Room EQ notch filters with delay?

Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.
Hi all,

I have some room modes I'd like to attenuate in my active X-O. The usual solution would be a notch filter designed to match the peak and Q of the measured room resonance.

Problem is, the standing waves in a room take some time to build up in a room. It appears quite evident using different stimuli, that the resulting "drone" in-room depends on the length of the exciting stimulus... A simple notch filter would just robotically cut the stopband frequencies by a determined amount of dB regardless of the actual standing wave situation - in essence I suspect it would sound "thin" on transients.

To prevent this from happening: Does anyone have a circuit handy whereby one could build an analog room EQ notch filter (typically an RLC notch with simulated inductor by op amp) that has a delayed onset? In essence a time windowed parametric EQ. The window delay could be tailored to the characteristics of the room at the specific frequency.

I'd imagine this should be possible using some kind of integrator as a servo for the op amp that simulates the inductor - in effect allowing it to work only if the signal in the pertinent frequency band persists for more than n cycles (or, a period say 1/2 of the RT(60) or something in that way).

I heard that the TACT room EQ system works in the time domain, I suppose they somehow do what I want to do, just in the digital domain and costing dearly. I essentially want an analog op amp circuit to the same effect.

Any input appreciated...
 
From your description of the slow onset, it sounds as if you have a high-Q room mode since the time behaviour of a high-Q resonance is characterised by a slow ramp up to the steady state gain. Conversely, a high-Q notch filter also shows the same slow onset.

Since room modes and notch filters are minimum-phase systems, that means you can predict the time response from the amplitude response, so if you can set up your notch filter to exactly cancel out your room resonance you'll find the slow onset of the room mode is exactly offset by the slow onset of the notch filter.

That does mean the notch filter needs to have the same frequency and Q as the room resonance. Slight mismatches are tolerable in the real world, however.

Long story short: don't worry about delaying your notch filter. Get it close to fixing the room and the problem solves itself.


Francois.
 
Excellent advice - I was just reading up and it dawned on me that the notch filter must have a delay, by design. But I couldn't quite figure out whether the time domain/frequency domain relationships of the filter (from the minimum phase requirement) would necessarily equate the ones of the room. Lack of a deeper understandig of the math involved...

As a sidenote, determining the actual steady state room response won't be a piece of cake however. With the usual low order MLS stimulus the peak looks quite low Q and harmless. Now, using multi cycle bursts or swept sines, this low Q low amplitude peak appears composed of several modes with extremely high Q. One seems to have corner frequencies of 100 and 104 Hz for instance... Curiously the exact f0 varies slightly with the day I measure it - I assume different humidity and barometric pressure could influence the actual speed of sound by a percent or so, meaning that the room modes will fall into slightly different frequencies as well.

Anyway, thanks for the explanation, I'll find a way to implement a "fair enough" solution.
 
Remember that anything non-steady state, like a swept sine or tone bursts, contains other frequency components. These will muddy the picture. If you do use a swept sine, slow down the sweep as much as possible to get a clearer idea of the actual room response. Also, if I recall MLS techniques properly, they're much like FFTs wherein the analysis frequencies are evenly spaced through the spectrum. Some MLS packages include a logarithmic chirp which might solve your problem by virtue of the chirp increasing resolution at low frequencies.

You're also correct in guessing that humidity and temperature slightly affect the room modes, since both of those change the speed of sound a little bit. Although, from what I remember of Singapore (apart from the excellent hawker stalls), isn't the place humid ALL the time?


Cheers,
Francois.
 
from what I remember of Singapore (apart from the excellent hawker stalls), isn't the place humid ALL the time?

Correct on both counts - but once you live here you discover the pleasant distinction of 80% vs 99.9% humidity ;)

The MLS works together with an appropriate FFT and appropriate sampling rate. The longer the stimulus, the higher the resolution, and as a side effect you also get a better S/N ratio on room modes.

For the family's sake I usually kept it short at about 90 ms (the signal is *not* a pleasant sound and I have to do it at night to keep ambient noise low). My software is bare bones though - no pre-emphasis of LF. There is a software called ETF that specifically targets room modes, and the pre-stimulation for room mode detection is in the second range... maybe I'll try that. Or single steady state sines in 1 Hz intervals...

Cheers,
 
Back to the original question. The logic is fairly solid, but there may be problems with assumptions. If you are trying to compensate for a mechanical room mode with an electrical notch filter, then "delays" are not equivalent. If you are trying to compensate for musical signals instead of true steady state excitement (really quite boring listening), then a steady state designed filter will almost always overcompensate for steady state measured build phenomena.

In the world were we practice this type of electical for mechanical compensation, the math gets the right correction and then we tune by ear until it doesn't sound wrong.

Reason, the only way to practically match the two requires impractical means.

Best of luck (and also check for temperature variations. Changes in structural flex depending upon ambient temperature may be a factor in the measured high resolution differences),

Mark
 
Helle there,

Influences from temperature & humidity are far less than the listening (measuring) positions in room, I guess.

Use a RTA & simply changing the mic location, you'll find the the response curve change a lot. Those resonances are also changing.

If you always sit on the sweet spot, then maybe it's OK to build a fixed notch filter. If not, it'll be painful to do it by OP-amp based filter. I tried it & not fun at all, although eventually it got the job done.

A digital EQ with convenient measurement/adjustment is highly recommended for this purpose. Behringer DEQ2496 is a good start. It's very cheap. I'm happy using it on my digital signal chain.
 
The effect of temperature and humidity on speed of sound was just a curiosity I observed, I am not too worried about that. The effect on structural properties may be there, and in addition here sometimes windows are all-open, sometimes closed, which must change the resonances. So, any filter must be a compromise. I'll be happy if I get rid of the 2 major loci of massive droning, at ca 100 and ca 50 Hz, for closed windows situation - only then is it quiet enough so that sound quality actually matters...

It is strange anyway that I should get these obvious modes even though I am already using dipoles. The room is large and irregular (L-shaped, bay window, minor variations in room height, plenty of glass doors and windows). That should if at all help acoustics. I suspect the main culprits are the resonant, undamped, uninsulated concrete construction, and the fairly even room height (ca. 3.3 m, explains the 50 and 100 Hz modes) . The vertical axis should fall ina dipole null, however, bounded by the floor at one end I don't have a dipole effect downwards on the z axis and likely excite the vertical mode through the floor. The floor is tiles on concrete fundament, so this is massive, doesn't resonate, but doesn't damp either. The ceiling is likely resonating like a drum but I can' do structural changes, I am renting...

Position in-room of both speakers and listeners / mic has a very large effect on measured FR. If, however, one succeeds in reasonably damping the underlying standing wave peaks, one should equally get rid of the corresponding nulls. The Mio $ question is, which measurement shows the necessary f0, a and Q of the required notch filter, and my original question, does the notch filter need a time correction?

So MarkMcK in your experience a complete electrical solution is subject to fine tuning.

What I am doing now is try to get a good idea of what to actually correct for. I use different means - slow sine sweeps and fast sweeps for qualitative evenness evaluation, MLS for FR measurement, and shaped tone bursts courtesy Siegfried Linkwitz Test CD for the exact f0 location (gives a much more drastic and exact peak than MLS). I will also give the ETF software a try - this is MLS based, it excites the room for 1 to 5 seconds, then does FFT of the signal in time slices to get an idea of the decay vs frequency. Supposedly this will leave the room modes intact while eliminating speaker anomalies. The crux is where to measure, and how much f0, a and Q as guessed from the decay parameters will quantitatively resemble the "actual" values. ETF recommends measuring in a node, i.e. corner, while exciting from the other end of the room. Maybe this is what leads to overestimation / overcorrection of peaks.

Step 2 then is to see where the results of the various methods qualitatively overlap.

Step 4 to guesstimate the values from a quantitative compromise of those measurements that were actually consistent.

Step 4, to build the filters and see if they sound right ;-). In practice I have an active XO and have built a small filter PCB where I can breadboard the notch filter by changing resistors in sockets around the op amp. So it's no big deal to experiment in the analog for me, the crux is just, what type/technique of measurement is actually "correct", do your ears tell you the truth or did you just get used to the anomaly, etc etc etc... Lucky this is a hobby and I can play around.
 
The problem with trimming by ear is that the person attached to the ear has to be able to recognize when the corrected sound isn't as wrong. This is not easy to do. That is why a relatively small number of companies get paid to do this for clients.

Being able to "hear" a room and tweak a calculated correction is a learned ability. You can learn through play. If you can afford it, I would second the suggestion of using a programmable digital filter for playing.

One good play program is to begin to learn the sound of what is wrong. Use the digital filter to introduce "known' anomolies into the response. Then after a few hundred iterations it will become easier to avoid those sounds when tweaking.

Best (and please have some fun with this),

Mark
 
MBK,

You probably solve your problems, but if you have time, please try shaped tone burst, video version, with titles...
It's available from:

http://www.bozoel.com/products/bozotest/index.html

This test has four sine periods in Hann window, long silence pauses (4s) between bursts,... etc.

We are still in working phase, but low frequency test is available, and if you have time, please share your experience.

Any opinions from others also will be highly appreciated.

Best regards,

boggy

p.s. "screen shot"

An externally hosted image should be here but it was not working when we last tested it.
 
Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.