Home Theater Forum and Systems banner

Not open for further replies.
1 - 5 of 5 Posts

Super Moderator
9,204 Posts
Discussion Starter #1
Minimal EQ, Target Levels, and a Hard-Knee House Curve

It’s time to re-think the way we equalize

Better late to the game than never

It’s been a little over a year since Sonnie dragged me into the 21st century, convincing me to start using REW and the Behringer DSP-1124 digital parametric equalizer (thanks again, Sonnie!). It’s been a peculiar combination of elation and frustration.

Certainly, it was a delight to ditch the time consuming manual method of plotting frequency response for REW’s ultra-slick, instant results.

The frustration? Well, things never sounded as good as they looked on the screen. Don't get me wrong, "with EQ" always sounded better than "no EQ." But I was never completely happy.

For instance, my favorite test track for checking bass, Basia’s “Ordinary People” from her London Warsaw New York CD. This track features a bass line that runs up and down two full octaves, so it’s great for determining (a) if a subwoofer has linear response, and (b) if it’s blending well with the mains.

Well, it just wasn’t right. The lowest notes didn’t come through the way I’m used to hearing (and feeling) them, and the upper notes sounded bloated and overemphasized. The subs were clearly “invading” the mains’ lower reaches, but turning them down for a better blend meant the lowest notes were even worse.

Every few months I’d get frustrated and run REW again and re-equalize – and end up with pretty much the same thing, albeit with a different set of filters.

I think REW’s terminal coolness probably had an adverse effect, discouraging me from taking the trouble to get to the root of the problem. After all, I probably told myself subconsciously, response that looks this good has to sound good! You’ve never had these powerful tools at your disposal, so you’re just not used to hearing "correct!"

I may have discontentedly left it at that, except that I recently noticed something rather distressing with the Basia track that had previously escaped my notice: Certain bass notes were coming through noticeably lower in volume than the notes before and after. And elsewhere, a note or two slightly hotter than those adjacent. What was up with that? I’d never had that problem before, when I was using much-less-precise 1/3-octave equalization.

Here is the baseline response I’ve been working with since we moved into this house a couple of years ago:

base graph.jpg
Baseline Response

I’ve never been able to get anything resembling an appropriate-looking graph with less than 7-8 filters. Here are a couple of equalized graphs, the second being more recent.

dual shivas in corner w-7 filters, doors open.jpg
Equalized Response with 7 Filters

reponse with 75 target - 8 filters - rew 6 db housecurve.jpg
Equalized Response with 8 Filters and 6 dB House Curve

eight filters a.JPG
Filter Panel for Above

Getting to the root of the problem
Certainly, they aren’t bad looking graphs. As the baseline graph shows, my measured response peaked substantially above the 75 dB Target Level. Naturally, I followed the more-or-less standard protocol and equalized towards the Target.

Helping numerous people with their equalizing issues on this Forum, I’ve noticed this is a familiar refrain. The problem is that equalizing towards the Target Level requires more filters than would be necessary if you say, just raised or lowered it to a good mid-way point between the worst peaks and depressions. In some extreme cases we’ve seen people attempting to equalize a fairly easy curve with 9, 10 or even 11 filters, in order to hit the Target Level. In my case, you can see by comparing the baseline graph to the 8-filter panel that the majority of filters used - five of the eight - were “spent” realigning the ~22-42 Hz area down to the Target Level.

“The problem” – is over-equalizing really a problem? Well, at the pro audio forums, where people use equalizers for a living, the prevailing wisdom is “use as little as necessary to get the job done.” Of course, equalizing a subwoofer in a small room is a lot different from what those guys are doing, tuning a full-range system in some big auditorium. So does the “keep it minimal” advice apply to us?

As we shall see - probably so.

For my latest re-equalizing attempt, I tried a new tact – raising the Target Level up between the peaks and valleys before equalizing. Here is the result.

response w moved target -3 filters and hard knee house curve.jpg
Equalized Response with 81 dB Target and 6 dB Hard-Knee House Curve

four filters a.JPG
Filter Panel

Yes, you’re seeing correctly: four filters. In my case, I got lucky in that my sagging response above ~45 Hz played right into the hands of the hard-knee house curve I wanted, which enabled me to use fewer filters. Furthermore, I decided to avoid equalizing close to the crossover point (above ~70 Hz or so in my case, with a 90 Hz crossover), since any equalizing up there will probably be blown out by the main speakers once they’re added. (Of course, that needs to be double-checked before locking down your filters, but it was true in my case.)

So - how does it sound? Absolutely fabulous. I’d have to say my bass response has never sounded this good! Most of the improvement I attribute to the hard-knee house curve, which we’ll discuss later. (For the impatient people who want the “how does it sound” details now, scroll down to the Hard-Knee House Curve post, “The Holy Grail of bass?” heading.)

And - remember the problem with the “hot and cold” bass notes in the Basia track? Well, that is totally and completely gone. Each and every note is now at an appropriate and consistent volume from one to the next! :jump:

Folks, there’s simply no that way realigning a house curve could have solved that problem. The uneven bass notes could only have been a by-product of poor equalization. Or more precisely, equalization that had no business being there to begin with.

Compelled to excess?
So - where have we gone wrong? In at least a few ways, in my opinion. We’ve already opened the door to the Target Level situation: If you don’t properly realign it after taking your baseline measurement, it may require a lot of unnecessary filters to instead realign your response curve. This is true if you equalize manually, like I do, or if you let REW auto-equalize. Wholesale level adjustment via a multitude of filters is poor use of an equalizer. The REW help files clearly indicate that the Target may need to be shifted before applying filters; I don’t know how we managed to get hung up on the 75 dB Target as a quasi-rigid standard.

To a lesser extent, I think our standard recommendation of not smoothing graphs is another problem stimulating the urge to over-equalize, since an unsmoothed graph exaggerates the peaks and valleys the graph displays.

Another big problem, I feel, is the window in REW that we use here. Early on we decided that all graphs presented for evaluation should have an amplitude spanning 45-105 dB (i.e. the vertical scale), in the interest of comparing “apples to apples,” as it were, from one Member’s graph to the next. Of course, that was and still is a laudable goal. From time to time people have presented graphs with a larger-amplitude window – say, 20-130 dB, which results in a response line that’s “squeezed” tighter – i.e., less exaggeration between the peaks and valleys. That Member would inevitably get an advisement from our resident experts that the window setting they were using was being generous to their response, making it appear better than it really is.

I submit that the opposite is a more accurate reality, that the narrow 45-105 dB window makes response look worse than it sounds. It generates a scary-looking graph that compels people to over-equalize in an effort to make things “look” correct.

Yes, a wider 20-130 dB window will visually flatten out the curve. But in my opinion, it’s a more accurate representation of what the response you’re seeing on-screen truly sounds like. And since the curve “looks” better going in, there’s less of a compulsion to equalize it to death.

For instance, my equalized response in the Hard Knee graph above, shown in a 45-105 dB window, still looks fairly ragged, but in “real life” it sounds as smooth as melted butter. In other words, the 45-105 dB window is a poor reflection of what I’m hearing.

This was actually one of my earliest reactions after EQing with REW: Visually, there was a huge improvement in the way my equalized response looked. But switching the equalizer in and out, it was obviously that the audible improvement was much more subtle than it appeared on-screen in the 45-105 dB window.

Super Moderator
9,204 Posts
Discussion Starter #2
On Minimal EQ, Target Levels, and a Hard-Knee House Curve

Graphic spin-doctoring?
Okay, it’s show and tell time. Let’s conduct a little virtual experiment to see if I can back any of this up.

First, here is a graph that conforms to Shack standards.

baseline 45-105.jpg
Baseline Response w/ 75 dB Target Level, 45-105 dB Window, No Smoothing

Wow, it’s a mess, isn’t it? I mean, response is all over the place! Let’s let REW apply some virtual filters between 20-70 Hz to clean it up.

45-105 predicted response 6 filters.jpg
Predicted Equalized Response w/ 75 dB Target Level, 45-105 dB Window, No Smoothing

6 filter panel.JPG
Filter Panel

There we go – six filters. Predictably, REW equalizes response down to the Target Curve, but things look a lot better, right? Once again, I didn’t attempt to do anything above ~70 Hz. Otherwise, I could’ve easily used another 2-3 filters, pushing the total up to eight or nine.
Okay, let’s try again, this time using some alternative REW settings that will hopefully enable less equalizing.

baseline 20-130.jpg
Baseline Response w/ Raised 77 dB Target Level, 20-130 dB Window, 1/6-Octave Smoothing

Things look much better already. Here's a cropped picture of both graphs, for easy comparision. Exact same measurement, but notice how much worse the top (45-105 dB) graphs looks.

baseline 45-105 cropped.jpg
baseline 20-130 cropped.jpg
(TOP) Baseline Response w/ 75 dB Target Level, 45-105 dB Window, No Smoothing
(BOTTOM) Baseline Response w/ Raised 77 dB Target Level, 20-130 dB Window, 1/6-Octave Smoothing

With a wider window, 1/6-octave smoothing and the Target Level raised, we can see that the 27-43 Hz range really needs no equalizing at all. So – let’s virtually equalize what’s left.

20-130 predicted response 2 filters.jpg
Predicted Equalized Response w/ Raised 77 dB Target Level, 20-130 dB Window, 1/6-Octave Smoothing

2 filter panel.JPG
Filter Panel

There ya go – two filters, one to help out the depression at ~50 Hz, the second to address a rise at ~43 Hz that the first filter caused.

I can imagine some readers are having conniption fits right about now, saying there’s no way a contrived, two-filter curve based on “nothing more than graph manipulation” will sound as good as the six-filter. Relax. Take a deep breath. I’m going to show you why most of those six filters are accomplishing absolutely nothing anyway – at least, nothing good. Here’s why.

How to recognize perfectly useless filters
First, take a closer look at the 6-filter panel. Notice that four of the six have only 2-3 dB adjustments. Anyone who has spent any amount of time using professional-grade 1/3-octave or parametric equalizers, and listening to the results, will tell you adjustments that slight are barely audible in the lower frequencies.

6 filter panel.JPG

Second, look at the filter bandwidth settings – 4, 5 and so on. Behringer uses a peculiar “xx/60” bandwidth designator for the Feedback Destroyer that no other manufacturer uses. If we translate that to the more traditional fractions of an octave designation, we see that a 5/60 filter width equals 1/12-octave, 4/60 translates to 1/15-octave, and 3/60 is 1/20-octave. These are all bandwidth settings that we commonly use when equalizing with the BFD. Indeed, it’s not unusual for REW to recommend a 2/60 filter when auto-equalizing, which is 1/30-octave.

People, I hate to break it to you, but these are notch filters. Notch filters are what they use in live-sound PA systems to eliminate acoustical feedback. You’ve probably heard feedback at concerts and other live performances; it’s basically a single ringing frequency that’s runaway in the system. In that situation the sound engineer wants to eliminate the offending frequency with minimal intrusion on adjacent, unaffected frequencies. Ultra-narrow notch filters, typically set at 1/6-octave or tighter, allow them to do this, essentially sucking a single note out of the frequency spectrum (or as close to that as they can possibly get). *

Apply this to our situation at home and what do we have? When you pepper your subwoofer with a menagerie of notch filters, from the perspective of a bass instrument you’re essentially equalizing single notes (each note of the musical scale is about 1/12 octave). If any of those super-tight filters happen to hit dead center on the fundamental of a bass note, you can literally push or pull them up or down in the mix. While 2-3 dB might not be readily audible with a broad filter, a razor-sharp adjustment hitting single notes can probably make enough of a difference to notice. I believe this why I was getting those uneven notes with the Basia CD track. There’s simply no other explanation I can think of.

Lest you think I’ve taken leave of my senses, I can tell you that I have actually done this with my bass guitar and an analog parametric EQ: Dial in a super-tight filter on the equalizer, cut several dB, and hit a note on the bass. Turn the frequency knob, and when you hit the right frequency it’ll suck the note right out. Move up or down a fret to the next note, volume is restored. Turn the frequency knob a little more, and that note is sucked out.

Make no mistake, folks, a parametric equalizer is a very powerful frequency-altering tool. A device that can excise specific musical notes with surgical precision should not be frivolously utilized.

Getting back to our six- vs. two-filter graphs, I will again contend that the ultra-narrow filters used in the six-filter graph absolutely will not sound better than the two-filter graph. Now, I’m generally a reasonable fellow, so I’ll concede that it’s possible – perhaps highly so - that the two-filter graph will not sound better than the six-filter. But if that’s the case, that means those extra filters aren’t delivering any audible improvement. So - what’s the point? Why pollute your subwoofer to all that extraneous processing? At what point does surgery become butchery?

In addition, as anyone who’s manually tweaked filters on-screen in REW can attest, the close-spaced filters we commonly use often end up conflicting with each other: For instance, a second filter partially eradicates what the first accomplished, so you have to go back and tweak it some more, or perhaps even add a third filter. This is a situation known as “equalizing the equalizer,” and generally it’s not considered to be a good thing.

Folks, improved subwoofer sound quality is actually pretty simple to achieve. It only requires response smoothing – i.e., minimizing the most serious peaks and valleys, which are typically addressed with wider filters. This will not necessarily result in the best-looking graph, but it will generally give the best results from a sonic perspective. Unfortunately, the tight filters REW typically recommends are not conducive to achieving the best sound quality (see next post for more on this).

Does accuracy matter to you?
Certainly, if your primary interest and use of your system is confined to movies, none of this matters much. But if music listening is important to you, and you’re using more than say, 4-5 filters, you owe it to yourself to go back and see if you can do the job with fewer, utilizing a wider 20-130 dB window in REW, realigning the Target Level after measuring, and using 1/6-octave smoothing (for manual filter adjustments). If you have a lot of ultra-narrow, low-amplitude filters, say less than 4/60 at 2-3 dB, chances are you can lose them, and possibly filters near the crossover point as well. See if you can create a single, broad filter to accomplish what REW auto-EQ recommends with 2 or 3, (see example in the next post). I think you’ll find in the end that the only filters you really need are the “severe” ones that are wider and deeper (say 4-5 dB and 6/60 or more). Fewer filters should also minimize the “equalizing the equalizer” situation, although admittedly a certain amount of that is often inevitable. Plus, I think you’ll find that when it’s all said and done, you’ll find the 20-130 dB window is a much better representation of what your subs actually sound like, than the 45-105 window.

For an even more dramatic improvement in sound quality, stay tuned for the discussion on the hard-knee house curve, following the next post.

* UPDATE: Since writing this piece, I have determined that the BFD’s bandwidth settings are highly inaccurate. In general they are much, much broader than most other parametric EQs – see this thread. For instance, a 4/60 filter should be about 1/15-octave, but in reality 4/60 is more like 1/8-octave on most other equalizers. Likewise, the BFD’s 10/60 setting, which should translate as 1/6-octave, is more accurately about 1/3-octave on other equalizers. So, bandwidth settings down to 4/60 are acceptable, but not smaller. Generally, most sub equalizing will stay in the 4/60 - 10/60 range (1/8-1/3 octave).

Super Moderator
9,204 Posts
Discussion Starter #3 (Edited)
On Minimal EQ, Target Levels, and a Hard-Knee House Curve

Tips for those who use REW’s auto-equalizing function

I think we can all agree that REW is a fabulous program that has made our lives much easier, and we’re indebted to John Mulcahy for not only creating it, but giving it away for free. But here’s a news flash that a lot of people aren’t aware of: REW is not designed to optimize frequency response for the purpose of improving sound quality. That’s right, folks – REW is designed to optimize time domain characteristics by addressing modal resonances, and the filters it recommends are optimized to that end.

The objective of the "modal equalizing" method is not so much to achieve the best-sounding low frequency performance as it is to achieve the best-looking waterfall graph, which is a "3D"-like presentation that shows low frequency ringing, i.e. the time it lakes the low frequencies in your room to fully decay.

There are a few problems with this technique, in my opinion and experience. For starters, the measuring and processing platform (REW in this case) tends to treat every peak in response as a room mode, whether it is or not. A true room mode will exhibit a slower rate of decay compared to surrounding frequencies, but the truth is, not every peak will exhibit this phenomenon. In other words, not every peak in response is a room mode (nor is every depression a null). Since it treats all peaks as modes, the processing platform will tend to "over equalize" - i.e., recommend a plethora of ultra-tight notch filters in order to achieve the desired waterfall graph.

The next issue I have with modal equalizing is that it addresses only peaks in response, but does nothing about any depressions that may be present. Depressions aren’t as annoying as peaks, but eliminating them usually makes an audible improvement in sound quality. We can see this with the DSPeaker Anti-Mode. The Anti-Mode is like a parametric equalizer and REW all in one package: It performs both the in-room measurements and equalization automatically. But you can clearly see in this thread how the Anti-Mode ignores depressions that could be addressed by a more traditional parametric equalizer like the BFD.

Third, the "modal equalizing” implemented in order to achieve the best-looking waterfall is only valid in the location where the measurement was taken. Move away from there and your waterfall graph is “ruined.” Not that that will especially make an audible difference, but listening to your system is what you do, not gaze at the waterfall graphs it generates. If you can’t hear an appreciable difference between the location where the waterfall graph was generated and the seat next to it - why over-equalize for a pretty waterfall to begin with?

The last issue has to do with what might be termed “system instability.” While the measurement platform is stable enough (REW for example), the transducers involved - the speakers and elements in the measurement microphones - are not. Their physical (and consequently electrical) properties are altered with changes in temperature, humidity, etc. As such, when you take a second REW reading six months or a year later you'll find it doesn't look quite like your original one. A waterfall graph generated today with last year's filters isn't going to look as good as it did back on the day you fine-tuned the filters for minimal ringing.

I first observed this phenomenon more than fifteen years ago when I got my AudioControl real time analyzer/pink noise generator. I would painstakingly set my 1/3-octave equalizers based on the RTA display, only to check it again a few months later and find, to my great dismay, that my response was no longer perfect, but off a little here and there. At least that's what the visual display was showing. There was no detectable (read audible) difference.

Indeed, check this graph comparing three measurements I took at two-month intervals (top and bottom traces shifted for the sake of clarity):

Three Baseline Readings, Each at Two Month Intervals

Note that the 42 Hz room mode stays pretty constant in level and shape, while everything above and below that point changes noticeably: Peaks and dips appear and disappear, they change in severity and shape, their frequency centers move back and forth, etc. REW would generate a different set of filters for each measurement.

In the end, the best any equalizer can do is reduce the ringing from a room mode and bring it back in line with the rest of the frequency spectrum's decay times (fortunately enough, that's all that's needed to improve sound quality). It isn't going to reduce the mode's decay time to something less than the rest of the frequency spectrum (as discussed and seen here).

So all things considered, I have my doubts that equalizing for the best waterfall reading will give the best sonic improvement. It seems to me that any improvement you may get by optimizing a waterfall graph could likely and easily be offset by poorer audible performance, due to over-equalizing. Thus it seems the best tact would be to apply the "modal equalizing" approach to the true room modes, to get their decay times in line with what the rest of the waterfall graph displays, and apply a more general, smoothing approach to the rest of the response curve.

So while you really can’t use REW to auto-equalize for response smoothing, you can manipulate the program in that direction, and then manually tweak the final touches yourself.

Optimize REW filters for response smoothing...
First, REW calibrate for the standard 75 dB Target in the Settings panel. Take your measurement sweep, then adjust your viewing window for a 20-130 dB setting, using the Graph Limits box (top right of screen), so that your response curve doesn’t look so scary. (NOTE: This is mainly for your benefit – REW will recommend its filters irrespective of widow resolution.)

Next, re-adjust the Target Level to a good midway point between your response peaks and depressions - or if your particular curve permits, align the Target to where it most closely tracks the bulk of your response curve.

Under Trace Adjustments, set the Smoothing option to 1/6-octave, which will help induce REW to focus more on the most serious problems instead of the minor ones. Then under Filter Tasks, limit the Find Peaks range to a bit below your crossover point. The response you’re seeing close to the crossover frequency is most likely duplicated by the mains, so it’s to no effect to equalize up there. (Of course, this needs to be verified by comparing graphs for sub only, mains only, and both together.)

Next run the Find Peaks, Assign Filters, Optimize PK Gain, and Adjust PK Gain functions.

...Then, manually tweak the filter
Click on the EQ Filters box to see the filters REW has recommended. The dotted line in the graph shows the predicted effect the filters will have on your response.

Tweak the Frequency, Gain and BW/60 settings in the EQ Filters panel to alter REW’s recommended settings. I usually first click the Gain control a few times, to pinpoint where the filter is operating. Watch out for broad areas on the graph, as REW usually recommends filters that aren’t wide enough or centered properly to cover them. If REW is recommending two filters for a broad area, see if you can get the same result with a single filter, altering the gain, frequency, and bandwidth settings.

Also look for any useless filters REW may have generated for minor problems. These will be ultra-narrow filters with minimal gain adjustments – say 4/60 or less and only 2-3 dB gain reduction. Typically these can be eliminated as they are providing no audible benefit.

Don’t be afraid to apply some boost to a depression, if needed. This is much preferable from a sound-quality perspective than applying a multitude of filters to bring everything else down to the level of the depression. As I mentioned before, using multiple filters for level adjustment is poor use of an equalizer.

A few examples
Let’s perform a few real-life experiments to put see the above recommendations in action. You can save the *.mdat file attached at the end of this post to your computer, and open it in REW and see for yourself how this works. Be sure and load the attached house curve as well, using the House Curve tab in the Settings window. Also, set the equalizer for “BFD Pro DSP1124P” in the Equalizer menu, and the Graph Limits for 45-105 dB “Top to Bottom,” and 10-200 Hz “Left to Right.”

We’ll start with Graph #1, dated and timed Sept. 27 14:40:24. Set the parameters in the Trace Adjustments, Target Settings, and Filter Tasks boxes as shown in this picture:

40-24 45-105 db trace - target - filter settings.JPG

First we’ll take a look at the equalizing REW will recommend with an unsmoothed measurement, after running the “Find Peaks,” “Assign Filters,” “Optimize PK Gains and Q,” and “Adjust PK Gains” adjustments in the Filter Tasks box. The dotted line represents the response REW predicts after equalization.

40-24 45-105 auto eq.jpg
Predicted Response and Filters Using REW Auto Equalizing

40-24 8 filter auto eq  panel.JPG
Filter Panel

Look at all those filters! As you can see from the EQ panel, six of the recommended filters are either ultra-narrow, with a bandwidth of 3 or less, or with minimal gain adjustments of 3 dB or less. Useless filters!

Now let’s see what happens to the same measurements when we make some changes to the REW parameters in the interest of minimizing useless equalizing. First we will change the graph from 45-105 db to 20-130 dB:

40-24 graph axis adjustment.JPG

40-24 45-105 db cropped.jpg
40-24 20-130 db cropped.jpg
(TOP)45-105 dB Window
(BOTTOM) 20-130 dB Window

Next we’ll add 1/6-octave smoothing:

40-24 smoothing adjustment.JPG

40-24 20-130 db cropped.jpg
40-24 20-130 db 1-6 smoothing cropped.jpg
(TOP) 20-130 dB Window w/ No Smoothing
(BOTTOM) 20-130 dB Window w/ 1/6-Octave Smoothing

You can see that the reading is mostly below the 75 dB Target, so we’ll re-align the Target down to 70 dB:

40-24 target adjustment.JPG

40-24 20-130 db 1-6 smoothing cropped.jpg
40-24 20-130 db 1-6 smoothing  70 db target cropped.jpg
(TOP) 20-130 dB Window, 1/6-Octave Smoothing, 75 dB Target
(BOTTOM) 20-130 dB Window, 1/6-Octave Smoothing, 70 dB Target

Next, go to Filters Tasks and change the “Find Peaks” range window from 20-90 Hz to 20-70 Hz.

40-24 find peaks adjustment.JPG

Now, run the “Find Peaks,” “Assign Filters,” “Optimize PK Gains and Q,” and “Adjust PK Gains” adjustments. The dotted line shows what effect REW’s recommended equalizing will have on response. Notice that using 1/6-octave smoothing, REW has generated only three filters:

40-24 w smoothing rew 3 recommended filters cropped.jpg
Predicted Response, 20-130 dB Window, 1/6-Octave Smoothing, 70 dB Target, REW Auto-Equalized

40-24 3 filter auto smoothing eq panel.JPG
Filter Panel

However, we can manually tweak these filters to get even better response, using wider filters than what REW recommended:

40-24 20-130 db w manual 2-filter eq cropped.jpg
Predicted Response, 20-130 dB Window, 1/6-Octave Smoothing, 70 dB Target, Manually Equalized

40-24 2 filter manual eq panel.JPG
Filter Panel

With the first filter, I shifted the center frequency down from 23 to 22.75 Hz, and opened up the bandwidth from 5/60 to 7/60. For the second filter, I shifted the frequency center up from 32.78 Hz to 38.3 Hz, and opened up the bandwidth from 5 to 14. These adjustments better addressed the response problems in question. The third filter at 66.40 Hz was useless, as it was equalizing well up into the above-70 Hz range.

Here are both response charts and filter panels side by side for easy comparison:

40-24 w smoothing rew 3 recommended filters cropped.jpg
40-24 20-130 db w manual 2-filter eq cropped.jpg
(TOP) Predicted Response REW Auto-Equalized Three Filters
(BOTTOM) Predicted Response, Manually Equalized Two Filters

40-24 3 filter auto smoothing eq panel.JPG
40-24 2 filter manual eq panel.JPG
Filter Panels

Here’s another example, using Measurement #2 from the attached file, dated and timed Sept. 27 14:45:19. First we’ll change the graph size from 45-105 dB to 20-130 dB:

45-19 45-105 db cropped.jpg
45-19 20-130 db cropped.jpg
(TOP)45-105 dB Window
(BOTTOM) 20-130 dB Window

Next we’ll add 1/6-octave smoothing:

45-19 20-130 db cropped.jpg
45-19 20-130 db 1-6 smoothing cropped.jpg
(TOP) 20-130 dB Window w/ No Smoothing
(BOTTOM) 20-130 dB Window w/ 1/6-Octave Smoothing

Then we’ll raise the Target from 75 dB to 77 dB:

45-19 20-130 db 1-6 smoothing cropped.jpg
45-19 20-130 db 1-6 smoothing 77 db target cropped.jpg
(TOP) 20-130 dB Window, 1/6-Octave Smoothing, 75 dB Target
(BOTTOM) 20-130 dB Window, 1/6-Octave Smoothing, 77 dB Target

Now, change the “Find Peaks” range window to 20-70 Hz, and run the “Find Peaks,” “Assign Filters,” “Optimize PK Gains and Q,” and “Adjust PK Gains” adjustments. The dotted line shows what effect REW’s recommended equalizing will have on response. Using 1/6-octave smoothing, REW has generated only four filters:

45-19 20-130 dB 4 auto filters cropped.jpg
Predicted Response, 20-130 dB Window, 1/6-Octave Smoothing, 77 dB Target, REW Auto-Equalized

45-19 eq panel 4 auto smoothing filters.JPG
Filter Panel

Again, we can manually tweak these filters:

45-19 20-130 db 3 manual filters cropped.jpg
Predicted Response, 20-130 dB Window, 1/6-Octave Smoothing, 70 dB Target, Manually Equalized

45-19 eq panel 3 manual filters.JPG
Filter Panel

Admittedly, response using three manual filters doesn’t look quite as good as REW’s four filters. However the differences are so slight it would be impossible to tell the difference, from an audible standpoint, so there’s no reason to use more filters than will get an audible improvement. Again, here are both graphs and EQ panels for easy comparison:

45-19 20-130 dB 4 auto filters cropped.jpg
45-19 20-130 db 3 manual filters cropped.jpg
(TOP) Predicted Response REW Auto-Equalized Four Filters
(BOTTOM) Predicted Response, Manually Equalized Three Filters

45-19 eq panel 4 auto smoothing filters.JPG
45-19 eq panel 3 manual filters.JPG
Filter Panels

Remember, the best way to improve your subwoofer’s sound is to merely smooth response – i.e. eliminate or at least minimize the most serious peaks and valleys. It shouldn’t require more than a few filters. If you find yourself needing too many, go back and start over with a different Target Level alignment.

Don’t forget, if you want to post your graphs on the Forum for questions or comment, be sure and change the window back to the 45-105 dB setting. Response won’t look as good, but ignore any urge to equalize further!

Below are the two REW files (*.mdat) for use with the above examples. Be sure and load the house curve before you start, as it determines the slope of the Target Curve, which in turn influences what filters REW uses.


Super Moderator
9,204 Posts
Discussion Starter #4
On Minimal EQ, Target Levels, and a Hard-Knee House Curve

A Hard-Knee House Curve

A better approach

In the opening post I noted how good my subwoofers sound since implementing a new house curve. If you’ve never heard of a hard-knee house curve, don’t feel bad. I invented the term just last week. :D

It’s been a long time coming, but the realization has dawned that what I’ll call the “REW house curve” most of us utilize produces a woefully inadequate result. (Okay, perhaps labeling it the “REW house curve” is a bit disingenuous since it’s user-created, but I had to call it something! :) )

To understand what I’m talking about, let’s take a look at a standard 6-dB REW curve and compare it to my hard-knee curve.

no house curve.jpg
90 Hz Cut-Off, No House Curve

6 db stock rew house curve target.jpg
90 Hz Cut-Off, 6 dB REW House Curve

6 db hard knee house curve target.jpg
90 Hz Cut-Off, 6 dB Hard-Knee House Curve

The stock REW curve is achieved by simply dialing down response at the crossover frequency by a specific value (see this post and following for details on how to create a house curve text file for REW). The “REW House Curve” graph above shows a 6-dB slope between 30-90 Hz, meaning the slope begins its downward tilt at 30 Hz and is reduced 6 dB at 90 Hz, compared to the “No Curve” graph. (I’ve long been a proponent of shelving a house curve at about 30 Hz, as anyone who’s seen my house curve article knows.)

The problem with the stock REW curve is that it gets a lot of “action” at the bottom of the slope (i.e. the crossover frequency), but very little at the top. Note in the REW graph that response at 40 Hz – 1/3-octave above 30 Hz – is reduced only 1 dB. The differential between 40 Hz and 60 Hz – a span of nearly 2/3-octave - is only about 2 dB.

As fully explained in the article linked above, idea of a house curve is to increase the level of lower frequency bass signals so that they will sound subjectively as loud as upper frequency signals, delivering bass response that sounds flat (as opposed to merely measuring flat). The REW curve’s “bulging” slope just doesn’t accomplish that: In most cases, in the small rooms we use, it’s going to take a lot more than a 1-dB difference every 10 Hz to achieve a good-sounding house curve.

Adding to the problem, if your bass response after equalization ends up not tracking the house-curved Target as well as it should, you can easily end up with the situation I discovered when I double-checked mine with 1/12-octave sine waves: I wasn’t getting any significant slope until well above 60 Hz. That’s right – I wasn't shelved at 30 where I wanted to be. Not good.

Here’s a visual example - note that response in the graph below will remain essentially flat to about 45 Hz, despite the house curve target that starts sloping downward at 30 Hz.

Poor house curve tracking.JPG
Poorly Tracked House Curve Target

With a hard-knee house curve, REW’s “bulging” curve is traded for a straight-line drop (see “Hard Knee” graph above). Response ramps down rapidly below the 30 Hz “hinge” and (in the case of the 6-dB slope depicted) is reduced 3 dB by 40 Hz, and fully 6 dB by 60 Hz, which is the mid-way point between the (30-90 Hz) beginning and ending boundaries. The result is more accurate-sounding bass once the mid-bass “bloom” you get with the rounded slope is eliminated.

The Holy Grail of bass?
Okay, enough techno-babble. How does my new minimal-filter, hard-knee response sound?

Absolutely fabulous!

With music, I can turn off the sub when a bass line gets up to the high notes (a benefit of my remote-controlled Chase RLC-1) and hear very little change (depending on how the bass instrument is EQ’d in the CD, of course). That means the subs have seamlessly transitioned to the main speakers - a big change from before, when the subs were swamping the mains’ lower reaches and over-emphasizing the high bass notes. The mid-bass “bloat” the REW house curve generates is gone. And unlike before, now I can feel the lowest notes as well as hear them. Taming the mid- bass bloom has also brought better detail to the lower notes, detail that was previously obscured. It’s subtle, but noticeable.

And as mentioned in the first post, minimal EQ filtering has eliminated the uneven levels of the bass notes that I was experiencing in the Basia track.

Overall, bass sounds more taut, substantial and authoritative. Finally, across the board, my subs are doing what they should: seamlessly handing off the high notes to the mains, underpinning the mid-bass with lower harmonics, and transitioning to full dominance of the lowest notes that are well below the range of the main speakers. This is the way it should be.

With movies, the depth-charge scene in U-571 hits me in the chest and vibrates the couch. That’s the effect I was getting in our previous home, using only my 1/3-octave equalizers, but it’s something I’ve never been able to achieve here at the home we moved into a year or so ago. I had attributed it to the fact that the total volume of our family room and adjacent areas increased from an already-huge 6000 cu. ft. to an absolutely cavernous 9000 cu. ft, and my marginal DIY throw-down subs were barely cutting the mustard in the old place. Nope – turns out it was all in the equalizing.

I may go back and tweak the slope a little, but overall I think I’m knocking at the door of the “Holy Grail” of bass response – no exaggeration there, folks. It sounds so-o-o fine! All I need now is a bit more extension, detail and dynamics – a bass lover’s work is never done – but as far as response goes, I think I’ve nailed it. It’s as smooth as syrup.

Bottom line, I think you’ll find a hard-knee house curve will make a world of difference if you’ve been using the REW curve, and minimizing equalizer filters will possibly make audible improvements as well. It’s easy enough to set up another memory in the BFD for the sake of comparison. I suggest getting a couple of 1/4” – 1/4” couplers and cheap guitar cables so you can move the BFD with you to the listening position. This will allow you to make instant A/B comparisons between the two sets of filters.

The only caveat, it may be disconcerting at first hearing a hard-knee house curve if you’re used to hearing the REW curve. The upper notes may sound a bit “thin” once the “bloom” gone, since they will be supported less by the subwoofer, but overall things should sound tighter and more refined. Give it a chance.

By the way, if you don’t have a copy of that Basia London Warsaw New York CD, or something like it, you’d be doing yourself a tremendous favor to get it. If nothing else, she’s easy on the eyes.


How to create a hard-knee house curve in REW
So - you may be wondering, how do you create a hard-knee house curve? Well, it is more difficult than the REW curve, as it requires more entries in the text file and some trial and error to maintain the straight-line slope.

The first step, naturally, is to determine how much of a slope your room needs. As I outlined in my house curve article, this is best established by playing pure sine wave tones at the crossover frequency and the point you want response to shelve, typically ~30 Hz. The idea is for the lower frequency (30 Hz) to sound as loud as the upper frequency (90 Hz, or whatever your crossover setting is).

I usually start with the upper frequency tone, and make a note of its SPL reading. Then play the lower-frequency tone and adjust the receiver’s volume so that it sounds as loud as the first tone did. This will be a higher SPL reading than the first; the difference between the two readings is your house curve slope. This is best done with the two front speakers and the sub playing in tandem, as that is the way you listen to your system. (Don’t add the rear speakers or any other digital soundfield processing, as that can result in inaccurate readings.) Naturally, make sure neither your upper or lower frequency is smack in the middle of a room mode or null – run REW to see.

After your slope differential is determined, the next step is to create a plain text file that you can load into the House Curve tab of REW’s Settings panel. This is the tricky part, as the file will require multiple entries to establish the straight-line slope we’re looking for, instead of the usual two entries for the standard REW curve.

Here are the values I used to achieve my 6-dB hard-knee slope (again, see this post and following if you’re unsure of the meaning of these values):

30 6.0
35 4.4
40 3.1
45 2.0
50 1.1
60 -0.1
70 -0.6
80 -0.5
90 0.0

After designating the beginning and ending values (90 Hz and 30 Hz in this case), the next value you’ll probably want to add is at the midway point of the slope (60 Hz in my case). From there I added values at each 10-Hz increment, with additional markers at 35 and 45 Hz. I found it best to hold a piece of paper up to the screen to make sure I was keeping the line straight as I experimented with different value settings, loading each of them into REW one at a time to see what they had accomplished.

Once you’ve finished your hard-knee file, load it into REW and re-equalize your subwoofer accordingly.

Thanks to Shack member and former Moderator Ayreonaut, whose house curve example on our REW Tips thread pointed me in the right direction for creating the necessary values to facilitate a hard-knee house curve.
1 - 5 of 5 Posts
Not open for further replies.