Acoustic Comb Filtering in Transportation Hubs

1. Project overview: what, where, who, and why

In late 2024, the CityLine Transit Authority approved an audio refresh for the main concourse of Riverton Central Station, a mixed-use transportation hub serving regional rail, metro, and bus. The space in scope was the main ticketing hall and adjacent retail corridor—approximately 94 m long × 28 m wide with a ceiling height varying from 7.5 m to 11 m. Daily foot traffic averaged 38,000 passengers, with peak-hour density high enough that staff routinely received complaints about announcements being “loud but unclear.”

The client team included the Transit Authority’s facilities PM, the station operations manager, and a safety compliance officer responsible for EN 54-16 related documentation (the station’s evacuation voice alarm was separate, but intelligibility and audibility requirements were still treated with comparable rigor). The integrator was NorthPier Systems. Our role (sonusgearflow.com) was to document the project and provide measurement support: baseline acoustics, intelligibility mapping, and validation after commissioning.

The trigger wasn’t simply “bad sound.” A renovation had replaced old suspended acoustic tiles with a visually open metal ceiling and large glass storefronts. The station looked modern, but the new geometry introduced repeatable comb filtering and localization issues that became obvious as soon as new digital signage and announcements increased overall messaging frequency.

2. Challenges and requirements at the outset

The most consistent complaint—confirmed quickly in walk tests—was that announcement consonants would “phase” or “swim” as passengers moved through the corridor. At some points, speech sounded thin and harsh; at others, it was oddly hollow. This is classic comb filtering: direct sound arriving along with a strong early reflection or a second nearby source with similar level but different path length.

Constraints and requirements were defined in a kickoff meeting and site survey:

Operational uptime: Station could only grant two night windows per week (00:30–04:30), plus one full Sunday shutdown for final cutover.
Noise floor variability: Measured ambient levels ranged from 62–68 dBA off-peak to 74–78 dBA during arrivals (rolling luggage, HVAC ramps, bus doors, crowd noise).
Target speech performance: The operations team requested intelligibility that was “clearly better than today.” We translated this into measurable goals: STIPA ≥ 0.50 in the ticketing hall and ≥ 0.45 in the retail corridor under typical operating noise, with a separate “night quiet” baseline for tuning.
Uniformity: Keep A-weighted SPL for announcements within ±3 dB across primary passenger pathways at ear height (1.5 m), without hotspots that would trigger complaints from retail tenants.
Integration: Reuse existing networked audio backbone (Dante) and paging logic in a Q-SYS core, but replace end-of-line loudspeakers and local amplification as needed.
Aesthetics and mounting: Minimal visual impact, no new floor stands, and all ceiling work had to respect sprinkler coverage and sightlines for CCTV.

Early measurements exposed the core acoustic issue. In multiple locations, there were strong reflections from glass shopfronts and the metal ceiling arriving within 6–18 ms of the direct sound, often within 3–6 dB of the direct level. That delay band is long enough to create frequency-dependent cancellations but short enough to smear articulation without being perceived as a discrete echo.

3. Approach and methodology chosen

The team chose a methodology focused on reducing competing arrivals rather than trying to “EQ away” comb filtering. Equalization can only optimize at one point; comb filtering changes rapidly with listener position because it’s governed by path length differences.

The approach had four pillars:

Source control: Replace widely spaced ceiling speakers that overlapped heavily with a layout that provides more consistent directivity and predictable coverage.
Time-domain management: Use zoning, delay, and level steering to ensure listeners receive a dominant first arrival from the nearest loudspeaker.
Acoustic mitigation where feasible: Add selective absorption at the most problematic reflection surfaces without undermining the architectural intent.
Verification by mapping: Validate outcomes with repeatable measurements: impulse responses, STIPA, and SPL mapping in the same points used for baseline.

Tools and instrumentation included a NTi XL2 with STIPA option, a Room EQ Wizard laptop rig for impulse response capture, a Focusrite Scarlett interface, and calibrated microphones (Earthworks M23 for IR work and a class-compliant measurement mic for quick checks). For modeling, the integrator used EASE Focus for line/column prediction; we supplemented with empirical checks because the concourse had enough reflective complexity that models were directionally helpful but not definitive.

4. Step-by-step execution narrative

Week 1–2: Baseline survey and symptom confirmation

The existing system was a grid of small ceiling speakers on 100 V lines installed during an earlier refurbishment. Spacing in the retail corridor averaged 10–12 m, with speakers mounted flush to a hard ceiling. In the ticketing hall, speakers were higher and unevenly distributed due to architectural constraints.

We established 32 measurement points at 1.5 m height: 18 along the retail corridor and 14 in the ticketing hall, marking each with discreet floor plan references (column numbers and storefront boundaries) so the team could repeat tests after each change.

Two early findings shaped everything that followed:

Overlap between adjacent speakers was extreme. In several corridor points, two ceiling speakers were within 2–4 dB of each other, producing audible combing as passengers walked.
Glass reflections created “shadow zones” where the direct-to-reflected ratio was poor. Impulse responses showed a strong secondary arrival at ~11 ms in front of a long glass storefront run.

Week 3–4: Design revision and equipment selection

The integrator proposed replacing the corridor ceiling grid approach with compact passive column loudspeakers mounted on structural columns at 3.2 m height, aimed down the corridor to control vertical spill and reduce ceiling excitation. In the ticketing hall, where mounting points were limited, the plan used distributed steerable columns to place energy on passenger areas while minimizing glass and ceiling hits.

Final loudspeaker choices were grounded in availability, directivity, and mounting practicality:

Retail corridor: 16 × TOA TZ-406B passive column speakers (black), tapped at 60 W each via 100 V distribution, mounted in pairs on alternating columns to maintain coverage continuity.
Ticketing hall: 6 × Renkus-Heinz IC Live Gen5 (ICL-F-RD) digitally steerable columns, each networked and individually tuned for beam steering.
Amplification: Powersoft Mezzo series for low-impedance where needed and Lab.gruppen 100 V amps for corridor lines, with monitoring on the existing Q-SYS system.
DSP/control: Existing Q-SYS Core 110f retained; new I/O and amp monitoring integrated over the station’s AV VLAN.

We also recommended targeted acoustic treatment: high-NRC wall panels (50 mm thick, fabric-wrapped) on two opposing surfaces in the ticketing hall where reflections were most dominant. The architect resisted large visible panels; the compromise was to place treatment above retail signage bands and on back walls behind ticket machines, totaling about 48 m².

Week 5–7: Installation and interim tuning (night windows)

Installation occurred across six night shifts, with daytime cordons only where lifts were positioned. Corridor columns made mounting predictable, but cable routing was the real schedule risk. The existing 100 V wiring was reused only where insulation and routing met current code; approximately 420 m of new plenum-rated cable was pulled to create cleaner zone boundaries and allow proper delay architecture.

Interim tuning followed a repeatable routine each night:

Polarity and basic level checks for each new zone using pink noise and an XL2 SPL check at a fixed reference point.
Impulse response capture at the nearest two measurement points to ensure the new “first arrival” dominance was improving.
Delay alignment between adjacent corridor zones, set so the downstream zone arrived after the upstream zone for listeners moving forward—preventing two equal arrivals at once.
High-pass filtering to reduce low-frequency buildup that contributed to muddiness in reverberant areas (corridor columns set around 120 Hz, steerable columns around 100 Hz depending on voicing).

Week 8: Commissioning and full intelligibility mapping

Final commissioning occurred during a Sunday shutdown. We ran STIPA tests in two conditions: “quiet station” (HVAC on, no passengers) and “operational simulation” using recorded station ambience played through a portable speaker array to reach a controlled 72 dBA average noise floor in the corridor.

The objective wasn’t to game the numbers; it was to establish that tuning held up when the signal-to-noise ratio was realistic. We used consistent announcement EQ and dynamics across the system, avoiding heavy compression that could raise background spill and annoy tenants.

5. Technical decisions and trade-offs made

Several decisions were debated, and the trade-offs are where the comb filtering lessons live.

Fewer sources vs. more sources: The old design tried to cover evenly with many ceiling speakers. In reflective transit spaces, that often creates multiple comparable arrivals. We accepted a slightly higher per-speaker SPL from fewer, better-controlled sources so the nearest loudspeaker dominated at the listener.
EQ restraint: We applied only gentle corrective EQ (broad bands, modest cuts) and avoided narrow notches to “fix” cancellations. Comb filtering nulls moved by 100–300 mm; chasing them with EQ would have caused unpredictable tonal swings.
Delay philosophy: In the corridor, we intentionally did not time-align all zones to a single reference. Instead, we used a forward-steering delay strategy: adjacent zones were offset by 8–15 ms depending on spacing so that any overlap favored a single apparent source rather than two simultaneous arrivals.
Beam steering vs. passive columns: Steerable columns were reserved for the ticketing hall, where ceiling height and reflections demanded precise vertical control. In the corridor, passive columns were more cost-effective and more robust to maintenance, and they provided sufficient directivity.
Acoustic treatment quantity: We would have preferred more absorption, especially on the largest glass areas. Budget and aesthetics limited us to 48 m². The mitigation helped, but it wasn’t a full acoustic redesign.

6. Results and outcomes with specific details

The most noticeable outcome during walk-throughs was that speech stopped “hollowing out” as passengers moved. Tonality still changed slightly—no transit hall is an anechoic chamber—but the moving comb-filter effect was dramatically reduced because two-speaker overlap was reduced and reflections were less energized.

Measured outcomes compared to baseline:

STIPA (operational simulation): Retail corridor improved from 0.38–0.46 (median 0.42) to 0.48–0.57 (median 0.52). Ticketing hall improved from 0.34–0.44 (median 0.39) to 0.47–0.60 (median 0.54).
SPL uniformity: Announcement level variation across primary pathways tightened from roughly ±6 dB to ±3 dB, with the largest remaining variance occurring near the open stairwell to the metro platforms.
Early reflection dominance: In the corridor point that had a strong ~11 ms reflection off glass, the reflected arrival dropped from about -4 dB relative to direct to approximately -9 dB after re-aiming and switching to column coverage.
Listener movement artifact: Subjective walk tests were backed by a simple repeatable check: we recorded a fixed STIPA test signal through a handheld recorder while walking a 20 m segment at a steady pace. Baseline recordings showed audible phasing; post-commissioning recordings did not exhibit the same sweeping coloration.

Operationally, the system also became easier to manage. The Q-SYS control pages were updated so station staff could select message priorities and apply a “Peak Mode” that raised paging level by +3 dB only in the ticketing hall and corridor (not in the quiet lounge), reducing tenant complaints compared with the previous global level bump.

7. Lessons learned and what could be done differently

Three lessons stood out—two technical, one procedural.

Comb filtering is usually a system geometry problem, not an EQ problem. The moment we reduced equal-level overlaps (speaker-to-speaker and speaker-to-glass), intelligibility climbed without aggressive processing.
Measurement points must match human pathways. Early drafts of the test grid included corners and edges that were easy to access but not where people complained. The most informative points were mid-corridor at ear height, 1–2 m off the glass line, where reflections were strongest.
Night work amplifies planning errors. Every missing bracket or unclear cable path cost a week because access windows were tight. A more detailed pre-install mockup (one corridor bay fully built and tuned before scaling) would have de-risked the schedule.

If we were to redo the project, we would push harder for an architectural acoustic package early—especially for the ticketing hall ceiling. Even an additional 0.4–0.6 s reduction in midband reverberation time (RT60) would have increased the intelligibility margin during peak noise. We would also request the station provide recorded ambient noise samples from different times of day before tuning, rather than relying on a simulated noise floor during commissioning.

8. Takeaways applicable to other projects

Transportation hubs are comb-filter factories: reflective materials, long sightlines, and a tendency to “add more speakers” when clarity is poor. For audio engineers and project managers tackling similar spaces, the following practices carried the most value at Riverton Central:

Prioritize first-arrival dominance. Design so the nearest loudspeaker is clearly louder than the next nearest—aim for at least 6 dB advantage in key pathways.
Control vertical dispersion. Ceiling reflections are often the strongest early reflections. Columns (passive or steerable) and careful aiming reduce energy wasted into hard ceilings.
Use delay to manage overlap, not to create a single “perfect” alignment. In corridors, a forward-delay strategy can reduce the perception of two simultaneous sources.
Keep EQ broad and conservative. Treat comb filtering as a spatial interference pattern; don’t try to notch out moving nulls.
Map before and after with repeatable points. A 25–40 point measurement grid tied to architectural references makes progress visible and prevents “tuning by opinion.”
Budget for selective absorption. Even small areas of treatment placed strategically (behind ticket machines, above signage bands, on opposing walls) can reduce early reflections enough to matter.
Plan for operations. Provide staff with modes that match real usage—peak, off-peak, maintenance—so the system stays intelligible without constant manual tweaks.

The main lesson from this station refresh is that comb filtering in transit environments is solvable when the team treats it as a coverage and time-domain problem. Once the sound system stopped fighting itself—fewer equal arrivals, less glass excitation, better-controlled directivity—speech became reliably understandable, and the station stopped receiving the “loud but unclear” complaints that had prompted the project in the first place.