Fixed Installations

A fixed installation is a spatial audio system that lives in one room, runs the same content for years, and is judged not by how it sounds for one engineer at the mix position but by how it behaves at 9:00 on a wet Tuesday in February when nobody technical is in the building. This is a different discipline from a live show. The art is no longer in the performance; it is in the design, the calibration, and above all the engineering for unattended reliability. Everything you learned in the earlier parts of this guide — panning laws, room acoustics, time alignment, room correction — still applies, but the constraints are inverted. There is no operator to ride a fader, no soundcheck before every audience, and no sweet spot because the audience is walking around or lying back staring at a dome.

This chapter walks through the whole arc: the contexts you will be asked to work in, how the constraints differ from live sound and post-production, how to cover a moving or standing crowd, the content and control models you will choose between, calibration and the daily power-on cycle, the reliability engineering that keeps the thing alive, and two worked examples. The recurring truth from the rest of the guide holds here too: most of the source material you are handed will be stereo or simple multichannel, and handling that gracefully — see stereo is already spatial — is half the job.

What fixed installations are

The label covers a surprisingly wide spread of contexts, and each one bends the design priorities in a different direction. The common thread is permanence: the system is commissioned once, signed off, and then expected to perform with minimal intervention.

The spread of contexts

Museums and galleries. Often the most acoustically hostile environments — hard plaster, stone, glass vitrines, polished concrete — with multiple sound zones in earshot of one another. The audience moves continuously and dwell time per exhibit may be 30 to 90 seconds. Intelligibility of narration, spill control between adjacent exhibits, and 24/7 reliability dominate. Audio is frequently triggered by proximity sensors or synchronized to short video loops.

Themed entertainment and rides. Dark rides, walkthrough attractions, and queue lines. Here audio is one channel of a larger show-control fabric that also drives ride motion, animatronics, lighting, and projection. Synchronization tolerances are tight (frame-accurate to video), safety interlocks are involved, and the audio designer rarely owns the master clock — they subscribe to it.

Planetariums and full-dome theaters. A hemispherical or tilted-dome screen with a seated, mostly stationary audience looking up. The challenge is matching a 360-degree visual field with a sound field that has credible overhead and horizon imaging, using a speaker array hidden behind an acoustically transparent perforated screen.

Immersive art and "experiences." The teamLab / Meow Wolf / projection-mapping category. Large, often reconfigurable rooms, generative or long-form content, frequently real-time and interactive. Budgets and ambitions vary wildly; the brief may change mid-build.

Houses of worship. Permanent reinforcement of speech and music for a standing/seated congregation in a room that is often deliberately reverberant. Speech intelligibility under heavy reverberation is the entire ballgame, and increasingly these rooms add immersive elements for music.

Brand and retail experiences. Flagship stores, showrooms, trade-show stands, brand "houses." Short content, high production polish, frequent content updates, and a hard requirement that it never embarrasses the brand by failing in public.

Takeaway

Across all of these, the audience is mobile or distributed and the operator is absent. Those two facts — no sweet spot and no human in the loop — are what make installation work its own discipline. Design for the area and for the failure modes, not for the ideal listener.

Audiences and dwell time

Design follows the audience. A planetarium audience is captive, seated, and present for 25 to 40 minutes, so you can invest in precise overhead imaging and a wide dynamic range. A museum visitor gives you under a minute and is constantly distracted, so you prioritize a fast, clear hook and consistent level everywhere in the zone. A retail visitor must not be fatigued over an 8-hour staff shift, which caps your long-term loudness. Write the dwell time and the audience posture at the top of every design document; nearly every later decision derives from them.

Defining constraints versus live sound

If you come from live sound, the mental shift is the hardest part. In live work the system exists for hours and is babysat by an engineer the whole time. In an installation the system exists for years and is babysat by nobody.

24/7 unattended operation

A retail or museum system may run $10$ to $14$ hours a day, $360$ days a year. That is on the order of $4{,}000$ powered-on hours annually. Consumer-grade gear rated for intermittent use will not survive it. You specify install-grade amplifiers (convection-cooled or with filtered, serviceable fans), media players designed for continuous duty (solid-state storage, no spinning disks, passive cooling where possible), and connectors that are locked, strain-relieved, and not within reach of the public. Thermal design matters: an amp rack in a sealed cupboard behind a wall will cook itself. Budget for airflow and measure the rack temperature during commissioning under a sustained load, not a 5-minute test tone.

Redundancy and longevity

The system must outlive its components. Plan for a service life of $7$ to $10$ years and assume individual parts will fail within that window. That means: specify parts that will still be buyable or substitutable, document everything, leave spares on site or on a service contract, and design so that a single failure degrades gracefully rather than killing the show. We return to specific redundancy techniques below, but it is a design axiom from day one, not a feature you add at the end.

Remote monitoring and integration

Because nobody is on site with test gear, the system must report its own health. This is where networking and integration stops being a convenience and becomes load-bearing. A modern install rides on an audio-over-IP backbone (Dante, AES67, Milan, or a proprietary equivalent) precisely because IP transport gives you device discovery, status telemetry, and remote control on the same cable as the audio. You want every amplifier, every media player, and every DSP to expose temperature, signal-present, fault, and link status to a central monitoring dashboard or building-management system (BMS), so that a fault generates a ticket before a visitor notices silence.

Dimension	Live sound	Fixed installation
Operating hours	Hours, attended	Years, unattended
Operator	Present, can intervene	Absent
Sweet spot	Mix position prioritized	None — cover the area
Soundcheck	Every show	Once at commissioning
Failure handling	Engineer reacts live	Automatic failover + alerting
Gear duty rating	Intermittent / touring	Continuous duty
Content	Live performance	Stored, looped, or generative
Tuning changes	Per venue, per night	Set once, drifts slowly
Master clock	FOH owns it	Often subscribes to show control

warning

The single most expensive lesson in installation work is treating it like a long live gig. A touring-grade system that you tuned beautifully on opening day will, with no monitoring and no redundancy, fail silently six weeks later and stay broken until a visitor complaint reaches a manager who reaches a contractor who reaches you. Design for the day you are not in the building.

The moving and standing audience problem

In stereo and most cinema work there is a sweet spot — the place where the stereo or surround image collapses correctly. In an installation, almost nobody stands there. People walk through, cluster at exhibits, or recline under a dome. Your panning and layout strategy must produce an acceptable result over an area, not a point. This reframes the technique choices from the techniques part.

Why classic stereo and VBAP struggle

Pairwise amplitude panning and its multichannel generalization VBAP place a phantom image using level differences between the two nearest speakers. The illusion depends on the listener being roughly equidistant from those speakers. Step off-axis and the precedence effect (the law of the first wavefront) pulls the image hard toward the nearer speaker — the phantom collapses into the closest box. For a single seated listener this is fine; for a crowd spread across a $15\,\text{m}$ wide gallery it means the "centered" voice is heard at the left wall by everyone on the left and at the right wall by everyone on the right. The image is only correct for the few people on the centerline.

DBAP for area coverage

Distance-Based Amplitude Panning (DBAP) is the workhorse of installation spatialization for exactly this reason. Instead of assuming a central listener and panning between a pair, DBAP treats all speakers as potentially active and weights each speaker's gain by its distance to the virtual source position, independent of where the listener is. A source near a physical speaker is loudest there and falls off with distance to the others. The result is a sound field where a virtual source has a stable location in the room that every listener localizes correctly by proximity, rather than a phantom that only works on a centerline.

A common gain model assigns speaker $i$ a weight that falls with its distance $d_i$ to the source,

$$g_i = \frac{d_i^{-a}}{\sqrt{\sum_{k} d_k^{-2a}}},$$

where the rolloff exponent $a$ (often near $a=1$ to $2$) controls how tightly the source localizes to nearby speakers, and the denominator normalizes total power. Increase $a$ for tighter, more localized sources in a dense array; decrease it for smoother, more diffuse blends in a sparse one. DBAP does not rely on phantom imaging between a pair, so it degrades gracefully as listeners move. The trade-off is that it gives you spatial position rather than crisp phantom precision, which is exactly the right trade for a moving crowd. Tools like RIPL (https://dam-audio.com/ripl) and most installation media servers implement DBAP or close relatives natively.

WFS for large, even fields

Where the budget and the architecture allow a dense line or grid of loudspeakers, Wave Field Synthesis physically reconstructs a wavefront so that a virtual source has a consistent direction for every listener in the field — the holy grail for a large, evenly populated space. WFS shines in long galleries, corridors, and walkthrough attractions where you want a sound object to appear to sit at a fixed point in space (or to travel along the visitor's path) regardless of where each person stands. The cost is the speaker count and the processing: spacing the elements at $\Delta x$ sets a spatial-aliasing frequency

$$f_\text{alias} \approx \frac{c}{2\,\Delta x},$$

so $0.25\,\text{m}$ spacing aliases above roughly $680\,\text{Hz}$ (with $c \approx 343\,\text{m/s}$). Above $f_\text{alias}$ the reconstruction degrades, but psychoacoustically the localization cues below it still dominate, so WFS remains convincing. Reserve WFS for projects that genuinely need uniform localization over a big area and can afford dozens to hundreds of drivers; for most installs DBAP over a sensible layout is the pragmatic answer.

Dome and 360 topologies

For domes and full-360 rooms, the speaker layout is itself the instrument. A planetarium typically uses rings: a horizon ring at the spring line, one or more elevation rings, and an apex (zenith) speaker, with subs decoupled from the rings (more on bass management below). Content is authored either as discrete channels mapped to the rings, as objects panned with VBAP/DBAP across the hemisphere, or as Ambisonics decoded to the specific dome layout. Ambisonics is attractive for domes because the content is layout-agnostic until the decode: you author a sound field once and decode it to whatever ring configuration the dome actually has, which is invaluable when the same show tours between domes of different speaker counts. The decode is tuned (with appropriate near-field compensation and max-$r_E$ weighting) for the dome radius and the seated audience zone.

Rule of thumb

If listeners move, pick the panner that ties a source to a place in the room (DBAP, WFS) rather than to a phantom between speakers (stereo, VBAP). If the room and budget are big and uniform, WFS; otherwise DBAP. For domes that must travel, author in Ambisonics and decode per-venue.

Content models

Independent of the panning technique, you choose a content model: how the audio is stored, generated, and triggered. This decision drives the entire control and reliability architecture, so make it early and explicitly.

Fixed channel beds

The simplest model: a multichannel file (anything from $4$ to $64$ channels) plays out, channel-to-speaker, on a loop or on trigger. There is no real-time panning — the spatial mix is baked in at authoring time, exactly like a surround master. This is rock-solid, trivially reproducible, and demands almost nothing of the playback engine (just sample-accurate multichannel playback). The downside is rigidity: the mix is tied to one specific speaker layout, and any change to the room means re-authoring. Channel beds are the right default for museum exhibits, retail loops, and any show where the layout is frozen and reliability trumps flexibility.

Looped object scenes

Here the content is a scene — audio objects plus their position metadata and trajectories — rendered in real time by a spatialization engine to the actual speaker layout. This is the object-based paradigm applied to a permanent room. The same scene file renders correctly to a $9$-speaker gallery or a $32$-speaker dome because the renderer knows the layout. You get layout independence, easy content updates, and the ability to move objects dynamically. The cost is a more complex, stateful playback engine that must itself be made reliable. This is RIPL's home territory and the model most immersive-art and dome installs choose.

Real-time and interactive

The most ambitious model: the spatialization responds to live inputs — visitor position from cameras or LiDAR, capacitive floor sensors, button presses, a phone app, or generative/procedural composition. Sound objects move with people, respond to gestures, or evolve algorithmically so the piece never exactly repeats. This delivers the strongest sense of agency and is the signature of high-end immersive art, but it is the hardest to make bulletproof: sensors drift and fail, latency must stay low, and there is no fixed "correct" output to validate against, which complicates automated testing.

Content model	Flexibility	Reliability	CPU / engine load	Best for
Fixed channel bed	Low (layout-locked)	Highest	Minimal	Museum exhibits, retail loops, signage
Looped object scene	High (layout-independent)	High (if engine is hardened)	Moderate	Domes, immersive art, multi-venue shows
Real-time / interactive	Highest	Lowest (most to go wrong)	High	Interactive art, responsive environments

A pragmatic pattern is to mix models: a hardened looped object scene as the baseline, with interactive layers that can fail to the baseline without taking down the whole experience. That way a dead sensor degrades the interactivity but never produces silence.

Loudness and dynamic-range targets

Whatever the model, set loudness targets up front. Borrow the discipline from broadcast and cinema even though no single standard governs installations. Measure integrated loudness with an ITU-R BS.1770 meter and pick a target appropriate to the context: an immersive theater can sit near cinema levels around $-27$ to $-23\,\text{LUFS}$ with wide dynamics, whereas a retail or museum environment where staff endure it all day should be flatter and quieter, commonly $-18$ to $-16\,\text{LUFS}$ with limited crest factor so peaks never startle. Crucially, fix the system gain at calibration so that a given file level always produces the same SPL in the room — the content's loudness and the system's gain are two separate decisions, and conflating them is a frequent source of "it got louder after the update" complaints.

Show control and triggering

In a live gig you press go. In an installation, something else presses go, on a schedule, on a sensor, or in lock with video and lighting. Audio is one subsystem in a show-control network, and choosing how it is triggered and synchronized is a core design task.

Transports and protocols

The common control transports, and when to reach for each:

• Timecode (LTC/MTC): The bedrock for tight A/V sync. When audio must lock frame-accurately to projection or a ride, slave everything to a single house timecode master. Tolerances of $\pm 1$ frame ($\approx 33\,\text{ms}$ at $30\,\text{fps}$, and you usually want tighter) are typical for lip-sync-critical content.
• OSC (Open Sound Control): The lingua franca for spatial engines and modern show control. Flexible, networked, human-readable address spaces, ideal for sending object positions, scene recalls, and state queries between a media server, the spatializer, and a show controller.
• MIDI (notes, MSC — MIDI Show Control): Still ubiquitous for simple cue triggering and for talking to lighting desks and older show-control hardware. Robust and dead simple, but limited bandwidth and resolution.
• Ableton Link: Tempo/phase sync across devices on a LAN for music-driven or generative work where a shared musical clock matters more than absolute timecode.
• Contact closures / GPIO and DMX/Art-Net: The plumbing that ties audio triggers to physical buttons, ride events, and the lighting world.

Sensors and interactivity

Interactive installs add an input layer: PIR/proximity sensors and beam-breaks (cheap, robust, binary presence), capacitive or pressure floors, computer-vision and LiDAR tracking (rich position data, more failure surface), RFID/NFC for personalized triggers, and visitor smartphones. The engineering rule is to debounce and validate every sensor input before it drives audio: a flickering beam-break should not machine-gun a cue, and a tracking dropout should hold last-known-position rather than slamming an object to the origin. Treat sensor data as untrusted and filter it, exactly as you would untrusted network input.

Synchronizing audio, video, and lighting

The cleanest architecture has one master clock and everything else slaved to it. Decide early who is master — usually the video server or a dedicated show controller — and make audio a disciplined slave. When audio free-runs and video free-runs on separate clocks, they drift apart over a long loop; even a $10\,\text{ppm}$ clock difference accumulates to a noticeable A/V offset over a $30$-minute show. On an audio-over-IP backbone, use PTP (IEEE 1588, the basis of AES67 and Milan) to discipline all audio devices to a sub-microsecond shared clock, and lock that domain to house timecode. Document the clock hierarchy on one page; it is the first thing the next engineer will need.

info

A surprising share of installation "audio bugs" are really sync or clocking bugs: pops from unlocked digital devices, A/V drift from free-running clocks, or cues firing twice from un-debounced sensors. Nail the clock architecture and the trigger validation first, and a whole class of intermittent, hard-to-reproduce faults simply never appears.

Calibration for the space and the daily power-on

Calibration is where the systems part of this guide pays off, but with two installation-specific twists: you tune for an area (and a hostile room) rather than a seat, and you must define what happens automatically every single morning.

Commissioning calibration

The commissioning procedure mirrors measurement and calibration but is performed across a grid of listening positions, not one mic at the sweet spot. A practical sequence:

1. Verify the install. Confirm every speaker is wired to the right output with correct polarity and no swaps. A polarity-flipped surround in a dome is inaudible on a sine sweep and ruinous on content. Use a polarity checker and a channel-ID test (pink noise stepping speaker by speaker) and physically confirm each one.
2. Set levels. With pink noise at a reference, set each speaker (or ring) to a matched SPL at the design reference position, then check uniformity across the visitor area with a sound-level meter. For a multi-zone museum, set each zone's reference SPL and verify spill into neighbors is below the design threshold.
3. Time-align. Apply delay and phase alignment: delay each speaker so wavefronts arrive coherently at the design region. For a listener at distance $d$ from a speaker, the propagation delay is $t = d/c$; with $c \approx 343\,\text{m/s}$, a $4\,\text{m}$ path is $t \approx 11.7\,\text{ms}$. Align the near speakers back to match the far ones so the array images as one source over the seating/standing zone.
4. Equalize for the room, not the seat. Apply EQ and room correction based on spatially averaged measurements across the grid. Correct broad, consistent problems (a room mode, a tonal tilt) and resist the temptation to notch position-specific dips that move as you walk — they are not really there for a moving audience.
5. Bass management and subs. Set crossovers and align the subs to the mains; place and possibly delay multiple subs to even out the modal field across the area, per subwoofers and bass management. Even bass coverage matters more than a deep-but-lumpy single position.
6. Verify with content. Play the actual show and walk the entire visitor path. Numbers get you close; the walk-through catches the rest.

Taming the room

Installations frequently live in hard, reverberant architecture — atria, stone galleries, glass retail spaces — and reverberation is the enemy of intelligibility and localization. EQ cannot fix a $2.5\,\text{s}$ reverberation time. The fix is acoustic treatment and, just as importantly, system design that fights reverberation: more, smaller, closer speakers raise the direct-to-reverberant ratio for each listener compared with a few loud far ones. Pushing the listener inside the critical distance — where direct energy exceeds reverberant energy — is the single most effective lever for clarity in a live room. This is why distributed, low-level, nearby loudspeakers beat a powerful central cluster in a gallery.

The daily power-on and automated checks

Every morning the building wakes up. Define that sequence explicitly:

• Sequenced power-up. Sources and DSP power first and settle, then amplifiers last, to avoid the turn-on thump that, repeated daily for years, kills drivers. Use sequenced power conditioners or amp standby triggered over the control network. Power-down reverses the order.
• Automated self-test. On boot, the system runs a brief routine: confirm every media player is online, every amp reports signal-present and no fault, network links are up, and clock is locked. Optionally play a sub-audible or pre-opening sweep and verify via a monitoring mic or the amps' load-monitoring that every speaker is actually producing sound.
• Scheduled start/stop. Content starts at opening and stops (or drops to a quiet idle) at close, on a calendar that knows about late nights and holidays.
• Heartbeat and alerting. The system reports "all healthy" to the monitoring dashboard every morning; absence of the heartbeat is itself an alert.

Drift over time

Nothing stays calibrated. Drivers age and their compliance changes, amplifier gains creep, foam surrounds rot, a cleaner unplugs a speaker, humidity and temperature shift the room's response, and someone "improves" a setting. Build in periodic re-verification: a quarterly or biannual service visit that re-runs the channel-ID and level checks, compares against the commissioning baseline, and re-walks the path. Keep the commissioning measurements and settings archived so drift is measurable against a reference rather than guessed at. A system with no baseline cannot be told whether it has drifted.

Reliability engineering

This is the section that separates installation professionals from everyone else. The goal is simple to state and hard to achieve: a single component failure must not produce silence or an obviously broken experience, and any failure must be detected and reported automatically.

Redundant playback and network

For high-value or high-traffic installations, deploy two media players or servers in a primary/backup pair. Approaches range from hot standby (the backup plays in sync, muted, and an automatic switch un-mutes it within milliseconds if the primary's heartbeat stops) to warm standby (the backup is running and takes over within a few seconds) to cold spare (a configured unit on the shelf, swapped in by staff). The right level depends on the cost of downtime: a flagship brand experience or a paid-ticket dome warrants hot standby; a single museum exhibit may be fine with a documented cold spare and a 24-hour service SLA.

Networks get the same treatment. Audio-over-IP standards support redundancy directly: Dante offers a fully separate primary/secondary network (seamless, glitch-free switchover), AES67/ST 2022-7 sends duplicate streams over two networks and the receiver picks intact packets, and Milan/AVB provides bandwidth reservation and fast reconfiguration. Run two physically independent switch fabrics for the audio so a single switch or cable failure does not mute the room.

Watchdogs, failover, and graceful degradation

A watchdog is a process whose only job is to notice when something stops. At minimum: a watchdog that restarts a crashed playback application, a hardware watchdog that reboots a frozen media player, and a supervisory layer that detects a dead primary and promotes the backup. Pair these with graceful degradation: if the interactive layer dies, fall back to the looped baseline; if one amplifier in a ring fails, the spatial image distorts but the show continues; if a sensor drops out, hold last-known state. The design question to ask of every component is: "When this fails — and it will — what does the visitor experience?" The acceptable answer is never "silence," and ideally is "they don't notice."

Content management and updates

Content changes over the install's life — seasonal campaigns, new exhibits, fixes. The mistake is editing files in place on the live machine via a remote desktop at 2 a.m. Instead: keep content under version control or at least dated, immutable releases; stage and validate updates on a backup/test rig before promoting them; have a one-click rollback to the last known-good package; and log who deployed what and when. Push content updates over the management network with verification (checksums) so a half-transferred file never goes live. Treat the show like software, because by now it is.

Failure	Detection	Mitigation	Visitor impact if handled
Media player crash	Watchdog / lost heartbeat	App auto-restart; promote hot standby	None to a few seconds
Amplifier channel dies	Amp fault telemetry	Spare amp / redundant amp; alert	Slight image shift; show continues
Network switch failure	Link-down trap	Secondary Dante/ST 2022-7 path	None (seamless)
Single driver failure	Load monitoring / service mic	DBAP redistributes; service ticket	Subtle; fixed at next visit
Sensor dropout	Input validation / timeout	Hold last state; fail to baseline scene	Reduced interactivity only
Power glitch	UPS + sequenced restart	UPS rides through; clean reboot	Brief, then auto-recovers

danger

No redundancy and no monitoring is the cardinal sin of installation work, and it is almost always a budget decision made by someone who will not be there when it fails. A silent exhibit in a paid attraction is a refund and a bad review; a dead system in a flagship store is a brand failure photographed and posted. If the budget cannot cover hot standby everywhere, cover it where downtime is most expensive, and put monitoring and alerting on everything — detection is cheap, and an undetected fault is the only truly unforgivable one.

Putting it together: a planetarium dome

Consider a commission for a $15\,\text{m}$ diameter tilted-dome planetarium seating $120$ people, used for both scientific shows and immersive music. We walk the whole pipeline.

Layout

Behind the perforated, acoustically transparent screen we hang three rings plus an apex: a horizon ring of $12$ speakers at the spring line, an upper ring of $8$ at roughly $45^\circ$ elevation, an apex/zenith speaker at the pole, and a lower fill ring of $4$ near the seating to anchor the front. That is $25$ full-range positions, complemented by $4$ subwoofers placed to even the modal field across the seating bowl and decoupled from the rings via bass management. Because the audience is seated and reclined, the design region is the seating bowl, not a single seat; we calibrate to that volume.

Content model and rendering

We choose looped object scenes authored in third-order Ambisonics, plus an object layer for discrete effects, both rendered in real time to the dome layout by a spatial engine (e.g. RIPL, https://dam-audio.com/ripl). Ambisonics is the right call because the same shows tour to other domes with different speaker counts: author the sound field once, and the per-venue decoder (tuned with near-field compensation for the $7.5\,\text{m}$ radius and max-$r_E$ weighting for the seated bowl) maps it to whatever array exists. Discrete fly-throughs and spot effects are objects, panned with VBAP over the hemisphere for tight localization. Loudness is targeted near $-24\,\text{LUFS}$ integrated with wide dynamics for the cinematic shows, and the system gain is fixed so the calibrated reference holds.

Control and sync

The dome's video server is the master clock, distributing LTC. The audio engine slaves to it via timecode for show playback and receives OSC scene-recall and object-position cues. All audio devices share a PTP clock domain locked to that timecode, so a $40$-minute show stays in lip-sync with the visuals without drift. Lighting (house lights, aisle lighting) is cued over DMX from the same controller. A pre-show "system ready" check runs before doors open.

Calibration

Commissioning measures a $5 \times 5$ grid across the seating bowl. Each ring is level-matched, then time-aligned: a horizon speaker at $7.5\,\text{m}$ from the dome center implies $t \approx 21.9\,\text{ms}$ of propagation, and the closer fill ring is delayed to match so the array images coherently across the bowl. EQ is applied from the spatial average; broad corrections only. The subs are aligned to the rings at the crossover and positioned to minimize seat-to-seat bass variation. Finally, the team plays reference shows and walks/sits across the bowl to confirm overhead imaging and horizon stability.

Redundancy

Two playback servers run in hot standby, the backup rendering in sync and muted, switching in under $50\,\text{ms}$ on heartbeat loss. The audio rides a redundant Dante fabric (primary/secondary switches). Amplifiers report fault and load telemetry to a monitoring dashboard that also watches server health and clock lock; loss of the morning heartbeat pages the service contractor. A documented cold spare amplifier sits in the rack room, and the show content is version-controlled with one-click rollback. Sequenced power-up brings DSP and servers up before amps each morning, and an automated self-test confirms all $25$ speakers and $4$ subs are alive before the first show.

The result is a dome that opens itself every morning, plays in lock with the visuals, sounds correct for every seat in the bowl rather than one, tours its content to other domes unchanged, and survives the failure of any single server, switch, or amplifier without going dark.

Putting it together: a museum gallery

A contrasting example shows how different the same principles look in a hostile, multi-zone space. A heritage museum wants four adjacent exhibits in one stone-walled hall (RT around $2.2\,\text{s}$), each with a $45$-second narrated audio loop triggered as visitors approach, and the zones must not bleed into one another.

The model is fixed channel beds — one stereo-or-quad loop per exhibit — because reliability and intelligibility trump flexibility here, and the layout is frozen. Each exhibit uses a small cluster of nearby, low-level directional speakers (or a focused beam / directional array) aimed down into the visitor zone, deliberately keeping listeners inside the critical distance so the high RT does not destroy intelligibility (see direct, diffuse and envelopment and reverberation). Spill is controlled by speaker directivity, tight level setting, and physical placement rather than by hoping EQ will help. Beam-break sensors trigger each loop, debounced so a hovering visitor does not retrigger it, and an idle visitor sees the loop fade out gracefully. A single multichannel media player drives all four zones over Dante to install amps, monitored centrally, with a documented cold spare and a daily auto-power and self-test. It is unglamorous and bulletproof — which is exactly right for a museum.

Common mistakes and limits

Designing for a sweet spot no one occupies

The recurring beginner error is tuning the system, and choosing the panning, for the central listening position — then discovering that nobody ever stands there. Phantom images collapse, levels are wrong at the edges, and the carefully-tuned response only exists at one point. Always design and calibrate for the area the audience actually occupies, with a panner (DBAP/WFS) and a spatially-averaged calibration that match that reality.

No redundancy and no monitoring

Covered above but worth repeating because it is the most expensive and most common failure: shipping an install with a single playback machine, no failover, and no health reporting. It works at handover and fails, silently, weeks later. Monitoring is cheap; the absence of it is what turns a five-minute fix into a week of bad reviews.

Ignoring the room

Dropping a great system into a $2\,\text{s}$-plus reverberant stone or glass room and expecting EQ to save it. It will not. Untreated reverberation smears localization and destroys speech intelligibility, and no amount of correction undoes it. Treat the room, raise the direct-to-reverberant ratio with more and closer speakers, and budget for acoustic treatment from the start — not as a value-engineered casualty.

Conflating content loudness with system gain

Letting each content update set its own level, so the room gets louder or quieter every time someone delivers a new file. Fix the system gain at calibration and hold all content to a measured loudness target; the two are separate decisions.

Under-engineering thermal and duty cycle

Specifying consumer or touring gear for continuous-duty service, or burying amps in an unventilated cupboard. The system runs thousands of hours a year; gear and cooling must be rated for it.

Limits

Be honest about what installations cannot do. There is no operator to rescue a show in real time, so anything that needs nuanced live mixing is out of scope — the mix must be right in advance and robust to drift. Truly uniform localization over a large area still demands WFS-scale speaker counts that most budgets cannot bear; DBAP is a pragmatic compromise, not physical perfection. Interactive systems trade reliability for agency, and the more responsive the experience, the more failure surface you accept. And no system survives years untouched: every install needs a maintenance contract, a baseline to drift against, and a human who comes back. The successful installation is not the one that is perfect on opening day; it is the one that is still working, and known to be working, three years later.

References

1. Ahnert, W. and Steffen, F. Sound Reinforcement Engineering: Fundamentals and Practice. Taylor & Francis, 1999. Foundational text on permanent reinforcement, intelligibility, and system design in real rooms.
2. ISO 3382 (Parts 1–3): Acoustics — Measurement of room acoustic parameters. International Organization for Standardization. Reverberation time, clarity ($C_{50}/C_{80}$), and other metrics used to characterize installation spaces.
3. ITU-R BS.1770: Algorithms to measure audio programme loudness and true-peak audio level. International Telecommunication Union. The loudness measurement basis applied to installation content targets.
4. ITU-R BS.2051: Advanced sound system for programme production. Channel and immersive layout definitions relevant to dome and multichannel beds.
5. Lossius, T., Baltazar, P. and de la Hogue, T. "DBAP — Distance-Based Amplitude Panning." Proceedings of the International Computer Music Conference (ICMC), 2009. The reference description of the panning law underpinning most installation spatialization.
6. AES (Audio Engineering Society) papers and conferences on sound system installation and audio-over-IP (e.g. AES67 / ST 2022-7 redundancy, Dante and Milan/AVB integration). Authoritative source for networked-audio reliability practice.
7. THX / Themed Entertainment Association (TEA) and industry guides on themed-entertainment audio and show control (timecode, MIDI Show Control, OSC integration with ride and projection systems).
8. Holman, T. Surround Sound: Up and Running, 2nd ed. Focal Press, 2008. Practical multichannel system setup, calibration, and bass management transferable to fixed multichannel installs.

← Back to Workflows