--> Lesson 2 — Impulse Response and Windowing | REW Pro Reference

Module 02 · Time–Frequency Analysis & Transfer-Function Isolation

Lesson 2 — Impulse Response and Windowing

Every frequency-response curve REW displays is the FFT of a time-gated impulse response. Master log-sweep deconvolution, negative-time distortion separation, asymmetric Left/Right window geometry, Frequency-Dependent Windowing (FDW), and the exact spectral validity boundaries that govern what you can legitimately diagnose or equalize.

Audience: Pro / Practitioner REW Views: Impulse · IR Windows · SPL & Phase · Captured Core Math: Δf ≈ 1 / Twindow FDW Basis: 96 PPO Cycle / Octave Gating

01 What This Is For

A logarithmic sine sweep in REW does not directly "measure a frequency response"—it excites the electro-acoustic path across frequency over time, captures the convolved pressure/voltage waveform, and deconvolves it via FFT processing into the complete complex transfer function (magnitude and phase) and its inverse FFT, the impulse response (IR). That single IR contains four physically distinct phenomena stacked sequentially along the time axis: harmonic distortion artifacts at negative time (t < 0), the direct sound arrival around t = 0, discrete early boundary reflections in the first few to tens of milliseconds, and late diffuse reverberation plus low-frequency modal ringing extending hundreds of milliseconds onward.

For a professional acoustician, loudspeaker designer, or calibrator, IR windowing is the primary analytical scalpel that separates those phenomena before the FFT runs. Unlike fractional-octave smoothing—which merely averages adjacent frequency bins after room reflections have already corrupted magnitude and phase—windowing physically gates time-domain arrivals before transformation. Without deliberate window discipline, you will routinely mistake a 4 ms console or floor comb-filter null for a loudspeaker crossover error, trust invalid sub-resolution magnitude ripples below 1 / Tright, or silently corrupt downstream calculations in Step Response, Minimum Phase, Trace Arithmetic, and EQ matching.

Strict Scope of Measurement Validity

Both the IR and its derived transfer function are valid only for one exact source position, microphone position, SPL/drive level, sample rate, bit depth, calibration file set, DSP/EQ state, and active IR window configuration. Changing the window does not alter the captured data, but it completely redefines what physical system—anechoic driver, early-field monitor, or steady-state room—the frequency and phase curves represent.

02 Steps — Concrete REW Workflow

Execute this sequence whenever capturing and dissecting a measurement where separating direct radiation, early reflections, and modal decay matters.

  1. Configure Sweep Length, Bandwidth, and Timing Reference for Clean Deconvolution

    Press Cmd+M (macOS) or Ctrl+M (Windows) to open Measure in SPL mode. Set sample rate to 44.1 kHz or 48 kHz for acoustic work (higher rates consume memory/CPU without improving acoustic accuracy below 20 kHz). Remember that REW's sweep actually traverses from half the Start Freq to twice the End Freq to ensure clean band-edge deconvolution.

    • Select 256k (default) or 512k / 1M sweep length: each doubling yields roughly +3 dB S/N and doubles the negative-time separation between t = 0 and harmonic distortion peaks.
    • Set drive level around -12 dBFS (or level established in Check Levels for ~75 dB SPL), verifying input headroom remains green (6–40 dB) with analog input RMS around -30 to -12 dBFS.
    • Enable a Loopback or Acoustic timing reference whenever comparing phase, delay, or performing Trace Arithmetic / Alignment Tool work across measurements. Choose Set t=0 at IR peak for full-range speakers, or IR start / delay estimate for bandwidth-limited subwoofers with broad peaks.
  2. Audit the Raw Capture Before Trusting Any Derived Curve

    Immediately after the sweep completes, inspect the Captured graph (which is non-live and unavailable once a measurement is closed/reloaded) and Measurement Info. Confirm there is zero input clipping (signals exceeding 0 dBFS permanently corrupt deconvolution), confirm clean sweep start/end margins, and verify sample rate, system delay, and active mic/soundcard calibration files.

  3. Inspect IR Anatomy, Negative-Time Distortion, and Early Reflections on ETC

    Switch to the Impulse graph. Toggle between linear %FS, logarithmic dBFS, and ETC (Energy-Time Curve). In the graph controls, set ETC smoothing to 0 ms so closely spaced boundary reflections are not blurred together.

    • Zoom to -50 ms through +50 ms around t = 0. Locate the earliest reflection spike on the ETC; note its arrival offset Δt (ms) and compute its excess acoustic path length: Δd = c · Δt (at 20 °C, 0.343 m/ms).
    • Zoom left into negative time (t < 0) in dBFS view to verify where the 2nd and 3rd harmonic distortion impulses sit relative to your intended Left Window boundary.
  4. Open IR Windows and Define Asymmetric Left & Right Windows

    Click the IR Windows toolbar button. Enable the Window indicators and Windowed trace overlay on the Impulse graph so you can visually verify exactly what energy survives the taper.

    • Left Window (pre-peak): Keep short—typically 1–3 ms for quasi-anechoic gating, or up to 10–125 ms only when accommodating linear-phase FIR pre-ringing or very low-frequency sub onset—never extending into negative-time distortion bursts.
    • Right Window (post-peak): For quasi-anechoic direct-sound gating, place the window end just before the earliest reflection spike on the ETC (e.g., 3.5 ms if a desk bounce hits at 3.9 ms). For full-room/modal analysis, extend to 500 ms (or longer for large reverberant spaces, adjusting IR truncation in Analysis Preferences if needed).
    • Window Type: Default to Tukey 0.25 for asymmetric acoustic gating—its flat top preserves direct arrival amplitude while cosine-tapering the boundary to suppress spectral leakage. Use Hann when strong energy still exists at the right edge and smooth sidelobe suppression outweighs amplitude flatness; reserve Rectangular strictly for periodic or fully decayed signals.
  5. Apply Windows and Expose the Sub-Resolution Validity Floor

    Click Apply Windows (or Apply Windows to All when comparing multi-angle spinorama/off-axis sweeps). Switch to SPL & Phase and enable Show response below window limit in graph controls. REW draws frequencies below the window resolution limit (≈ 1 / Tright) as a dashed trace—an unmistakable visual boundary separating valid spectral measurements from under-resolved low-frequency interpolation.

  6. Configure Frequency-Dependent Windowing (FDW) When Full-Bandwidth Context Is Needed

    In IR Windows, check Add frequency dependent window and specify width in cycles (e.g., 6 cycles for tight early-field/direct isolation, 15 cycles for psychoacoustic room integration) or octaves. Click Apply Windows. REW converts the response to 96 PPO log spacing and applies a window whose duration scales inversely with frequency, capped by your fixed Right Window at low frequencies.

  7. Cross-Check Downstream Dependencies Before Saving or Exporting

    Remember that Step Response (the integral of the windowed IR), Minimum Phase / Excess Phase generation, Group Delay, Trace Arithmetic (e.g., nearfield/farfield Merge), and EQ predictions all inherit the active IR window. Clone measurements (or save separate window presets via the API/workflow notes) whenever maintaining both gated quasi-anechoic and full-room modal views of the same sweep.

macOS Pro Operational Notes & API Window Automation
  • Clock & Driver Integrity: macOS uses Java audio drivers only (ASIO and WASAPI exclusive are Windows-only). Configure 24-bit depth and exact sample rate in Audio MIDI Setup before launching REW. Never use multiple-sweep synchronous averaging with a USB microphone or separate input/output devices—clock drift smears the high-frequency IR peak and ruins windowing.
  • Permissions & Controls: REW's internal volume controls are disabled on macOS; adjust gain in Audio MIDI Setup or your interface mixer. If macOS drops mic permission after an update, reset via tccutil reset Microphone before starting REW.
  • Scripted Batch Windowing via Local API: Launch REW with its localhost-only API server (default port 4735) to query or batch-update IR window parameters under the /measurements endpoints.
macOS Terminal — Launch REW with API Server Enabled (localhost:4735)
open -a REW.app --args -api

03 Reading the Output — Graph & Panel Interpretation

Windowing ties the time domain and frequency domain into a single coupled system. Reading any one REW view in isolation leads to misdiagnosis; professionals cross-read the Impulse / ETC, IR Windows, SPL & Phase, and Captured views simultaneously.

Impulse & ETC View

Time Domain

Displays the linear or logarithmic IR, the Hilbert envelope (ETC), the Step Response (integral of the windowed IR), the overlaid Left/Right Window shape, and the resulting Windowed trace.

  • ETC spikes (0 ms smoothing): Discrete early reflections appear as sharp peaks at +Δt. Multiply Δt by 0.343 m/ms to match physical boundary paths (desk, floor, sidewall, ceiling).
  • Negative time (t < 0): Contains harmonic distortion impulses separated by the log-sweep deconvolution, plus any causal pre-ringing from linear-phase DSP/FIR filters.
  • Step Response: Because REW integrates the windowed IR, truncating the right window too early visibly chops the low-frequency return tail of the step response.

SPL & Phase View

Windowed FFT

Shows magnitude and phase computed exclusively from the windowed IR. Always inspect this alongside the active window duration.

  • Solid vs. Dashed Trace: With Show response below window limit enabled, dashed segments warn that frequency resolution (≈ 1 / Tright) is insufficient to resolve true peaks/dips.
  • Windowing vs. Smoothing: Shortening the Right Window removes reflection-induced comb notches from both magnitude and phase; fractional-octave smoothing only averages already-contaminated bins.
  • Over-short window artifacts: Truncating while the direct driver/port decay is still energetic produces wavy truncation ripple and underestimates high-Q resonances.

Captured & Measurement Info

QA & Metadata

Captured displays the raw generated sweep and uncorrected input waveform from the last measurement; Measurement Info logs timing reference, system delay, sample rate, and calibration files.

  • Flat-topped sweep envelope: Indicates ADC clipping—the resulting IR will show broad noise-floor elevation and spurious pre-ringing that no window can fix.
  • Noise floor vs. tail: If the captured signal sinks into ambient noise before the IR finishes decaying, lengthening the Right Window past the noise-floor intercept only injects random noise into the FFT.

IR Windows Panel

Gate Control

Defines Left Window width/type, Right Window width/type, window reference time, and optional FDW (cycles or octaves).

  • Asymmetric design: Left window stays tight to reject pre-peak noise and negative-time distortion; right window sets the reflection gate and low-frequency validity floor.
  • No live auto-commit: Modifying values does not update graphs until you press Apply Windows (a frequent cause of misreading stale curves).

Window Parameter & Validity Reference Matrix

Window Strategy Typical REW Settings Valid Low-Freq Floor (fmin) What You Can Validly Read What Becomes Invalid / Hidden
Tight Quasi-Anechoic Gate Left: 1.5–2 ms Tukey 0.25
Right: 3.5–5 ms Tukey 0.25
FDW: Off
~200–285 Hz
(1000 / Tright)
Direct on/off-axis speaker response, baffle diffraction, tweeter/mid crossover summation, direct phase & GD above fmin. All bass/room modes, port tuning, cabinet LF rolloff, and baffle-step transition below fmin (dashed).
Early-Reflection / Midfield Gate Left: 5 ms Tukey 0.25
Right: 15–25 ms Tukey 0.25
FDW: Off
~40–67 Hz Direct sound + first-order boundary interference (SBIR, desk/floor bounce) without late reverberant wash. True steady-state modal peak amplitude if room modes ring >100 ms; sub-40 Hz accuracy.
Psychoacoustic FDW Left: 10–125 ms
Right: 500 ms
FDW: 6–15 cycles
Full bandwidth (capped by 500 ms → 2 Hz) Perceived direct+early tonal balance at mid/HF alongside high-resolution LF modal structure on one unified curve. Steady-state diffuse room power at HF (FDW attenuates late HF reverberation, tilting HF down vs unwindowed in-room SPL).
Full-Room Modal Window Left: 10–125 ms
Right: 500–1000+ ms
FDW: Off
1–2 Hz True steady-state in-room modal peaks/nulls, subwoofer-to-room coupling, LF EQ matching (pair with Variable smoothing). Direct speaker anechoic behavior at mid/HF (heavily obscured by dense comb-filtering and reverberant energy).
Interactive DSP Sandbox · Real 8192-pt Radix-2 FFT

IR Window, Negative-Time H2 & FDW Simulator

Synthetic 2-way monitor + small room IR (fs = 32,768 Hz): direct arrival at 0 ms, simulated short-sweep H2 distortion spike at -18 ms, desk bounce at +3.9 ms, sidewall at +8.6 ms, rear wall at +16.4 ms, and room modes at 52 Hz & 98 Hz. Drag the controls to see real-time time-gating, dashed sub-resolution limits, and FDW cycle narrowing.

Valid Freq Floor (1/TR) 286 Hz
Max Included Path Δd 1.20 m
Reflections Inside Gate None (Quasi-Anechoic)
FDW Gate @ 100 Hz / 4 kHz Fixed 3.5 ms
Negative-Time H2 Risk Clear (-18 ms outside)
Time Domain: Impulse / ETC Envelope (dBFS, -25 ms to +25 ms zoom) + Active Window
ETC Envelope Active Window H2 (-18 ms) Reflections
Frequency Domain: Windowed SPL Response (30 Hz – 12 kHz, log scale)
Unwindowed Room (150 ms) Active Valid (≥ fmin) Below Window Limit (Dashed)

Diagnostic Signatures — Valid vs. Misread Windowed Outputs