Waterfalls

I've read some posts lately that I haven't really agreed with.

Point # 1 is regarding the effectiveness of equalization at low frequencies (15Hz-100Hz). The question posed is how can a parametric filter possibly correct room resonances. The assertion being that an EQ filter only lowers the relative SPL level in the room at that frequency, and as a result the ringing may be reduced since it drops into the noise, but it can't really correct the problem. Sorry, I don't agree..............................

Point # 2 deals with the practice of applying REW's smoothing feature to measured low frequency responses before addressing filter creation.

One of the most useful aspects of REW, and the one that seems to get ignored the most (especially with regard to the above) are the waterfall plots. Actually, I spend more time looking at LF waterfall and LF decay than I do frequency response graphs. They tell you so much more about the response of your room. How could they not, they add another dimension. The frequency response graphs completely ignore time. The effectiveness of smoothing can't be examined without a waterfall plot.

I performed a couple experiments that I think addresses the two points above.

Point # 1.
We've all looked at waterfall plots. They use the familiar horizontal axis of frequency and the vertical axis of SPL level that the frequency response graphs use, but they add a third dimension of time.

The waterfall is derived from the impulse response by shifting the impulse response window to the right by a proportion of the time range to generate each succeeding slice. It has to be generated before you can observe it, so you need to click the 'generate waterfall' for the measure you want to see..

The slice slider on the waterfall graph page shows you slices of time from the first response at zero time and then time is increased as you move to slice 30. So if you have the Time Range (ms) set to 300 ms and the resolution Window set to 300ms, each slice is 10msec after time zero. This shows what the microphone hears as time moves on from the initial sweep. If you have a resonance that tends to decay very slowly, you'll see it in the waterfall. This is what we're trying to reduce with the equalizer filters.

Whenever you create a waterfall plot of a measure, you can set the slider to zero and then slowly move it out to 30. This is the decay of the signal in the room. It will tell you quite a bit where the room resonances are.

Anyway, to better demonstrate that the behaviour of the BFD EQ filters have an effect in the time domain as well as the frequency domain (that matches the modal response of a room), I connected the BFD into a loopback cable of my soundcard and used this setup to take frequency response measurements.

Below is the result of a measurement of the BFD (with the filters bypassed) using the waterfall plot. The two dimensional frequency plot would show a simple flat line of course, but the waterfall plot adds the time dimension. Each slice in the waterfall is 10msec. After 130msec, the return is down in the noise.

Waterfall plot of a BFD with all filters bypassed.

Now I enter a single filter into the BFD. It is a 40Hz filter with a Gain of +15dB and a Bandwidth (BW) of 10, and I do a response measurement.

Below is the expected frequency response of the BFD with the single filter added.

Frequency Response plot of a BFD using a single filter of (40Hz, Gain +15dB, BW 10)

But now I look at the resulting waterfall plot of that single filter below.

Look familiar? Sure it does.

It looks like a room mode resonance of any REW measurement at subwoofer frequencies. And it should. It's because the EQ filter, just like the modal resonances of a room, has a time response that acts like a 2nd order biquad. If I apply an EQ filter with the same Q and opposite gain of a room mode, I would completely counteract the effect of the mode. See the time component of the filter (just like a room mode). It rings out, and still isn't in the noise after 300msec. You see, EQ filters don't just affect level. This is why they're so effective at equalizing at modal frequencies below 100Hz. Yes, it is listening position dependant, and only valid at the point where the response was measured, but because of the long wavelengths of low frequencies, the region around that area is fairly large. This is in opposition to higher frequencies where equalization is a bit of a waste of time, since the effective region is so small that eq is impractical.

Waterfall plot plot of a BFD using a single filter of (40Hz, Gain +15dB, BW 10)

As a side note, you can see what a completely terrible idea it is to add a gain filter to boost the level of a sub at low frequencies. You do nothing more than emulate a room mode at the gain frequency

So, if I enter a second filter into the BFD of (40Hz, -15dB, 10BW) to counteract the existing filter of (40Hz, +15dB, 10BW), the result is the measured response below. The room mode is completely nullified.

Waterfall plot plot of a BFD using a two filters of
(40Hz, Gain +15dB, BW 10) & (40Hz, Gain -15dB, BW 10)

And so, if we applied a counteracting filter in the BFD that matched a room's modes, the effects of the resonance is completely removed.

continued in the next post.......................