Today I decided to tackle that dip in the living room left channel around 100 Hz. I assumed that this was also related to the dip in the stereo-correlated pink noise at 111 Hz. (This assumption now looks wrong.)
Crucial to my adjustment method is what I like to think of as "adjust-to-model." In principle, I am determining faults in the system performance, and either fixing them (if possible) or "correcting" them through some kind of sensibly determined system adjustment. The principle is fairly loose, I often don't know exactly what is causing this or that bulge or dip in the frequency response, I might simply label it "unknown resonance."
This is a vastly different philosophy, however, compared to the what-I-view-as-naive adjust-to-measurement school, which now seems to dominate the professional sound industry. Endless devices over the past 40 years and still automatically measure your system. Most take a fairly primitive adjust-to-measurement approach. Actually, my TACT units are way more sophisticated than that, and I still don't trust them. TACT uses a proprietary hybrid approach. I'd prefer things be completely out in the open and customizeable. Failing that, as all automated systems are, I'm fine with fully manual. I take measurements, devise some kind of model, and correct it.
In this case, I was convinced that the circa 100 Hz dip in the left channel was caused by an error in the delay compensation. I spent several days over the past two years fine tuning the delay times for woofers, panels, and tweeters. Most of this time was spent optimizing the right channel. When it came to the left channel, I had little choice but to accept the same values, as until a few days ago I hadn't figured out how to set different delays in each channel of my Behringer DEQ's, which is the only place I can do it currently.
But now, I've figured out how to do that, so I got to work right away twisting the delay time for the left subwoofer only, to match it with the panels so there would be no "cancellation" I assumed was causing the dip at 100 Hz.
Well, no such luck, I turned the knob forwards and backwards by several milliseconds and there was no signficant change in the level as measured by my phone in the listening chair (I usually haven't been so exacting, I might even just listen to things on the floor and assume that's close enough. But more and more it isn't.)
So if this notch near 100 Hz was NOT caused by cancellation, what was causing it? It probably wouldn't be a room mode, or it would affect the other channel, and in the other channel I'm actually appling a large parametric EQ cut at 100 Hz (something I tried removing to make the stereo pink noise response better...but unflattening the response in one channel to make the stereo pink noise response better is not the path to Hi Fi...it's making the stereo balance at those frequencies Even Worse...as I decided quickly a few weeks ago after naively trying this adjust-to-measurement). I immediately discovered the weakness was in the panel response, not the woofer response.
It didn't take too much thinking to begin wondering if it was boundary cancellation from the rear wall behind the Acoustats. I though about how much distance the sound wave would have to travel at 100 Hz for 180 degrees of phase change. That would be 1/2 wavelength, or about 5.5 feet. My speakers are about 3.5 feet from the wall, making the sum about 7 feet. That's pretty close, and perhaps all the extra stuff piled in this area reduces the "effective" distance.
I got out the oscillator and did some, well, lots of sweeping. Note to self: despite my recent calibration, I can't really trust the markings on my Genrad oscillator. I ultimately determined frequencies of interest by trying to cancel them with 1/10 octave notch filters on the EQ. From this, it became apparent that the frequency of the most obvious notch is actually around 93 Hz. That would correspond to about 6 feet. Perhaps that's actually the effective "average" distance of membrane to wall given all the junk in the corner, or some other influence is bending the cancellation notch, but the measured notch seems too close to where the cancellation notch should be, and where the cancellation notch should be, perhaps around 78 Hz, there's actually a measured peak (around 81 Hz) that's I'm cancelling for (see later view of all left channel eq's).
In any case, given that delay has no effect, and the speaker positions are almost entirely determined by other considerations, including practical considerations, there's not much I can do except apply EQ to this notch anyway. AND, since it's a notch in the panel response, it's essentially pointless to try correcting it in the panel EQ (and I only have mono EQ for the panels now, anyway). Nevertheless, I thoughtlessly tried correcting the notch in the panel response, and it was nearly useless, with 6 dB of boost causing about 1dB gain in the measured response.
Fortunately, since this is very close to the 100 Hz crossover frequency (in fact, 93 Hz is technically IN the subwoofer range) I can, should, and must correct the problem simply by boosting the subs. So, I simply added a high Q peak to the sub response at 93 Hz. And it seemed to have considerable effect, though nothing like 1-to-1. I didn't try to cancel out the notch as measured at the listening chair, but I raised the level a few dB. Above all, one doesn't want to overcompensate, and especially when adding boosts!
Even with 1/10 octave bandwidth, setting this boost higher than a few dB seemed to make 100 Hz (where I previously thought the notch was) too loud.
I proceeded to spend 90 minutes sweeping the left channel 20-200 Hz, and tweaking existing notches mostly. It was clear I needed to add a new EQ notch around 80 Hz, so I did, taking much of the burden off the notch at 71 Hz which I could then lighten up a bit. Sweeping an oscillator by hand is FAR more internally revealing than even looking at a 1/6 octave RTA display.
In the end, I think I made the left channel bass sound pretty smooth. I then looked at RTA and decided to make some additional refinements, then back to oscillator, and ultimately back to RTA.
I did the best smoothing job on both the swept and 1/6 octave measured response with 90 minutes of sweeping and measuring. The result being this
There still appears to be a slight depression at 100 Hz. But this slight depression is not obvious when sweeping the oscillator, in fact 100 Hz "sounds" louder than frequencies below it. In my interpretation, we are looking at essentially flat response from 50Hz to 20kHz, within 2dB (when averaged by usual methods) except around 30Hz, where the additional weight may be helpful. I've almost always had a far more elevated bass response than this, only getting enamored with "electrostatic bass" in the last couple years, in other words essentially flat "room curve" except now below 40 Hz. Also notice an extremely gradual "tilt" downwards with frequency. Many designers over the years have come to appreciate this downward tilt as being sonically beneficial. In my case, the downward tilt is extremely small, about 2dB down at 10kHz (it used to be identical to 1kHz until I added a new EQ there last week, to make the highs flatter from 3kHz-20kHz). As good as it gets! (Well, for now anyway. Until tomorrow's adjustment.)
You can see the plateau from 90 Hz on down which led me to add a second notch around 80 Hz as well as the existing one at 71 Hz. There was a larger roller coaster without notching 80 Hz down. When sweeping the oscillator, you hear the exact inflection points (even if you don't know exactly what they are with my not-remaining-calibrated oscillator) and can sweep the EQ's as narrow as possible to cancel peaks resulting from modes and resonances.
I am always disinclined to provide "boost" for various reasons. Since I run my digital processors close to 0dB there is essentially no extra headroom. If there was a pure tone at the exact frequency, the boost could cause digital clipping. However, at 93 Hz, the subwoofer drive is already being attenuated by about 5dB because of the LR4 crossover at 100 Hz. So I think adding back in 2dB of boost is fairly (if not perfectly) safe. (Perfect safety may require as much as 6dB of headroom WHENEVER you apply a boost or do anything tricky, then you can be sure sample-overs or other peculiar waveforms aren't going to cause clipping either. Countering one EQ against another isn't perfect protection from clipping, I've discovered on numerous occasions. I'll have to see how this current boost goes in practice, but it seems like it ought to be mostly OK, with possible clipping once every month of Sundays a small price to pay for solid and true bass with no annoying notch near 100 Hz.)
Here are the latest left channel EQ's:
Sadly, all this work didn't seem to remove the dip or notch in the stereo-correlated pink noise response for both channels. So, all this work, and I didn't address that problem at all, which seems to be above 100 Hz and not below it anyway.
Crucial to my adjustment method is what I like to think of as "adjust-to-model." In principle, I am determining faults in the system performance, and either fixing them (if possible) or "correcting" them through some kind of sensibly determined system adjustment. The principle is fairly loose, I often don't know exactly what is causing this or that bulge or dip in the frequency response, I might simply label it "unknown resonance."
This is a vastly different philosophy, however, compared to the what-I-view-as-naive adjust-to-measurement school, which now seems to dominate the professional sound industry. Endless devices over the past 40 years and still automatically measure your system. Most take a fairly primitive adjust-to-measurement approach. Actually, my TACT units are way more sophisticated than that, and I still don't trust them. TACT uses a proprietary hybrid approach. I'd prefer things be completely out in the open and customizeable. Failing that, as all automated systems are, I'm fine with fully manual. I take measurements, devise some kind of model, and correct it.
In this case, I was convinced that the circa 100 Hz dip in the left channel was caused by an error in the delay compensation. I spent several days over the past two years fine tuning the delay times for woofers, panels, and tweeters. Most of this time was spent optimizing the right channel. When it came to the left channel, I had little choice but to accept the same values, as until a few days ago I hadn't figured out how to set different delays in each channel of my Behringer DEQ's, which is the only place I can do it currently.
But now, I've figured out how to do that, so I got to work right away twisting the delay time for the left subwoofer only, to match it with the panels so there would be no "cancellation" I assumed was causing the dip at 100 Hz.
Well, no such luck, I turned the knob forwards and backwards by several milliseconds and there was no signficant change in the level as measured by my phone in the listening chair (I usually haven't been so exacting, I might even just listen to things on the floor and assume that's close enough. But more and more it isn't.)
So if this notch near 100 Hz was NOT caused by cancellation, what was causing it? It probably wouldn't be a room mode, or it would affect the other channel, and in the other channel I'm actually appling a large parametric EQ cut at 100 Hz (something I tried removing to make the stereo pink noise response better...but unflattening the response in one channel to make the stereo pink noise response better is not the path to Hi Fi...it's making the stereo balance at those frequencies Even Worse...as I decided quickly a few weeks ago after naively trying this adjust-to-measurement). I immediately discovered the weakness was in the panel response, not the woofer response.
It didn't take too much thinking to begin wondering if it was boundary cancellation from the rear wall behind the Acoustats. I though about how much distance the sound wave would have to travel at 100 Hz for 180 degrees of phase change. That would be 1/2 wavelength, or about 5.5 feet. My speakers are about 3.5 feet from the wall, making the sum about 7 feet. That's pretty close, and perhaps all the extra stuff piled in this area reduces the "effective" distance.
I got out the oscillator and did some, well, lots of sweeping. Note to self: despite my recent calibration, I can't really trust the markings on my Genrad oscillator. I ultimately determined frequencies of interest by trying to cancel them with 1/10 octave notch filters on the EQ. From this, it became apparent that the frequency of the most obvious notch is actually around 93 Hz. That would correspond to about 6 feet. Perhaps that's actually the effective "average" distance of membrane to wall given all the junk in the corner, or some other influence is bending the cancellation notch, but the measured notch seems too close to where the cancellation notch should be, and where the cancellation notch should be, perhaps around 78 Hz, there's actually a measured peak (around 81 Hz) that's I'm cancelling for (see later view of all left channel eq's).
In any case, given that delay has no effect, and the speaker positions are almost entirely determined by other considerations, including practical considerations, there's not much I can do except apply EQ to this notch anyway. AND, since it's a notch in the panel response, it's essentially pointless to try correcting it in the panel EQ (and I only have mono EQ for the panels now, anyway). Nevertheless, I thoughtlessly tried correcting the notch in the panel response, and it was nearly useless, with 6 dB of boost causing about 1dB gain in the measured response.
Fortunately, since this is very close to the 100 Hz crossover frequency (in fact, 93 Hz is technically IN the subwoofer range) I can, should, and must correct the problem simply by boosting the subs. So, I simply added a high Q peak to the sub response at 93 Hz. And it seemed to have considerable effect, though nothing like 1-to-1. I didn't try to cancel out the notch as measured at the listening chair, but I raised the level a few dB. Above all, one doesn't want to overcompensate, and especially when adding boosts!
Even with 1/10 octave bandwidth, setting this boost higher than a few dB seemed to make 100 Hz (where I previously thought the notch was) too loud.
I proceeded to spend 90 minutes sweeping the left channel 20-200 Hz, and tweaking existing notches mostly. It was clear I needed to add a new EQ notch around 80 Hz, so I did, taking much of the burden off the notch at 71 Hz which I could then lighten up a bit. Sweeping an oscillator by hand is FAR more internally revealing than even looking at a 1/6 octave RTA display.
In the end, I think I made the left channel bass sound pretty smooth. I then looked at RTA and decided to make some additional refinements, then back to oscillator, and ultimately back to RTA.
I did the best smoothing job on both the swept and 1/6 octave measured response with 90 minutes of sweeping and measuring. The result being this
There still appears to be a slight depression at 100 Hz. But this slight depression is not obvious when sweeping the oscillator, in fact 100 Hz "sounds" louder than frequencies below it. In my interpretation, we are looking at essentially flat response from 50Hz to 20kHz, within 2dB (when averaged by usual methods) except around 30Hz, where the additional weight may be helpful. I've almost always had a far more elevated bass response than this, only getting enamored with "electrostatic bass" in the last couple years, in other words essentially flat "room curve" except now below 40 Hz. Also notice an extremely gradual "tilt" downwards with frequency. Many designers over the years have come to appreciate this downward tilt as being sonically beneficial. In my case, the downward tilt is extremely small, about 2dB down at 10kHz (it used to be identical to 1kHz until I added a new EQ there last week, to make the highs flatter from 3kHz-20kHz). As good as it gets! (Well, for now anyway. Until tomorrow's adjustment.)
You can see the plateau from 90 Hz on down which led me to add a second notch around 80 Hz as well as the existing one at 71 Hz. There was a larger roller coaster without notching 80 Hz down. When sweeping the oscillator, you hear the exact inflection points (even if you don't know exactly what they are with my not-remaining-calibrated oscillator) and can sweep the EQ's as narrow as possible to cancel peaks resulting from modes and resonances.
I am always disinclined to provide "boost" for various reasons. Since I run my digital processors close to 0dB there is essentially no extra headroom. If there was a pure tone at the exact frequency, the boost could cause digital clipping. However, at 93 Hz, the subwoofer drive is already being attenuated by about 5dB because of the LR4 crossover at 100 Hz. So I think adding back in 2dB of boost is fairly (if not perfectly) safe. (Perfect safety may require as much as 6dB of headroom WHENEVER you apply a boost or do anything tricky, then you can be sure sample-overs or other peculiar waveforms aren't going to cause clipping either. Countering one EQ against another isn't perfect protection from clipping, I've discovered on numerous occasions. I'll have to see how this current boost goes in practice, but it seems like it ought to be mostly OK, with possible clipping once every month of Sundays a small price to pay for solid and true bass with no annoying notch near 100 Hz.)
Here are the latest left channel EQ's:
Sadly, all this work didn't seem to remove the dip or notch in the stereo-correlated pink noise response for both channels. So, all this work, and I didn't address that problem at all, which seems to be above 100 Hz and not below it anyway.
No comments:
Post a Comment