Home Theater Forum and Systems banner

1 - 20 of 39 Posts

·
Registered
Joined
·
31 Posts
Discussion Starter #1
Hi,
I have been experimenting with full range equalisation with REW for a few days, and I'd like to discuss the topic of Target curves for a stereo 2.0 system. Strangely enough, I wasn't satisfied at all with the proposed target curve in REW, and I ended with a somewhat different one.

I came to REW and parametric equalisation since I replaced my hifi speakers (Dynaudio Gemini, then Kef R-300) with Neumann KH-120 monitors. They are active monitors with a flat response until 50 Hz, which triggers two room modes at 55 and 70 Hz that are absolutely unbearable.
The hifi speakers had an attenuated response at these frequencies, and I could play with the bass reflex opening, but with the Neumann, although the mids and highs sound much more natural, for low frequencies, equalisation couldn't be avoided.

The question of the target curve is especially interesting with the Neumann because they are supposed to sound completely flat in anechoic conditions, or that's what they claim. Here is an independent measurment of these speakers : http://kenrockwell.com/audio/neumann/kh-120-a.htm

Therefore what I measure with REW at the listening position should reflect the room gain more than the speakers response.

I use a Umik microphone in vertical position, with the standard calibration file (the Umiks are not calibrated one by one, instead, the file is supposed to fit a whole batch with the same serial number).

First, here are the measurements before equalisation. These are the average of 6 measurement for each speakers, separated by about 40 cm from each other, so as to match a wide range of possible listening positions.





All of the next curves (except the 0.9 db slope ones) differ from the reality because of the activation of -1 dB treble correction directly on the speakers, that isn't taken into account in the predicted result. I set the treble at -1 according to my liking, as the user manual explains that the treble level must be set according to the kind of acoustic - bright or damped- the speakers are installed in.
Anyway, here is the measured effect of the switch. This is nothing to worry about. It might even be lower than the accuracy of the Umik (don't take the general look of the curve into account, the tone controls of the preamplifier were mistakenly turned on during both measurements).



I first targeted the curve proposed by REW, which lead to this :





On the graph, it looks ok. But in real life, there seem to be a good lack of low frequencies. The sound is thin... and at the same time, the bass, below 100 Hz are too loud.

Then I decided to try the following : since the Neumann are supposed to sound neutral, let them do, and just equalize the low frequencies flat.
I decided to ignore the high frequency part of the curve, and to align the target curve with the measured curve in the 1 - 2 kHz range, where they have exactly the same slope.

Following the measured curve towards the low frequencies, I tried this : a flat target curve with a knee at 400 Hz, then -1.8 dB / octave (the original slope proposed by REW). The target level for bass equalisation being given by matching the measured and target curves between 1 kHz and 2 kHz.





And wow ! That sound much better. Although the bass are not perfect yet.

I then tried another idea : to completely follow the Neumann measured response in the mid-highs and continue with the same slope across the whole frequency range. That leads to a 0.9 dB / octave curve. Here is the predicted result :





Aaaarg ! That sounds terrible ! Muffled.

I came back to the previous curve, with the knee at 400 Hz, and I tried to move the knee to 200 Hz (more bass), or 600 Hz (less bass). The result were audible inferior. The sound seems balanced when the knee is at 400 Hz.

This week, I went further and tried to fill the notches in the 100 - 200 Hz range. That lead to a great improvement. The bass were cleaner, and more natural.

Listening to my new equalisation today, I found that there was still a little annoyance in the low mids, especially in choral music, but also in metal. I again tried to move the knee of the target curve to the right, but that was not good, as the mids sound too loud on solo voices.
I wondered if the problems didn't come from the remaining little oscillations between 300 and 1 kHz. They might very well be high order room modes. The speakers are standing 80 cm away from the rear wall, which causes two dips, one just below 100 Hz and the other just below 300 Hz. Maybe there are other ones.

I then carefully equalized them, checking in the overlay view of REW the 6 separate measurements to ensure that the corrections I was doing were valid for a wide enough set of listening positions.
The corrections are small, but I think that the problem in the low-mid is now reduced. The current result sound incredibly well balanced. Here are the predicted psychoacoustic curves :





And the filters used :





For the time being, the correction is done in Foobar2000's convolver, but I should soon get my MiniDSP and load the correction for all sources. I can't wait !


Now I'd like to hear about other people's experiments with the target curve. Do you actually feel the need to increase the bass level below 100 Hz ? Are you satisfied with the low-mids (200 - 800 Hz) ?
 

·
Moderator
Joined
·
1,662 Posts
Therefore what I measure with REW at the listening position should reflect the room gain more than the speakers response.
Welcome Pio.
I understand what you are saying, but in reality, the room/speaker response is inseparable. The "speaker response" you refer to is the on axis onset taken at some distance. There are many more 3 dimensionally, the bass being pretty much omni at LF. If we compared it to a gradient loudspeaker with the exact same anechoic/free field response that you are referring to as the "speaker" response, the room response would be completely different. A subtle point often missed.

This week, I went further and tried to fill the notches in the 100 - 200 Hz range. That lead to a great improvement. The bass were cleaner, and more natural.
By now you should have read JJ's work.:smile: It is generally not advisable to fill notches, as these are points of energy storage in the room (where the mic is). Filling these "points" tends to create excess elsewhere (there goes your "flat" onset :smile:)....but if the EQ was used judiciously..and you are happier with the sound, ok. Just beware of headroom issues if you are applying several db of boost at those frequencies.

For the time being, the correction is done in Foobar2000's convolver, but I should soon get my MiniDSP and load the correction for all sources. I can't wait !
Which MiniDSP, existing 2x4 or new 2x4HD?

cheers,
 

·
Registered
Joined
·
31 Posts
Discussion Starter #3
Thank you for the response, ajinfla

By now you should have read JJ's work.
What is JJ's work ? I've read the threads by Wayne A. Pflughaupt about house curves, and also Floyd Toole's book about Loudspeakers and rooms.

:smile: It is generally not advisable to fill notches, as these are points of energy storage in the room (where the mic is). Filling these "points" tends to create excess elsewhere (there goes your "flat" onset :smile:)...
I understand, although I wonder if this is also true for acoustic cancellations occurring at the speaker location (the 80 and 280 Hz dips caused by the front wall direct reflections).

Anyway, this is where I used the "overlay" option in REW. First, in the averaged curve, I identify a dip that I would like to fill. Then I reset all smoothing and go to the overlay view. I ask for all the individual curves measured for the speaker (6 in my case).
If the dip is present in all 6 measurements -> I fill it (in consequence, the sound may become worse outside of the measurement zone).
If the dip results of the averaging of various behaviors -> I equalize according to the individual curve that has the least pronounced dip.
If one of the curves recorded at equal distance of the speakers (one of the possible exact sweet spots), or if two curves recorded sideways have no dip at all, I don't equalize this part.

Then, there is the second problem with dips: our audition is more sensitive to peaks than to dips. That's where the Psychoacoustic smoothing option is useful. It draws a curve that gives precedence to peaks over dips.
So, after having selected the dips that are eligible for equalization, and setup the right frequency and Q in the filters section, I fine-tune the amount of correction with the Psychoacoustic view so that half of the curve is below the target, and half is above.
Doing the same thing with any other kind of smoothing would emphasize the regions where the curve oscillates around a mean value over regions where it is smooth.

Just beware of headroom issues if you are applying several db of boost at those frequencies.
Yes. I disabled the autolevel option in Foobar's convolver and set its gain to -8 db, as my highest boost is +8 dB.
I hope that there is an easy way to do this with the MiniDSP, as I don't see any master gain in the filters section of REW.

Which MiniDSP, existing 2x4 or new 2x4HD?
I ordered the cheap 2x4 with the Advanced 2x4 software. I'll just have to reduce my equalization to 6 filters per channel, as it is the maximum that the MiniDSP can do, if I understand well.
 

·
Premium Member
Joined
·
1,562 Posts
...Now I'd like to hear about other people's experiments with the target curve. Do you actually feel the need to increase the bass level below 100 Hz ? Are you satisfied with the low-mids (200 - 800 Hz) ?
We all know there is only one correct house curve. :sarcastic: It's the one we each prefer with our particular setup, measuring practices and chosen set of program material. :)
It's the one we choose is dependent on:
> Speakers used
> Room acoustics
> Chosen program material
> Smoothing type applied (psy, 1/6, var, etc)
> Mic orientation
> Averaging type (MMM, Multiple Sweeps, etc)
> EQ type and practices
> Our personal preferences

Now to answer your questions.
'Other Experiments':
Attached is versions 19-36 of the house curves I have tried. There are maybe 5 different measurement and averaging schemes that are represented there so there is no real direct comparison between them. I am currently using #31 as a reasonably good compromise for most material. It tends to be too sharp on some material for me, but is a good compromise.

'Bass Level below 100 Hz':
I leave 20-100 Hz target flat as shown. Note though that the relative level compared to the midrange varies depending on the House curve roll-off. More roll-off results in stronger bass. I just chose the bass range as the reference range for my trial curves. It made sense to me as it is a flat portion of the curves. Some of my earlier curves did not have this same standard and thus more confusing to display here.

'200-800 Hz':
No, not really. For me 100-600 is the range that is very difficult for me to get right. I have room influences that can only be solved with positional or room acoustic changes that I am not willing to make. At least in my case it seems the lower and higher freqs are relatively easy to control with EQ.

House Curves.png
 

·
Registered
Joined
·
31 Posts
Discussion Starter #5
Thank you for sharing your data and experiences. I'm a bit relieved that I'm not the only one to choose a flat extension into the bass below 100 Hz.

I have pictured three curves together : your average one, the one used in REW, and what I ended with after my calibration (including the -1 switch on the speakers).



They seem quite different, but the vertical scale is wide. In fact, the difference between them is typically 1dB, which is small.
 

·
Moderator
Joined
·
1,662 Posts
What is JJ's work ?
I thought you were the mod from HA. This JJ, this work, in particular: A Low-Complexity, Fast-acquiring Perceptually Tuned Room Correction Algorithm.
Yes Toole's work too, in particular: http://www.aes.org/tmpFiles/elib/20160707/17839.pdf

I understand, although I wonder if this is also true for acoustic cancellations occurring at the speaker location (the 80 and 280 Hz dips caused by the front wall direct reflections).
Not sure what you mean here. There should be no notches "at the speaker location", only away from speaker where the speaker/room response is measured, at least the pressure at that point.

If the dip results of the averaging of various behaviors -> I equalize according to the individual curve that has the least pronounced dip.
If one of the curves recorded at equal distance of the speakers (one of the possible exact sweet spots), or if two curves recorded sideways have no dip at all, I don't equalize this part.
Then, there is the second problem with dips: our audition is more sensitive to peaks than to dips. That's where the Psychoacoustic smoothing option is useful. It draws a curve that gives precedence to peaks over dips.
So, after having selected the dips that are eligible for equalization, and setup the right frequency and Q in the filters section, I fine-tune the amount of correction with the Psychoacoustic view so that half of the curve is below the target, and half is above.
Doing the same thing with any other kind of smoothing would emphasize the regions where the curve oscillates around a mean value over regions where it is smooth.
Ultimately if the bass sounds better to you with the EQ you have applied and there is no audible clipping of the signal (the speaker should handle those frequency boosts as they would not be excursion limited in those ranges, more amplifier power limited), then "correct" should not matter. Only your preference.

Yes. I disabled the autolevel option in Foobar's convolver and set its gain to -8 db, as my highest boost is +8 dB.
I hope that there is an easy way to do this with the MiniDSP, as I don't see any master gain in the filters section of REW.
Good, I figured you might know how to use Foobar:smile:. I believe the Minidsp has limiter functions, but you could always split 8db into 4db gain and 4db cut across the main bandwidth.

I ordered the cheap 2x4 with the Advanced 2x4 software. I'll just have to reduce my equalization to 6 filters per channel, as it is the maximum that the MiniDSP can do, if I understand well.
Ok.

cheers
 

·
Registered
Joined
·
31 Posts
Discussion Starter #7
I thought you were the mod from HA. This JJ, this work, in particular: A Low-Complexity, Fast-acquiring Perceptually Tuned Room Correction Algorithm.
Indeed I am. But I have been inactive for years in Hydrogenaudio. Thanks a lot for the link !

Not sure what you mean here. There should be no notches "at the speaker location", only away from speaker where the speaker/room response is measured, at least the pressure at that point.
I was talking about the distance between the speaker and the front wall behind it. If it is 80 cm, a 100 Hz signal will bounce back on the wall and pass again through the speaker location, but with opposite polarity, as 160 cm is roughly half the wavelength of 100 Hz. In this case, shouldn't the sound level decrease in all the room, like when you put two speakers playing with opposite polarity front to front ?
 

·
Moderator
Joined
·
1,662 Posts
Indeed I am. But I have been inactive for years in Hydrogenaudio. Thanks a lot for the link !
You coulda fooled me :smile:
https://hydrogenaud.io/index.php/topic,111267.msg916570.html#msg916570
Welcome on the links.

I was talking about the distance between the speaker and the front wall behind it. If it is 80 cm, a 100 Hz signal will bounce back on the wall and pass again through the speaker location, but with opposite polarity, as 160 cm is roughly half the wavelength of 100 Hz. In this case, shouldn't the sound level decrease in all the room, like when you put two speakers playing with opposite polarity front to front ?
It's a bit more complex than that as the speaker is omnidirectional in this range, so there will be a great many reflections and delays combining with the direct response to get that pressure minima you measure. All that should concern you is what is heard by your two ears...and what they prefer. A pressure mic should be used as a guideline, but as already noted, it may to be advisable not to try complete fill of those points of pressure minima/velocity maxima with more energy, as that more energy ends up elsewhere as well. Luckily those frequencies you are boosting should not overtax the driver. If the result to your ears is better bass...better yet.:smile:

cheers
 

·
Registered
Joined
·
31 Posts
Discussion Starter #9
By the way I've got a question. I read, and noticed, that it is better to equalize the lowest frequencies with both speakers active, because two independent correct equalizations don't sum up to give a correct one in stereo.

At what frequency should be the transition between common equalization (bass) and independent equalization (higher frequencies) ?
 

·
Registered
Joined
·
31 Posts
Discussion Starter #10
Thanks ajinfla,

About channel independence, I've run a small test : I turned my head to the left and the right while playing a sinewave with a software sine generator. It seems to me that the direction of the sound becomes perceptible between 70 and 100 Hz.
 

·
Moderator
Joined
·
1,662 Posts
By the way I've got a question. I read, and noticed, that it is better to equalize the lowest frequencies with both speakers active, because two independent correct equalizations don't sum up to give a correct one in stereo.
At what frequency should be the transition between common equalization (bass) and independent equalization (higher frequencies) ?
How far apart are the speakers? I would also use global EQ as you suggest, rather than independent CH.

Thanks ajinfla,
About channel independence, I've run a small test : I turned my head to the left and the right while playing a sinewave with a software sine generator. It seems to me that the direction of the sound becomes perceptible between 70 and 100 Hz.
Try an impulsive signal instead, something with a sharp onset.

cheers,
 

·
Registered
Joined
·
31 Posts
Discussion Starter #12
How far apart are the speakers? I would also use global EQ as you suggest, rather than independent CH.
The speakers are 130 cm apart from each other.
I experimented with full stereo, partial stereo and full mono correction, but, except for the two peaks below 100 Hz, it seems that the result depends more on the way the corrections are made (they are manual above 100 hz) than on the stereo or mono choice.

Anyway there is not much difference in the left are right measurements above 100 Hz, as we can see looking at both curves, so it shouldn't matter.

Left_and_Right.png

I'm currently satisfied with a set of filters that are identical below 100 Hz and different above. I'm currently tuning the general amount of bass with an extra low shelf filter of -2 dB starting at 200 Hz.

There was too much bass, but anything I do at and above 400 Hz to modifiy the target curve decreases the quality of the medium frequencies. I prefer not correcting anything above 300 Hz and deal with the bass level alone, which gives me a strange shaped house curve, with a small peak at 400 Hz.
 

·
Super Moderator
Joined
·
9,210 Posts

By the way I've got a question. I read, and noticed, that it is better to equalize the lowest frequencies with both speakers active, because two independent correct equalizations don't sum up to give a correct one in stereo.
As far as I know that doesn’t apply for main-channel speakers so much as people using multiple subwoofers in different locations. I’ve had good luck equalizing the lower frequencies of bookshelf-sized speakers.


At what frequency should be the transition between common equalization (bass) and independent equalization (higher frequencies) ?
I would recommend matching EQ above ~400 Hz. Indepent EQ above that point does weird things to the imaging. If that’s what you’ve been doing, it could account for your not being happy with your results in that range.

Regards,
Wayne

 

·
Registered
Joined
·
31 Posts
Discussion Starter #14
Ah thanks,

Here are the filters I'm currently using. It should be ok as long as mono / stereo is concerned.

Filter #1 and #2 have been generated by the software from the measurements made with both speakers, then copied here.

17-Filters-Left.png

18-Filters-Right.png
 

·
Registered
Joined
·
31 Posts
Discussion Starter #16
Oops ! I completely forgot about phase !

Here is the file. I've removed all individual measurements to make it smaller to upload.
The filters used are the ones that are active in the L and R measurements (Left and Right). The inactive filters are obsolete.
The St measurement (left + right speakers) is no more used.
 

Attachments

·
Premium Member
Joined
·
1,562 Posts
The attached file contains averages. They do not include the IR and thus no phase information. I was looking for 2 single sweep measurements at the LP; one each of the L and R mains. I just thought you may have that at hand.

From the discussion, I was just thinking of the common situation in the 200-800 Hz range where is difficult or inadvisable to EQ due to possibly creating more issues than resolving. I was just interested in seeing if you have as much trouble with strong reflections in this area as I do. I don't anticipate there is any helpful info to be found that would suggest a best EQ approach for in that range. I am on board with the general thoughts above regarding whether to / how to EQ in this area. It was just more for my continued learning.
 

·
Registered
Joined
·
31 Posts
Discussion Starter #18
The attached file contains averages. They do not include the IR and thus no phase information. I was looking for 2 single sweep measurements at the LP; one each of the L and R mains. I just thought you may have that at hand.
Sweep measurements or impulse measurements ?
I didn't do impulse measurements. How do you do this, in two words ?

I was just interested in seeing if you have as much trouble with strong reflections in this area as I do.
I'm sure I do. Ceramic floor, concrete ceiling and a bare concrete wall 50 cm on the left of the left speaker ! These solid surfaces absorb no acoustic energy. The sound completely bounce back on them.
 

·
Premium Member
Joined
·
1,562 Posts
I was just looking for 2 sweep measurements. Don't go to any trouble. It was just a thought.

To explain:
Given a normal sweep measurement, REW automatically calculates the IR and thus the phase trace is available. If several sweeps are taken in the listening area and the 'Average the Responses' button is selected then the IR/Phase data is lost in the average, i.e., the SPL alone is averaged without regard to the IR/phase of the original sweeps.

It is possible to manually average several sweep measurements in REW using 'Trace Arithmetic'. That process will retain the IR/phase data in the resulting average, but that is quite a bit more complicated to do. It is also considered unnecessary for the purposes of EQ. It does have other uses however.
 

·
Registered
Joined
·
31 Posts
Discussion Starter #20
Aaah Ok.

Here are the individual measurement for the center front listening position (attached file at the bottom of the message). I don't know how to interpret the phase.

Don't go to any trouble. It was just a thought.
I'm interested in any info. I've just received my MiniDSP, so I could easily perform a measurement with the correction loaded. Here is the result :

19-Final-LS-3.png

A huge bump at 300 Hz. I then went into new trials to see if I could get something more balanced, but if I try to remove the bump, it doesn't sound right. I just rose the low shelf from -3 dB to -2 dB to get a bit more bass below 200 Hz.

My hypothesis, for the time being, is that the excess at 300 Hz comes from reflections in the room, and if I try to equalize them, the direct sound becomes unnatural. I've read that we can, rather unconciously that conciously, distinguish the direct sound from the reflections (through directional clues, and precedence effect / masking). The microphone doesn't distinguish between them. It records the sum of everything.

It would mean that this bump would sort of "sounds right" as long as real audio sources would be expected to sound this way in my living room.

It could also mean that I did it all wrong from the beginning 0:)

Time will tell. I can't stand unbalanced sounds for a long time. If something is really wrong, I should get tired of it after some days.
 

Attachments

1 - 20 of 39 Posts
Top