Module 03 · Complex Transfer Functions & Coherent Summation
Lesson 3 — Phase, Minimum Phase and Time Alignment
Magnitude is only half of the acoustic transfer function. Without valid phase data, locked timing references, clock-skew compensation, and minimum/excess-phase separation, you cannot distinguish EQ-able room modes from un-invertible cancellations—nor can you integrate multi-way speakers and subwoofers with predictable complex summation.
01 · Operational Rationale
What This Is For
Two acoustic systems can exhibit identical steady-state magnitude curves at the listening position while sounding
and summing completely differently. The complete linear time-invariant (LTI) description of a source-to-microphone
path is the complex transfer function
H(f) = |H(f)| · ejφ(f).
In professional calibration—whether aligning a studio monitor/subwoofer crossover, phase-matching a multi-subwoofer
array, diagnosing comb filtering from console/boundary reflections, or deciding where IIR EQ is physically valid—the
phase term φ(f) and its negative frequency derivative, group delay τg(f),
govern whether your corrective decisions solve the physics or merely mask one symptom while degrading time-domain
clarity and spatial consistency.
Working with phase in REW requires discipline at capture time. Unlike magnitude, phase is hypersensitive to
unreferenced buffer latency, asynchronous A/D–D/A clock drift (ubiquitous with USB measurement microphones on macOS),
and careless t=0 shifting. A 0.5 ms timing error barely alters a smoothed SPL trace, yet it rotates
phase by 180° at 1 kHz and turns an 80 Hz subwoofer alignment measurement into guesswork. Mastering REW's timing
reference architecture, minimum-phase decomposition, and complex-vector Alignment Tool converts crossover integration
and room diagnostics from trial-and-error knob turning into deterministic engineering.
- Separate invertible from non-invertible anomalies: Decomposing measured phase into minimum phase and excess phase (and reading excess group delay) tells you immediately which peaks can be cleanly flattened in both frequency and time with minimum-phase IIR PEQ, and which notches are non-minimum-phase cancellations that must be solved via alignment, placement, or acoustic treatment.
- Eliminate crossover cancellations without burning headroom: Using a locked timing reference (loopback or acoustic chirp + clock adjustment) preserves the true inter-channel phase relationship so REW's Alignment Tool can compute exact delay, polarity, and gain for coherent broadband summation.
-
What goes wrong when phase is ignored: Boosting a crossover cancellation dip or boundary SBIR
notch with EQ wastes amplifier/driver headroom, elevates harmonic distortion, rings in the waterfall, and
worsens off-axis summation. Aligning by SPL peak alone at a single frequency often lands a full 360° cycle late,
summing at
fcwhile tearing comb notches above and below the crossover. -
What goes wrong when timing references are sloppy: Measuring mains and subs without a timing
reference, moving the reference channel mid-session, running USB mics without clock-drift adjustment, or running
Estimate IR delayindependently on each trace destroys relative time-of-arrival data and invalidates both Alignment Tool predictions and vector averages.
| Engineering Problem | Primary REW Tool / View | Diagnostic Metric | Physical Decision |
|---|---|---|---|
| Can this +9 dB peak or −12 dB dip be fixed with IIR PEQ? | SPL & Phase → Generate Minimum Phase; Group Delay |
Excess Group Delay (EGD) flatness near 0 ms | Flat EGD → cut with IIR PEQ. Sharp EGD spike/reversal → non-minimum phase; move source/mic, align, or treat. |
| Subwoofer & mains show a notch through the 80 Hz crossover | All SPL → Alignment Tool (Impulse & Phase modes) |
Phase-slope match, 1/3-oct IR cross-correlation, Aligned sum |
Apply delay to earlier source (usually mains) + polarity flip if needed; validate with physical combined sweep. |
| High-frequency phase wraps erratically with a USB mic on macOS | Preferences → Analysis + Measurement Info |
Clock offset in ppm via dual acoustic timing chirps |
Enable Adjust clock with acoustic ref; verify consistent ppm across consecutive sweeps. |
| Need true absolute electro-acoustic latency of a DSP monitor | Measure with Loopback timing reference |
System delay (ms) in Measurement Info & Impulse onset |
Subtract acoustic time-of-flight (d / c) to isolate internal DSP + converter latency. |
| Vector averaging multiple measurements collapses high frequencies | All SPL → Cross corr align / Time align |
Inter-trace arrival skew and excess phase dispersion | Time-align traces before vector averaging, or use RMS + phase / spatial RMS average for multi-seat work. |
02 · Concrete REW Workflow
Steps: Phase-Valid Capture, Decomposition, and Alignment
Follow this sequence end-to-end whenever phase accuracy, group delay interpretation, or multi-source summation is the objective. Shortcuts in Steps 1–4 cannot be repaired in post-processing.
Stage A — Establish a Phase-Valid Timing Reference & Clock Lock
-
Lock audio I/O sample rate and hardware routing before launching REW.
On macOS, open Audio MIDI Setup and lock both input and output devices to the same sample rate
(48 kHz or 44.1 kHz for acoustic work) and 24-bit depth.
Launch REW, open Preferences → Soundcard, and explicitly select the named input and output
devices—never leave either set to
Default Device. Verify channel assignments and configure Output Channel Mapping if addressing a multichannel interface or HDMI receiver. -
Select the timing reference architecture matched to your hardware topology.
Open Measure (⌘M on macOS / CtrlM on Windows)
and choose one of two reference modes:
-
Loopback as timing reference (preferred when I/O shares a master clock): Physically patch a
spare interface output (
Ref output) back into a spare interface input (Ref input) on the same audio interface or word-clock-locked converter. In Preferences → Soundcard, assign the loopback input and output channels. Loopback eliminates all OS buffer, driver, and interface latency jitter, yielding true absolute electro-acoustic system delay. -
Use acoustic timing reference (required for USB mics, HDMI/AVR paths, or split I/O): Select a
dedicated, full-range reference speaker (typically the Left main tweeter—never a subwoofer or band-limited driver)
as the
Ref output. REW emits a short high-frequency chirp (~5–20 kHz) through that speaker immediately before the measurement sweep and lockst=0to its acoustic arrival at the microphone. Never change the reference speaker or move the microphone across a set of measurements you plan to align.
-
Loopback as timing reference (preferred when I/O shares a master clock): Physically patch a
spare interface output (
- Enable clock-drift compensation for asynchronous I/O (USB mics / separate devices). If your input and output run on independent quartz oscillators (e.g., UMIK-1/2 + audio interface or HDMI on macOS), open Preferences → Analysis and check Adjust clock with acoustic ref. When active, REW plays the high-frequency timing chirp both before and after each sweep, measures the exact sample count elapsed between chirps against the nominal sweep length, calculates the fractional clock offset (in ppm), and resamples the capture prior to deconvolution. Keep Number of sweeps = 1—REW disables coherent multi-sweep averaging for USB mics and split-clock devices because inter-sweep phase drift would cancel high frequencies.
-
Configure sweep bounds to protect the minimum-phase calculation, then verify timing lock.
Set Sweep length to
256k(or512kin noisy rooms for +3 dB S/N per doubling). Set sweep frequency span wider than the band you intend to analyze: for a subwoofer crossing at 80 Hz, sweep at least 15 Hz – 500 Hz (not 20–120 Hz) so the upper roll-off tail is cleanly captured for REW's Hilbert-transform minimum-phase solver. Run a test capture of your reference speaker, then inspect:- Captured graph: confirm healthy headroom (6–40 dB green) and zero clipping on both chirp and sweep.
- Measurement Info (⌘I): verify
Timing refis logged, checkSystem delay(near0.000 msfor the acoustic ref speaker itself; positived/c + DSP latencyfor loopback), and confirm a plausible, repeatableClock adjustmentvalue (typically within ±10 to ±100 ppm).
macOS uses Java audio drivers exclusively (no ASIO or WASAPI exclusive). While Java drivers allow
mixing a USB mic input with a separate DAC/HDMI output, those two devices never share a sample clock—making
Acoustic timing reference + Adjust clock with acoustic ref mandatory for valid phase work.
Remember that REW's internal volume sliders are disabled on macOS; set preamp/output gain in
Audio MIDI Setup or your interface mixer, and never load a soundcard calibration
file when measuring with a USB microphone. If macOS silently blocks mic capture after an update, quit REW and run
tccutil reset Microphone in Terminal before relaunching.
1. Loopback protection: Always verify +48V phantom power is OFF on the loopback
input channel and mute hardware direct monitoring on that channel before connecting the cable; phantom voltage can
damage line outputs and unmuted monitoring triggers an instant full-scale feedback loop.
2. Do NOT independently shift t=0 on alignment pairs: Running
Estimate IR delay → Shift IR individually on Main and Sub traces destroys their shared timing-reference
origin. Keep original captures locked to the timing reference; if you want delay-removed phase for single-speaker
inspection, operate on a copy or note the exact offset.
Stage B — Capture Sources & Decompose Minimum vs. Excess Phase
-
Capture each source individually at identical gain, sample rate, and mic position.
With the microphone fixed at the listening position and the timing reference untouched, measure each subsystem alone
(e.g.,
L Main (HPF),R Main (HPF),Sub 1 (LPF), plus an initial unaligned baseline ofL Main + Sub 1). Use identical sweep level (e.g.,-12 dBFScalibrated to ~75 dB SPL) across all captures so relative levels reflect true system balance. -
Inspect phase with transport delay removed (single-source analysis).
Select a single speaker measurement, open SPL & Phase, and ensure smoothing is set to
None or 1/48–1/24 oct (heavy smoothing smears rapid phase transitions).
If using a loopback reference, raw phase will wrap densely due to acoustic flight time
τ0. Open Controls → Estimate IR delay to read the cross-correlation delay between the measured IR and its minimum-phase equivalent. To inspect phase cleanly without corrupting an alignment source trace, either duplicate the measurement first (All SPL → Actionsor save/reload copy) or note that applying Shift IR subtracts a pure constant delayτ0—flattening the linear phase ramp without altering the underlying transfer function's minimum-phase or excess-phase curvature. - Generate Minimum Phase and inspect Excess Phase & Excess Group Delay. In SPL & Phase → Controls (or Group Delay → Controls), click Generate minimum phase. REW computes the causal, invertible minimum-phase transfer function corresponding to the measured magnitude response (taking into account your active IR windows and tail settings). Enable the Minimum phase and Excess phase traces on SPL & Phase, then switch to the Group Delay graph and enable Minimum group delay and Excess group delay. Optionally apply light smoothing (1/12 or 1/6 oct, or Variable smoothing for broad trends) only after inspecting unsmoothed excess group delay spikes so sharp cancellations are not hidden.
Stage C — Align Speakers and Subwoofers in the Alignment Tool
-
Load the timing-referenced pair into the Alignment Tool.
Switch to All SPL, check the unshifted individual traces you wish to integrate (e.g.,
L MainandSub 1), and click Controls → Alignment tool. Assign the reference speaker (typicallyL Main) to trace A and the subwoofer (or secondary driver) to trace B. Place the frequency cursor at the acoustic crossover frequencyfc(the frequency where the two SPL traces actually intersect—not merely the nominal number dialed into the AVR or DSP). -
Resolve 360° cycle ambiguity first, then refine phase and polarity.
Do not blindly click "Align phase at cursor" on an unknown delay offset—phase alignment alone is blind to
integer-cycle wraps (
±N / fc, e.g., ±12.5 ms at 80 Hz). Execute in this order:-
Select Impulse alignment mode with the cursor at
fc. REW applies a 1/3-octave bandpass filter centered atfcto both IRs and cross-correlates their envelopes and waveforms to locate the true arrival offset without cycle wrap. Evaluate both normal and inverted polarity candidates. -
Switch to Phase alignment mode and compare Align phase slopes at cursor against
Align phase at cursor. Visually verify that the adjusted phase trace of B tracks
the slope of trace A through the entire overlap band (roughly
0.5·fcto2·fc), not merely touching at a single point. - Toggle Invert polarity on B and test fine delay (±1–2 ms) while watching the live Aligned sum curve. Adjust Gain on B only if a level mismatch shifts the acoustic crossover point away from target.
-
Select Impulse alignment mode with the cursor at
-
Generate the predicted sum, translate delay to hardware, and verify physically.
Click Aligned sum in the Alignment Tool to create a new measurement representing the exact complex
vector sum
A + Badj(and optionally Aligned copy to inspect B's shifted phase/impulse directly). Inspect the generatedAligned sumin SPL & Phase, Group Delay, and Impulse / Step. Next, enter the delay, polarity, and gain into your DSP or AV processor—paying strict attention to sign conventions (if REW calls for a negative delay on the sub, you must add positive delay to the mains!). Finally, run a physical combined sweep (L Main + Sub 1) and overlay it against REW'sAligned sumin All SPL. Close agreement (±0.5–1.0 dB through the crossover) proves your timing chain and DSP implementation are exact.
03 · Graph & Metric Interpretation
Reading the Output
Knowing which buttons to click is trivial; reading the coupled behavior across SPL & Phase, Group Delay, Impulse, and the Alignment Tool is where professional diagnostics live. Every anomaly leaves a correlated signature across magnitude, phase, excess group delay, and time.
3.1 SPL & Phase + Excess Group Delay: Minimum Phase vs. Non-Minimum Phase
On the SPL & Phase graph, wrapped phase cycles between +180° and −180°
(vertical jump lines are wrap discontinuities, not physical steps; toggle Unwrap phase to inspect
continuous slope). Once bulk transport delay τ0 is removed, compare the measured phase
φmeas(f) against REW's computed φmin(f):
-
Minimum-phase region (EQ-able): Measured phase tracks minimum phase closely;
φexcess(f)is flat or varies only slowly, and Excess Group Delay (EGD) sits flat near0 ms(or at a constant offset equal to any unremoved pure delay). Isolated axial room-mode peaks almost always fall here: applying a complementary minimum-phase PEQ cut flattens SPL, straightens phase, and shortens modal ringing in the waterfall simultaneously. - Non-minimum-phase region (NOT EQ-able): A sharp cancellation notch (SBIR, floor bounce, overlapping out-of-phase reflections, or misaligned crossover) introduces a transfer-function zero outside the unit circle. Measured phase diverges abruptly from minimum phase—typically swinging by ~180° to 360° across the notch—while Group Delay and Excess Group Delay erupt into tall positive or negative spikes. Negative group delay does not violate causality; it reflects destructive interference reshaping the narrowband wave envelope so its peak emerges earlier than the undisturbed carrier. Attempting to boost this notch with minimum-phase EQ cannot invert the excess-phase rotation and only dumps excess energy into the room.
3.2 Group Delay: Reading Smooth Slopes vs. Constant Offsets
Group delay is defined as τg(f) = −(1 / 2π) · dφ(f)/df. When reading the
Group Delay panel in REW, distinguish three completely different physical phenomena:
-
Uniform vertical offset across frequency: Pure unremoved time delay
τ0. If a subwoofer sits on a flat EGD plateau at+8.5 msrelative to a timing-referenced main speaker at0 ms, that plateau directly quantifies the subwoofer's relative arrival lag (acoustic path + DSP latency). - Smooth, monotonic low-frequency rise: Normal minimum-phase group delay of high-pass roll-offs (sealed enclosure = 2nd order, ported box = 4th order, subsonic protection filter + crossover LPF/HPF). A 25 Hz ported subwoofer naturally rises toward 15–25 ms near tuning—this is minimum-phase group delay and vanishes from the Excess Group Delay trace.
- Narrow spikes (positive or negative) on both GD and EGD: Multi-path interference (room reflections, crossover phase mismatch, cabinet edge diffraction). Notice how checking Excess Group Delay strips away the benign minimum-phase enclosure/filter rise so only true alignment errors and reflections remain visible.
3.3 Impulse Response, Step Response, and Measurement Info
The Impulse graph (viewed in % FS or dBFS) and Measurement Info
panel anchor your time-domain sanity checks:
-
IR Peak vs. IR Start on subwoofers: A tweeter impulse peaks sharply within tens of microseconds of
onset. A subwoofer low-passed at 80 Hz has a broad, slow-rising waveform whose peak arrives 5–10 ms after the
first acoustic energy. That is why REW offers
Set t=0 at IR startfor band-limited sources and whyEstimate IR delayuses cross-correlation against the minimum-phase IR rather than naive peak picking. - Polarity verification: On a time-aligned full-range speaker, a positive direct impulse peak followed by a positive initial step response indicates absolute positive polarity. For a sub vs. main pair, however, acoustic polarity at the crossover depends on crossover order (e.g., LR2 requires 180° inversion; LR4 sums in-phase), so rely on the Alignment Tool's complex summation rather than forcing identical raw IR peak polarity.
-
Measurement Infoaudit: Always checkSystem delay,Timing ref, andClock adjustment. WithAdjust clock with acoustic refactive, consecutive sweeps on the same USB mic should report nearly identical ppm offsets (e.g.,-28.4 ppm,-28.9 ppm,-28.1 ppm). A sudden jump to+410 ppmmeans room noise or a strong reflection corrupted one of the two timing chirps—discard and re-measure.
3.4 Alignment Tool: Phase Matching vs. Phase-Slope Matching (The 1-Cycle Trap)
The single most common professional error in sub/main integration is aligning phase at fc while
being 1 full cycle (360°) off. At 80 Hz, one period is T = 1 / 80 = 12.5 ms.
If the subwoofer is 12.5 ms late relative to the mains, both drivers have identical phase at 80.0 Hz and
produce a tempting +6 dB sum right at the cursor—yet because the extra 12.5 ms adds an extra
−360° × 0.0125 s = −4.5°/Hz of phase slope, the sub and main slide 90° out of phase just 20 Hz away
(at 60 Hz and 100 Hz) and hit 180° destructive cancellation 40 Hz away (at 40 Hz and 120 Hz), hollowing out the
crossover skirts and smearing bass transient impact.