Auditorium High-Density Design for 2,000 Devices
An 1,850-seat university auditorium hosts a keynote with 1,300 attendees and a projected peak of 2,000 associated devices, mixing laptops, phones, live-streamed video, and two-way Q&A. The room is tiered with a 9 m ceiling at the stage, a 4 m deep balcony soffit, and concrete-and-steel construction that keeps most RF inside the room. The current design is six ceiling APs, 2.4 and 5 GHz, 40 MHz channels, and automatic power and channel selection. During a pilot event, 5 GHz utilization peaked at 82%, 2.4 GHz retries hit 60%, attendees under the balcony reported -74 dBm and dropped video, and the event Wi-Fi team needs a defensible design before next semester.
1. Size the 5 GHz radio count. Show the client-count math, compare it to the throughput math, and explain which constraint governs.
Read the worked answer
Client count first: assume 85% of 2,000 devices land on 5 GHz after aggressive band steering = 1,700 clients. At a design ceiling of 50 clients per radio for streaming workloads, that is 34 radios. If each AP has one 5 GHz radio, that is 34 APs; round to 36 for N+1 coverage under the balcony and at the rear.
Throughput check: 1,700 clients at ~1 Mbps average offered load = 1.7 Gbps aggregate. A 20 MHz 2x2 Wi-Fi 6 radio realistically delivers 250-300 Mbps of goodput at high MCS, so ~7 radios would carry the average. Even if 300 attendees stream at 3 Mbps each (900 Mbps), throughput is not the binding constraint.
Conclusion: client count and airtime efficiency govern, not bandwidth. That is why the plan is many low-power radios rather than fewer high-power ones, and why 2.4 GHz cannot be part of the capacity answer: 400 devices spread over 3 channels is 130+ per channel, which is why retries hit 60%.
2. The room is RF-contained. What channel width, channel reuse, and power strategy do you run, and why is 40 MHz wrong here?
Read the worked answer
40 MHz is wrong on two counts. It halves the usable 5 GHz channel pool (about 25 to 12, and only 8 non-DFS), and inside one open room every AP hears every other AP, so each additional MHz of width is co-channel interference you cannot spatial-reuse away. In a single line-of-sight room, two APs on the same channel hurt each other regardless of distance.
Plan:
- 20 MHz only, with 40/80 disabled at the WLAN and RRM level.
- Treat the room as a set of RF zones, not a grid: use directional antennas aimed down the seating rake so each zone's energy lands on seats and decays before reaching the next zone. Reuse a channel only between zones that are physically far apart and screened by the balcony or side walls.
- Fixed channel and power plan instead of RRM. RRM optimizes for coverage and neighbor count; it cannot see 2,000 devices in one room, and a mid-event channel change is a self-inflicted outage.
- Lowest power that still yields -65 dBm at the seat, typically 8-14 dBm with directional antennas, to keep the reuse distances short.
3. How do you serve the under-balcony seats, and what validation proves the design before the next event?
Read the worked answer
Under-balcony seats sit in the balcony's shadow, so ceiling APs 13 m up cannot reach them. Add APs mounted on the balcony soffit or on the balcony face, aimed back and downward at low power, with directional antennas so the energy covers the shadowed seats without blasting the main floor. Verify the mounting surface and the soffit material actually pass RF.
Validation, in order:
- AP-on-a-stick survey with the final antenna models at seat height, including under the balcony, confirming -65 dBm and 25 dB SNR at every seat for the target rates.
- Fixed-plan validation: confirm the chosen channels are non-overlapping in the as-built positions and measure co-channel interference between zones.
- Load test with at least 100 real clients and a video stream from several locations, including under the balcony, measuring retries, airtime utilization, and per-client throughput.
- Capacity rehearsal at partial attendance before the event, with monitoring thresholds set so the team can react live. A predictive model alone cannot be signed off for a room with this geometry.