DiyMobileAudio.com Car Stereo Forum banner

1 - 20 of 98 Posts

·
Registered
Joined
·
291 Posts
Discussion Starter · #1 ·
I recently sprung for the Behringer ECM8000 and the 1/24 version of TruRTA :D It's a great tool, but raises some questions for me about how my system is behaving.

Here's the setup, the Behringer plugged into the MicMate stuck in the headrest.


Here's the plot of the right mid only, flat EQ, no TA. The background plot is after smoothing.


Here's the plot of the left mid only, flat EQ, no TA. The background plot is after smoothing.


Overlayed to see the difference in response, which really isn't that terrible I guess...


Here's the summed L+R plot, still flat EQ with no TA, background is smoothed.


The huge dip at 200Hz cannot be fixed with the EQ in the 880 because the I think the Q is too wide...it just pulls the whole curve up from 100-300Hz or so. I can't adjust the peaks on either side because those frequencies aren't present in the EQ. TA doesn't do anything to the plots, neither does changing the volume. I've read through tyroneshoes' excellent thread and tried some things from that thread to no avail. I have some similar issues ie, pink noise sounds very different from the left to right speakers, summed or not. It's interesting to me that both speakers played alone have similar curves, considering where they're mounted, yet sound so very different.

Is there a guideline like "plus or minus x-dB over x-octaves is audible"? I've done some reading on the equal loudness curve, but I still don't know if the peaks and dips are too narrow (or wide, or deep) to even be heard. I can say the tonality of pink noise of the left is night and day from the right. Frequency sweeps "change sides" multiple times. How does one correlate what the plots show with what's being heard? Sorry if this is a stupid question, but I just can't get my head wrapped around this at all.
 

·
Registered
Joined
·
524 Posts
The 200hz dip is probably to blame on location - something you cannot fix with the drivers in their current location. Since the flaw in apparent in both doors when played independently, I would look at door skin resonances or reflections off of the lower dash as a culprit.

Next are the baffles and ID OEM mids. They are sealed with wood glue and bolted to the doorskin with 1.25" bolts, nuts, and washers. For some reason I can't find the pics of these things mounted, but they're VERY solid.
Also try taking a left measurement by your left ear when you are sitting in the truck, and likewise with the right side. Measuring FR without taking diffraction caused by your body in to account is silly.
 

·
Registered
Joined
·
66 Posts
Just FYI, I've found it about 1000 x easier to make comparisons by clicking off the bar graph button at the top so you get a trace line instead. That way dips on the background traces are viewable.

10db is perceived as half or 2x as loud, so anything that major should be really noticeable. When I EQ for left and right, a 2db change is audible, a 4db difference is pretty big and a 6 or 8db difference left/right is major. I know your question was more focused to the over-all response curve, but that might serve as some reference to you.

To actually calibrate your program, you'll have to expose the microphone to a known db level and use the db calibration in the program, or it will more than likely be off. I haven't done this with mine and I just take what I get with a major grain of salt.

If you can't adjust a certain frequency accurately, it is probably because there is a cancellation and I think you will "hear" that if it's real. I put "hear" in quotes because you can't hear what isn't making noise, but that was also the point, the loss will be there and you won't hear it. I've always found that when you have a hole in your response, you'll go round and round with eq, tones, level setting, etc trying to get that hole back, without even knowing what is really wrong.

I am not totally convinced the holes that show up in the graph are always present however, but I can't currently prove that hunch right or wrong. I was setting subwoofer levels one night about a year ago with my old setup and when I upped the sub level, the graph went backwards (showed less bass) which was clearly not the case by my ears. I can't say I wasn't listening to other portions of the bass increasing, but I'm pretty sure it was a bogus result. I have had better results since then however with totally predictable and responsive measurements, but that time I just frustrate myself to no end.

Have you tried with your tweeters hooked up yet?
 

·
Registered
Joined
·
110 Posts
I recently sprung for the Behringer ECM8000 and the 1/24 version of TruRTA :D It's a great tool, but raises some questions for me about how my system is behaving.

The huge dip at 200Hz cannot be fixed with the EQ in the 880 because the I think the Q is too wide...it just pulls the whole curve up from 100-300Hz or so. I can't adjust the peaks on either side because those frequencies aren't present in the EQ. TA doesn't do anything to the plots, neither does changing the volume. I've read through tyroneshoes' excellent thread and tried some things from that thread to no avail. I have some similar issues ie, pink noise sounds very different from the left to right speakers, summed or not. It's interesting to me that both speakers played alone have similar curves, considering where they're mounted, yet sound so very different.

Is there a guideline like "plus or minus x-dB over x-octaves is audible"? I've done some reading on the equal loudness curve, but I still don't know if the peaks and dips are too narrow (or wide, or deep) to even be heard. I can say the tonality of pink noise of the left is night and day from the right. Frequency sweeps "change sides" multiple times. How does one correlate what the plots show with what's being heard? Sorry if this is a stupid question, but I just can't get my head wrapped around this at all.
This isn't a stupid question. In fact, I've had the same question for at least 5 years, and there's a lot of misinformation out there, and I learned a lot of things the hard way. I ended up reading about 10 key articles published in JAES and everything kind of fell into place. A lot of the following will sound like voodoo, and I'm sure the people who repeat the same misinformation will chime in soon, but here goes:

a) you absolutely must take several measurements in slightly different locations and average them together to get a valid measurement. This is called spatial averaging. Geddes in an older JAES article found that taking one random time-averaged (not spatial) RTA measurement gave you a measurement accuracy window of about +/- 8dB or so...which basically is almost useless. You'll find that just moving your mic a few inches changes your RTA curve completely. He found that you required a minimum of 6 measurements in 6 strategic locations to yield a precision of +/- 0.5dB.
b) A flat RTA with a pink noise signal is NOT desirable. It will sound overly bright. Ignore the IASCA metric. Everybody who has done RTA work will say "flat sounds terrible." The reason, very briefly, is that you're measuring the power response, which includes direct and reflected sound. A flat power response sounds too bright. What you want is a flat ANECHOIC response. You can't measure anechoic response without sophisticated software. TrueRTA doesn't do it. So the next best thing is to model the power response curve (what you measure) after what a flat anechoic response speaker would sound like. This has been studied. It is a downsloping line from about 100-400Hz to 10-20kHz at a rate of 1-1.7db/octave. That's your target. Some people have also demonstrated that a power reponse that starts at around 2kHz or so and downslopes about 1-2dB/oct towards 20kHz is also desirable, but the study that supported it was not as authoritative.
c) some peaks/valleys in your response are caused by reflections and other non-time-coherent processes, and are what is called "non-minimum-phase." What that means is that no matter how much you equalize, it will make little to no difference in the actual measured response. If this is the case, there's nothing you can do other than move your listening position or the speakers.
d) if you're equalizing with a 1/3-octave-band equalizer or some other fixed-frequency equalizer, you might be doing more harm then good. You can actually generate high Q high value peaks/valleys that measure well to a 1/3-octave RTA, but are only seen with higher resolutions (like 1/6 or 1/12 octave). The problem is that in order to eq out a peak or valley, you need to have the correct frequency and Q. A parametric equalizer would be a better tool.
e) to answer your question, low Q (wide) peaks/valleys are more audible than high Q (narrow) aberrations, given the same amplitude. this has been studied. i don't have specific numbers.
f) In Olive's study, a large double-blind study with 70 loudspeakers covering $100 to tens of thousands, showed that listeners definitely preferred speakers with a flat, smooth response, and extended low frequency cutoff. Smooth meant very little deviation from flat when examined at 1/20th octave resolution.
g) based on Olive's paper, and a few others, 1/3 octave resolution is NOT the ideal resolution to examine loudspeakers, because it doesn't reveal low-Q high amplitude variations that "average" out to a smooth 1/3 octave response.
h) Keep in mind that the default ECM8000 calibration file that comes with trueRTA is not very accurate. the ECM8000 has a lot of sample variation, so each one should ideally be calibrated.

Hope this is a good starting place. I'm sure this posted will be followed up by someone saying "yo, RTA is a good starting place, but i just trust my ears to tune," which will be followed by "yo, I use the RTA, but flat sounds terrible," which will be followed with "yo, RTA is crap." Good luck.

SG
 

·
Registered
Joined
·
66 Posts
This isn't a stupid question. In fact, I've had the same question for at least 5 years, and there's a lot of misinformation out there, and I learned a lot of things the hard way. I ended up reading about 10 key articles published in JAES and everything kind of fell into place. A lot of the following will sound like voodoo, and I'm sure the people who repeat the same misinformation will chime in soon, but here goes:

a) you absolutely must take several measurements in slightly different locations and average them together to get a valid measurement. This is called spatial averaging. Toole in an older JAES article found that taking one random time-averaged (not spatial) RTA measurement gave you a measurement accuracy window of about +/- 8dB or so...which basically is almost useless. You'll find that just moving your mic a few inches changes your RTA curve completely. He found that you required a minimum of 6 measurements in 6 strategic locations to yield a precision of +/- 0.5dB.
b) A flat RTA with a pink noise signal is NOT desirable. It will sound overly bright. Ignore the IASCA metric. Everybody who has done RTA work will say "flat sounds terrible." The reason, very briefly, is that you're measuring the power response, which includes direct and reflected sound. A flat power response sounds too bright. What you want is a flat ANECHOIC response. You can't measure anechoic response without sophisticated software. TrueRTA doesn't do it. So the next best thing is to model the power response curve (what you measure) after what a flat anechoic response speaker would sound like. This has been studied. It is a downsloping line from about 100-400Hz to 10-20kHz at a rate of 1db/octave. That's your target. Some people have also demonstrated that a power reponse that starts at around 2kHz or so and downslopes about 1-2dB/oct towards 20kHz is also desirable, but the study that supported it was not authoritative.
c) some peaks/valleys in your response are caused by reflections and other non-time-coherent processes, and are what is called "non-minimum-phase." What that means is that no matter how much you equalize, it will make little to no difference in the actual measured response. If this is the case, there's nothing you can do other than move your listening position or the speakers.
d) if you're equalizing with a 1/3-octave-band equalizer or some other fixed-frequency equalizer, you might be doing more harm then good. You can actually generate high Q high value peaks/valleys that measure well to a 1/3-octave RTA, but are only seen with higher resolutions (like 1/6 or 1/12 octave). The problem is that in order to eq out a peak or valley, you need to have the correct frequency and Q. A parametric equalizer would be a better tool.
e) to answer your question, low Q (wide) peaks/valleys are more audible than high Q (narrow) aberrations, given the same amplitude. this has been studied. i don't have specific numbers.
f) In Olive's study, a large double-blind study with 70 loudspeakers covering $100 to tens of thousands, showed that listeners definitely preferred speakers with a flat, smooth response, and extended low frequency cutoff. Smooth meant very little deviation from flat when examined at 1/20th octave resolution.
g) based on Olive's paper, and a few others, 1/3 octave resolution is NOT the ideal resolution to examine loudspeakers, because it doesn't reveal low-Q high amplitude variations that "average" out to a smooth 1/3 octave response.
h) Keep in mind that the default ECM8000 calibration file that comes with trueRTA is not very accurate. the ECM8000 has a lot of sample variation, so each one should ideally be calibrated.

Hope this is a good starting place. I'm sure this posted will be followed up by someone saying "yo, RTA is a good starting place, but i just trust my ears to tune," which will be followed by "yo, I use the RTA, but flat sounds terrible," which will be followed with "yo, RTA is crap." Good luck.

SG
Awesome post! I agree completely.

I think Navone said to hold the mic in the listening position and rotate it for about 30 seconds while gathering averages to equal out the response.

I think in theory, flat should be ideal, but in reality you are picking up reflections and other problems that your ears may be tuning out. Sometimes highly compressed music forms (like crappy recordings and MP3's) sound better when you do some lame eq settings like the classic "smiley face" or whatever, it's only because you're trying to doctor up something that sounds god-awful to begin with though.. IMO.

I use the RTA to find problems I'm having trouble hearing. It's a good visual for me to compare to my aural results, but it's far from something as bone head as setting resistance with a volt meter and variable resistor though (which is how some people try to use them). It would be nice if it were.
 

·
DIYMA Founder
Joined
·
4,757 Posts
Very good response from SmellyGas. Should probably be stickied somewhere. In short, RTA is a tool of limited usefulness for the purpose you're describing.

In addition I'd like to add the following:

1. Equal loudness curves - It doesn't mean what you think it does. You DO NOT TUNE so that every single frequency is "equally" loud to your ear!! That's not natural. Do not confuse your ear's sensitivity to the sensitivity of a speaker which is REPRODUCING a recording, and not evaluating (listening) to one.

2. Consider not only using spatial averaging, but also frequency adaptive windowing. In short, the measurement window length changes with frequency. So at higher frequencies the window would be relatively short and progressively longer as you go down. The result of this is a more anechoic response which would window out more reflections instead of averaging them in. Playing with the window length yields dramatically different responses, and it does take some play time to get good results. (for example you don't want to window out high amplitude early reflections which are known to affect tonality)

3. I'm not certain how effective equalisation is beyond 1/3rd octave. IIRC the ear doesn't differentiate between sounds that are less than 1/3rd octave in distance. Can't cite it, and could be wrong in how that applies to tuning however but in my experience I haven't found any significant qualitative differences using convolvers with very fine frequency correction and with 1/3rd octave correction.
 

·
Registered
Joined
·
4,916 Posts
b) A flat RTA with a pink noise signal is NOT desirable. It will sound overly bright. Ignore the IASCA metric. Everybody who has done RTA work will say "flat sounds terrible." The reason, very briefly, is that you're measuring the power response, which includes direct and reflected sound. A flat power response sounds too bright. What you want is a flat ANECHOIC response. You can't measure anechoic response without sophisticated software.

I was wondering why I was getting an overly bright system after tuning. :(
 

·
Registered
Joined
·
113 Posts
This is from Linkwitz's website about an equalizer for a headphone. I too have made binaural recordings and when I play them back on headphones that are equalized to sound flat to my ear they sound the most realistic. This is why I am confused when I am told that you shouldn't tune flat to your ear. I currently tune so that if I were to play a sine sweep it would "sound" constant in volume to my ear, no dips where it sounds like it fades out for a little and no peaks where it starts to get louder. Am I missing something here?

The equalizer is formed by inserting a passive network between the earphone amplifier output and the connector to the ER-4S. The headphone amplifier's open circuit output voltage is Vs and its output impedance is R1. The earphone impedance is 100 ohm and symbolized by R2 in the schematic below.



The insertion loss of the network will be 1.6 dB when a value of 20 ohm is assumed for R1. The inductor L must have sufficiently low dc resistance R3 compared to R1 to achieve the necessary notch depth. The inductance value determines the width of the notch. Larger values make it narrower. The notch frequency is then adjusted by C. The optimum values for L, C, and R3 are determined empirically by adjusting them for a constant amplitude sound while a sinewave generator's frequency is changed.
 

·
Registered
Joined
·
110 Posts
2. Consider not only using spatial averaging, but also frequency adaptive windowing. In short, the measurement window length changes with frequency. So at higher frequencies the window would be relatively short and progressively longer as you go down. The result of this is a more anechoic response which would window out more reflections instead of averaging them in. Playing with the window length yields dramatically different responses, and it does take some play time to get good results. (for example you don't want to window out high amplitude early reflections which are known to affect tonality)
This is a highly desirable technique when used in listening rooms, where you have early reflections and late reflections/reverberation. Only the first 30ms or so is "registered" as the primary sound, based on the Haas effect. I believe MLSSA uses this technique. However, in one JAES paper, they showed that in the CAR (not a room), 90% of the sound is actually received within the first 10ms, because the sound that reaches the listener (in the driver's seat) is direct + reflected only a short distance. (remember, sound traves approx 1ft in a millisecond). The volume of the reflected sound is often just as loud as that of the direct sound, in fact. I have not been able to find a method to use time gating to approximate an anechoic response in the car.

3. I'm not certain how effective equalisation is beyond 1/3rd octave. IIRC the ear doesn't differentiate between sounds that are less than 1/3rd octave in distance. Can't cite it, and could be wrong in how that applies to tuning however but in my experience I haven't found any significant qualitative differences using convolvers with very fine frequency correction and with 1/3rd octave correction.
It was widely accepted for a long time that 1/3 octave is the desired resolution to achieve smoothness and the best SQ. However, this was specifically studied in Olive's JAES paper and it was disproven. Yes, 1/3 octave smoothness correlates with listener preferences, however, it does NOT correlate nearly as well as smoothness at 1/20th octave resolution. They found that loudspeakers with better smoothness at the higher resolution (1/20th vs. 1/3rd) actually sounded better to a large group of listeners over a large, diverse sample of loudspeakers. The range that correlated best was around 100Hz or so up to 10-16kHz, depending on the metric.

Do you remember the Consumer Reports loudspeaker ratings that were based on 1/3 octave power-response (=RTA)?...the one that ranked Bose #1?? Well Olive took 13 loudspeakers tested by CR and did double-blind listening tests with a group of listeners. There was ZERO correlation between CR's ratings and listener preferences. Astonishing. Basically, a flat 1/3-octave power response did not mean people thought the speaker sounded good. How come?
Two reasons, and this was studied in an elegant design in one of Olive's JAES papers.
#1 - 1/3 octave response was not accurate enough. when the same loudspeakers were evaluated for smoothness at 1/20th octave, there was a different rank order! which means 1/3 octave resolution failed to reveal problems.
#2 - a flat POWER response, which is what the RTA measures, is again, NOT desirable. Speaker that are preferred in blinded tests have a downsloping power-response, not flat/horizontal. Speakers that are preferred have a flat ANECHOIC response, but that isn't what CR was testing!

My own personal take on it is that smoothness at 1/3 octave is still very important, and that a downsloping power/RTA response is desirable. However, if you have extra processing power, it may be desirable to correct peaks/valleys at higher than 1/3-octave resolution, PROVIDED that you have measured them accurately and they actually exist (i.e. good spatial averaging). Things get tricky in a car at higher frequencies because the spacing of nodes and antinodes are very close, sometimes within a few inches, so moving your head (or measurement mic) slightly may cause or eliminate peaks/dips. So it's very had to equalize them out reliably. Plus, if you're using spatial averaging, a peak to your left ear and a dip to your right ear may actuall measure FLAT. This may be why I think there's only so much you can do with EQ, even if you're using a convolver. (I'm using DRC+convolver with good results).
 

·
Registered
Joined
·
110 Posts
This is from Linkwitz's website about an equalizer for a headphone. I too have made binaural recordings and when I play them back on headphones that are equalized to sound flat to my ear they sound the most realistic. This is why I am confused when I am told that you shouldn't tune flat to your ear. I currently tune so that if I were to play a sine sweep it would "sound" constant in volume to my ear, no dips where it sounds like it fades out for a little and no peaks where it starts to get louder. Am I missing something here?
Honestly, if it sounds good to you, just do it. Tuning by ear for equal loudness, assuming you do this accurately, will yield a measured RTA response that follows an equal-loudness curve (an example would be the Fletcher-Munson...there are many of them). This is similar to what Whitledge did in his Magic Bus, although he didn't do this by ear. There was a long thread over on DIYaudio, and nobody agreed that an EQ curve that followed an equal-loudness curve was desirable. It also is at odds with the findings of Toole and Olive in published JAES articles. I skimmed the relevant article on Linkwitz's site, and I'm not convinced he advises people to eq to flat by ear, he just says that you can hear large peaks and dips easily and you can EQ them out. The electronic LCR circuit he describes does just that - it can equalize a single peak/dip.

The equalizer is formed by inserting a passive network between the earphone amplifier output and the connector to the ER-4S. The headphone amplifier's open circuit output voltage is Vs and its output impedance is R1. The earphone impedance is 100 ohm and symbolized by R2 in the schematic below.
I actually used to own a pair of ER-4S. However, I had heard some better and more comfortable 'phones for the same price or less, so I sent them back. At any rate, headphones are a completely different ball game. The actual eardrum-heard "frequency response" curve is way different from the measured microphone freq resp because of manipulation from your external ear (auricle), the ear canal, and vibration through your head. Your actual ear-drum heard response is also different if you have a free-field source vs. a diffuse-field source, which tends to radiate sound from all directions. An external (non-canal) headphone radiates sound onto your auricle and canal similar to how a distant diffuse-field would, and so most good headphones use a diffuse-field equalization. Using canal phones is a different story. Basically, you would want to the frequency response in the ear canal that is generated by a flat distant source, and reproduce it with the ER4's (canal phones). Because people have different ear canal anatomies, there may be differences among individuals, and this would only reflect an "average" equalization that might be universally appealing.

The short version of the above is that headphone equalization and listening is a completely different entity from listening to car loudspeakers. A lot of it isn't very generalizable back and forth. The only thing I think is worth looking at more ist hat some people argue that the vehicle listening environment replicates a "diffuse-field" environment, which implies that a free-field to diffuse-field equalization should be applied. I've tried this and it sounds okay (improvement) but my convolver made everything sound even better, and I'm too lazy to experiment more. There is an ISO curve that equalizes free-field to diffuse field. This may be the true target curve (applied on top of the downsloping curve that starts at 100-400Hz or so) in a car. I bet there are some car companies or car audio manufacturers that have discovered this in their labs, but aren't sharing it...it hasn't been published in any journal that I've found and it's my own theory, so take it for what it is. :)

SG
 

·
Registered
Joined
·
291 Posts
Discussion Starter · #12 ·
The 200hz dip is probably to blame on location - something you cannot fix with the drivers in their current location. Since the flaw in apparent in both doors when played independently, I would look at door skin resonances or reflections off of the lower dash as a culprit.

Also try taking a left measurement by your left ear when you are sitting in the truck, and likewise with the right side. Measuring FR without taking diffraction caused by your body in to account is silly.
I will definitely do this. At the minimum, now I can compare what the response looks like with me sitting in the cabin versus the graphs I already have. Do you think it's worth trying a nearfield measurement with the door open? Maybe that will help rule out some interior reflections?
 

·
Registered
Joined
·
291 Posts
Discussion Starter · #13 ·
Just FYI, I've found it about 1000 x easier to make comparisons by clicking off the bar graph button at the top so you get a trace line instead. That way dips on the background traces are viewable.

10db is perceived as half or 2x as loud, so anything that major should be really noticeable. When I EQ for left and right, a 2db change is audible, a 4db difference is pretty big and a 6 or 8db difference left/right is major. I know your question was more focused to the over-all response curve, but that might serve as some reference to you.

To actually calibrate your program, you'll have to expose the microphone to a known db level and use the db calibration in the program, or it will more than likely be off. I haven't done this with mine and I just take what I get with a major grain of salt.

If you can't adjust a certain frequency accurately, it is probably because there is a cancellation and I think you will "hear" that if it's real. I put "hear" in quotes because you can't hear what isn't making noise, but that was also the point, the loss will be there and you won't hear it. I've always found that when you have a hole in your response, you'll go round and round with eq, tones, level setting, etc trying to get that hole back, without even knowing what is really wrong.

I am not totally convinced the holes that show up in the graph are always present however, but I can't currently prove that hunch right or wrong. I was setting subwoofer levels one night about a year ago with my old setup and when I upped the sub level, the graph went backwards (showed less bass) which was clearly not the case by my ears. I can't say I wasn't listening to other portions of the bass increasing, but I'm pretty sure it was a bogus result. I have had better results since then however with totally predictable and responsive measurements, but that time I just frustrate myself to no end.

Have you tried with your tweeters hooked up yet?
This is good info, I'm going to try tackling the calibration within the next few days. I was under the impression that the USB setup didn't require it, but it can't hurt to try. I don't know of a way to cal the mic at this time. It would be great if that's what's causing all of this though :)

Regarding what's been bolded in your response above, I have to admit that I don't have a very trained ear. I've been listening to music my whole life and like it to sound as good as possible, but this is the first time I've used a tool to evaluate anything. I feel like I'm chasing my tail because of the learning curve associated with this, and I think my struggle is finding the comparison between what I like and what it's supposed to sound like.
 

·
Registered
Joined
·
291 Posts
Discussion Starter · #14 ·
This isn't a stupid question. In fact, I've had the same question for at least 5 years, and there's a lot of misinformation out there, and I learned a lot of things the hard way. I ended up reading about 10 key articles published in JAES and everything kind of fell into place. A lot of the following will sound like voodoo, and I'm sure the people who repeat the same misinformation will chime in soon, but here goes:

a) you absolutely must take several measurements in slightly different locations and average them together to get a valid measurement. This is called spatial averaging. Toole in an older JAES article found that taking one random time-averaged (not spatial) RTA measurement gave you a measurement accuracy window of about +/- 8dB or so...which basically is almost useless. You'll find that just moving your mic a few inches changes your RTA curve completely. He found that you required a minimum of 6 measurements in 6 strategic locations to yield a precision of +/- 0.5dB.
b) A flat RTA with a pink noise signal is NOT desirable. It will sound overly bright. Ignore the IASCA metric. Everybody who has done RTA work will say "flat sounds terrible." The reason, very briefly, is that you're measuring the power response, which includes direct and reflected sound. A flat power response sounds too bright. What you want is a flat ANECHOIC response. You can't measure anechoic response without sophisticated software. TrueRTA doesn't do it. So the next best thing is to model the power response curve (what you measure) after what a flat anechoic response speaker would sound like. This has been studied. It is a downsloping line from about 100-400Hz to 10-20kHz at a rate of 1db/octave. That's your target. Some people have also demonstrated that a power reponse that starts at around 2kHz or so and downslopes about 1-2dB/oct towards 20kHz is also desirable, but the study that supported it was not authoritative.
c) some peaks/valleys in your response are caused by reflections and other non-time-coherent processes, and are what is called "non-minimum-phase." What that means is that no matter how much you equalize, it will make little to no difference in the actual measured response. If this is the case, there's nothing you can do other than move your listening position or the speakers.
d) if you're equalizing with a 1/3-octave-band equalizer or some other fixed-frequency equalizer, you might be doing more harm then good. You can actually generate high Q high value peaks/valleys that measure well to a 1/3-octave RTA, but are only seen with higher resolutions (like 1/6 or 1/12 octave). The problem is that in order to eq out a peak or valley, you need to have the correct frequency and Q. A parametric equalizer would be a better tool.
e) to answer your question, low Q (wide) peaks/valleys are more audible than high Q (narrow) aberrations, given the same amplitude. this has been studied. i don't have specific numbers.
f) In Olive's study, a large double-blind study with 70 loudspeakers covering $100 to tens of thousands, showed that listeners definitely preferred speakers with a flat, smooth response, and extended low frequency cutoff. Smooth meant very little deviation from flat when examined at 1/20th octave resolution.
g) based on Olive's paper, and a few others, 1/3 octave resolution is NOT the ideal resolution to examine loudspeakers, because it doesn't reveal low-Q high amplitude variations that "average" out to a smooth 1/3 octave response.
h) Keep in mind that the default ECM8000 calibration file that comes with trueRTA is not very accurate. the ECM8000 has a lot of sample variation, so each one should ideally be calibrated.

Hope this is a good starting place. I'm sure this posted will be followed up by someone saying "yo, RTA is a good starting place, but i just trust my ears to tune," which will be followed by "yo, I use the RTA, but flat sounds terrible," which will be followed with "yo, RTA is crap." Good luck.

SG
I need to go out to the truck with a printout of this. Tons of info, I'll need to read it a couple more hundred times to let it sink in.

I think I'll start with your point c) in bold, it seems reasonable to determine if this is the case before pulling what little hair I have left out trying to tune, when I need to move the speakers instead.
 

·
Registered
Joined
·
110 Posts
This is good info, I'm going to try tackling the calibration within the next few days. I was under the impression that the USB setup didn't require it, but it can't hurt to try. I don't know of a way to cal the mic at this time. It would be great if that's what's causing all of this though :)
There are two things you can calibrate: the sound card (or USB interface for you) and the microphone. You generally calibrate the sound card by using a loopback device (lineout to line in). There is very little benefit though, since most sound cards are already very flat. The more important thing to calibrate is your microphone. The ECM8000 is very accurate, perhaps +/- 1 dB from 20Hz all the way up to about 1-2kHz or so, but it can vary by as much as 6dB or so once you go higher than that, and even worse when you measure off-axis. Calibrating the mic involves sending it in to a professional, and having them measure the mic, and send you a file with the actual response of the mic. This file can be plugged into TrueRTA. It generally costs upwards of $50 for this service.

Regarding what's been bolded in your response above, I have to admit that I don't have a very trained ear. I've been listening to music my whole life and like it to sound as good as possible, but this is the first time I've used a tool to evaluate anything. I feel like I'm chasing my tail because of the learning curve associated with this, and I think my struggle is finding the comparison between what I like and what it's supposed to sound like.
Here's an easy way to use the RTA to improve your system. It will be far from perfect, but it should at least help.
1) Use the ECM8000 mic calibration file that is included with TrueRTA. Load it.
2) Take at the very minimum 4 averaged readings at your headrest location, spaced out over the width of about a foot or so. Save each one to a memory location in TrueRTA. Then use the "average" function and average the 4 memory tracings.
3) Do this for each channel (left/right) individually, with the other one muted, provided you can make EQ adjustments to each channel separately.
3) Make EQ adjustments off the averaged responses
4) Adjust the LEVELSS of your individual speakers so that you have a roughly downsloping RTA response. If you have a passive crossover setup, then you can increase the tweeter attenuation if you have this option.
5)Since you have a 1/3-octave equalizer, you're better off looking at the 1/3-octave smoothed RTA plot. Pick an arbitrary target straight line from ~300Hz or so down to 10kHz at 1db/octave or so.
6) Decide how much deviation in dB there is between the actual response and the target line for each band on your EQ
7) Set your eq accordingly to "counteract" each deviation.
8) Do the average of 4 measurements again, and compare to the pre-EQ plot. Make sure you seen an improvement.

This is a good way to start, and it doesn't really require more than a basic understanding of RTA's.

SG
 

·
Registered
Joined
·
110 Posts
I think I'll start with your point c) in bold, it seems reasonable to determine if this is the case before pulling what little hair I have left out trying to tune, when I need to move the speakers instead.
Not all peaks/valleys are non-minimum-phase, but it's dangerous to try to correct the non-minimum-phase valleys because this can overdrive your speaker without actually improving frequency response. Also, in a car, it isn't really practical to move the speaker locations, especially since the new location might also have issues as well. Good luck.

SG
 

·
Registered
Joined
·
291 Posts
Discussion Starter · #17 ·
Very good response from SmellyGas. Should probably be stickied somewhere. In short, RTA is a tool of limited usefulness for the purpose you're describing.

In addition I'd like to add the following:

1. Equal loudness curves - It doesn't mean what you think it does. You DO NOT TUNE so that every single frequency is "equally" loud to your ear!! That's not natural. Do not confuse your ear's sensitivity to the sensitivity of a speaker which is REPRODUCING a recording, and not evaluating (listening) to one.

2. Consider not only using spatial averaging, but also frequency adaptive windowing. In short, the measurement window length changes with frequency. So at higher frequencies the window would be relatively short and progressively longer as you go down. The result of this is a more anechoic response which would window out more reflections instead of averaging them in. Playing with the window length yields dramatically different responses, and it does take some play time to get good results. (for example you don't want to window out high amplitude early reflections which are known to affect tonality)

3. I'm not certain how effective equalisation is beyond 1/3rd octave. IIRC the ear doesn't differentiate between sounds that are less than 1/3rd octave in distance. Can't cite it, and could be wrong in how that applies to tuning however but in my experience I haven't found any significant qualitative differences using convolvers with very fine frequency correction and with 1/3rd octave correction.
This clarifies a lot of things for me, especially the equal loudness curves. You're absolutely right that I was thinking in term of "to my ear".

I'm not too sure if I fully understand what frequency adaptive windowing is, but could I just manually change the frequency scale in TrueRTA and take the response data in say 500hz chunks, then use something else to graph the exported data? Or would that skew the curve too much?

It sounds like a big problem of mine is confusing speakers and ears when it comes to tuning. I can only adjust the speakers, but I don't have an RTA for my ears for a comparison :D
 

·
Registered
Joined
·
4,916 Posts
Calibrating the mic involves sending it in to a professional, and having them measure the mic, and send you a file with the actual response of the mic. This file can be plugged into TrueRTA. It generally costs upwards of $50 for this service.

Do you know anyone in Los Angeles that can calibrate my Mic?
 

·
Registered
Joined
·
291 Posts
Discussion Starter · #19 ·
There are two things you can calibrate: the sound card (or USB interface for you) and the microphone. You generally calibrate the sound card by using a loopback device (lineout to line in). There is very little benefit though, since most sound cards are already very flat. The more important thing to calibrate is your microphone. The ECM8000 is very accurate, perhaps +/- 1 dB from 20Hz all the way up to about 1-2kHz or so, but it can vary by as much as 6dB or so once you go higher than that, and even worse when you measure off-axis. Calibrating the mic involves sending it in to a professional, and having them measure the mic, and send you a file with the actual response of the mic. This file can be plugged into TrueRTA. It generally costs upwards of $50 for this service.

Here's an easy way to use the RTA to improve your system. It will be far from perfect, but it should at least help.
1) Use the ECM8000 mic calibration file that is included with TrueRTA. Load it.
2) Take at the very minimum 4 averaged readings at your headrest location, spaced out over the width of about a foot or so. Save each one to a memory location in TrueRTA. Then use the "average" function and average the 4 memory tracings.
3) Do this for each channel (left/right) individually, with the other one muted, provided you can make EQ adjustments to each channel separately.
3) Make EQ adjustments off the averaged responses
4) Adjust the LEVELSS of your individual speakers so that you have a roughly downsloping RTA response. If you have a passive crossover setup, then you can increase the tweeter attenuation if you have this option.
5)Since you have a 1/3-octave equalizer, you're better off looking at the 1/3-octave smoothed RTA plot. Pick an arbitrary target straight line from ~300Hz or so down to 10kHz at 1db/octave or so.
6) Decide how much deviation in dB there is between the actual response and the target line for each band on your EQ
7) Set your eq accordingly to "counteract" each deviation.
8) Do the average of 4 measurements again, and compare to the pre-EQ plot. Make sure you seen an improvement.

This is a good way to start, and it doesn't really require more than a basic understanding of RTA's.

SG
smellygas said:
Not all peaks/valleys are non-minimum-phase, but it's dangerous to try to correct the non-minimum-phase valleys because this can overdrive your speaker without actually improving frequency response. Also, in a car, it isn't really practical to move the speaker locations, especially since the new location might also have issues as well. Good luck.

SG
Thanks a lot for writing the step by step, I will do it sometime tonight. I didn't think of swapping a current set of problems for a potential set of completely different ones by changing speaker locations! I'm glad I haven't torn my door apart yet :D
 
1 - 20 of 98 Posts
Top