Some measurements (and discussion)

[DrWho]

I finally got around to obtaining a calibrated measurement system and just thought I'd share some measurements. I'm using a Behringer ECM8000 microphone, connected to a Behringer UB802 mixer (channel 1 panned left) and then the Left Tape Out is being fed into the Aux2 input on the front panel of my Audigy 2 ZS Pro.

I'm also using my computer as the source for all my music/movie playback being fed digitally from the Audigy into a Denon AVR 3300. The receiver is powering all of my speakers except the LR mains which are being powered via the Denon's preouts by an HK 630.

The subwoofer pre out is being fed into the Right Tape In. The tape channel is being routed to the Ctrl Room and the Right Ctrl Room Out is being fed into a Crown XLS 402 that is stored in a closet down the hall (that way I don't have to listen to the fan). The reason for this is to solve an impedance mismatch between the Denon and the Crown.

For mains, I'm running a pair of 1985 Klipsch Chorus II. My center channel is a Klipsch Synergy SC-1 and my rears are a pair of old Klipsch Heresy. For subwoofage I'm using a pair of DIY subs...Ascendant Audio Assassin 12", each in its own 3.5 cubic foot sonotube enclosure tuned to 16Hz.

I do not own a BFD, nor do I plan on it (more on that later).

Man that was a mouthful, now for the measurements (might as well put it in the next post)...

[DrWho]

Figure 1
Left Chorus II:


Figure 2
Right Chorus II:


Figure 3
Subwoofage:


Figure 4
Combined Response:


Figure 5
And finally, all 4 overlaid:


Figure 6
All 4 smoothed to 1 octave:
(to better illustrate a later point).

[DrWho]

Figure 7
Total system, spectral plot:


Figure 8
Total system, waterfall:

[DrWho]

The first thing I found interesting when measuring was the slight decay in the combined HF response as you go up in frequency - as seen in Figure 6. It looks as if this is the result of the insane comb-filtering in the higher frequencies (which is for the most part inaudible). However, I don't think I've seen this behavior on any other measurements taken in a similar fashion. Is this possibly the result of the "Post-Ref Window Width"? I know my room isn't symmetrical, but the mic was positioned equidistant from each speaker - I just find it odd that there seemed to be some out of phase interactions between the speakers. Anyone care to comment on how the windowing is being applied? I've noticed that I can change it after the fact, but haven't done any measurements to see if changing it during the measurement makes a difference...Being labelled as post-ref leads me to believe it's not, which then makes me wonder how exactly the data is being retrieved???

One thing I was curious about was the response of the Chorus II mains below 30Hz (where they are showing a huge increase in output). I know for a fact that these speakers roll-off at 18dB/octave below 40Hz, which combined with the HPF of the receiver should severely reduce the output down that low. At first I thought it was crosstalk, but to turn the subwoofer off I've been pulling the signal cable out of the mixer. I've also calibrated everything so I don't think it's that either...Ultimately, I'm just wondering how accurate my subwoofer measurements actually are below 30Hz.

The other thing I wanted to comment on is the time-response of the system. How exactly are these spectral and waterfall plots being generated?

One thing I find of interest is that the peaks and dips in the response are moving around through time. It's kinda hard to see, but Figure 7 is showing an intial flat response at 35Hz, but then a dip after a while. Likewise, you see dips at 50 and 70Hz sliding up in frequency...

Another interesting thing to point out are the resonances at 27 and 55Hz. Trying to EQ down the response in the 27Hz region (where there is a peak in the frequency response) is going to be in vain because it's not going to change the duration of the resonance.

The previous two paragraphs are just two reasons I would be very hesitant to use aggressive EQ to tame the response of a room. There's a bunch more, but I figured I would stick to things we can see in a common program that we're all using. I would be very interested in comparing the before and after spectral and waterfall plots for those implementing EQ. Who knows, perhaps I might change some opinions?

And lastly I just wanted to thank whoever is reposible for creating this great FREE program It's always interesting to see what your system is doing right and not so right.

It's late and I realize that my writing is not at its best at the moment, but I was hoping to stir some conversation and get a better idea of what all is going on - which should hopefully improve interpretations. I apologize if I'm asking questions asked elsewhere, but I did look first

Anyways, good night and I hope everyone had a merry Christmas. Best wishes for the New Year.

[JohnM]

Hello Mike,

Glad you like REW.

Altering the impulse response windows doesn't affect the measurement process. The final result of the measurement is the impulse response (IR). The IR is then windowed to select a portion of it and put it through an FFT to derive the corresponding frequency response. The pre and post ref windows select the portion of the impulse before and after the peak respectively. For assessing the system frequency response you would generally adjust the window widths to include all of the response up to the point where it disappears into the noise of the impulse - it is best to have the IR Y axis set to dB to see this more clearly.

The spectral decay plots and waterfalls are generated by starting with a window of (say) 300ms width (or whatever has been selected on thos graph pages) and applying it to the initial part of the IR, then moving it along to generate the frequency response plots for successively later parts of the IR.

A general roll-off at high frequencies is normal (and usually desirable). High frequency content is more easily absorbed and attenuated by furnishings in the room.

Resonances addressed by EQ show changes in both frequency and time domains. The resonance is essentially a 2nd order system, as is a typical parametric EQ filter. If the centre frequency, gain and Q of the filter are accurately matched to those of the resonance then the poles of the resonance are cancelled by the zeroes of the filter, which fixes both the frequency and time responses (in fact you can't change one without changing the other).

At low frequencies there can be a number of external influences which affect the measurement, such as noise from air conditioning ducts, traffic, trains etc. Parts of the waterfall which end up in a ridge which doesn't decay further would usually indicate some external source, which might be the case for the 27Hz peak.

Best regards,

John

[brucek]

If the centre frequency, gain and Q of the filter are accurately matched to those of the resonance then the poles of the resonance are cancelled by the zeroes of the filter, which fixes both the frequency and time responses (in fact you can't change one without changing the other).

Whenever anyone starts discussing poles and zeros of transfer functions, my eyes glaze over.

I know John correctly refers to the filters as second order effect, but in my little brain I see the BFD as simply reducing first order stimuli by lowering the frequency specific voltage to the sub itself. As a result the second order effect isn't as great. I'll leave all the fancy math to John - I trust he knows what he's doing.

Anyway DrWHo, I think if you play around with a BFD and see the effect on decay that results from lowering the level of a resonance in your response, you'll be sold. You'll not change the fact that there is a resonance, but if you target that specific frequency, the resonances effect isn't as pronounced, and its audible time will be lowered. You simply won't hear it.

brucek

[DrWho]

Is it possible to gain access to the raw data in the measured impulse response? It would be fun to drop it into Praxis (even the free version) where it can be convolved into any of the other views.

The resonance is essentially a 2nd order system, as is a typical parametric EQ filter. If the centre frequency, gain and Q of the filter are accurately matched to those of the resonance then the poles of the resonance are cancelled by the zeroes of the filter, which fixes both the frequency and time responses (in fact you can't change one without changing the other).

I understand this is only true in a minimum phase environment. I would be very interested to see some before and after plots of the time domain using the BFD if anyone is willing to make a few. Or if they already exist I would love to be pointed in their direction.

I've been reading some excerpts from Sound System Engineering, by Davis that discuss how EQ is only appropriate for dealing with the direct sound - not anything induced by the acoustics of the room. That's not to say that EQ can't "help", but it seems that acoustical treatment is the better way to go? Doesn't everything go to the second you move your mic a foot in either direction of the initial measurement used to determine the EQ?

Thanks again

[brucek]

I understand this is only true in a minimum phase environment.

Well that's correct, and aren't we discussing EQ of low frequencies? The wavelengths of any frequencies that we would care to equalize in a normal HT room would be sufficiently large to be considered minimum phase, and so first reflections arrive at any listening position at less than a cycle. EQ'ing those frequencies would result in the desired effect. My understanding is that this is why equalizing full range isn't effective.

I've been reading some excerpts from Sound System Engineering, by Davis that discuss how EQ is only appropriate for dealing with the direct sound - not anything induced by the acoustics of the room.

Likely was a discussion of full range?

brucek

edit:

Is it possible to gain access to the raw data in the measured impulse response?

You can Export the IR in the file pull down menu as WAV or TEXT.......

[DrWho]

brucek wrote:

Well that's correct, and aren't we discussing EQ of low frequencies? The wavelengths of any frequencies that we would care to equalize in a normal HT room would be sufficiently large to be considered minimum phase, and so first reflections arrive at any listening position at less than a cycle. EQ'ing those frequencies would result in the desired effect. My understanding is that this is why equalizing full range isn't effective.

Interesting - I've never heard that one before. I've been asking around (to those who know way more than myself), but in the meantime I did some research too. Seems it's a more widely argued point than I originally imagined...I think this article hit both sides pretty well:
http://www.prosoundweb.com/studyhall/lastudyhall/eq.php

Definetly some food for thought.

Thanks

[muse77]

There is not much I can add to what's been said already but I read the article that you linked to DrWho.
I found out that I am like the monkey they mentioned. I have a BSR EQ-3000, it came with a mic and built in pink noise generator. I have in the past did exactly like they said turn the noise on and adjust the sliders while watching the bar graph on the EQ. With RoomEQ I feel that I've climbed down out of the trees and became a Neanderthal. With the great info and I'll be a modern man yet.


To walk a mile takes a step at a time.

Bryan