← Back · Lesson 1 · Lesson 3

Lesson 2 — Transfer Function & Phase Analysis

Purpose. The frequency response alone is an incomplete description of a system. Two speakers (or two room positions) can show nearly identical magnitude responses but have radically different phase behaviour, causing different timbre, different perceived size, and different subwoofer interaction. This lesson teaches you to read the full transfer function — magnitude, phase, group delay, excess phase — and to make correct analytical decisions about which regions can be EQ-corrected and which cannot.

Step-by-step workflow

  1. Always capture the transfer function, not just the magnitude. In REW, run the measurement using Sweep → Save (not RTA), then open the graph. The default view shows magnitude and phase on separate windows or the Phase tab. Enable the Group Delay and Excess Phase graphs (via Graphs → Add or by clicking the tab bar). Do not make EQ or treatment decisions from magnitude alone; the phase and group-delay information determines whether a correction is physically possible.
  2. Identify minimum-phase regions. In the Excess Phase graph (measured phase minus minimum-phase prediction), read the value at each frequency. A value of 0° means that frequency band is minimum phase. A value of −180° means that band is exactly one half-wave non-minimum phase. A region where excess phase is flat (constant slope) is minimum phase and that region can be inverted by EQ. A region where excess phase is actively decreasing with frequency is non-minimum phase and cannot be fully corrected by EQ — any EQ boost applied there will increase the total delay in that region and reduce the perceived clarity. Peaks in the magnitude response are typically minimum phase; EQ can tame them. Sharp dips are often non-minimum phase and are not fixable by boost — they represent cancellation events between direct and reflected sound or between driver interactions.
  3. Read group delay to understand temporal behaviour. Group delay is the slope of the phase curve. A flat group-delay curve over a band means all frequencies in that band experience the same delay — the band is minimum phase. A rising or falling group delay over a band means different frequencies arrive at different times — a time distortion. In a room, group delay varies with position (the direct sound path changes). In a loudspeaker, group delay peaks at driver crossover resonances and in the baffle region. A high-frequency group delay of 1–3 ms above 5 kHz is normal for a single driver; above 10 ms at any frequency above 1 kHz is a sign of a serious problem (severely non-minimum phase or measurement window error).
  4. Use the Pole-Zero plot for modal diagnosis. REW's Pole-Zero graph plots the real/imaginary complex-plane poles and zeroes of the room response. Room modes appear as pole pairs; each pole's position in the plane (real part = damping, imaginary part = frequency) tells you the decay time of that mode. A poorly damped mode (pole close to the real axis) will produce a long-ringing, sharp peak. A well-damped mode (pole farther from the real axis) has a shorter ring time. A zero in the plane corresponds to a null in the frequency response. Matching a zero with an EQ peak (at the same frequency, in the right position) does not cancel the mode (this is a common misconception); instead, place acoustic treatment or adjust the subwoofer position so that the speaker's own pole and the room's mode are less aligned. See Lesson 4 for practical treatment. Never use EQ to "chase" a pole — EQ is not a pole shifter. Instead, EQ can smooth the magnitude effect of a peak that you cannot treat, which is acceptable for broad peaks but aggressive at narrow Q modes.
  5. Build the EQ from minimum-phase regions only. When designing a parametric EQ in REW (Equalizer → New): (1) for each significant peak, check the corresponding excess-phase value. If it is in a minimum-phase region (excess ≈ 0°, or excess group delay ≈ 0°), you can apply a notch or gain-cut Q matching the peak width. For a broad peak in a 60 Hz room mode (Q ~2), apply a 6 dB cut at 60 Hz with Q 1.5–2.0. For a narrow Q ~7 peak at 82 Hz, a 3–6 dB cut at Q ~4–6 is appropriate. (2) For dips: if the dip is less than −6 dB and the surrounding excess phase indicates a minimum-phase region, a small boost (≤ +3 dB) is acceptable. If the dip is wider than −10 dB or the phase indicates non-minimum phase behaviour (excess phase decreasing at that frequency), a boost is wrong — it will amplify the cancellation and waste headroom. Correct dips at the source (placement, room treatment, source/sub alignment). (3) Apply smoothing to the final response; REW's built-in smoothing (1–3 Hz or 1–3 Hz in frequency) helps reveal the real trends and hides measurement noise. Do not use smoothing as a reason to ignore a real dip — it only helps you see the structure behind the noise.
What the graphs mean.
Magnitude (dB vs. frequency): The system's amplitude gain per frequency. Peaks = resonance; dips = cancellation. The 20 Hz–20 kHz range shows what a system passes. A flat, smooth response (±1–2 dB) is excellent for critical listening; ±3 dB is acceptable; ±5 dB is a clear problem that should be addressed. The Captured graph (not the processed response) shows the signal as captured before any EQ; it must be clean before any correction is applied.

Phase (degrees vs. frequency): The angular shift of each frequency component. Phase is only meaningful for a single source-and-mic pair; two measurements with the same magnitude can have phase offsets of 180° that are irrelevant to timbre but relevant to coherence with other channels or speakers. The absolute phase value is arbitrary; only the shape and slope (group delay) and the minimum-phase deviation (excess phase) carry diagnostic value. The phase curve is discontinuous at 360° boundaries; REW plots it modulo 360°. If the phase curve jumps by 360° mid-band, that is a normal wrap, not an error.

Group delay (ms vs. frequency): The time offset of each frequency. A flat group delay curve = time-coherent signal (all frequencies arrive together — minimum phase). A rising group delay with frequency = later arrival of higher frequencies (a common room/position artifact when reflections dominate at HF). A narrow spike in group delay at a particular frequency = the phase slope at that frequency is steep — a resonance or cancellation event. A group delay of 0 ms means the signal was measured at the reference point (or with zero window). A measured group delay of 100 ms across the full range means either the system genuinely delays 100 ms (unlikely except in very long pipes) or your window is misconfigured — check the left-window cursor (it should be before the direct-sound peak; moving it earlier adds artificial delay).

Excess phase (degrees vs. frequency): Measured phase minus the minimum-phase prediction. 0° = minimum phase (correctable by EQ). Decreasing excess phase (going more negative with frequency) = non-minimum phase (not fully correctable). A step change in excess phase at a specific frequency = a pole-zero at that frequency. A slowly varying excess phase = a broad non-minimum phase region. This is the most diagnostic graph for deciding "can I EQ this region or not?"

The Borel-Cartheodorus / minimum-phase theorem, in practical terms. If a causal, stable, minimum-phase system has a magnitude response in some frequency band, its phase is uniquely determined. This means that if you know the magnitude and the system is minimum phase, there exists an EQ filter that, when applied, will bring the output to zero (or to your target). This is the theoretical basis for EQ-based correction of a non-minimum phase response — it is only valid when the system is minimum phase in the relevant band.

In the non-minimum-phase case, there is no filter in the forward direction that can produce zero at the cancellation frequency without also producing zero at some other frequency or in the time domain becoming non-causal. A room null at 55 Hz is created by destructive interference: the direct sound arrives with +180° relative to the reflection. If you boost 55 Hz with EQ, you are amplifying the signal before the cancellation occurs, so the cancellation still happens — in fact, boosting 55 Hz will increase the level of the direct sound arriving at the mic at a different phase than the reflection, but it will also increase the level of the direct sound, which may not help if the dip is in a reflection-dominated region. More concretely: if the dip at 55 Hz is 12 dB and is caused by a −180° cancellation between the direct sound and a strong reflection, boosting 55 Hz by +12 dB in the signal input does not restore 0 dB at the listening position because the reflected signal at 55 Hz is still −6 dB below the direct sound and still at 180° phase. The boost amplifies both signals by +12 dB (the direct and the reflection both get boosted), so their relative phase and relative amplitude do not change. The dip remains at 12 dB. The only way to reduce the dip is to change the relative phase or amplitude of the reflection — by moving the speaker, moving the listener, adding absorption in the reflection path, or changing the source level relative to the reflection (e.g. by placing the sub closer to the listener). This is why "just boost the dip" is a mistake: it wastes headroom, can distort the source, and does not improve the measured or perceived response. EQ is valid for taming resonance peaks (which are minimum-phase), not for restoring cancellation dips (which are fundamentally non-minimum-phase in the local region).

Group delay intuition for minimum vs. non-minimum phase. A minimum-phase system has the smallest possible group delay for a given magnitude. If you add a pure delay to a minimum-phase signal, its phase curves down linearly (a constant group delay) and its magnitude is unchanged. But a non-minimum-phase region introduces extra group delay that cannot be removed by EQ. A system with a 20 Hz resonance that has a +3 dB peak and a −3 dB dip 40 Hz later will have a non-minimum phase region between 20 and 40 Hz. Any EQ applied to reduce the +3 dB peak will reduce the magnitude at 20 Hz but will increase the group delay in the 20–40 Hz band, meaning the time delay of the 25 Hz component increases relative to the 100 Hz component — this is audible as a loss of "snap" in the low-mid range, and it worsens subwoofer-main interaction because the sub's own 30 Hz peak and the room's 25 Hz region now have a time offset that was not present originally. This is the subtle cost of EQ-correcting a minimum-phase peak that is adjacent to a non-minimum-phase region — it is not zero-cost.

Q1: You measure a 20 Hz peak of +18 dB. The excess phase at 20 Hz is 0° and flat between 15 and 25 Hz. What EQ correction is appropriate and why?
A1: Apply a parametric notch of roughly 6–12 dB at 20 Hz with Q ≈ 2–3. The excess phase is 0° (minimum phase) so an EQ will reduce the peak without introducing time delay. A 6–12 dB reduction is appropriate because the peak is resonance, not a cancellation event. The Q should match the modal bandwidth — a room mode at 20 Hz in a typical room is Q ≈ 2–3, so the EQ Q should be similar to avoid over-correcting at neighboring frequencies.

Q2: You measure a 40 Hz dip of −15 dB. The excess phase at 40 Hz is −90° and decreasing from −60° at 30 Hz to −120° at 50 Hz. Can you fix this with EQ? Why or why not?
A2: No. The excess phase is actively decreasing with frequency — this is a non-minimum-phase region. The dip is a cancellation event. EQ will not restore it and any boost will waste headroom without improving the measured response at the listening position. Correct the cause: move the subwoofer, add bass-trap absorption in the reflection path, or adjust the room geometry (speaker placement relative to a corner or wall). EQ can supplement but not replace positional or treatment-based correction for a non-minimum-phase dip.

Q3: The group delay graph shows a flat 1 ms delay from 20 Hz to 1 kHz, then a sharp spike to 8 ms at 3 kHz, then back to 1 ms by 5 kHz. What does this tell you, and what is the first thing to check in REW?
A3: The spike at 3 kHz corresponds to a steep phase slope — a resonance or cancellation event near 3 kHz. The first thing to check is the window: a too-short right-window (e.g. a 5 ms total window when the reflection tail extends to 30 ms) will exclude reflections from the FFT and shift the apparent group delay. Extend the right cursor to at least 10 ms and re-read. If the group delay still spikes after window adjustment, the spike is a genuine 3 kHz resonance (likely a baffle interaction, driver resonance, or a room standing wave at 3 kHz). It may not require correction if the peak in the magnitude is less than 2 dB; otherwise, if the corresponding excess phase is in a minimum-phase region (flat), a small EQ cut is appropriate. If the phase indicates non-minimum-phase behaviour (excess phase decreasing at 3 kHz), the 3 kHz problem is a cancellation or baffle interaction and should be treated at the source (speaker placement, baffle edge treatment, or a room-mode correction) rather than by EQ.

→ Continue to Lesson 3: Impulse Response & Windowing