How can noise-canceling headphones make things quieter?

Start with something familiar

The soft cushion and the earcup form a physical barrier. Even with the electronics off, they can change how sound gets through. Fit, materials and frequency matter.

Sense a useful signal

An outer microphone can provide information about the disturbance. An inner microphone can check pressure near the listening region. They serve different roles; neither magically identifies everything a listener wants to hear.

Make another pressure change

The driver is a small loudspeaker. Electronics shape its signal so the secondary sound can oppose some unwanted pressure at a target. This is active noise cancellation, often shortened to ANC.

Watch the total

The three traces are pressure variations at the same place: the original sound, the added sound, and their point-by-point sum. A flat zero means no variation in this ideal example, not a vacuum.

Keep a setting, then challenge it

A tiny delay is a different fraction of a slow cycle and a fast cycle. Keep the gain and phase fixed while changing frequency. A setting that helps one component can reinforce another.

Move a probe in a separate experiment

Our two fixed plane waves can oppose at one point and reinforce elsewhere. Retuning moves the node. A headphone normally moves with the wearer’s head, so this bare-probe experiment is not an earcup fit simulation.

Ask whether there is enough time

For an unexpected arrival, processing and secondary sound travel both take time. A response that arrives later cannot erase pressure that already arrived. A known repeating tone presents a different prediction problem.

Meet an actual measurement

Real speakers, microphones and acoustic spaces have frequency-dependent responses. The evidence view shows original stored coefficients from a measured path, with its sample rate and processing history attached.

A stated phase convention

The added pressure is g cos(2πft + θ − 2πfτ). Here θ is its actual phase before the explicit delay τ. The original pressure is cos(2πft). At gain g = 1, delay 0 and θ = 180°, their sum is ideally zero.

Add pressures before calculating their level

The residual-to-original amplitude ratio is R = √(1 + g² + 2g cos(θ − 2πfτ)). Equal-frequency correlated pressures must be added before squaring. Adding their separate dB values would give the wrong result.

Relative dB is not a loudness percentage

For this single-frequency comparison, the change is 20 log₁₀ R. With correct opposition and gain 0.8, R = 0.2, or about −13.98 dB relative pressure. That is not a claim of 80% less perceived loudness. Exact zero is shown explicitly, not mislabeled 0 dB.

One delay, several outcomes

With gain 1, phase 180° and delay 0.25 ms, the model gives about −16.09 dB at 100 Hz, −2.32 dB at 500 Hz, +3.01 dB at 1000 Hz and +6.02 dB at 2000 Hz. These are analytic examples, not the response curve of a retail headphone.

Two components do not average in decibels

For equal-amplitude 100 Hz and 1000 Hz components, average their mean-square contributions over common complete cycles. At the quarter-millisecond setting, their combined change is about +0.053 dB. Strong bass reduction does not guarantee a reduction of the total.

A defined spatial model

Two equal-amplitude plane waves travel in opposite directions. With cancellation tuned at x₀, the residual amplitude ratio is 2|sin(2πf(x − x₀)/c)|. We fix c = 343 m/s as an illustrative value. Nodes repeat at half-wavelength intervals; this is not a universal spherical quiet bubble.

Shorter wavelength, narrower node region

Around a node, the region with at least 10 dB pressure reduction has half-width λ/(2π) asin(10^(−10/20)/2). It is about 8.67 cm at 100 Hz and 0.867 cm at 1000 Hz in this particular geometry. Those are model dimensions, not measured headphone fit tolerances.

A causal budget includes the acoustic leg

For our authored example, noise travels 30 mm in 87.46 microseconds. Secondary sound travels 10 mm in 29.15 microseconds. That leaves 58.31 microseconds for processing. A 40-microsecond delay can arrive early; an 80-microsecond delay arrives about 21.69 microseconds late. Early arrival alone does not prove a correct filter.

Feedforward, feedback and hybrid

Feedforward uses reference information, often sensed outside. Feedback uses a residual measurement, often inside. Hybrid architectures combine information. Designs vary, and a microphone’s response need not equal the response at the eardrum.

Why flipping every sample is insufficient

Using the additive convention e = d + S·y, with d = P·x and y = W·x, formal cancellation suggests W = −P/S wherever defined. A usable controller must also address causality, stability, conditioning, limits and changing paths. The ratio is not a ready-to-run safe controller.

A measured path is one part of the system

PANDAR’s modified QC20 hardware had the original Bose ANC electronics removed. One stored record has 8,192 samples at 48 kHz per channel. Its preprocessed coefficients retain the electronic backend here and have unspecified stored units. They are not a noise recording, pressure in pascals, hearing sensitivity or achieved cancellation.

Pressure nodes do not account for all energy

This lesson calculates pressure superposition. It does not calculate the complete pressure/particle-velocity energy flow. A node is not evidence that acoustic energy or the outside source vanished everywhere.

A steady engine hum

Predictable low-frequency components can be more forgiving to control than arbitrary fast changes. The particular device, paths, fit and algorithm still determine actual performance.

A headset in ambient mode

The same broad components can intentionally relay surroundings instead of opposing them. Trace the goal and information route before assuming that every listening mode means silence.

Engineering a controller

Engineers measure paths and test residuals at relevant targets. Our real impulse-response download shows why a plausible waveform and an attractive cutaway are not enough to establish product performance.

The total pressure trace gets smaller. What happened?

• The room’s air vanished
• Two pressure variations partly opposed each other at the target
• The outside source was removed

Electronics off, cups still on: what can still affect sound?

• The cushion and enclosure
• An unpowered controller making anti-noise
• Nothing works without a battery

The added pressure is exactly opposed but only 80% as large. What remains?

• 80% of the original pressure amplitude
• Nothing
• 20% of the original pressure amplitude

Keep a 0.25 ms delay. Why does changing 100 Hz to 1000 Hz change the result?

• The delay occupies more of each faster cycle
• High-frequency sound becomes electricity
• The driver must have moved farther away

Fixed sources make one point quiet. Is every nearby point equally quiet?

• Yes; sound was deleted
• No; relative phase can change with position
• Yes, if the diagram is green

An unpredictable disturbance arrives before the secondary signal. Can a later response erase that arrival?

• Yes, with more processing
• Yes, electronics can send it backward in time
• No; usable information and arrival time matter

What does an ambient-listening mode intentionally do?

• Relays surrounding sound through an electronic listening route
• Turns the cushion into empty space
• Guarantees every warning will be heard

Does an ANC label alone prove suitable hearing protection for a noisy workshop?

• Yes; all ANC protects equally
• No; rating, fit and actual exposure matter
• Yes, if playback is turned up

Primary, secondary and feedback paths in measured ANC hardware

Original ICA 2019 paper. Modified QC20 acoustics, with original Bose control electronics removed; measured fit and direction variation.

Original measured-path database and reuse license

RWTH-owned archive. MIT license retained. Our local extraction preserves raw stored values and the documented electronic backend.

Source direction, non-causal delay and feedforward performance

Original analytical and experimental study. Supports timing constraints, not a universal frequency cutoff.

Accessible author-university record for the timing study

Institutional paper record and full text.

Feedback microphone and listener responses differ

Original controller optimization and dummy-ear experiment; stability and noise enhancement constraints.

Feedback control can also affect wanted audio

Original compensation study. ANC is not guaranteed to mathematically leave all program audio untouched.

Passive isolation and powered cancellation have different roles

Manufacturer explanation used for basic product operation, not comparative performance claims.

Ambient listening serves a different goal

Official mode description. Generic relaying of surrounding sound does not guarantee every warning will be heard.

ANC alone is not a hearing-protection rating

NIOSH distinguishes consumer cancellation from products labeled with a noise reduction rating; fit and exposure remain relevant.

Sound exposure and hearing

Agency background. This silent exploration does not prescribe playback levels or test a learner’s hearing.

Historical phase-opposition proposal

US2043416A, published June 9, 1936, with 1933 German priority. A patent documents a proposal, not every claimed practical result.

Actual external headphone photograph

Florian Fuchs/Wikipedia/CC-BY-SA 3.0. Whole photograph resized to WebP. QC25 exterior, not the PANDAR hardware or our generic internals.