  The mic and cal go to 20 KHz and my horn tweeter’s spec says -6 dB @ 21 KHz http://www.jblselenium.com/marcas/upload/d7d4f6a4591c3ae4da51ad778c71e3e1.pdf

My previous targets did not quite sound right in the very top end.  As an expereiment, upping Audiolense's sample rate and increasing the sweep frequency has uncovered more frequency response for my target to better match to.

Subjectively, this has resulted in a smoother sounding and extended top end.  I listen to headphones everyday and use that as my listening reference when comparing to my speakers. 

Is 30 KHz enough?  I could not say.  I can only relate what I did and heard in my experiment. 

Cheers,

Mitch

[Brad]

Here is an example of Bernt's description below: Is Linear Phase Really Ideal?

[OlavK]

I have finally got the chance to 

1. Test a flat target curve, with 10dB roll-off at 22.5kHz (no treble or bass boost prevention)

2. Optimizing the use of multiseat measurements (close to sweet), leading to rejecting two really off response measurements

3. Using gentle method B pre-ringing control

I have a quite treated room, some would say heaivly damped. The new curve sounds much more lively. I'll try it for a period.

Is there any risk in ticking off "treble boost prevention"? Like blowing my tweeter?

Best regards,

Olav

kl. 17:17:37 UTC+2 fredag 31. august 2012 skrev Mitch følgende:

From the Audiolense help file:

----------------

"We often get questions about why the filters attenuate the output so much. The short answer is that in order to avoid digital clipping, the average sound pressure level has to be attenuated by, typically 6-10 dB to give enough headroom for the single frequency that needs the most amplification. The frequency response of the correction filters – as they appear in the main form will give a 100% accurate picture of how the filters will attenuate the output. The sound pressure level of the evaluation charts doesn’t give the right picture in this regard.

If you see excessive attenuation in the correction filter, you should look for unnecessary boosting of single frequencies. Typically this can occur in the top treble when the treble boost limiting is disabled, and the measured speaker response drops before the target response. Such cases can easily be negotiated by activating the treble boost limiter or by making a target that doesn’t give as much boost"

-----------------

Depending on which Convolver you are using, you can make up the gain.  For example, if using convolverVST, it has a gain slider that you make the gain back up by adding 10 dB for example. Protip - press the shift key when moving the slider.  Or if you are using JRiver's Convolution engine, you can enable "normalize filter volume".  If that is not enough gain, then a nice free gain VST plugin is: http://www.bluecataudio.com/Products/Product_GainSuite/ 

With respect to the B&K curve, if it sounds dull to you, then change the target.  Do you have "prevent treble boost" checked on or off?  If on, try the same procedure with the treble boost off.  Some folks prefer flat out till the natural roll-off of the tweeter (i.e. no -3 dB at 2KHz like in the B&K curve).  Other folks use the "Sean Olive" target at the top of slide 24 here: https://docs.google.com/file/d/0B97zTRsdcJTfY2U4ODhiZmUtNDEyNC00ZDcyLWEzZTAtMGJiODQ1ZTUxMGQ4/edit?pli=1&hl=en    This curve is more of a slope starting at 0 dB at 20 Hz and then sloping to -10 dB at 20 KHz.  I am listening to and liking this "slope"  right now. Finally, depending on how dead or live your room is will also have an impact on the target used.  For example, my really live room liked the B&K curve.  But when I added acoustical treatments, I found it sounded a bit dull and started to experiment.

The best peice of advice I got from Bernt is to experiment.

Hope that helps.

Cheers,

Mitch

 

 

 

[Brad]

Here are the room response targets built into Audyssey MultEQ Pro. The dip in response is at 2 kHz. They call it midrange compensation. Human hearing is very sensitive around 2 kHz. Also, most speakers have a narrowing of beam width and correspondingly reduced power response in this frequency region. Reduced power response can cause room correction tools to boost the frequency response excessively in this region.

I use a 2 to 3 dB dip at 2 Khz in my Audiolense target. I find it lets me listen to live concert recordings at close to live music volume levels without the sound becoming harsh or causing listener fatigue.

My current midrange horns are exponential in design and as such do narrow in beam width with frequency. I will be building new constant directivity horns in the near future now that I can easily correct them with Audiolense. Maybe they will not need the dip.

Notice the trend that bigger room volumes need increasing roll off in the target response.

[Mitch Global]

Hey Brad thanks for sharing this.  Yes, it says what Bernt says.  Very interesting.   I am going to give min phase a try. 

Is there any info on the threshold level in dB (what are the vertical units of measure in the impulse window?) we can hear the  onset of ringing?   Would it be considered the same as http://www.ethanwiner.com/audibility.html Part 2?  Can we identify the threshold in Audiolense's impulse response window?  Is it frequency dependent?

Cheers Mitch

[Mikkel Gylling Hangaard]

Hello again everyone,

 

This is becoming one of the most excting topics in the forum. I've tried following Brads advice on reducing the freq and time windows. I wanted to try 96khz recording but that causes aliasing in the output signal (for some reason).

 

Anyway, I've attached some pictures of my measurements, since I have difficulties removing some pre-ringing in my left and right front-channels. Strangely it doesn't show in the front center-channel.

I was hoping some of you with your expertise could help me.

 

 

Best regards,

Mikkel

2012/9/14 Mitch Global <mitch...@gmail.com>

--

[Mikkel]

Just as an additional detail: I've never heard pre-ringing so I have no idea how it sounds. To this date I have never heard anything unusual from any kind of recorded material. So basically, is the preringing seen in the pictures in my last post worth fighting?

 

 

Best regards,

Mikkel

[Bill Street]

Hi Mikkel,

 

Looking at your last pic for your correction procedure designer, I see the 10 Hz setting for TTD is 5 cycles while the frequency correction setting is 3 cycles. I may be wrong, but my understanding is the TTD window should never be larger than the frequency window.

 

As for the visual pre-ringing in your impulse response pic, I think that is on the high side and you should try and reduce it. My first approach would be tweaking the hf roll off of your target. I would think this would give the most noticeable improvement (may end up being relatively extreme tweaking).. Maybe try eliminating some of the anchors in you target's hf area as well? Possibly try disabling the beta pre-ringing suppression you're using and see if it makes any difference, at least visually.

As for hearing pre-ringing, I've never heard it myself in music playback. I mentioned in a previous post that for me, the one test that I will hear even the slightest pre-ringing is with male dialogue. I have a test disc I use that has male voice saying things like "right channel, left channel, etc" and it will always show up in this. It's also a pretty good test in general for tonality/accuracy of the filter, since we all have a pretty good idea how a male speaking voice should sound.

 

Good luck,

 

Bill

--

[OlavK]

Mikkel, at least in the low frequency region, you would know if you had it :-)

I have problems <100Hz, and it creates a distinct whooping sound before the actual sound comes. In some music it is more pronounced. Try the intro of Beat It on Michael Jackson's Thriller album. The kick drum would be unbearable to listen to...

I have no idea if I have such problems in the high register, but I will make sure to read this thread thoroughly to learn and absorb!

Best regards,

Olav

[Mitch Global]

+1 to what Bill says.  In addition to what Bill pointed out in your correction procedure designer, you may want to try reducing the TTD subwindow from 500 milliseconds @ 10Hz to 25 milliseconds for example.  It's just an experiment as some folks get good results with long TTD windows, others with short windows....

 

Cheers, Mitch

[Brad]

There are some answers here:

linear phase eq

The Audio Expert

[Mikkel Gylling Hangaard]

Thank you very much for your response, Bill. Your suggestions seem to work. I've attached the results.

 

I made a new measurement with speaker-by-speaker measurement, reorienting the mic towards each speaker to obtain the on-axis frequency response. Prior to that I did a short all-in-one to obtain the delay settings.

 

After that I reduced the time domain window as suggested (made no real difference, but no reason to have a long window if it makes no difference, I guess).

Then I modified the target curve to follow the natural roll-off a bit more.

 

One odd thing I notice is that the back left speaker is delayed a bit compared to the others (if you look at the impulse response). Any suggestions to why that happened? I tried to change the delay for the channel but it made no difference.

 

 

Thank you once again for your input. Anyone, feel free to comment on what possible faulty settings etc. you see.

 

 

Best regards,

Mikkel

2012/9/14 Bill Street <stree...@gmail.com>

[Brad]

Hi Mikkel,

 

I have not used speaker-by-speaker with Audiolense, but I did for many years using DEQX processors and software. You are moving the correction balance from room correction towards speaker correction. This may be a good thing in your room. As you get closer to the speaker you will find the target curve will need less reduction of high frequencies. You may want to try a shorter frequency measurement window to take more room reflections out of the measurement. Always look at the filter response too, no need to post them but sometimes they can get improperly designed with weird shapes due to room and speaker interactions that the math just can't handle well. Just make some minor changes and try again.

 

Your target has an abrupt low frequency roll off. Plot the simulated group delay of the subwoofer channel and let's take a look at it. You will probably see the roll-off in the filter response is introducing a big frequency dependant delay which will shift the phase response up in the listening region of your subwoofer.

 

Look Here: Group Delay

[Mikkel Gylling Hangaard]

Hi Brad,

 

First, which measurement technique would you recommend? My reason for taking the measurements as I have was to reduce filter-work and the rapid roll-off in the last octave (10-20khz).

 

Anyway, I've attached the group delay for the sub, and indeed there seems to be a few rather large delays. .. any thoughts?

I have an idea that you will suggest a less steep roll-off. If so my question is at what freq. I should begin the roll-off and where it should end? The current design just follows the natural roll-off (if I remember correctly).

 

Best regards,

Mikkel

2012/9/16 Brad <hul...@mac.com>

[Mikkel Gylling Hangaard]

Ups, forgot the simulated response!

Here you go.

 

 

Mikkel

2012/9/16 Mikkel Gylling Hangaard <mikkel.gyll...@gmail.com>

[Brad]

Hi Mikkel,

 

Which measurement technique is best?  If your seating is close to a wall, measuring nearer to the speakers may be the best thing to do as measuring near a wall may cause some radical filtering to happen. Most speakers have a big difference between the power response and frequency response. The closer they are the better for measuring in the seats. All-at-once sure is easy and the speakers will all be the correct loudness and timing in the seats without another adjustment. You need a mic with good omni-directional response to do all-at-once measurements.

 

You can click on the chart in Audiolense to rescale the simulated delay graph of your sub so it is easier to read. It looks like you may have a difference in delay from 20 to 80 Hz of nearly 80 milliseconds. I noticed that in mine also with a target that rolls off in the bass. I just changed my target to be flat in the bass and I limited the boost to 6 dB or you can check the box to prevent bass boost. Music will probably never over stress your subwoofer but movie sound effects may. Use the bass roll-off target for playing movie special effects to protect your subwoofer if needed. Group delay is an issue with music but not with a bomb going off in a movie.

 

I'll post an update on what settings I'm using in my room. I had not closely looked at the filter responses and other speaker responses and I ended up needing much more time domain correction for my right main speaker. That's when I found the group delay issue and some weird filtering of one of my subs.

 

Brad

[Bernt Ronningsbakk]

Hi OlavK,

 

There is no significant risk of blowing the tweeter from unchecking this box. But study the correction filter in the top before you save just in case. If there’s a boost up there that you don’t like you need to rework your target or check the box, or both.

 

Kind regards,

 

Bernt

--

[Mitch Global]

Brad, thanks.  I am going to order up Ethan's book, looks good.  Can anyone comment on this book?  http://www.amazon.com/Introduction-Digital-Filters-Audio-Applications/dp/0974560715/ref=sr_1_11?s=books&ie=UTF8&qid=1347718093&sr=1-11&keywords=audio+digital+signal+processing

 

I changed to a minimum phase filter.  Attached are pics of target matching to high frequency roll-off, simulated frequency and impulse responses.

 

So far so good.

 

Cheers,

 

Mitch

[Bernt Ronningsbakk]

Hi Mitch,

 

At the risk of being picky, this is not a minimum phase filter. It is a mixed phase filter that corrects the sound towards a minimum phase target. Big difference.

 

You will get a minimum phase filter if you correct full range speakers without TTD correction. That’s what in Audiolense is regarded as a pure frequency correction.

 

I haven’t read the book but it looks like to be covering some of the most relevant topics with regards to Audio.

 

Here’s a free e-book that you may want to check out first: http://www.dspguide.com/

[Brad]

Mitch,

That book appears to be online for free here: Introduction to Digital Filters with Audio Applications

Brad

[Mitch Global]

Hi Bernt,

 

Thanks for pointing that out. For full range speakers, is there any difference/advantage to using a minimum or linear phase target?

 

In the case of full range speakers, is the only way to get a minimum phase filter is not to use TTD?

 

Thanks for the link to dspguide. 

 

Cheers,

 

Mitch

[Mitch Global]

Brad, thanks so much for sharing this!  Cheers, Mitch

[Bernt Ronningsbakk]

Yes there is a difference between using minphase target and linphase target. It will always be measurable but not always audible.

 

It seems like some people prefer min phase target while others prefer linphase target. Music taste may have something to do with it too, but I think it is more important that different systems respond different to the two. Sometimes minphase target works better. Sometimes linphase works best. But the jury is still out on this. And the jury is you guys, btw J

[Mikkel Gylling Hangaard]

My preliminary and intuitive feeling is that linear sounds more open and min. phase more focused and a tiny bit duller. But I'm not sure, though.

 

 

Best regards,

Mikkel

2012/9/20 Bernt Ronningsbakk <bernt.ron...@lyse.net>

[Bill Street]

Hi Mikkel,

 

Good to hear some of the suggestions helped. Main thing is, I hope it sounds better for you. I have no experience with multichannel setup and audiolense. Sorry I can't offer any suggestions there.

 

I've found something interesting (at least for me) while experimenting with Prering Suppression A (Repair) from Betatesting tab in the correction procedure designer. I've attached an image combining screenshots of my simulated frequency response. The top simulated response is with no Prering Suppression A being used. The lower one is using identical settings, target, etc., for everything else, except Prering Suppression A is enabled, using the settings shown. The upper Filtered Measurement is identical in both cases.

 

Of interest to me is the differences between the two results caused by Prering Suppression A. Notice the lower one (with prering suppression enabled remember). The simulated response is much tighter to the target. Notice how much more tightly aligned the right and left channels are as well, especially from about 60Hz up to around 800Hz.

 

Notice in the top simulated response, with no Prering Suppression, notice the dip in right channel (blue line) around 150Hz compared to the left channel (pink line) and especially the original filtered measurement above.There is no corresponding dip in the filtered measurement. I have spent a lot of time tweaking trying to get rid of that (and other less obvious, similar anomalies), with only limited success. I was never able to get rid of those completely. The best results were by reducing the window in the Time Domain Correction window, which to me always seemed like a trade off. I thought I was trading off more time domain correction in order to compensate for those anomalies. I considered these anomalies an issue because of reading in Audiolense help guide that a main thing to watch for in the simulated response/correction filter was we don't want to see dips where there weren't dips in original response, or peaks where there were none in the original. These "anomalies" as I call them, were these unwanted dips and peaks.

 

Now notice in the lower response how that dip was eliminated, as well as most of the others. It would take careful comparison between the two responses to really see all of them, but to me, it was pretty obvious. This was achieved strictly by enabling Prering Suppression A, all other settings identical. At this point, this filter is the best "sounding" one I've come up with. Comparing with and without the Prering Suppression enabled seems to give a better overall presentation. I'm no expert in audio but you can see that filter is _090 and my target is 43LP (linear phase by the way). That's a lot of tweaking. At this stage my filters and target were pretty much zeroed in, but enabling Prering Suppression made a noticeable improvement to me.

 

I'd be very interested in hearing any others experiences or comments along these same lines.

 

Bill