Car Audio Explained, Part 8: Mic Placement, Target Curves and EQ (Andy Wehmeyer's Lessons)

4 days ago

36 min read

This is Part 8 of 19 in the series Andy Wehmeyer on Car Audio, Explained for Beginners. Start here · Download the full guide as a PDF

Unofficial guide. Andy Wehmeyer's points are paraphrased from his public forum posts, blog, tuning guides, videos and podcasts, with a link to each source; the plain-English explanations were written for this guide. Not written, reviewed or endorsed by Andy Wehmeyer or Audiofrog.

3.4 Mic placement, spatial averaging and smoothing

What this part is about: When you tune a car system, a measurement microphone stands in for your ears, and software shows how loud each pitch (frequency) is at that spot. This part covers where to put that mic, why he averages a few nearby mic positions, and how much to "smooth" (blur) the graph so you see what you actually hear.

3.4 #1 · Measure from outside the car with the door shut; the target assumes an empty car.

Strap the mic to the headrest at the listening position, run the extension cable out and close the door. Insisting on exact listening conditions "doesn't make any sense" (a chair maker doesn't need you in the chair); it also spares you 20–60 minutes of pink noise and keeps the laptop from blocking speakers. [RARE] Sources: UMI-1 video @11:02 · Tuning PDF v1 (~2018)

In plain English: To tune, you play test sounds and a mic at your listening spot reports what arrives there. He says: strap the mic to the headrest where your head would be, run its extension cable out of the car, close the door and run the measurement from outside. Some people insist on measuring under exact listening conditions, with themselves in the seat. He says that "doesn't make any sense", comparing it to a chair maker who doesn't need you sitting in the chair to build it. The target curve (the shape you tune toward) assumes an empty car. Measuring from outside also spares you 20–60 minutes of pink noise (a hiss-like test signal), and keeps the laptop out of the car where it could block speakers. Many people believe the "right" way is to measure with yourself in the seat; he disagrees.

3.4 #2 · Don't sit in the seat waving the mic around your head.

Your head blocks short wavelengths, so the far-side mic reads less treble, the average shows missing highs, and you add needless HF boost; results also vary day to day, and no published target (his included) assumes a body in the car. If you must, wave from the back seat, or measure the difference your body makes and fold it into the target. [RARE] Sources: DIYMA 2021-09-02 · DIYMA 2020-10-08

In plain English: Some tuners sit in the seat and move the mic around their head while measuring. He says don't. High notes (treble) have short wavelengths, and a head is big enough to block them. So whenever the mic is on the far side of your head from a speaker, it reads less treble. The averaged result then shows highs that seem to be missing, and you add high-frequency (HF) boost you don't need. Results also vary from day to day, and no published target curve, including his, assumes a person in the car. If you must do it, he says to wave the mic from the back seat, or measure how much difference your body makes and build that difference into your target.

3.4 #3 · What moving the mic does, by frequency.

His REW test (5 positions inside a 7-in circle at the driver's head): below about 200–220 Hz the curves are nearly identical; 200 Hz–1 kHz varies a lot but in broad (low-Q) features; above 1 kHz a small move changes narrow (high-Q) features by 25 dB or more. Sources: Audiofrog blog Jan 2021 (single vs multi-mic) · DIYMA 2016

In plain English: Hz (hertz) measures pitch: low numbers are bass, high numbers are treble, and 1 kHz is 1,000 Hz. He tested what happens when the mic moves a little. Using REW (Room EQ Wizard, a free measurement program), he measured 5 spots inside a 7-inch (about 18 cm) circle at the driver's head. Below about 200–220 Hz the 5 curves were nearly identical. From 200 Hz to 1 kHz they varied a lot, but in broad, gentle bumps ("low-Q" means wide). Above 1 kHz, a small move changed narrow, sharp features ("high-Q") by 25 dB or more. A dB (decibel) is the unit of loudness difference, and 25 dB is a very large change. The lesson: the higher the pitch, the more an inch or two of mic movement changes the fine detail on the graph.

3.4 #4 · Narrow peaks and dips above about 1 kHz are a comb filter you don't hear; leave them.

They come from two or more sources (L and R, two drivers, a speaker and its reflections) arriving at different times; head movement averages them, so "We don't hear them unless we play a sine wave." Any speaker above ~100 Hz measured at one point "will measure like crap." [RARE] Sources: Audiofrog blog Jan 2021 (single vs multi-mic) · DIYMA 2021-08-06 · DIYMA 2021-08-26

In plain English: A comb filter is what you get when the same sound reaches one spot two or more times, slightly out of step. At some pitches the copies add up (peaks); at others they cancel (dips). On a graph it looks like the teeth of a comb. He says in a car the copies come from left and right speakers, two drivers playing the same notes, or a speaker and its own reflections off glass and panels. Because your head is always moving slightly, your hearing averages these out: "We don't hear them unless we play a sine wave" (a single pure tone). He adds that any speaker above about 100 Hz measured at one single point "will measure like crap." So in practice: leave those narrow peaks and dips above about 1 kHz alone. Many people believe every peak and dip on the graph should be EQ'd away; he disagrees.

3.4 #5 · One mic with 1/6-octave smoothing is good enough for production tuning.

A spatial average correlates better with hearing, but after MS-8 Harman found a smoothed single-mic ("spectral") average about 90% as accurate, so the MS-2 used one mic; his 2021 test matched the 5-position average closely (exceptions near 2.6 kHz, 1.2 kHz, 602, 445, 334 Hz), and one live mic lets you EQ in real time. [RARE] [Conflict: in 2013 he called one mic worse than a corrected multi-mic average] Sources: Audiofrog blog Jan 2021 (single vs multi-mic) · DIYMA 2017-06-26 · DIYMA 2013

In plain English: Smoothing blurs the graph so you see the trend instead of every wiggle; 1/6-octave smoothing averages each point over a band one-sixth of an octave wide (an octave is a doubling of pitch, such as 500 to 1,000 Hz). A spatial average combines measurements from several mic spots. He agrees a spatial average matches hearing better. But after the MS-8 (a Harman/JBL car processor that used a head-worn two-mic headset), Harman found a smoothed single-mic average (a "spectral" average, meaning averaged across frequency) was about 90% as accurate, so the later MS-2 used one mic. His own 2021 test matched the 5-position average closely, except near 2.6 kHz, 1.2 kHz, 602, 445 and 334 Hz. One live mic also lets you change EQ and watch the result in real time. The conflict: in 2013 he wrote that a single calibrated mic is worse than a spatial average of several known, corrected mics. Here he says one smoothed mic is good enough for production tuning. Both statements are his, from different years.

3.4 #6 · A spatial average is for high-frequency magnitude only; below about 1 kHz use a single point.

An average throws away phase, and phase matters far more below 1 kHz, where long wavelengths make every position look alike anyway; above that, the average represents hearing better. A single mic is fine below about 800 Hz. [RARE] Sources: DIYMA 2017-06-28 · DIYMA 2016 · DIYMA 2017-10-30

In plain English: Magnitude means how loud each frequency is. Phase means timing: where each wave is in its up-and-down cycle. Phase decides whether two speakers add together or cancel. An average of several mic positions keeps the loudness but throws away phase. He says phase matters far more below 1 kHz, and down there the waves are long (about a foot at 1 kHz, and many feet in the bass), so every position looks alike anyway and averaging gains nothing. Above 1 kHz, the average represents what you hear better. He also says a single mic is fine below about 800 Hz. So in practice: use a spatial average only for loudness above about 1 kHz, and use a single mic position below that and for anything involving phase. Many people treat the average as the best measurement for everything; he limits it to the highs.

3.4 #7 · Five or six positions are enough; hundreds add nothing.

Rotate the mic by hand around a ~7-in circle at the head, one measurement per spot; 15–20, let alone 1,000, are unnecessary. Measuring spots where nobody listens (windshield, A-pillar, dash) only dilutes the seats that matter. [RARE] Sources: DIYMA 2021-08-06 · DIYMA 2016 · DIYMA 2017-06-28

In plain English: He moves the mic by hand around a circle of about 7 inches (18 cm) at head position, taking one measurement at each spot. Five or six spots are enough; 15–20, let alone 1,000, are unnecessary. Some methods also measure spots where nobody's ears are, such as near the windshield, the A-pillar (the post on each side of the windshield) or the dash. He says that only dilutes the data for the seats that matter. Many people assume more mic positions always means a more accurate result; he says the gain stops after a handful.

3.4 #8 · Average magnitudes; never sum mics or average phase.

Wiring mics together or mixing a multi-input interface gives a phase-dependent sum (a delay-and-sum array is a beam-former), not an average; switch quickly between mics or average afterwards (SMAART, SysTune, REW). Averaging phase is meaningless: speakers at −180°/0° and at −90°/+90° both null. [RARE] Sources: DIYMA 2020-10-08 · DIYMA 2021-08-17 · DIYMA 2017-06-27

In plain English: Some people try to make a spatial average by wiring several mics together or mixing them in a multi-input audio interface. He says that gives a sum, not an average, and the sum depends on phase: where the mic signals are out of step, they partly cancel. Mics delayed and added together form a "beam-former" (a mic array that listens hardest in one direction), not an average of what your head would hear at each spot. The right way is to switch quickly between mics, or measure each one and average afterwards in software (SMAART, SysTune, REW). Averaging phase is also meaningless. His example: speakers at −180° and 0° average to −90°; speakers at −90° and +90° average to 0°. The averages differ, yet both pairs are exactly half a cycle (180°) apart, so both cancel (null). The averaged number hides that.

3.4 #9 · Two practical ways to tune to an average.

Either average six responses in REW's "All SPL" view and tune that (slow, not real time), or find the one mic position whose response best matches the average and tune live there. Either way, don't boost nulls; REW's Auto EQ can then fit the filters. [RARE] Sources: DIYMA 2021-08-09

In plain English: He gives two ways. First: measure six positions in REW, view them together in its "All SPL" view (SPL means sound pressure level, i.e. loudness), average them, and tune to that averaged curve. This is slow because you can't watch changes live. Second: find the single mic position whose response best matches the average, leave the mic there, and tune live. Either way, don't boost nulls (deep dips caused by cancellation). Then REW's Auto EQ (a feature that calculates EQ settings for you) can work out the filters.

3.4 #10 · Move the mic before you fix a narrow dip or ripple.

Measure while slowly moving the mic (his example: a 3 kHz dip), or take a few readings around a 7-in circle: if the feature persists, fixing it may help; if it vanishes, fixing it probably makes things worse. Precision there may be "a nothingburger with extra tomatoes." [RARE] Sources: DIYMA 2011 · DIYMA 2021-03-26 · DIYMA 2021-05-07

In plain English: Before you EQ a narrow dip or a wavy ripple, check whether it's real for your ears or just an accident of that exact mic spot. Measure while slowly moving the mic (his example was a dip at 3 kHz), or take a few readings around a 7-inch (18 cm) circle. If the feature stays, fixing it may help. If it vanishes, fixing it probably makes things worse. He jokes that being precise about it may be "a nothingburger with extra tomatoes", meaning a lot of fuss over nothing. Many people EQ whatever one measurement shows; he says test it first.

3.4 #11 · Smoothing: coarser up high, finer down low; zoom in only to find a cause.

His numbers vary by year: 1/3 octave above 3 kHz, 1/6 from 800 Hz–3 kHz, 1/12 below (2017); 1/3 above 1 kHz, 1/6 below (2020); 1/3–1/6 high and 1/6–1/12 low (2022, see #12). Read an L+R sum at 1/3 first. Example: 1/3 octave showed a 6 dB rise at 300–600 Hz that 1/12 revealed as a 20 dB peak at 500 Hz plus a dip. [Era: numbers shifted, rule didn't] Sources: DIYMA 2017-06-26 · DIYMA 2020-07-08 · DIYMA 2012–13

In plain English: Smoothing blurs the graph. Coarse smoothing (1/3 octave) blurs a lot; fine smoothing (1/12 octave) shows more detail. His rule: coarse for the highs, fine for the lows, and zoom in only when you are hunting for the cause of a problem. His exact numbers changed by year. 2017: 1/3 octave above 3 kHz, 1/6 from 800 Hz to 3 kHz, 1/12 below. 2020: 1/3 above 1 kHz, 1/6 below. 2022: 1/3 to 1/6 for the highs and 1/6 to 1/12 for the lows. He also says to first read left and right playing together (L+R) at 1/3 octave. His example: at 1/3 octave a 6 dB rise appeared across 300–600 Hz; at 1/12 it turned out to be a 20 dB peak at 500 Hz plus a dip. The numbers moved over the years; the rule didn't.

3.4 #12 · Never tune at 1/24 or 1/48 octave, and don't match L and R that finely.

Very high resolution is for studying a speaker apart from the room (anechoic or gated), not for EQ; in a car you get lost chasing an inaudible comb. [RARE] Sources: DIYMA 2022-03-16 · DIYMA 2021-08-26

In plain English: 1/24 and 1/48 octave are extremely fine settings that show every tiny wiggle. He says that much detail is for studying a speaker on its own, apart from the room: in an anechoic chamber (a room whose walls absorb echoes) or with a "gated" measurement (one that stops recording before reflections arrive). It is not for EQ. In a car you get lost chasing comb filtering you can't hear. He also says not to match left and right that finely. Many people assume more detail always means a better tune; he disagrees.

3.4 #13 · The in-car RTA sums direct and reflected sound, so a "flat" reading can sound wrong.

Off-axis reflections can carry a dip near a crossover; EQing the combined reading flat makes the direct sound bright and fatiguing. Below 1 kHz an out-of-phase reflection can pin the image to a speaker even when the RTA matches (why far-off pillar pods cause headaches). [RARE] Sources: Tuning PDF v2 (~2023) · DIYMA 2022-02-23

In plain English: An RTA (real-time analyzer) shows how loud each frequency is at the mic. In a car, the mic hears the direct sound from a speaker plus its reflections off glass, dash and seats, all mixed together, and it can't tell them apart. He names two problems. First, reflections of sound that left the speaker off to the side (off-axis) can carry a dip near the crossover (the point where one driver hands the sound to another). If you EQ the combined reading flat, you also boost the direct sound, which makes it bright and fatiguing. Second, below 1 kHz, a reflection that arrives out of phase with a speaker's direct sound can "pin" the image (where the sound seems to come from) to that speaker, even when the RTA shows left and right matching. He says this is why pillar pods (speaker housings on the windshield posts) placed far from the nearby surfaces cause tuning headaches: the farther a speaker is from a surface, the lower in pitch that combing begins.

3.4 #14 · Measure at normal listening level, and only where you'll listen.

Measuring too loud degrades results (the MS-8 calibrated near conversation level); a mic left in the console doesn't verify a driver-seat calibration, since they are two different curves. Sources: Tuning PDF v1 (~2018) · DIYMA 2011

In plain English: Measuring too loud makes the results worse. He notes the MS-8 ran its calibration at about conversation level. Also, measure only where people will actually listen. A mic left sitting in the center console can't check a calibration done for the driver's seat, because the sound at those two spots is different: they are two different curves.

3.5 Target curves

What this part is about: A target curve is the shape you tune toward: how loud the bass, midrange and treble should be compared with each other at your seat. This part covers his target, where it came from, why it isn't flat, and how to use it sensibly.

3.5 #1 · The Audiofrog target: about +10 dB of bass below ~60 Hz, a transition to flat midrange between ~60 and 160 Hz, and a gentle treble tilt to −6 dB at 20 kHz.

Use −3 dB at 20 kHz if you want more highs. Cars need roughly 6–12 dB more bass than mids and highs; he builds the lift as a 12 dB/octave low shelf at about 60 Hz, adjusted to taste. Sources: Tuning PDF v1 (~2018) · DIYMA 2021-03-26 · DIYMA 2017-07-05

In plain English: Audiofrog is his company. Its target has bass below about 60 Hz roughly 10 dB louder than the midrange. Between about 60 and 160 Hz the curve eases down to flat. The midrange stays flat, and then the treble tilts gently down until it is 6 dB lower at 20 kHz (the top of human hearing). If you want more highs, end the tilt at −3 dB instead. He says cars need roughly 6–12 dB more bass than mids and highs. He builds that bass lift with a "low shelf" (an EQ setting that raises everything below a chosen frequency by a set amount), placed at about 60 Hz with a 12 dB-per-octave slope and adjusted to taste.

3.5 #2 · The so-called "Harman curve" began as his own installer curve.

He used it for years in retail, brought it to Harman, where it became the MS-8 curve because no Harman curve existed yet; Harman research later "kind of confirmed" it with small preference tweaks. All good-sounding car targets share this shape, but it is "not the right shape for larger rooms." He disputes that it came from David Clark. [RARE] Sources: DIYMA 2020-10-14 · Audiofrog blog Mar 2021 (target curves)

In plain English: Harman is the large audio company (owner of JBL) where he worked. People call this target shape the "Harman curve." He says it began as his own curve from his years as a retail installer. He brought it to Harman, where it became the curve in the MS-8 processor because Harman had no curve of its own yet. Harman's research later "kind of confirmed" it, with small tweaks based on listener preferences. He says all good-sounding car targets share this shape, but it is "not the right shape for larger rooms" such as a home living room. He disputes the claim that the curve came from David Clark (of DLC Design), allowing only that Clark may have contributed to an earlier target in Harman's factory-audio division.

3.5 #3 · Target-curve wars are pointless: the popular curves differ by about 2 dB.

Normalized at 1 kHz, the Audiofrog and "Half Whitledge" curves differ by about 2 dB at 20 Hz, 1.3 dB around 200 Hz (an octave wide) and 2 dB at 20 kHz, the same three tweaks his research found. People claiming one is "MUCH better" have failed to hit either. [RARE] Sources: Audiofrog blog Mar 2021 (target curves)

In plain English: Enthusiasts argue online about which target curve is best. He compared his Audiofrog curve with another popular one, the "Half Whitledge" curve, lined up so both read the same at 1 kHz. They differ by about 2 dB at 20 Hz, 1.3 dB around 200 Hz (over a band about an octave wide) and 2 dB at 20 kHz. Those are small differences, and he notes they are the same three tweaks his research had found. His conclusion: people who claim one curve is "MUCH better" have failed to hit either. Many people believe picking the right target is the key to good sound; he says the popular choices are nearly the same.

3.5 #4 · The numbers drifted over the years; the shape never did.

2008: +6 to +9 dB below 60 Hz, flat 160 Hz to ~3 kHz, then about −3 dB/octave. 2009: highs falling gently from ~1 kHz. MS-8 (2011): +9 dB bass, flat 160 Hz–1 kHz, −6 dB at 20 kHz. 2018 PDF: +10 dB. He says the curve "has made my customers happy for 25 years" and pleases ~90% of listeners. [Changed view: numbers only] Sources: DIYMA 2008 · DIYMA 2011 · DIYMA 2010

In plain English: Over the years he quoted slightly different numbers for his target, but the overall shape stayed the same. 2008: bass 6 to 9 dB up below 60 Hz, flat from 160 Hz to about 3 kHz, then falling about 3 dB per octave (each doubling of frequency). 2009: highs falling gently from about 1 kHz. MS-8 (2011): 9 dB of bass boost, flat from 160 Hz to 1 kHz, and 6 dB down at 20 kHz. 2018 tuning PDF: 10 dB of bass. He says the curve "has made my customers happy for 25 years" and pleases about 90% of listeners. The [Changed view] tag applies to the numbers only, not to the idea.

3.5 #5 · Why the treble tilts down instead of staying flat: an in-car measurement is essentially the speaker's power response.

The mic sums direct and reflected sound, so the target is what a room-friendly speaker's power response looks like, plus bass lift (he cites Toole and Harman's speaker-and-room work). EQing that sum flat makes the direct sound too bright; with LR4 between mid and tweeter the top usually needs only a slight tilt. [RARE] Sources: DIYMA 2009 · Tuning PDF v2 (~2023) · Audiofrog blog Jan 2021 (crossovers)

In plain English: A speaker's "power response" is its total sound output in every direction, not just straight ahead. In a car, the mic picks up direct sound plus many reflections at once, so he says what it measures is essentially the power response. His target is what the power response of a speaker designed to sound good in rooms looks like, plus extra bass. He cites Floyd Toole's and Harman's research on speakers and rooms. If you EQ that combined measurement flat, the direct sound ends up too bright. He adds that with an LR4 crossover between the midrange and the tweeter (a common, fairly steep crossover type), the top end usually needs only a slight downward tilt to meet the target. Many people believe a perfect system should measure flat; he says not in a car.

3.5 #6 · Start the bass rise around 160–200 Hz, not 350 Hz, and roll the top off gradually from about 2 kHz.

A rise starting at 350 Hz sounds thin; about 5 dB too much above 2 kHz sounds bright but not piercing. Harman's early in-car tests (Sean Olive's group, 2009) preferred gradual attenuation above about 2 kHz over a longer slope starting lower. [RARE] Sources: DIYMA 2013 · DIYMA 2009

In plain English: Two details of the shape. Bass: start the rise around 160–200 Hz. He says a rise starting as high as 350 Hz sounds thin (the source doesn't spell out why). Treble: roll off gradually from about 2 kHz. About 5 dB too much treble above 2 kHz sounds bright, but not piercing. He cites Harman's early in-car listening tests (Sean Olive's research group, 2009): listeners preferred a gradual drop above about 2 kHz over a longer slope that started lower.

3.5 #7 · The target's low end is also the design spec for his subwoofers.

Audiofrog sub manuals show a modeled in-car response averaged over about 30 cars, and in the recommended box it closely matches the target's low end (unless the car is a Smart ForTwo or a school bus). He starts from the in-car target and designs the woofer for a sensible box, which is why textbook Qtc 0.707 rules don't apply. [RARE] Sources: DIYMA 2021-09-04 · DIYMA 2021-08-26

In plain English: How much bass a subwoofer makes depends on its box and on the car it's in. Audiofrog's sub manuals show a modeled in-car response averaged over about 30 cars. In the recommended box, that response closely matches the low end of his target (unless the car is a Smart ForTwo or a school bus, i.e. extremely small or extremely large). His approach is to start from the in-car target and design the woofer to suit a sensible box. That, he says, is why textbook "Qtc 0.707" rules don't apply. Qtc is a number describing how a woofer behaves in a sealed box; 0.707 is the textbook value for the flattest bass measured outside a car. Many people believe a sub should be designed and boxed to that textbook value; he designs for the response inside the car instead.

3.5 #8 · Preferences change the target, never the process.

The target is just where the objective part ends; the best one is the one that matches your (or the customer's) taste most often. A wider bass-to-mid transition (to 300 Hz or beyond) adds impact at some cost to midrange detail; a bigger bass boost makes bass-in-front harder to achieve. Sources: Tuning PDF v1 (~2018) · DIYMA 2020-10-03

In plain English: He says the objective, measurable part of tuning ends at the target; choosing the target is a matter of taste. The best target is the one that matches your taste (or your customer's) most often. A different preference changes the target, never the process. Two trade-offs he gives: stretching the bass-to-midrange transition up to 300 Hz or beyond adds impact, at some cost to midrange detail. A bigger bass boost makes "bass in front" harder to achieve, meaning bass that seems to come from the front of the car, where the music's stage is, rather than from the sub behind you.

3.5 #9 · Head-worn and binaural mics need their own target with more treble roll-off.

Flat on a head mic is "way too bright"; binaural and spatial-array data differ mainly in HF roll-off above about 2.5 kHz, from head masking. Adding EQ peaks and dips for "HRTF" to raise the stage misunderstands HRTF, which is about moving heads and moving sounds. [RARE] Sources: DIYMA 2013 · DIYMA 2017-06-26 · DIYMA 2020-10-14

In plain English: A binaural or head-worn mic sits at your ears, so your head shapes what it picks up. He says these mics need their own target with more treble roll-off: tuning flat with a head mic is "way too bright." Binaural data and spatial-array data (from several free-standing mic positions around the head) differ mainly in the high-frequency roll-off above about 2.5 kHz, because the head blocks (masks) short wavelengths. HRTF (head-related transfer function) describes how your head and ears change sound coming from different directions. Some people add EQ peaks and dips to imitate it, hoping to raise the stage (make the music seem to come from higher up). He says that misunderstands HRTF, which is about moving heads and moving sounds.

3.5 #10 · A shared target curve means nothing without the method used to measure it.

A curve that won "best sound" at the 2013 Eurofinals has "way too much midbass" if measured with one mic and pink noise, and results also depend on tweeter directivity; always state mic, averaging and signal. [RARE] Sources: DIYMA 2016 · DIYMA 2012

In plain English: A target curve posted online means little unless you know how it was measured. His example: a curve that won "best sound" at the 2013 Eurofinals (a car-audio sound competition) has "way too much midbass" if it was measured with one mic and pink noise. The result also depends on how directional the tweeters are. So, he says, always state the mic, the averaging and the test signal you used. Many people copy a winner's curve expecting the same result; he says that doesn't work without the method.

3.5 #11 · Tweeterless systems: drop the treble tilt, and let the aim decide the HF EQ.

Full-range drivers aimed at you keep highs in the direct sound: tight, spaceless images, and the boost you add makes it bright. Off axis they need more boost and sound dull. Aimed at the windshield (as OEMs do), the glass becomes "a tweeter without having to pay for a tweeter"; still, every car sounds better with tweeters. [RARE] Sources: Tuning PDF v2 (~2023) · DIYMA 2021-12-02

In plain English: Some systems have no tweeters and use small full-range drivers for mids and highs. For these he says: drop the downward treble tilt from the target, and let the speakers' aim decide the high-frequency EQ. He explains that such drivers send their highest notes in a narrow beam. Aimed at you, the highs arrive mostly as direct sound: tight images with no sense of space, and the treble boost you add makes it bright. Aimed off to the side, they need more boost and sound dull. Aimed at the windshield, as carmakers (OEMs) do, the glass becomes "a tweeter without having to pay for a tweeter." Still, he says every car sounds better with tweeters.

3.5 #12 · REW house-curve settings that approximate his target:

LF Rise Start 160 Hz, LF Rise End 60 Hz, LF Rise Slope about 12, HF Fall Slope about 1.6 (units not stated). Testgear also supplies target files for each analyzer. Sources: DIYMA 2014 · Testgear site

In plain English: REW (Room EQ Wizard) lets you describe a target, which it calls a "house curve," with a few settings. He gave these to approximate his target: LF Rise Start 160 Hz, LF Rise End 60 Hz, LF Rise Slope about 12, HF Fall Slope about 1.6. LF means low frequency (bass) and HF means high frequency (treble). So the bass starts rising at 160 Hz and stops rising at 60 Hz. He didn't state the units for the two slopes. His Testgear site also supplies ready-made target files for each analyzer.

3.6 EQ: what to fix, what to leave, and how

What this part is about: EQ (equalization) means turning particular frequencies up or down to correct the sound. This part covers which problems EQ can fix, which it can't, and the habits he uses to apply it well.

3.6 #1 · Separate left and right EQ is mandatory.

You tune each side to the target from its own position; DSPs that link L/R EQ can't run his process. At Harman, a German automaker that forced identical L/R filters (citing "phase problems") got bad-sounding cars. [RARE] Sources: Tuning PDF v1 (~2018) · DIYMA 2015

In plain English: In his process, the left speakers and the right speakers are each tuned to the target separately, each measured from its own position. So a DSP (digital signal processor: the box that handles EQ, crossovers and delays) that links left and right EQ, forcing the same settings on both, can't run his process. He recalls that at Harman, a German automaker forced identical left and right filters, citing "phase problems," and got bad-sounding cars. Many people believe left and right should get identical EQ; he says each side needs its own.

3.6 #2 · Matching left to right matters more than hitting the target.

Match within a couple of dB; if one side has a dip you can't fix, put the same dip on the other side to protect the center. After matching, deviate only where needed to stop the image wandering ("The windshield is the devil for everything but a center speaker"). [RARE] Sources: Tuning PDF v2 (~2023) · DIYMA 2014

In plain English: He says left and right matching each other, within a couple of dB, matters more than both hitting the target. If one side has a dip you can't fix, put the same dip on the other side, to protect the center (keep sounds meant to be in the middle, like a singer's voice, centered). Once they match, depart from the match only where needed to stop the image wandering, meaning sounds drifting sideways as the pitch changes, which the car's shape and reflections can cause. "The windshield is the devil for everything but a center speaker," he says. Many people treat hitting the target as the main goal; he puts left-right matching first.

3.6 #3 · Treat peaks as peaks: put the target at the dips and cut down to it.

Don't read the peak tops as the speaker's level or the dips as "it won't play that low"; slide the target onto the dips like "a router template" and cut each peak by its excess (dip at 57.5 dB → target there; a 64 dB peak at 1.5 kHz → cut ~7 dB), then restore level with amp input sensitivity. You'll find more bandwidth and standard crossovers work. [RARE] [Conflict: PDFs set midrange at 0 dB instead] Sources: UMI-1 video @16:00 · DIYMA 2021-08-20 · DIYMA 2021-08-06

In plain English: A measured response is full of peaks and dips. He warns against two misreadings: treating the tops of the peaks as the speaker's true level, and reading the dips as "it won't play that low." Instead, slide the target curve down until it sits on the dips, like "a router template" (a guide a woodworker traces along), then cut each peak by however much it sticks up above the target. His example: a dip reads 57.5 dB, so the target goes there; a peak at 1.5 kHz reads 64 dB, so cut about 7 dB there. Then restore the overall loudness with the amplifier's input sensitivity (gain) setting. He says you'll find the speaker plays a wider range (more bandwidth) than you thought, and standard crossover settings work. The conflict: his tuning PDFs instead line up the target so the midrange sits at 0 dB, rather than sliding it down to the dips.

3.6 #4 · An amp with built-in DSP takes that trick away.

With no analog gain stage after the DAC, once the DSP outputs are at maximum that's all the level you get, so you must boost dips as well as cut peaks. [RARE] Sources: DIYMA 2022-03-03

In plain English: Some amplifiers have a DSP built in. In those, there is no analog gain stage after the DAC (the chip that turns the digital signal back into an audio signal), so there is no amplifier gain to turn up afterwards. Once the DSP outputs are at maximum, that's all the level you get. So with these amps you must boost the dips as well as cut the peaks, instead of only cutting as in the previous point.

3.6 #5 · Use channel levels for broad regions first; EQ only what's left.

Raise a channel a little for more 80 Hz, then remove remaining peaks. Boost on a band plus lower channel gain is the same as a cut. Sources: Tuning PDF v1 (~2018) · DIYMA 2022-03-16 · DIYMA 2015

In plain English: Each speaker channel has its own overall level. He says to use those channel levels to fix broad regions first, then use EQ only for what's left. His example: raise a channel a little to get more 80 Hz, then remove the peaks that remain. He also notes that boosting one EQ band while lowering that channel's overall gain is the same as a cut, because the response ends up with the same shape.

3.6 #6 · Compute a parametric filter's Q from the peak's −3 dB bandwidth.

Q = center frequency ÷ the width between the points 3 dB below the peak, read at 1/12 octave. Worked example: peak at 474 Hz, −3 dB at 398 and 571 Hz → Q 2.73, refined by ear to −12 dB at Q 1.9. (The energy definition, 2πf × stored ÷ lost, gives similar, not identical, values.) Sources: Tuning PDF v2 (~2023) · DIYMA 2017-09-02 · DIYMA 2021-12-22

In plain English: A parametric EQ filter has three settings: its center frequency, its gain (how much it boosts or cuts) and its Q (how narrow it is; higher Q means narrower). He shows how to work out Q from a measurement read at 1/12-octave smoothing. Find the peak's center frequency, then the two points on either side where the response is 3 dB below the top. Q = center frequency ÷ the width between those two points. His example: a peak at 474 Hz with −3 dB points at 398 and 571 Hz, so Q = 474 ÷ (571 − 398), which he gives as 2.73. He then refined it by ear to a 12 dB cut at Q 1.9. He notes the energy definition of Q (2π times frequency, times energy stored, divided by energy lost) gives similar, but not identical, values.

3.6 #7 · Keep EQ broad, especially above 1 kHz.

Don't chase narrow, position-dependent features; avoid parametric Q above about 6. Treble cuts are fine when wide (his examples: −4.5 dB at 10 kHz and −4 dB at 15 kHz, both Q 1.4). [Conflict: the PDFs, even v2, still show a Q 10 cut at 1.6 kHz] Sources: Audiofrog blog Jan 2021 (single vs multi-mic) · Tuning PDF v2 (~2023)

In plain English: Higher Q means a narrower filter. He says to keep EQ broad, especially above 1 kHz, where narrow features change with every small move of the head (see 3.4). Don't chase those narrow, position-dependent features, and avoid parametric Q above about 6. Cutting treble is fine when the cut is wide: his examples are −4.5 dB at 10 kHz and −4 dB at 15 kHz, both at Q 1.4. The conflict: his tuning PDFs, even the 2023 version, still show a narrow Q 10 cut at 1.6 kHz.

3.6 #8 · Cut rather than boost; near a crossover boost the low-passed driver, not the high-passed one.

Boosting the high-pass driver near its crossover overstresses it, and boosting in or near a crossover hole is often a bad idea. Bass boosts cause most clipping: after the crossover it shows in one channel, before it (or at the A/D) in all. Tweeter crackle that appears when you turn the sub up is digital clipping from too much boost. [RARE] Sources: Tuning PDF v1 (~2018) · DIYMA 2021-07-08 · DIYMA 2021-11-14

In plain English: He prefers cutting peaks to boosting dips. A crossover splits the music between drivers: a high-pass filter sends only higher notes to a smaller driver such as a tweeter, and a low-pass filter sends only lower notes to a bigger one. If you must boost near the crossover point, boost the low-passed driver, not the high-passed one; boosting the high-passed driver near its crossover overstresses it. Boosting in or near a crossover hole (a dip where the two drivers don't add up well) is often a bad idea. Bass boosts cause most clipping (harsh distortion when a signal is pushed past the most the electronics can handle). Boost applied after the crossover clips in one channel; boost before it, or at the A/D (the input's analog-to-digital converter), clips in all channels. Tweeter crackle that appears when you turn the sub up is digital clipping from too much boost.

3.6 #9 · Never significantly boost a null; the nicer graph is "PROOF that reflected sound matters."

A 12 dB boost into a null clips the amp and overdrives the speaker; the mic improves only because extra reflected energy fills the dip, and it sounds "nasty." Leave a −6 dB cancellation dip alone. [RARE] Sources: DIYMA 2012 · DIYMA 2011 · DIYMA 2021-08-09

In plain English: A null is a deep dip where sound waves cancel, for example a speaker's direct sound and a reflection arriving out of step. He says a 12 dB boost into a null clips the amp and overdrives the speaker. If the graph then looks better, it's only because extra reflected energy fills the dip at the mic: that nicer graph is "PROOF that reflected sound matters," and it sounds "nasty." Leave a −6 dB cancellation dip alone. Many people boost any dip the analyzer shows; he says never boost these significantly.

3.6 #10 · "Does it respond to EQ?" sorts the three kinds of dip.

A box or driver peak is minimum phase, so EQ fixes magnitude and phase together (car systems are often close enough). A dip from a driver's own reflection can't be EQ'd. A dip that appears only with both sides playing is two speakers interacting: a phase or delay fix, not EQ. [RARE] Sources: DIYMA 2021-11-02 · DIYMA 2016

In plain English: He sorts dips with one question: does it respond to EQ? First kind: a peak from the speaker box or the driver itself. That is "minimum phase," a technical term meaning its loudness and timing errors are tied together, so EQ fixes both at once, and he says car systems often behave close enough to this. Second kind: a dip caused by a driver's own reflection, which EQ can't fix. Third kind: a dip that appears only when both sides play together. That is the two speakers interacting, and it needs a phase or delay fix, not EQ.

3.6 #11 · Find the surface behind a reflection null with path arithmetic, then accept it.

The null sits where the reflected path is half a wavelength longer: about 11 in extra for 600 Hz; a surface 1 ft away gives ~280 Hz, 6 in ~550 Hz. One reflection makes a deep narrow dip, many make wide shallow ones (more audible). A 300–500 Hz dip is usually the door glass, the price of width; "It's always something." [RARE] Sources: DIYMA 2021-11-06 · Audiofrog blog Jun 2021 (dispersion and reflection) · Audiofrog blog Dec 2020 (Time Alignment Part 2)

In plain English: A reflection travels a longer path than the direct sound. When the extra distance equals half a wavelength, the reflection arrives exactly out of step and cancels that frequency, making a null. He uses that to work backward to the surface causing it: about 11 inches (28 cm) of extra path for 600 Hz; a surface 1 foot (30 cm) away gives a null near 280 Hz, and one 6 inches (15 cm) away near 550 Hz. One reflection makes a deep, narrow dip; many reflections make wide, shallow dips, which are more audible. A 300–500 Hz dip is usually the door glass. He calls it the price of width: elsewhere he explains that mids low in the doors give a wide stage, and the side-glass dip is the trade-off. So you accept it. "It's always something."

3.6 #12 · Aiming isn't an EQ substitute, but in a 2-way the aim decides whether to EQ the crossover dip.

A 6-in is near-omni below ~500 Hz, a 2-in below ~2.7 kHz, a 1-in tweeter's aim matters only above ~10 kHz. Aimed on axis, the 2-way's hole is in the reflections: ignore it. Aimed off axis, it's in the direct sound: you may EQ it, at the passenger's expense. A 3-way removes the trade. [RARE] Sources: Audiofrog blog Jun 2021 (dispersion and reflection) · DIYMA 2021-11-30

In plain English: Aiming a speaker doesn't replace EQ. But in a 2-way system (a woofer plus a tweeter on each side), the aim decides whether to EQ the dip at the crossover. Speakers spread sound almost evenly in all directions (near-omni) at low frequencies and beam at high ones. His figures: a 6-inch (15 cm) driver is near-omni below about 500 Hz, a 2-inch (5 cm) below about 2.7 kHz, and a 1-inch (25 mm) tweeter's aim matters only above about 10 kHz. Aimed at you (on axis), the 2-way's crossover hole is in the reflections, so ignore it. Aimed away (off axis), the hole is in the direct sound, so you may EQ it, at the passenger's expense. A 3-way (adding a midrange between the two) removes that trade-off.

3.6 #13 · Dash speakers have too much bass, not too little treble: EQ by cutting.

The dash adds about 6 dB and the windshield about 6 dB more at low frequencies (depending on rake); cutting it lowers power, excursion and distortion. Don't boost the natural HF roll-off. A vent-mounted mid's peaks just need turning down, then the tweeter can cross at 2.5 kHz LR4. [RARE] Sources: Audiofrog blog Jun 2021 (dispersion and reflection) · DIYMA 2021-09-20

In plain English: Speakers in the dash get extra bass from the surfaces around them. He says the dash adds about 6 dB, and the windshield about 6 dB more, at low frequencies (how much depends on the windshield's angle, or "rake"). So a dash speaker has too much bass, not too little treble, and the fix is to cut the excess. Cutting also lowers the power needed, the cone's movement (excursion) and distortion. Don't boost the speaker's natural treble roll-off. A midrange mounted in an air-conditioning vent just needs its peaks turned down; then the tweeter can cross at 2.5 kHz with an LR4 crossover. Many people would boost the treble of a dash speaker; he says cut the bass instead.

3.6 #14 · A 250 Hz "resonance" is usually a reflection peak to EQ, not a panel to deaden.

Play a 250 Hz tone: no buzz or rattle means constructive interference from a cabin dimension, so just cut it. [RARE] Sources: DIYMA 2021-09-07

In plain English: Many people blame a peak around 250 Hz on a vibrating panel and add sound-deadening material (mats that stop panels vibrating). He says it is usually a reflection peak instead. His test: play a 250 Hz tone. If nothing buzzes or rattles, it's constructive interference (sound waves adding up) caused by one of the cabin's dimensions, so just cut it with EQ.

3.6 #15 · A 60–80 Hz dip in one midbass is usually a cross-car mode: cross above it to the sub instead of EQing.

In a thread about a subwoofer dip he traced the same kind of dip to two paths (sub→mic and sub→far side→mic) differing by ~7 ft; moving the mic to the other seat also helps confirm it. "Crossing midbass at 40 Hz isn't the way to get great midbass." [RARE] Sources: DIYMA 2022-03-16 · DIYMA 2021-12-02

In plain English: A "mode" is a frequency where sound bouncing between two surfaces reinforces or cancels itself; a cross-car mode forms across the width of the cabin. He says a 60–80 Hz dip in only one midbass (the driver that plays upper bass, usually in a door) is usually this. Instead of EQing it, set the crossover above the dip and let the subwoofer play that range. To confirm (he gave this test in a thread about a similar dip from a subwoofer): compare the distance from the sub to the mic with the distance from the sub to the far side of the car and on to the mic; a difference of about 7 feet (2.1 m) points to this mode. Or move the mic to the other seat. His view: "Crossing midbass at 40 Hz isn't the way to get great midbass."

3.6 #16 · A ~2 kHz dip in a dash center channel is fine, even helpful.

Ear spacing puts a dip near 2 kHz in the familiar stereo phantom center, so a center with that windshield dip sounds natural; L/C/R "voicing" matters less because you EQ anyway. [RARE] Sources: DIYMA 2021-09-22

In plain English: A center channel is a speaker in the middle of the dash. In the dash it may show a dip near 2 kHz from its reflection off the windshield. He says that's fine, even helpful. With ordinary stereo, a voice in the middle is a "phantom center" created by the left and right speakers together, and the spacing of your ears already puts a dip near 2 kHz in it. So a center speaker with that windshield dip sounds natural and familiar. He adds that matching the "voicing" (tonal character) of left, center and right speakers matters less in a car, because you EQ them anyway.

3.6 #17 · The usual car faults: a broad peak near 600 Hz and too much midbass/lower midrange.

The 600 Hz peak makes a car sound loud; because midbass is "always too loud," a correct tune often sounds quieter. In-car sub graphs include cabin gain, so cut a boomy peak 6–10 dB, not the 15 dB shown. A real customer tune: cut a little near 1 kHz and between 50–80 Hz. [RARE] Sources: DIYMA MS-8 FAQ · DIYMA 2015 · DIYMA 2018-06-27

In plain English: The faults he finds most often: a broad peak near 600 Hz, and too much midbass and lower midrange. The 600 Hz peak makes a car sound loud. Because midbass is "always too loud," a correct tune often sounds quieter. "Cabin gain" is the extra low bass a small, closed space like a car adds. In-car graphs of a sub already include it, so for a boomy peak, cut 6–10 dB, not the full 15 dB the graph shows. One real tune he did for a customer: cut a little near 1 kHz and between 50–80 Hz.

3.6 #18 · Make sure EQ reaches above 10 kHz; he fought for it at Harman.

Engineers there called HF EQ invalid; he argued it improved cars and the MS-8 shipped with it. If an analyzer's suggested EQ stops at a "brick wall" at 10 kHz, its algorithm may not go that high; a far-off-axis tweeter needs boost up there. [RARE] Sources: DIYMA 2015 · DIYMA 2015 (off-axis tweeter)

In plain English: At Harman, engineers called high-frequency (HF) EQ invalid. He argued it improved cars, and the MS-8 shipped with it. If an analyzer's suggested EQ stops dead at 10 kHz, like a "brick wall," its algorithm may simply not go that high. A tweeter mounted far off axis (pointing well away from you) needs boost up there. So in practice: make sure your EQ can reach above 10 kHz.

3.6 #19 · Parametric EQ for fixing, graphic EQ for drawing.

A 31-band EQ doesn't do what its sliders show (raising every band 12 dB isn't a flat +12 dB; adjacent pairs rarely render as drawn); 1/3-octave EQ "is insufficient for any real correction," and an EQ with only ±6 dB range can't tune a car. Graphics suit shaping an already-corrected curve. [RARE] Sources: DIYMA 2009 · DIYMA 2009 (31-band) · DIYMA 2016

In plain English: A graphic EQ has fixed bands with sliders, such as 31 bands spaced 1/3 octave apart. A parametric EQ lets you set each filter's exact frequency, width (Q) and gain. He says a 31-band graphic doesn't do what its sliders show: raising every band 12 dB doesn't give a flat +12 dB, and neighboring bands set in opposite directions rarely come out as drawn. He says 1/3-octave EQ "is insufficient for any real correction," and an EQ with only ±6 dB of range can't tune a car. So use parametric EQ to fix problems; a graphic suits shaping a curve that has already been corrected. Many people tune with a 31-band graphic; he says it isn't enough for correction.

3.6 #20 · Use auto-EQ, but limit each channel to its passband.

REW's auto-EQ is "damn good" and saves time; but run against the full target on every channel it flattens your crossover slopes. Auto-EQs need phase, so they build minimum phase from the average and must cap Q and boost to avoid boosting nulls; pasting a single-mic phase into a spatial average improves the model. [RARE] Sources: DIYMA 2016 · DIYMA 2021-04-17 · DIYMA 2017-06-27

In plain English: Auto-EQ is a feature that calculates EQ filters for you to match a target. He calls REW's auto-EQ "damn good" and a time saver. But if you run it against the full target on every channel, it pushes each driver toward the target everywhere and flattens the crossover slopes you set. His explanation points to the fix in the heading: limit each channel's EQ to its passband (the range that channel is meant to play). Auto-EQs also need phase, which an average doesn't have, so they build a "minimum phase" estimate from the average. That estimate hides the regions that don't behave that way, such as cancellations, so the software must cap Q and boost to avoid boosting nulls. Pasting the phase from a single-mic measurement into a spatial average improves that model.

3.6 #21 · Hiss after auto-EQ on a factory source is EQ boost, not a ground fault.

If the head unit rolls off the highs, auto-EQ boosts the top end to compensate and you hear the boost; don't tear the install apart. [RARE] Sources: DIYMA 2021-08-08

In plain English: Some factory head units (the car's original radio) roll off the highs. Auto-EQ then boosts the top end to compensate, and you hear that boost as hiss. He says this is EQ boost, not a ground fault (a wiring problem that causes noise), so don't tear the installation apart looking for one.

3.6 #22 · After tuning, fix only what a notepad and the RTA confirm.

Note track and time whenever something annoys you; later play that passage on the RTA, then check the pink-noise curve there: a peak gets fixed, otherwise it's the recording. Cutting everything occasionally annoying ends in a "smile" curve; peaks are more audible than dips. [RARE] Sources: DIYMA 2021-08-30

In plain English: After the main tune, don't keep tweaking at random. Whenever something annoys you while listening, write down the track and the time. Later, play that passage with the RTA running, then check your pink-noise curve at the same frequencies. If there's a peak there, fix it; if not, the problem is in the recording. Cutting everything that's occasionally annoying ends in a "smile" curve (bass and treble up, midrange scooped out, like a smile drawn on the graph). He notes that peaks are more audible than dips.

3.6 #23 · EQ removes linear differences: cone "sound," L/C/R matching, road noise.

Linear timbre differences between cone materials EQ away (distortion doesn't); EQ is a fine way to match L, C and R without identical speakers. Road noise is an EQ-only change (he kept a preset for a pebbled stretch near El Paso); crossovers, delays and gains stay put. Sources: DIYMA 2016 · DIYMA 2021-01-02 · DIYMA MS-8 FAQ (road noise)

In plain English: "Linear" differences are differences in frequency response (how loud each pitch is), which EQ can change. Distortion means the speaker adds sounds that weren't in the signal, and EQ can't remove that. He says the timbre (tonal "sound") differences between cone materials are linear and EQ away; distortion doesn't. So EQ is a fine way to match left, center and right speakers without making them identical. Road noise is also a job for EQ only: he kept a preset for a pebbled stretch of road near El Paso. Crossovers, delays and gains stay put. Many people insist that left, center and right speakers must be identical to match; he says EQ handles it.

The full series

Download the complete guide as one PDF (351 pages, every point with its explanation).