Phase in NEXUS and Omnissiah: What We Correct and Why

1. 1Level. Peaks and broad imbalances are corrected with minimum-phase IIR filters. For a minimum-phase problem, such as a room mode's peak at the seat, fixing the level fixes its timing too. Deep nulls are left alone. A null is a reflection cancelling the direct sound, and a boost raises both by the same amount.
2. 2Alignment. Every output gets a delay and a polarity, measured at the listening position. By default the sub is aligned to the mains. Set Time-Align Tops to All speakers and left and right are aligned to each other as well, for rooms where they sit at different distances.
3. 3Phase through the crossover, optionally. With Phase Correction on, Omnissiah fits all-pass filters so each speaker's phase follows the reference through the band they share. Their corners are at or below 1.5 kHz. It is off by default.

The delay is found by matching phase across the overlap rather than by lining up impulse peaks. Through a narrow crossover there is a ladder of other delays that also add, alternating in polarity, each rung costing a little more summed level (about 1 dB, then about 3 dB, at 80 Hz). Omnissiah takes the shortest delay that brings the phases within 10 degrees, which keeps the added latency lowest, at the summation cost of that rung.

Two speakers playing the same octave sum at your seat, and if one arrives late they partly cancel. Through a 4th-order crossover at 80 Hz, a sub 2 ms late costs 1.5 dB, 4 ms costs 6.5 dB and 5 ms costs 11.7 dB. A sub one metre further back than the mains is about 2.9 ms late before its own filters are counted. Acoustic treatment changes reflections, not the arrival time of each speaker's direct sound, so it leaves this dip in place.

A crossover cancellation is a broad dip, about an octave wide. Bücklein (1981) found broad response deviations audible at about 1 to 2 dB and narrow ones from about 3 to 5 dB, with dips harder to hear than peaks of the same size. A 3.2 dB dip an octave wide sits above that broad figure, though 80 Hz lies below the bands Bücklein found most sensitive, 200 to 600 Hz and 2 to 6 kHz.

Fixing the dip costs only delay. The speaker that arrives first is held back until the late one catches up, so a sub 3 ms late adds 3 ms to the mains and nothing else changes. NEXUS holds up to 8.5 ms per output. If a room needs more, alignment is skipped with a warning rather than applied wrongly.

Left and right are polarity-checked as standard and time-aligned only when you ask. Two identical speakers at equal distances already match, and above 1 to 2 kHz the remaining difference between them comes from reflections that move with your head, which is why the all-pass pass keeps its filter corners at or below 1.5 kHz.

All-pass filters change phase without changing level, so they look like the natural tool for flattening a speaker's group delay at low latency. Two exact results show that they cannot.

1. 1A stable all-pass filter only ever adds delay. Its group delay is zero or positive at every frequency. So it cannot pull late bass forward. It can only hold back everything above the bass until the bass catches up.
2. 2Each second-order all-pass section adds the same total amount of delay, wherever it is tuned and however sharp it is. Add up the delay it gives each frequency across the whole audio band and the total is always one second-hertz. Lifting a band of B hertz by D seconds needs at least D times B sections.

The crossover row is feasible and already about 0.2 ms against a threshold of about 1 ms. The bass rows need 79 and 400 sections, and NEXUS has 15 filters per output. Optimised cascades of four sections confirm the count. A two-way from 200 Hz up goes from 0.10 to 0.07 times the threshold, and a ported two-way from 4.23 to 3.76 with the threshold held at 3.2 ms below 500 Hz.

An FIR filter could flatten the bass group delay, at 19 ms for an 80 Hz crossover and up to 82 ms for a ported box, with pre-ringing wherever the correction and the acoustics disagree. Linear-Phase Room Correction: The Trade-Offs covers that trade, and the mechanism is in Where the Latency in Linear-Phase Room Correction Comes From.

• Single-speaker group delay. A modelled sealed two-way is 0.14 to 1.29 times the threshold depending on the bass reading, and a ported two-way 0.36 to 4.23 times. Removing either takes 31 or 82 ms of FIR latency or 79 to 400 all-pass sections.
• Room phase above 1.5 kHz. A 5 cm head movement changes a path's phase by 105 degrees at 2 kHz, so a correction fitted at the microphone does not hold at your ears.
• Modal ringing as a target. A minimum-phase cut matched to a room mode shortens its decay as a side effect, but nothing in the optimiser aims for decay, and the app does not yet show a decay view to prove it.
• Every seat but one. Calibration is measured at one microphone position by default.

Phase Correction is off by default. Turn Phase Correction on when the tuner still shows the needles apart after alignment. The extra pass makes calibration take longer.

Re-measure after applying. In the tuner, show the sub and mains snapshots together. Their needles should sit side by side and the verdict should read Locked. The summed response should no longer show a dip around the crossover. If the needles move apart after applying, something in the signal chain differs from what was measured.