Divisible Training Room: Open-Mic Feedback and Pumping
A 12 m × 8 m training room can be divided by an operable partition. Each half has six ceiling microphones, four ceiling loudspeakers, a DSP, and a codec; when the partition is open, the control system combines both halves into one room and keeps all 12 microphones and eight loudspeakers active. Users report a hollow, distant sound and acoustic feedback at moderate speech levels, while remote participants hear pumping and abrupt level changes. The original integrator set every microphone input gain so the DSP meters sat just below clipping and placed a 0 dBFS limiter on the program output. There is no automixing, and both halves use the same far-end reference. The room is used for hybrid training with a presenter walking around and audience questions. The customer wants the room to "just work" in both configurations without local audio expertise. You are asked to assess the design and propose a fix.
1. What is the most likely technical cause of the feedback and pumping, and what measurements would you take before changing settings?
Read the worked answer
The symptom pattern points to excessive open-microphone gain rather than a single faulty microphone. Twelve live ceiling mics summed into one mix raise the effective noise floor and reduce gain before feedback; running inputs near clipping removes the headroom needed for a presenter who moves. The 0 dBFS output limiter then compresses peaks, which explains the pumping remote participants hear. A shared far-end reference is also suspect: if each half's loudspeaker feed is not represented correctly in the AEC reference, the echo canceller adapts to the wrong signal and can pump.
Before adjusting, measure rather than guess:
- Log DSP input and output levels with normal and loud speech, noting peaks and average levels.
- Measure gain before feedback with one mic open, then with all mics open, using a known source and slowly raising gain.
- Check the AEC reference bus against the actual loudspeaker feed, including zone gain and delay.
- Capture an impulse response or at least verify loudspeaker delay and coverage between halves.
- Confirm whether automixing is bypassed and whether unused mics stay open.
This separates gain-structure failure from acoustic feedback and AEC misconfiguration.
2. Recommend a gain-structure and zoning strategy for combined and divided modes.
Read the worked answer
Treat gain structure as a chain, not a single master fader. Set each microphone preamp so normal speech peaks leave roughly 20 dB of headroom below clipping. Do not use the output limiter to control program level; set it only as a protective ceiling above expected peaks. Insert an automixer so only mics with speech gate open, and apply NOM (number of open mics) attenuation so the combined gain falls as more mics open.
Use the partition state to drive logic:
- Divided mode: process each half independently, with its own loudspeaker zone, delay, and AEC reference derived from that half's program bus.
- Combined mode: mute or heavily attenuate mics in an unoccupied half, and use one combined program bus as the AEC reference. Do not feed two different references to one canceller.
- Keep loudspeaker zones aligned to the coverage area; if both halves cover the same seats, delay and level must be coordinated to avoid comb filtering.
Verify that the combined mode does not simply sum both rooms' open mics. The goal is fewer effective open mics and a stable reference, not a lower master fader.
3. How would you commission and verify the fix in both room configurations?
Read the worked answer
Commission each mode as a separate preset and test the transitions. In divided mode, open one mic at a time and confirm that local speech is intelligible without feedback and that the far end hears stable level. In combined mode, repeat with the partition open, walking the presenter path and using audience questions from both halves. Measure gain before feedback with one mic and then with the automixer active; the combined result should not collapse when a second mic opens.
For the far end, check echo return loss enhancement (ERLE) during double-talk and confirm that the AEC reference changes correctly when the partition state changes. Listen for pumping while someone speaks continuously and another person interrupts. Verify that unoccupied-half mics are muted by the partition sensor, and test what happens if the sensor fails: the default should be safe, not all mics open at high gain.
Document input gains, automixer thresholds, limiter settings, zone delays, and both presets. Leave a simple user interface with only volume and mode selection, and include a commissioning report showing measured levels and intelligibility checks.