Purpose
REW's primary method is a logarithmic sine sweep → FFT → transfer function + impulse response. The sweep gives 30-40dB better S/N than an MLS click and critically separates harmonic distortion to negative time in the IR. Everything else — SPL, phase, waterfall, RT60 — is a derived view of that one complex dataset.
Workflow — The 7 Concepts to Lock In
- Signal vs System: RTA shows signal spectrum (source+room combined). Sweep measurement computes the system's transfer function. They are not interchangeable. RTA cannot separate direct from reflected.
- Sweep = Both Domains: One log sweep yields frequency response and impulse response. TF = FFT(IR).
- Distortion Separation: Because sweep frequency rises log-wise, harmonics appear at predictable negative-time offsets. REW windows them out of the linear IR.
- Resolution Tradeoff: FFT resolution = sampleRate / FFT length. Longer FFT = finer low-freq resolution but worse time resolution. Same tradeoff appears in windowing.
- Validity Scope: A measurement is valid only for that exact mic position, source position, level, cal files, window, and processing state. Move the mic 15cm and HF above 2kHz is a different measurement.
- Phase Matters: Two systems can have identical |H(f)| but completely different phase/group-delay and sum differently with a sub.
- EQ Limits: EQ can tame minimum-phase peaks (often modal). It cannot fill non-minimum-phase dips (cancellation). Boosting them just burns headroom.
Graph Interpretation
SPL & Phase
Magnitude + phase = complete TF. Look for broad trends before narrow wiggles. Smoothing hides modal truth — use variable smoothingVariable smoothing: ~1/48 oct at LF, ~1/3 oct at HF. Preserves modal detail. for EQ work.
Impulse + ETC
First large spike = direct sound. Later spikes = reflections. ETC (envelope) makes them obvious. Distance = Δt × 343 m/s.
Waterfall / Spectrogram
Shows decay. A room mode is a horizontal ridge that persists. RT60 is derived from this via Schroeder integral.
RTA (Live)
Useful for live pink PN tweaking. Cannot window out room. Requires averaging; periodic noise avoids windowing artifacts.
Deeper Detail — Convolution, Minimum Phase & Why Dips Don't EQ
Convolution: Output = Input * IR (convolved). TF describes that convolution in frequency domain: Y(f) = X(f) · H(f). That's why IR and TF are duals.
Minimum Phase: For a given |H(f)|, the minimum-phase version has the smallest possible group delay and is invertible. Drivers and crossovers are often near-minimum-phase. Rooms are mixed-phase: some bands are invertible (peaks), some aren't (sharp dips from cancellation). REW's Generate Minimum Phase lets you compare measured phase to minimum phase. Flat excess group delay = minimum-phase region = EQ-able.
Room Modes: Axial modes at f = c/2 × (nx/Lx). They ring (high T60) and are often minimum-phase at the peak — hence cut filters work. But the adjacent dip is interference — boosting it just drives the mode harder.
Log Sweep Trick: Distortion products are at multiples of instantaneous frequency, so they arrive "early" in the deconvolved IR (negative time). REW excludes them from the linear IR calculation.
Check Your Understanding
You measure a sharp -18dB dip at 110Hz that doesn't move when you EQ +9dB into it. Excess group delay spikes at that frequency. What's happening?