Lesson 04 • Time-Frequency & Reverberation Analysis
Lesson 4 — Decay, RT60 and Room Acoustics
Steady-state magnitude hides the time dimension that actually governs translation, intelligibility, and low-end tightness. This module dissects REW's full decay toolkit—ETC, Filtered IR, RT60, RT60 Decay (T60M), Clarity (C50/C80/D50/TS), Spectral Decay (CSD), Waterfall (Fourier & Burst Decay), and Spectrogram (Fourier, Wavelet, Airy/Morlet CWT)—to separate physical acoustic defects from DSP-solvable resonances.
01 • Engineering Rationale
What this is for
Two control rooms—or two adjacent seats in the same theater—can exhibit virtually identical 1/6-octave smoothed SPL curves while sounding diametrically opposite: one dry, articulate, and phantom-center locked; the other smeared, one-note in the sub-bass, and fatiguing in the upper midrange. The steady-state FFT integrates direct sound, early specular reflections, modal overhang, and diffuse reverberation into a single complex sum. Without decomposing energy across both frequency and arrival/decay time, you cannot distinguish a boundary interference cancellation from a low-Q absorption dip, nor can you distinguish a high-Q minimum-phase modal ring from a broadband reverberant tail.
For a professional, REW's decay and room-acoustics views serve three concrete engineering decisions:
- Drawing the acoustic-treatment vs. EQ boundary: Minimum-phase modal peaks can be attenuated in magnitude—and partially shortened in decay—using targeted parametric or Modal EQ filters. Non-minimum-phase boundary cancellations, specular comb-filtering, and broadband reverberant tails cannot be solved with EQ. Decay plots expose which anomalies require mineral wool, tuned membrane traps, diffusion, or repositioning before touching a single biquad.
- Diagnosing early vs. late energy regimes independently: In small rooms (<200 m³), acoustic behaviour splits sharply across frequency and time. Discrete early reflections (0–25 ms) degrade imaging and tonal neutrality via comb filtering (ETC); sparse standing waves below the Schroeder frequency dominate bass overhang (Burst Decay, Wavelet Spectrogram, RT60 Decay, Modal Analysis); and statistical decay above a few hundred Hz dictates reverberation balance and speech/music intelligibility (RT60, Clarity).
-
Avoiding false diagnostics caused by transform and measurement artifacts: Default Fourier waterfalls routinely fabricate apparent "bass ringing" out of window-length smearing. Standard ISO 3382
T30estimates routinely fail below 200 Hz in domestic rooms due to low modal density, forward-filter impulse ringing, and noise-floor contamination. Knowing which REW view—and which mathematical kernel inside it—matches the physical regime prevents expensive mistreatment.
Single-Position Validity & Spatial Averaging
Every impulse response—and every decay metric derived from it—is strictly valid only for the exact source position, microphone position, sweep level, sample rate, calibration state, and IR window used. In the modal region, moving the microphone 30 cm across a pressure node can drop an axial mode's apparent SPL by 15 dB and corrupt its T20/T30 regression slope. Always corroborate single-seat decay findings across 3–6 positions using Overlays → RT60, Overlays → Clarity, and Overlays → ETC before specifying structural acoustic treatment.
02 • REW Execution Protocol
Steps: End-to-End Decay & Room Acoustics Workflow
Follow this sequence on a calibrated full-range sweep. Every downstream decay calculation inherits the signal-to-noise ratio (SNR), IR truncation ceiling, and t=0 accuracy of Step 1.
-
Configure IR truncation and capture a high-SNR full-range sweep.
- Before measuring reverberant rooms (>150 m³ or expected RT60 >0.8 s), open Preferences → Analysis and increase IR truncation so REW does not clip the late decay tail prior to Schroeder integration.
- Press Cmd+M (Ctrl+M on Windows) to open Measure. Set Start Freq to 10–20 Hz and End Freq to 20,000 Hz (remember REW sweeps from half the start frequency to twice the end frequency).
- Select at least a 512k (or 1M) sweep length at 48 kHz. Each doubling of sweep length gains ~3 dB of SNR—critical for keeping the noise floor ≥45 dB below peak so
T30and Lundeby estimation remain uncontaminated. - Set acoustic playback level to 75–85 dB SPL (higher when room HVAC/ambient noise is elevated, while keeping input headroom safely green at 6–40 dB). Enable an acoustic or loopback timing reference if you will compare reflection timing or multi-channel overlays.
- Avoid multi-sweep averaging when using a USB mic or separate input/output audio devices (clock drift smears HF decay). Instead, use a single longer sweep.
-
Verify
t=0and map specular early reflections in Impulse → ETC.- Switch to the Impulse graph, enable the ETC trace, and open graph Controls (gear icon).
- Set ETC smoothing to 0 ms (default smoothing obscures tightly spaced early reflections). Switch the Y-axis to dBFS or dB relative to peak (range 0 to −40 dB) and zoom the X-axis to −2 ms to +30 ms.
- Verify
t=0sits cleanly on the direct-sound arrival peak (use Estimate IR delay or Offset t=0 if misaligned; wrongt=0invalidates C50/C80/D50 and reflection path math). - Log every discrete spike above −20 dB within the first 20–25 ms. Multiply each spike's arrival time
Δt(in ms) by the speed of sound (0.343 m/msor1.125 ft/msat 20 °C) to compute its exact extra path length and triangulate the reflecting boundary with a laser tape. - Open Filtered IR and step through 1/3-octave or 1-octave bands (e.g., 500 Hz, 1 kHz, 4 kHz) viewing ETC and the Schroeder integral. Toggle between forward, time-reversed, and zero-phase filtering to verify whether an early low-mid reflection is genuine or a filter-ringing artifact.
-
Evaluate statistical reverberation in RT60 and verify regression quality.
- Open the RT60 graph. Select 1/3-octave bands first to catch narrow band anomalies, then toggle 1-octave bands for standard ISO-aligned room summaries.
- In Controls, confirm Filter order is set to 6 (standard starting point) and verify the filtering method.
- Display EDT, T20, T30, and Topt simultaneously. Open the data/legend readout (or inspect Filtered IR per band) to check correlation values and curvature—never accept a
T30bar whose regression line is bent by noise floor or double-slope coupling. - Use Topt as your primary small-room decay metric where
T20andT30diverge due to limited SNR or strong early reflections.
-
Generate the high-resolution frequency-domain decay model in RT60 Decay.
- Open the RT60 Decay graph tab.
- Click Generate to build the narrowband decay slices, then click Calculate RT60 model. REW fits an exponential decay plus noise-floor model directly in the frequency domain.
- Return to the RT60 graph and enable the T60M overlay to compare the frequency-domain decay model directly against Schroeder-derived
EDT,T20,T30, andTopt.
-
Inspect early-to-late ratios in Clarity and aggregate across positions.
- Open the Clarity graph and inspect C50 (speech, dB), C80 (music, dB), and D50 (definition, %), plus TS (center time, ms) from the RT60/Filtered IR metrics table.
- Repeat the sweep across 3–5 listening positions. Open the Overlays window (Overlays → RT60, Clarity, and ETC) and compute the spatial Average RT60 trace (exporting as needed) to separate room-global decay from seat-specific modal artifacts.
-
Analyze time-frequency modal overhang across Spectral Decay, Waterfall, and Spectrogram.
- Open Spectral Decay, set the frequency range to 20–500 Hz, and press Generate. Enable CSD mode (anchors the right window edge and advances only the left edge) to compare slice progression without shifting right-edge truncation artifacts.
- Open Waterfall. First inspect Fourier mode (20–300 Hz, 300–500 ms time range, 50–60 dB vertical span, 200–300 ms window length for LF resolution). Then switch mode to Burst Decay (period axis, 0–30 periods, 1/6- or 1/12-octave bandwidth) to evaluate resonance Q independently of frequency.
- Open Spectrogram. Switch from Fourier to Wavelet (or Morlet CWT / Airy CWT) for wideband 20 Hz–10 kHz analysis with adaptive time-frequency resolution. Enable the peak energy time overlay and compare normalized vs. absolute amplitude scaling.
-
Extract discrete modal poles in EQ → Modal Analysis.
- Open the EQ window and expand Modal Analysis.
- Enter the room's physical length, width, and height, set the search range (typically 20–250 Hz), adjust analysis length and noise threshold, and click Find Resonances.
- Inspect the Resonances table (exact modal frequencies, SPL, and extracted
T60) alongside the Pole-Zero plot. Cross-reference every slow-decaying resonance against your Burst Decay and T60M plots before choosing between physical bass trapping and Modal EQ.
macOS Measurement & Performance Notes
On macOS, REW uses Java audio drivers only (no ASIO or WASAPI exclusive). Stick to 44.1 kHz or 48 kHz at 24-bit configured in Audio MIDI Setup before launching REW—higher sample rates (up to 192 kHz+, or up to 1536 kHz in V5.30+) multiply RAM consumption and Wavelet/Waterfall calculation latency without improving room-acoustic accuracy below 20 kHz. Adjust output/input gain in Audio MIDI Setup or your interface mixer (REW's internal volume sliders are disabled on macOS). If a macOS update blocks mic capture, reset permissions in Terminal prior to launching REW:
tccutil reset Microphone
open -a REW.app --args -api
03 • Analytical Interpretation
Reading the output: Decoding Room Behaviour Across Views
No single REW decay graph tells the whole acoustic truth. Professionals triangulate across time-domain envelopes, Schroeder integrals, narrowband frequency-domain decay models, and time-frequency transforms. Below is what each view physically represents and how to read it without falling for display artifacts.
A. Impulse, ETC & Filtered IR
The raw impulse response oscillates positively and negatively, making reflection amplitudes difficult to gauge visually because a reflection can land near a zero-crossing. The Energy-Time Curve (ETC) computes the analytic signal envelope (via the Hilbert transform) of the IR, revealing the true instantaneous energy envelope on a dB scale.
- Direct arrival (
t = 0 ms): Normalized to 0 dB. Any pre-ringing beforet = 0indicates linear-phase/FIR crossover processing upstream, not room acoustics (recall that log-sweep harmonic distortion artifacts appear at negative time well before the impulse, isolated from the linear IR). - Specular reflection spikes (1–25 ms): Sharp, discrete spikes standing clearly above the surrounding floor indicate coherent first- or second-order reflections from rigid boundaries (desk, sidewalls, ceiling, floor, front/rear walls). In critical listening rooms, early reflections within the first 15–20 ms should ideally sit at least −15 to −20 dB below the direct sound (ITU-R BS.1116-3 calls for ≤−10 dB over 1–8 kHz within 15 ms, with professional control rooms targeting −15 to −20 dB).
- Path-length triangulation: Every millisecond of delay equals
0.343 m(1.125 ft) of excess acoustic path over the direct speaker-to-mic line. A spike at4.0 mslies on the ellipsoid of constant path-length difference1.37 m; checking the specular mirror point on sidewalls, ceiling, floor, and console surfaces immediately identifies the offending plane. - Filtered IR & Schroeder integral: Broadband ETC is dominated by high-frequency energy because tweeters produce narrow impulse peaks. Use Filtered IR (octave or 1/3-octave) to inspect band-limited ETC spikes and the backward-integrated Schroeder curve. Look for a clean, linear decay slope down to the noise-floor knee—if the Schroeder curve turns abruptly horizontal before −35 dB,
T30is noise-corrupted.
B. RT60 and RT60 Decay (T60M)
Strictly defined, RT60 is the time required for sound energy in a diffuse field to decay by 60 dB after source cessation. Because acoustic measurements rarely achieve 70+ dB of usable dynamic range across every band, REW computes backwards-integrated Schroeder decay curves (with Lundeby noise-floor estimation) and extrapolates linear regression slopes to 60 dB:
- EDT (Early Decay Time): Extrapolated from 0 to −10 dB. Because human perception of reverberance weights the initial drop heavily, EDT correlates more strongly with subjective "liveness" than T30. If
EDT << T30, the listening position is dominated by direct sound (nearfield geometry or high-directivity horn). IfEDT > T30, strong early reflections are propping up the first 10 dB before decay accelerates. - T20 & T30: Extrapolated from −5 to −25 dB (×3) and −5 to −35 dB (×2) respectively. Skipping the initial 5 dB eliminates direct-sound spike bias.
T30requires at least 45 dB of SNR in-band (10 dB margin below −35 dB) to avoid noise-tail flattening. - Topt (Optimized RT60): REW selects the start point (based on the direct-to-reverberant intersection) and end point (above the Lundeby noise knee) dynamically to maximize linear fit. In small domestic and studio rooms (<200 m³),
Toptis routinely more reliable thanT20orT30. - Curvature & Correlation: High curvature (typically >10% deviation between T20 and T30) exposes double-slope decay—caused by coupled rooms (open hallway/alcove), flutter echo between untreated parallel surfaces, or two closely spaced modes beating against each other. Low correlation (
r²dropping well below 0.95) flags an unreliable Schroeder fit. - RT60 Decay (
T60M): Instead of wide 1/3-octave time-domain bandpass filters (which smear individual modes and ring at low frequencies), REW's RT60 Decay panel fits an exponential decay + noise-floor model directly in the frequency domain. OverlayT60Mon the RT60 plot to inspect high-resolution modal decay times without filter-bandwidth bias.
Target Reverberation Guidelines & The Small-Room Cutoff
As baseline design targets from the REW reference guidelines: small rooms <50 m³ typically target around 0.3 s, while larger rooms up to 200 m³ target 0.4–0.6 s, with decay remaining fairly uniform across frequency (a mild, smooth rise below 200 Hz is normal and preferred over an overly dead, dull top end caused by thin foam). Crucially, statistical RT60 is not physically meaningful below a few hundred Hz in small domestic rooms where discrete standing waves—not a diffuse sound field—govern decay. Below the Schroeder frequency, read T60M, Burst Decay, Wavelet Spectrogram, and Modal Analysis instead of octave-band T30.
C. Clarity Metrics: C50, C80, D50, and TS
Clarity views quantify the balance between useful early-arriving energy (direct sound plus early integrating reflections) and late reverberant masking energy. Unlike T30, which discards the direct sound and ignores level balance, Clarity directly reflects both room decay and listening distance / speaker directivity.
| Metric | Boundary / Formula | Physical Meaning | Professional Interpretation & Benchmarks |
|---|---|---|---|
| C50 (dB) | 10 log10(E[0–50 ms] / E[>50 ms]) |
Early-to-late ratio for speech intelligibility | 0 dB = equal early/late energy. Studio control rooms & dialogue stages typically read > +6 to +12 dB across 500 Hz–4 kHz. Below +2 dB signals noticeable vocal smearing. |
| C80 (dB) | 10 log10(E[0–80 ms] / E[>80 ms]) |
Early-to-late ratio for musical clarity / transparency | Concert halls target −2 to +2 dB; small mastering/hi-fi rooms typically read +6 to +15 dB. A sharp dip in C80 at 125–250 Hz flags slow-decaying upper-bass overhang even when mid/HF RT60 looks flat. |
| D50 (%) | (E[0–50 ms] / E[0–∞]) × 100% |
Definition (percentage of total energy arriving within 50 ms) | Algebraically locked to C50: 50% = 0 dB, 80% = +6.0 dB, 90% = +9.5 dB. High speech intelligibility generally requires D50 > 50–65%; studio mixing positions typically exceed 80–90%. |
| TS (ms) | ∫ t·h²(t)dt / ∫ h²(t)dt |
Center Time (energy-weighted mean arrival time) | Immune to the sharp 50 ms / 80 ms step-function boundary artifact. Lower TS = tighter, clearer sound (<25–35 ms in tight control rooms; 60–120 ms in concert halls). |
The t=0 Trap in Clarity Measurements
Because C50, C80, and D50 integrate starting at t=0, a shifted time origin (for example, leaving 15 ms of acoustic propagation delay unremoved) truncates the early integration window from 50 ms down to 35 ms—artificially depressing Clarity across every band. Always verify t=0 sits accurately on the IR arrival before citing Clarity numbers.
D. Spectral Decay (CSD), Waterfall (Fourier & Burst Decay) & Spectrogram
Whereas RT60 reduces each band to a single scalar number, time-frequency decay surfaces expose exact modal frequencies, non-exponential beating, and energy storage across time. Understanding each mode's mathematical windowing prevents misdiagnosis:
- Spectral Decay & CSD Mode: Plots 2D frequency-response slices evaluated at successive time offsets into the IR (e.g., 0 ms, 40 ms, 80 ms, 160 ms). In standard sliding-window mode, both left and right window edges advance together. In CSD (Cumulative Spectral Decay) mode, REW anchors the right window and moves only the left window forward in time. CSD eliminates shifting truncation-ripple patterns across slices, though because the remaining window narrows on later slices, very-low-frequency resolution gradually coarsens toward the back of the plot.
- Waterfall — Fourier Mode: A 3D stack of windowed FFT slices across milliseconds. Its critical vulnerability is the time-frequency uncertainty tradeoff: a 500 ms window resolves closely spaced bass modes cleanly along the frequency axis, but smears rapid energy drops along the time axis (making even an instantaneous delta pulse appear to "decay" over the window rise/length). Conversely, a 100 ms window tracks fast decay sharply but blurs frequencies below ~50 Hz into wide humps. Always inspect the Window and Rise Time settings before judging a Fourier waterfall.
-
Waterfall — Burst Decay Mode: Replaces the millisecond time axis with a periods (cycles) axis using constant-percentage-bandwidth (constant-Q) filtering. For any damped second-order resonance, the number of cycles required to decay by 60 dB is
N60 ≈ 2.2 × Q(orN30 ≈ 1.1 × Qfor a 30 dB span)—completely independent of frequency. On a millisecond waterfall, a 45 Hz mode looks 3× slower than a 135 Hz mode of identical Q; on a Burst Decay plot, both extend the exact same number of periods, exposing true high-Q ringing immediately. - Spectrogram — Fourier, Wavelet, Airy CWT, Morlet CWT & Burst Decay: A top-down time-frequency color map (typically frequency on X, time on Y, amplitude as color). While fixed-window Fourier mode suffers the same single-window compromise as a Fourier waterfall, REW's Wavelet, Morlet CWT, and Airy CWT modes scale the analysis window inversely with frequency—yielding sharp millisecond timing at 2 kHz (to spot early reflections and driver group delay) simultaneously with high frequency selectivity at 40 Hz (to track narrow modal tails). Enabling the Peak Energy Time trace overlays effective low-frequency group delay right on top of the modal overhang.
E. Cross-View Diagnostic Matrix: Translating Signatures into Action
Use the table below to synthesize multi-view evidence into an unambiguous physical diagnosis rather than reacting to any single trace in isolation.
| Observed Multi-View Signature | Underlying Acoustic Mechanism | Correct Engineering Intervention |
|---|---|---|
Narrow SPL peak + long ridge in Burst Decay (>15 cycles) + elevated T60M + matching room-dimension pole in Modal Analysis |
High-Q axial or tangential room mode (typically minimum phase at pressure antinodes) | Primary: corner/pressure-zone bass trapping (deep porous or tuned membrane) or multi-sub cancellation. Secondary: targeted cut or Modal EQ filter (matching target T60) verified across seats. |
| Deep narrow SPL notch (−12 dB) + rapid decay in Spectrogram/Waterfall + excess group delay spike | Non-minimum-phase destructive cancellation (SBIR from front/sidewall or standing-wave node) | Do not boost with EQ (wastes headroom, adds ringing, leaves null intact). Reposition speaker/sub, add broadband absorption at the canceling boundary, or deploy a second time-aligned sub. |
Uniform T30 (≈0.30 s) at 1–4 kHz, but EDT > T30, depressed C50, and discrete ETC spikes >−12 dB at 2–10 ms |
Strong first-order specular reflections (desk, sidewalls, ceiling) causing comb filtering and image shift | Treat exact mirror points identified by Δt × 0.343 m using ≥100 mm broadband porous panels (or re-angle desk/monitors). Avoid blanket thin foam that kills HF RT60 without touching 250–800 Hz reflections. |
High T20/T30 curvature (>15% divergence), non-linear Schroeder slope, and periodic evenly spaced ETC spikes |
Flutter echo between parallel reflective surfaces or coupled reverberant space (adjacent hallway/alcove) | Break parallelism with scattering/diffusion or treat at least one opposing surface in the flutter axis; isolate or damp coupled cavities. |
Elevated T30 across all bands with low correlation below 125 Hz, while T20/Topt are significantly shorter and Schroeder bends flat at −28 dB |
Insufficient measurement SNR (noise floor contaminating late Schroeder tail) rather than actual room reverb | Re-measure with 512k/1M sweep at higher SPL, silence HVAC, check input gain/headroom, and rely on Topt + T60M. |
Broadband high RT60 (>0.6 s in a <50 m³ room) across 250 Hz–8 kHz with depressed C50/C80 and slow Spectrogram fade everywhere |
Under-damped reflective room (insufficient total Sabine absorption area) | Increase broadband absorption coverage evenly across frequency (deep porous panels + rear-wall bass trapping/diffusion) while monitoring HF decay so top octave does not collapse below midband RT60. |
Topt hovering smoothly at 0.25–0.35 s from 250 Hz–8 kHz with no >2:1 bass-to-mid ratio; C50 > +8 dB; Burst Decay clearing within 10–14 periods across modal bands; Schroeder correlation >0.97 above 200 Hz.