IEC 61063 Octave-Band and Fractional-Octave-Band Filters: The Frequency Sieves Behind Every Noise Measurement

IEC 61063 Octave-Band and Fractional-Octave-Band Filters — The Frequency Sieves of Acoustic Engineering

IEC 61063:1991 | First Edition | TC 29 Electroacoustics | ~1,800 words

1. Beyond the Single-Number Sound Level: Why Frequency Analysis Matters

Stand next to a humming transformer and your sound level meter reads 72 dB(A). But that number conceals a critical question: is the dominant energy at 100 Hz (magnetostriction from the core), at 1200 Hz (cooling fan blade-pass), or at 6.3 kHz (partial discharge corona)? Each points to a different physical mechanism and a radically different mitigation strategy. The dB(A) alone cannot tell the difference. This is exactly the problem that IEC 61063 octave-band and fractional-octave-band filters solve — partitioning the audio frequency range into standardized “buckets” so engineers can read a sound the way chemists read a spectrum.

Published by IEC TC 29 (Electroacoustics), IEC 61063 — together with ISO 266 “Preferred frequencies for acoustical measurements” and IEC 61672 “Sound level meters” — forms the measurement-toolchain trinity of applied acoustics. The standard defines center frequencies, passband widths, stopband attenuation, phase response limits, and permissible tolerances for octave-band (1/1-octave) and fractional-octave-band (primarily 1/3-octave) filters, covering implementations from analog RC active networks to multi-rate digital FIR filter banks.

The fundamental idea: An octave-band filter passes energy between a lower and upper cutoff frequency whose ratio is exactly 2 — mirroring the musical octave. A 1/3-octave filter divides each octave into three equal geometric sub-bands, each with a cutoff ratio of 2^(1/3) ≈ 1.259. This logarithmic spacing naturally aligns with the way human hearing perceives pitch and loudness across frequency, making the resulting spectra both physically meaningful and perceptually relevant.

2. Inside the Filter Specifications

2.1 The Base-10 Center Frequency System

IEC 61063 and ISO 266 jointly define a precise center frequency grid anchored at 1000 Hz and expanded using powers of 10. This is not an arbitrary choice — it ensures that frequency numbers fall on clean, engineering-friendly values that are easy to read, communicate, and label on instrument panels:

Octave-band centers: f_m = 1000 x 10^(3n/10) Hz, yielding the familiar sequence: 31.5, 63, 125, 250, 500, 1000, 2000, 4000, 8000, 16000 Hz — ten bands covering the entire audible range relevant to most industrial measurements.
1/3-octave centers: f_m = 1000 x 10^(n/10) Hz, giving 31 bands from 20 Hz to 20 kHz with center values such as 100, 125, 160, 200, 250, 315, 400, 500, 630, 800, 1000 Hz, etc.
Band edge frequencies for any band: lower edge f₁ = f_m / 2^(1/2b), upper edge f₂ = f_m x 2^(1/2b), where b = 1 for octave and b = 3 for 1/3-octave.

Octave-Band vs. 1/3-Octave-Band Filter Key Parameters
Parameter	Octave-Band (1/1)	1/3-Octave-Band
Edge frequency ratio f₂/f₁	2.0 (1 octave)	2^(1/3) ≈ 1.259
Relative bandwidth (f₂-f₁)/f_m	~70.7% of center frequency	~23.16% of center frequency
Typical number of bands	10 (31.5 Hz to 16 kHz)	31 (20 Hz to 20 kHz)
Frequency resolution	Coarse — good for quick surveys	Fine — suitable for noise source identification
Typical applications	Room acoustics, noise rating curves	Product noise diagnostics, tonality assessment
Example center frequencies	125, 250, 500, 1k, 2k, 4k Hz	100, 125, 160, …, 1k, …, 5k Hz
Equivalent Noise Bandwidth (ENB)	0.707 x f_m	0.232 x f_m

2.2 Filter Classes: Precision Is Only Half the Story

IEC 61063 defines three performance classes. Unlike sound level meter classes where the entire instrument chain is graded, here the classification focuses specifically on the filter section:

IEC 61063 Filter Class Specifications
Specification	Class 0	Class 1	Class 2
Center frequency accuracy	≤ 0.5%	≤ 1.0%	≤ 1.5%
Passband ripple (1/3-octave)	≤ 0.1 dB	≤ 0.3 dB	≤ 0.5 dB
Stopband attenuation (1/3-octave lower limit)	≥ 70 dB	≥ 60 dB	≥ 50 dB
Linear operating range	≥ 100 dB	≥ 80 dB	≥ 60 dB
Anti-aliasing requirements	Extremely stringent	Stringent	Basic
Compatible sound level meter	Class 0 / Class 1 SLM	Class 1 SLM	Class 2 SLM
Typical use case	Laboratory reference, calibration standards	Precision field measurements, regulatory compliance	General survey, trend monitoring

One aspect consistently overlooked in field practice: the anti-aliasing filter is the Achilles’ heel of any digital real-time analyzer. Before the ADC samples the microphone signal, an analog low-pass filter must suppress energy above the Nyquist frequency (half the sample rate). If this filter lets through energy at, say, 30 kHz while sampling at 48 kHz, that energy folds back (aliases) into the 18 kHz band and corrupts the entire top end of the 1/3-octave spectrum. IEC 61063 demands that aliased contributions in any band remain at least 60 dB below the genuine signal — a requirement that pushes analog anti-aliasing filter design to its practical limits in Class 0 instruments.

Engineering reality — Class 1 vs. Class 2 in the field: Many engineers assume Class 1 filters will always deliver better results than Class 2. But when background noise fluctuates by more than 3 dB during a measurement sequence, or when the microphone’s own frequency response deviates by more than 1 dB across the band of interest, the Class 1 filter’s precision advantage is completely buried under systematic errors. On a factory floor, moving the microphone 10 cm typically changes the reading more than the difference between Class 1 and Class 2 filters. The lesson: spend at least as much effort on microphone placement, wind screening, and background noise documentation as on the filter class specification.

2.3 Filter Shape and Adjacent-Band Isolation

The central design challenge is selectivity — capturing all the energy within the target band while reliably rejecting crosstalk from neighboring bands. IEC 61063 defines filter shape not by a single gain specification at the center frequency, but through a family of attenuation limits that collectively constrain the entire magnitude response:

Passband tolerance mask: Between f₁ and f₂, the magnitude response must stay within a specified envelope (e.g., -0.3 dB to +0.3 dB for 1/3-octave Class 1).
Stopband rejection rate: At least 30 dB/octave (octave-band) to 60 dB/octave (1/3-octave Class 1), ensuring that adjacent-band leakage contributes less than 0.1% of the total energy in the target band.
Effective bandwidth verification: The Equivalent Noise Bandwidth (ENB) — the width of a perfectly rectangular filter that would pass the same total noise power — must agree with the nominal bandwidth within stated tolerances. This serves as an integral check of the entire filter response shape.

In the analog domain, classical realizations stack multiple Butterworth or Cauer (elliptic) stages to achieve the required selectivity with manageable component count. In the digital domain, FIR (Finite Impulse Response) filters dominate because their precisely linear phase response avoids group-delay distortion across the band — critical when analyzing transient sounds. The price is filter length: a Class 1 1/3-octave FIR filter typically requires 200 to 1000 taps, with corresponding computational cost in real-time multi-band analysis.

3. Field Applications and Common Pitfalls

3.1 Key Application Domains

Industrial noise assessment: Per ISO 9612 and the ISO 11200 series, 1/3-octave spectra are the standard tool for identifying dominant noise sources. A gearbox generates narrow-band energy at the gear-meshing frequency (number of teeth x shaft RPM); a fan produces tones at the blade-pass frequency (number of blades x RPM); an electric motor radiates at slot-harmonic frequencies. Plotting the 1/3-octave spectrum immediately reveals which component dominates. The same spectrum, when combined with A-weighting or C-weighting, yields NR (Noise Rating) and NC (Noise Criteria) curves for evaluating workspace acoustic quality.

Building acoustics: Octave bands from 125 Hz to 4 kHz are the canonical frequency range for measuring reverberation time (RT60), airborne sound insulation (Rw, DnT,w), and impact sound pressure level (Ln,w). The 1/3-octave resolution — required by ISO 16283 for field measurements and ISO 10140 for laboratory tests — captures the modal behavior of rooms and reveals standing-wave patterns that octave-band data would smear into invisibility.

Environmental noise monitoring: Airport, railway, and highway noise assessments routinely use 1/3-octave spectra to evaluate tonality — a key penalty factor in ISO 1996-2 for subjective annoyance. When a single 1/3-octave band exceeds both adjacent bands by 5 dB or more, a significant tonal component is deemed present, and a tonal adjustment (typically +3 to +6 dB) is added to the rating level.

Audio engineering: Real-time 1/3-octave analyzers (RTA) are the primary tool for live sound system equalization and feedback identification. In room acoustics, 1/3-octave spectra underpin NCB (Balanced Noise Criteria) and RC (Room Criteria) rating calculations for HVAC system noise.

3.2 The Five Most Common Measurement Errors

Mistake 1 — Mismatched time weighting and filter settling time. Every filter has a finite settling time — the interval it requires to reach a steady reading after a change in the input signal. For low-frequency 1/3-octave bands (e.g., 25 Hz center), settling time can exceed 500 ms. If you sweep through the spectrum using “Fast” time weighting (125 ms integration), the low-band readings are systematically low because the filter has not yet stabilized. Fix: use “Slow” weighting (1 s) for bands below 100 Hz, or better yet, measure in Leq (equivalent continuous level) mode with at least 10 seconds of integration per band.

Mistake 2 — Ignoring background noise contamination per band. In field measurements, the target noise and background noise add on an energy basis in each frequency band. If the target source contributes only 3 dB more than the background in a given band, the actual level requires a 1.8 dB correction. Below 3 dB signal-to-noise ratio, the band is essentially unusable. IEC 61063 requires manufacturers to declare the self-noise floor of each filter band. Always record a background spectrum with the noise source turned off — this is not optional, it is fundamental.

Mistake 3 — Using octave bands where only 1/3-octave will do. Octave bands pool multiple tonal or narrow-band components into a single wide bucket, smearing the diagnostic signature. A 2000 Hz bearing whine and a 1050 Hz gear-mesh tone both fall into the 1000 Hz octave band and appear as a single blended peak. In a 1/3-octave spectrum they separate cleanly into the 1000 Hz and 2000 Hz bands — a distinction that directly determines which component to repair.

Mistake 4 — Neglecting microphone and preamplifier frequency response limits. The filter is only as good as the signal entering it. If the measurement microphone rolls off above 10 kHz (common in moderately priced Class 2 instruments) or exhibits rising low-frequency noise below 30 Hz, the resulting spectrum reflects the microphone, not the sound field. The fix: verify the entire measurement chain — microphone, preamplifier, cabling — against a pistonphone or electrostatic actuator across the full frequency range of interest.

Mistake 5 — The sample-rate trap in digital analyzers. When you reconfigure a digital RTA to a lower sample rate (to save storage or processing power), the anti-aliasing filter cutoff is automatically lowered. If your 1/3-octave analyzer was verified at 48 kHz sampling and you subsequently switch to 16 kHz, the 16 kHz and 20 kHz bands may contain aliased garbage. Practical rule: never analyze above 40% of the configured sample rate unless you have independently verified alias rejection.

3.3 Selecting the Right Instrument — A Decision Framework

The market spans from USD 500 entry-level sound level meters to USD 50,000 multi-channel analyzers. Here is a practical selection logic:

Real-time or post-processed? Real-time analysis requires hardware filter banks or DSP engines that continuously update all bands — essential for on-site surveys where you need to move the microphone and see the spectrum change live. Post-processing from audio recordings is viable for stationary measurements and offers unlimited re-analysis options, including fractional-octave resolutions (1/6, 1/12, 1/24) that no portable RTA supports.
“True” filtering or FFT-synthesized? For stationary signals, FFT-based energy integration over frequency bins can accurately synthesize the response of ideal IEC 61063 filters. But for transient or impulsive noise (blasting, impact hammers, gunshots), only a properly designed real-time digital or analog filter bank delivers results compliant with the standard’s time-domain requirements.
Is Class 1 mandatory? There are exactly three scenarios where Class 1 filters are genuinely required: legal evidence collection, contractual compliance verification, and inter-laboratory comparison studies. For day-to-day engineering diagnostics, trend analysis, and internal QC, Class 2 is perfectly adequate — and the money saved is better spent on a higher-quality microphone and calibration accessories.

Beware of “app-store acoustics.” Numerous smartphone noise apps now claim to provide “real-time 1/3-octave RTA” functionality. The overwhelming majority work by taking a 1024-point or 2048-point FFT and simply distributing the resulting linear-spaced frequency bins into the nearest 1/3-octave band. This is not compliant with IEC 61063. The fatal flaw: at low frequencies, the FFT bin resolution is far too coarse. A 1024-point FFT at 48 kHz sample rate yields 46.9 Hz per bin — so the entire 25 Hz 1/3-octave band (which spans roughly 22.4 to 28.2 Hz) is covered by barely one FFT bin. Measurement errors at 25 Hz and 31.5 Hz can easily reach 5 to 10 dB. If you must use a smartphone for screening, at minimum verify the app’s response against a calibrated 1 kHz tone and check that all 1/3-octave bands above 125 Hz read within 1 dB of the known input level.

4. FAQ

When should I use octave-band analysis versus narrowband FFT?: Think of octave/1/3-octave analysis as the “survey map” and narrowband FFT as the “microscope.” Octave spectra compress the entire audible range into 10-31 data points — ideal for rapid identification of the general spectral shape and for computing standardized metrics like sound power level and noise rating curves. Narrowband FFT resolves hundreds or thousands of equally spaced spectral lines, making it the tool of choice for pinpointing specific rotating-machinery fault frequencies (bearing defect frequencies, gear-mesh sidebands). Best practice: start with a 1/3-octave overview to identify the frequency regions of interest, then zoom in with narrowband FFT for root-cause diagnostics. The two approaches are complementary, not competing.
What is the relationship between IEC 61063 filters and A/B/C/D weighting?: A/B/C/D weighting networks are single broadband filters — they produce one number (e.g., dB(A)) representing the sum of all frequency content after applying a frequency-dependent weighting curve. IEC 61063 filters produce a spectrum — one number per frequency band. In a modern sound level meter, both operations happen: the 1/3-octave filter bank resolves the signal into bands, then A-weighting coefficients are applied to each band, and the weighted band levels are summed to produce the A-weighted sound level. This two-step architecture is standard in all digital sound level meters compliant with IEC 61672.
Why does my 1/3-octave spectrum bounce around at low frequencies?: Three mechanisms typically combine to produce this effect: (1) Room mode dominance — at low frequencies, standing waves between reflective surfaces create large spatial variations in sound pressure level, easily exceeding 10 dB from one measurement position to another. (2) Insufficient integration time — the 25 Hz 1/3-octave band requires nearly 500 ms to settle; “Fast” weighting captures only 125 ms and will show jumpy readings. Switch to Leq mode with at least 10 seconds of averaging. (3) Microphone low-frequency artifacts — structure-borne vibration transmitted through the microphone housing and wind-induced pressure fluctuations produce spurious low-frequency output that the analyzer cannot distinguish from real acoustic energy. Always record a 1-minute background spectrum with the noise source turned off, using the same measurement chain and microphone position.
How do digital real-time analyzers implement 1/N-octave filter banks efficiently?: The dominant modern approach is the multirate filter bank architecture, not a brute-force bank of N independent FIR filters. The core idea: successively decimate (downsample by 2) the input signal, applying a half-band low-pass filter at each stage. The outputs from successive decimation stages correspond to progressively narrower frequency bands. By combining outputs from different stages, all octave and 1/3-octave bands can be synthesized with computational complexity of O(N log N) rather than the O(N^2) of N independent filters. This is how portable analyzers achieve 1/12-octave or even 1/24-octave real-time resolution on modest DSP hardware — a feat that would be computationally prohibitive with a naive implementation.

At its core, IEC 61063 exists to answer a single, indispensable question: When something is too loud, where on the frequency axis is the problem? Without an answer to that question, noise control engineering is guesswork — adding absorption where the problem is actually structural, or damping where the issue is actually aerodynamic. The octave-band and fractional-octave-band filters defined by this standard are the acoustic engineer’s equivalent of the prism in optics: they decompose the complex waveform arriving at the microphone into its spectral constituents, making the invisible structure of sound visible. Used correctly, they transform noise from an annoyance to be endured into a diagnostic signal to be acted upon.