Head-Related Transfer Functions (HRTF), Harman Target Curves, Crossfeed Networks, and Binaural Spatial Audio

Why does a loudspeaker with a ruler-flat anechoic frequency response ($20\text{ Hz}$ to $20\text{ kHz} \pm 0.5\text{ dB}$) sound natural, balanced, and lifelike in a treated room, whereas a headphone engineered with an identical flat frequency response measured at the eardrum sounds intolerably dark, muffled, and devoid of upper-midrange clarity? Conversely, why do traditional stereo recordings played through headphones collapse the three-dimensional soundstage into a narrow, artificial line trapped directly between the listener’s ears?

The answers lie in the complex physical acoustics of the human head, torso, outer ear (pinna), and ear canal—collectively mathematically described by the Head-Related Transfer Function (HRTF). When we listen to loudspeakers or live acoustic instruments in an open room, sound waves diffract around our shoulders, reflect off the intricate ridges of our pinnae, and resonate within our ear canals before striking the tympanic membrane. Headphones bypass these external anatomical structures entirely, injecting sound directly into the ear canal entrance. This comprehensive engineering reference examines the physics of HRTF, Free-Field and Diffuse-Field equalization standards, the Harman Target Curve, binaural localization cues (ITD and ILD), analog/DSP crossfeed networks, and head-tracked spatial audio rendering.

Before an acoustic wave travelling through free space reaches the human eardrum (tympanic membrane), its spectral amplitude and phase are systematically modified by four distinct anatomical acoustic structures acting in series:

1. Torso and Shoulder Reflections ($100\text{ Hz – }1,500\text{ Hz}$): Acoustic waves arriving from elevated or lateral angles bounce off the listener’s chest and shoulders, creating subtle comb-filtering notches and low-mid reinforcement that provide vertical elevation cues.
2. Head Diffraction and Acoustic Shadowing ($500\text{ Hz – }20,000\text{ Hz}$): Because the human head has an average diameter of approximately $17.5\text{ cm}$, sound waves with wavelengths shorter than the head’s width ($f > 1,500\text{ Hz}$) cannot diffract cleanly around the cranium, casting a high-frequency acoustic shadow on the far (contralateral) ear while causing pressure doubling ($+6\text{ dB}$) on the near (ipsilateral) ear.
3. Pinna Cavities and Concha Resonance ($2,000\text{ Hz – }12,000\text{ Hz}$): The folds of the outer ear—specifically the helix, antihelix, tragus, and concha bowl—act as directionally dependent acoustic resonators and delay lines. The concha cavity exhibits a prominent quarter-wavelength resonance centered around $4,000\text{ Hz to }5,000\text{ Hz}$ providing $+8\text{ to }+10\text{ dB}$ of natural acoustic gain, while sharp spectral notches between $6\text{ kHz}$ and $10\text{ kHz}$ shift in frequency depending on whether a sound source is in front of, behind, or above the listener.
4. Ear Canal Quarter-Wave Resonance ($2,500\text{ Hz – }3,500\text{ Hz}$): The human external auditory meatus (ear canal) is an acoustic tube approximately $25\text{ mm}$ long and $7\text{ mm}$ in diameter, closed at the inner end by the eardrum. Acting as an open-closed quarter-wavelength pipe ($f = c / 4L$), it produces a powerful natural resonance peak centered around $2.7\text{ kHz to }3.2\text{ kHz}$ that boosts acoustic pressure at the eardrum by $+10\text{ to }+15\text{ dB}$.

Combined Anatomical Acoustic Gain at the Eardrum (0° Frontal Incidence):
- 20 Hz to 500 Hz:    0 dB (Wavelengths much larger than head/torso; zero diffraction)
- 1,000 Hz:          +2 dB to +3 dB (Onset of head/torso baffle step)
- 2,700 Hz – 3,200 Hz: +12 dB to +17 dB Peak (Primary Ear Canal 1/4-Wave + Concha Resonance)
- 4,500 Hz – 6,000 Hz: +8 dB to +12 dB Plateau (Concha Secondary Mode)
- 8,000 Hz – 10,000 Hz: Sharp Pinna Destructive Interference Notch (-5 dB to -15 dB)

Mathematical Definition of Left/Right HRTF:
H_L(r, theta, phi, f) = P_eardrum_L(r, theta, phi, f) / P_freefield_center(r, f)
H_R(r, theta, phi, f) = P_eardrum_R(r, theta, phi, f) / P_freefield_center(r, f)

Because our auditory cortex hears every sound in daily life through this $+15\text{ dB}$ at $3\text{ kHz}$ anatomical filter, our brain calibrates its internal perception of “flat, neutral timbre” to expect that exact $3\text{ kHz}$ boost at the eardrum. When a headphone is placed over the ear (and especially when an In-Ear Monitor is inserted directly into the ear canal, bypassing the pinna and shortening the ear canal), that natural anatomical boost disappears. Consequently, a headphone must have a deliberate $+10\text{ to }+15\text{ dB}$ rise in its measured eardrum response between $1.5\text{ kHz}$ and $5\text{ kHz}$—known as “Ear Gain” or “Pinna Compensation”—just to sound tonally flat to a human listener!

Since headphones cannot be tuned to a flat line on an ear simulator, acoustic standards organizations spent decades attempting to define the ideal reference target curve at the eardrum (Drum Reference Point, DRP):

Defined in early IEC standards, the Free-Field target assumes the headphone should replicate the acoustic pressure at the eardrum produced by a single flat loudspeaker positioned directly in front of the listener ($0^\circ$ azimuth, $0^\circ$ elevation) inside an echo-free anechoic chamber. In practice, headphones tuned strictly to the Free-Field curve sound excessively bright,shouty, and glaring in the $3\text{ kHz}$ region because real-world listening rooms do not behave like anechoic chambers with single point-source speakers.

Introduced by Gunther Theile and the IRT in the 1980s, the Diffuse-Field target models the average eardrum response in a reverberant room where sound waves arrive with equal probability and intensity from all $360^\circ$ directions simultaneously. While Diffuse-Field tuning tamed the harsh $3\text{ kHz}$ glare of Free-Field headphones (inspiring classic studio monitors like the Etymotic ER4S and early Sennheiser HD600), a pure Diffuse-Field target still sounds noticeably bass-light and overly clinical to modern mixing engineers.

Beginning in 2012, Dr. Sean Olive and Todd Welti at Harman International conducted landmark double-blind psychoacoustic research to solve the target curve dilemma once and for all. First, they measured the in-room eardrum response of high-end, anechoically flat studio monitor loudspeakers calibrated inside a reference semi-reflective listening room. They discovered two critical deviations from pure Diffuse-Field theory:

1. Room Gain and Early Reflections: In a real room, low frequencies reinforce off room boundaries while high frequencies are absorbed by furnishings and air, creating a smooth downward-sloping target from bass to treble of approximately $-1\text{ dB per octave}$ ($-10\text{ dB}$ total tilt from $20\text{ Hz}$ to $20\text{ kHz}$).
2. Missing Whole-Body Tactile Bass: When listening to loudspeakers, humans perceive low-frequency bass ($20\text{ Hz – }120\text{ Hz}$) not only through their eardrums, but also through somatosensory tactile vibration across their chest, skin, and skeletal structure. Because headphones stimulate only the eardrums and supply zero chest vibration, a headphone with a flat low-frequency response sounds subjectively thin. To compensate for missing tactile bone/chest conduction, controlled listener preference studies proved that most trained and untrained listeners require a $+4\text{ to }+6\text{ dB}$ low-bass shelf below $105\text{ Hz}$ (for over-ear headphones) and up to $+7\text{ to }+9\text{ dB}$ (for in-ear monitors) to match the perceived bass impact of flat speakers in a room.

When reviewing headphone frequency response graphs, engineers must verify which acoustic measurement coupler was used, as different ear simulators have vastly different acoustic impedances:

• IEC 60318-4 (“711” Coupler / GRAS 43AG / KB5000): The historic industry standard occluded-ear simulator. It models the inner half of the ear canal using a cylindrical metal cavity with tuned acoustic damping volumes. While highly accurate up to $8\text{ kHz}$, its rigid metal walls produce an artificial, exaggerated half-wave resonance spike at $8\text{ kHz to }10\text{ kHz}$ when measuring IEMs, and its input acoustic impedance deviates from real human flesh above $10\text{ kHz}$.
• ITU-T Rec. P.57 Type 4.3 (Brüel & Kjær Type 5128): Introduced to match the exact 3D geometry and soft tissue compliance (Shore hardness) of the average human pinna and full-length ear canal down to $20\text{ kHz}$. Because the B&K 5128 accurately models the acoustic impedance of human skin and cartilage across the entire audible spectrum, target curves developed for older 711 couplers cannot be directly overlaid onto B&K 5128 measurements without a fixture-specific transfer function conversion.

How does the human brain determine the exact azimuth (horizontal angle), elevation (vertical angle), and distance of a sound source using only two eardrums? Defined by Lord Rayleigh’s classic Duplex Theory of Sound Localization and modern spectral cue research, the auditory system relies on three primary mechanisms:

1. Interaural Time Difference (ITD) — Dominant Below 1,500 Hz:
   ITD = (r / c) * (theta + sin(theta))   [Woodworth Spherical Head Formula]
   - At 0° (dead center): ITD = 0 microseconds
   - At 90° (hard left/right): Maximum ITD ≈ 650 to 700 microseconds
   - Human neural resolution can detect ITD shifts as small as 10 microseconds (≈ 1° azimuth).

2. Interaural Level Difference (ILD) — Dominant Above 1,500 Hz:
   - High-frequency wavelengths (< 22 cm) are shadowed by the cranium.
   - Creates up to 20 dB to 30 dB of attenuation at the far ear above 4 kHz.

3. Monaural Pinna Spectral Cues — Dominant for Elevation & Front/Back Disambiguation:
   - Because a sound at +30° front-right and +150° rear-right can have identical ITD and ILD (the "Cone of Confusion"), the brain analyzes high-frequency pinna notches (6 kHz – 14 kHz) and subtle micro-head rotations to lock in vertical elevation and front-versus-rear placement.

Over $99\%$ of commercial music is mixed on stereo loudspeaker pairs arranged in an equilateral $60^\circ$ triangle ($\pm 30^\circ$ off-center). When you listen to loudspeakers in a room, your left ear hears the Left speaker first, followed roughly $220\text{ to }300\text{ microseconds}$ later by the sound of the Right speaker diffracting around your head (with its high frequencies attenuated by about $6\text{ dB}$ by head shadowing). This natural acoustic crosstalk between speakers and ears is called Interaural Cross-Correlation.

When you play that exact same stereo recording through headphones, the Left headphone cup feeds only the left ear, and the Right headphone cup feeds only the right ear ($100\%$ channel separation). If a vintage jazz recording has a saxophone panned hard left and drums panned hard right, the brain receives an unnatural sensory contradiction: an infinite Interaural Level Difference ($\text{ILD} = \infty\text{ dB}$) paired with zero Interaural Time Difference ($\text{ITD} = 0\text{ }\mu\text{s}$) and zero head-shadowed crosstalk. Unable to reconcile these impossible physical cues, the auditory cortex collapses the spatial image inside the skull—a fatiguing phenomenon known as In-Head Localization (IHL) or "three-blob lateralization."

While simple RC crossfeed networks alleviate extreme hard-panned listening fatigue, true holographic 3D headphone imaging requires full Binaural Synthesis via Head-Related Impulse Responses (HRIRs). If an audio object (such as a violin or cinema sound effect) is convolved in real time with the left-ear and right-ear HRIRs measured for a specific spatial coordinate $(r, \theta, \phi)$, the listener perceives the sound emanating from outside the headphones in three-dimensional space.

However, static binaural audio suffers from two fundamental limitations when played back on standard headphones:

1. Generic vs Personalized HRTF Mismatch: Because every human being has uniquely shaped pinnae, ear canal dimensions, and head width, listening through a generic HRTF (such as a KEMAR or Neumann KU100 dummy head) often causes front-to-back reversals—where sounds intended to appear directly in front of the listener instead sound as though they are hovering behind their neck or slightly above their forehead. Modern spatial audio engines utilize smartphone LiDAR/photogrammetry scans of the user's outer ears to select or synthesize a personalized HRTF profile that matches their exact concha and pinna geometry.
2. Absence of Vestibulo-Acoustic Head Tracking: In the real world, we constantly make unconscious micro-rotations of our head ($1^\circ$ to $3^\circ$). If a sound is genuinely in front of us and we rotate our head $5^\circ$ to the left, the ITD shifts by $40\text{ }\mu\text{s}$ in favor of the right ear and the pinna notches shift predictably—instantly confirming to our brain that the sound source is anchored externally in front of us. In static headphones, when you turn your head, the entire soundstage rotates rigidly with your skull, instantly breaking the externalization illusion. By embedding 6-axis Inertial Measurement Units (IMUs—accelerometers and gyroscopes) inside the headphone and updating the binaural convolution engine at $> 100\text{ Hz}$ with less than $30\text{ ms}$ motion-to-photon/audio latency, dynamic head tracking anchors virtual loudspeakers in fixed room coordinates, eliminating front-back confusion and creating convincing out-of-head externalization.

When calibrating headphones for mixing, mastering, or critical audiophile listening using a multi-band Parametric Equalizer (PEQ), engineers divide the frequency spectrum into three distinct acoustic zones based on wavelength vs ear cavity physics:

• Zone 1 — Modal Piston Region ($20\text{ Hz – }800\text{ Hz}$): Wavelengths are much larger than the earcup cavity. Measurements on industry-standard fixtures (GRAS 43AG or B&K 5128) are highly repeatable and predictable here. Minimum-phase low-shelf and peaking filters with low-to-moderate $Q$ ($0.5$ to $2.0$) can be applied with confidence to dial in exact sub-bass extension and remove mid-bass bloat ($150\text{ Hz – }300\text{ Hz}$).
• Zone 2 — Pinna / Ear Gain Region ($1,000\text{ Hz – }5,000\text{ Hz}$): Dominated by pinna coupling and ear canal resonance. Broad, gentle peaking filters ($Q < 1.5$) should be used to match the target ear gain slope around $3\text{ kHz}$, ensuring vocals and snare drums sit neither too recessed nor too shouty.
• Zone 3 — High-Frequency Cavity Standing Wave Region ($6,000\text{ Hz – }20,000\text{ Hz}$): Above $6\text{ kHz}$, half-wavelength and quarter-wavelength standing waves form between the headphone driver and the listener's unique eardrum geometry. A sharp notch at $8.5\text{ kHz}$ on a laboratory measurement rig may shift to $7.4\text{ kHz}$ or $9.6\text{ kHz}$ on your actual head! Therefore, engineers should never apply high-Q ($Q > 4$) narrow peaking filters above $6\text{ kHz}$ based purely on someone else's measurement graph. Instead, use a manual sine-wave sweep or pink noise band-pass filter on your own ears to locate and tame individual sibilance peaks, or use a broad high-shelf filter ($Q = 0.71$) above $8\text{ kHz}$ to adjust overall treble air.