← Back to all lessons · Next: Transfer Function & Phase

Lesson 1 — Signal, Spectrum & Resolution

Purpose. Understanding the physical limits that determine what any single REW measurement can actually resolve. Sample-rate caps, FFT bin width, and window duration set the maximum frequency you can measure and the smallest frequency detail you can resolve. Knowing these before setup prevents wasting capture time on unusable data.

Step-by-step workflow

  1. Set sample rate before measuring. In Preferences → Signal Generator or Metering, confirm the sample rate matches the recording device. For low-frequency modal work in a large room, 48 kHz (Nyquist 24 kHz) is sufficient. 44.1 kHz gives a slightly longer maximum useful low-frequency window; 96 kHz doubles the Nyquist and is rarely needed for room acoustics. The highest capture frequency is always sample rate ÷ 2.
  2. Choose FFT size deliberately. The frequency resolution of every spectrum display is sample rate ÷ FFT length. At 48 kHz with a 65536-point FFT: 0.73 Hz bin width. At 48 kHz with 8192 points: 5.86 Hz. Shorter FFTs are faster in real time but coarser at low frequencies. For RTA work where ⅔-octave grouping applies anyway, 4096 or 8192 points is adequate. For capturing sub-bass modal detail, use 65536 or higher.
  3. Check bit depth and full-scale (FS) headroom. In the Captured or real-time meter view, ensure the highest peak is well below 0 dBFS. A signal clipped at full scale corrupts the entire measurement. A conservative target: peak signal at −12 to −6 dBFS, leaving margin so that the sweep's amplitude modulation does not clip. If the captured waveform shows flat-top peaks, reduce input gain or lower sweep amplitude.
  4. Set window position on the captured impulse response. Open the captured view. The left cursor sits before the direct-sound arrival; the right cursor sits after the last relevant decay tail. The window length directly sets your lowest credible frequency: resolution ≈ 1 ÷ window duration. A 10 ms window gives a useful floor around 100 Hz. For measuring the 40–80 Hz band of a small treated room, extend the right cursor to at least 50–100 ms. For a large hall measuring 20 Hz, you may need 500 ms+ and accept the resolution cost.
  5. Use Periodic Noise or Log Sweep based on your need for full transfer function. If you only need a broadband RTA view of the room for placement or treatment decisions, periodic noise + fractional-octave RTA is sufficient and faster. If you need the full transfer function (magnitude AND phase, valid over a continuous band), log sweep is required. This distinction is revisited in Lesson 4.
What the graphs mean.
Captured (unprocessed) view: The raw waveform of the last signal. A clean sinusoidal sweep that never exceeds full scale indicates valid capture. Clipped edges mean corrupted data; the measurement cannot be trusted regardless of EQ or smoothing applied afterward. A latency curve plotted on top shows the system delay between playback and capture — any step or discontinuity in the delay curve indicates a problem (gain staging, DAW routing, interface buffer-size change mid-session, or an unstable timing reference).

Frequency response (magnitude): dB above/below full scale per frequency. The smooth low-frequency region is dominated by room modes; sharp mid-band ripples are typically standing waves, diffraction, or speaker resonance. A dip to −20 dB or below at low frequency is a cancellation event — EQ cannot restore it (Lesson 2 and 4).

RTA view (fractional octave): Groups multiple FFT bins into ⅓-octave or ⅙-octave bands. Smoother than the FFT view and more perceptually meaningful. It is not a true room measurement; it combines direct and reflected energy. Use it for placement and live treatment decisions, not for EQ derivation.

Nyquist theorem. A signal sampled at f_s Hz can be faithfully reproduced up to f_s⁄2. Frequencies at or above f_s⁄2 fold (alias) back below it. At 48 kHz, a 30 kHz tone aliases to 18 kHz. An audio interface's analog anti-aliasing filter must attenuate content above roughly half the Nyquist rate (24 kHz) well before the digital sampling clock. Most interfaces are well-specified, but a cheap interface with a poor AAF will produce spectral artifacts that appear in the FFT as spurious low-frequency peaks. If you see repeating "ringing" structures at specific frequencies in a capture, suspect AAF distortion before suspecting the room. The remedy is increasing sample rate (48 → 96 kHz) which raises the Nyquist cutoff and eases the AAF's job — but only if the interface's AAF bandwidth permits it. A 96 kHz interface may still have an AAF that rolls off at 38 kHz; a 48 kHz interface with a good AAF is often more faithful at 40 kHz than a cheap 96 kHz unit. Always verify with the Captured view before trusting a frequency response plot.

Windowing and frequency resolution — why 10 ms ≠ 100 Hz simply. A rectangular window of duration Δt has a first null at 1/Δt in frequency domain. But the usable resolution (first null from centre, i.e. the half-width of the main lobe) is 1/(2Δt) for rectangular, roughly 1.1/Δt for Hann, etc. In REW the "window duration" in the window dialog refers to the total selected width (left + right), so for a symmetric window of total 10 ms, the resolution is approximately 100 Hz. The frequency-dependent window (FDW) used in REW adjusts width per frequency bin: at high frequencies the window is short (high time resolution, low time-domain information), at low frequencies it is long (low time resolution, good frequency detail). FDW is the default and is correct for almost all room measurements. If you disable FDW and use a flat rectangular window across all frequencies, your low-frequency resolution improves but your high-frequency resolution degrades significantly (you will lose phase detail above ~2 kHz), which is almost never desirable for speaker or system measurement. For room-only work at low frequencies, FDW or a simple rectangular window with sufficient total length both work; for measuring a loudspeaker's own frequency and phase response, keep FDW on.

When to use which FFT size in practice. RTA live view: 4096 points (48 kHz → 11.7 Hz bin width, but RTA groups into ⅓-octave anyway). Room mode identification (40–120 Hz): 65536 points (0.73 Hz/bin). Subwoofer alignment measurement (sub to main, low-freq focus): 65536 or 131072. Full-range speaker measurement (20 Hz – 20 kHz) with smooth phase: 65536 points is typically sufficient. RTA at 48 kHz and 4096 points groups into ⅓-octave bins which average over ~100 bins at low frequencies; you do not need a 65536-point FFT for live EQ decisions — the RTA smoothing already integrates out individual bin noise. This is a key reason RTA is useful: it is robust to low averaging counts because the octaves themselves average the spectrum.

Q1: At 48 kHz with a 10 ms total window, what is the approximate lowest resolvable frequency?
A1: Approximately 100 Hz (1 ÷ 0.010 s).

Q2: You are measuring a 30 Hz mode in a room. Why is a 10 ms window insufficient?
A2: 10 ms gives a 100 Hz resolution floor; the 30 Hz peak cannot be represented in a bin narrower than ~100 Hz, so it would appear smeared or absent in the frequency response. Use a ≥ 35 ms window (33 Hz or better) for reliable sub-40 Hz measurement.

Q3: The captured view shows a clean sine sweep but the RTA shows a "hole" at 4 kHz. What is the first thing to check?
A3: Whether the hole is a standing-wave cancellation (frequency stays stable when mic moves slightly) versus an AAF/resonance artifact (does not move with position). AAF artifacts in a capture would also show up as non-sinusoidal features in the Captured waveform; a room null would show consistent peak/dip structure in the ETC.

→ Continue to Lesson 2: Transfer Function & Phase Analysis