How to Find a Flanking Path Without Opening the Wall
Why this matters
The expensive version of this job is exploratory demolition: open the ceiling, open the wall, find nothing, close it up, bill the customer for the drywall. The cheap version is a sequence of measurements in which you change one path at a time and hold everything else constant, and read the flanking path off the differences. It needs a sound level meter, a substitute source you can switch on and off, and a record disciplined enough that the arithmetic closes at the end. What follows is that record, field by field, with what each line is for and what breaks when it is missing. Get to the end and you know which path owns the complaint and how much any remaining work on the others can possibly return.
Before anything runs
You are about to deliberately operate equipment and deliberately generate high levels, so the hazards here are ones your own instructions create.
- Running equipment. The machine has to run for this test. Guards stay on, nobody stands inside the guarding line or near a belt drive or coupling while it turns, and any adjustment, any hand on the machine, happens only after it is shut down, isolated at its disconnect and locked and tagged with stored mechanical energy relieved, per 29 CFR 1910.147. If you need to open an electrical enclosure, that work is carved out of 1910.147 and belongs to 29 CFR 1910.333(b)(2). Equipment under building-automation control can start on a schedule with nobody at the panel, so verify it will not restart before you put hands on it.
- The levels you generate. Wear hearing protection in the source room whenever the machine or the test source is running, and station nobody there for the duration. 29 CFR 1910.95, Table G-16, sets the general-industry permissible exposure at 90 dBA over 8 hours, measured A-weighted on slow response, and triggers a hearing conservation program at an 85 dBA 8-hour time-weighted average; the construction counterpart is 29 CFR 1926.52.
- Above a ceiling. Set the ladder to the rules that apply to your work, 29 CFR 1910 Subpart D in general industry or 1926 Subpart X in construction, and do not disturb thermal system insulation or surfacing material of unknown vintage. Presume it contains asbestos until sampled and leave it alone; the control is non-disturbance and sampling, not a dust mask, under 29 CFR 1910.1001 in general industry and 29 CFR 1926.1101 for construction work.
- Temporary seals. Never seal, gasket or block a fire door assembly, a rated opening protective, or any door in a required exit route, which are governed by NFPA 80 and NFPA 101 in the editions your authority having jurisdiction adopted, binding the building owner and reaching you through the permit, and by 29 CFR 1910.36 and 1910.37 for exit routes. Never restrict an opening that supplies combustion or make-up air to a fuel-fired appliance while that appliance can fire; shut it off and lock it out first, because restricting its air can spill combustion products into the room.
The record, line by line
Line 1: geometry and positions. Sketch the two rooms, mark the source position, mark each receiving position, and note what is continuous between them: ceiling grid, plenum, slab, deck, duct, conduit, pipe sleeve. Where the job needs a defensible field result rather than a diagnostic one, take the positions and spatial averaging from ASTM E336 in the edition the specification or the adopted code cites, and say on the sheet which you did; a diagnostic sweep is not a compliant test and should not be reported as one.
Line 2: how every level is written. A number on this sheet is not a measurement unless it carries five things in the same line: the quantity (pressure, not power), the weighting (A, C or unweighted), the reference (20 micropascals for pressure), the bandwidth (overall or the named band), and the time basis (fast, slow, equivalent-continuous over a stated duration, or peak). Two people comparing "46 dB" without those have compared nothing.
Line 3: background, source off. Read the receiving room with everything off. This is the field everyone skips and it decides whether the rest of the sheet means anything. If your measured level sits at least 10 dB above background, the correction is under half a decibel and you may ignore it, though print it anyway. Between 3 and 10 dB, apply it:
corrected = 10 log10 (10^(total/10) - 10^(background/10))
Below 3 dB of margin the result is not a value at all. Write it as an upper bound of the combined reading minus 3 dB, with one inequality sign, and stop treating it as a number.
Line 4: total, machine running. The complaint level. Corrected per line 3.
Line 5: airborne only, using a substitute source. Put a loudspeaker driven with broadband noise in the source room and switch the machine off. A loudspeaker excites the air and essentially nothing else, so what arrives next door is the airborne share alone. This line only works with its correction: the loudspeaker will not produce the same source-room level the machine did, so you must record the source-room level under both and normalize. The offset is only valid as a single overall number if the two spectra match band by band; check 125 Hz through 4000 Hz before you apply it, and if they do not match, normalize per band instead.
Line 6: the split. Structure-borne share equals total minus airborne share, done on energy, not by subtracting decibels. If the two shares do not add back to the measured total, one of the earlier lines is wrong.
Line 7: one path closed, retest. Temporarily seal a single suspected airborne path, subject to every restriction above, and re-read. The difference is that path's contribution, extracted the same way. Close one path per reading. Two at once and you have two unknowns and one equation.
Line 8: the ceiling on what is left. For each remaining path, the most that perfecting it can return is 10 log10 (1 + its energy divided by everything else's energy). This is the flanking card's gate and it is what turns the sheet into a decision.
The sheet, filled in
A tenant office above a plant room. All levels are re 20 micropascals, A-weighted, overall, equivalent-continuous over 60 seconds.
| Line | Reading |
|---|---|
| Background, receiver, all off | 32 dB |
| Total, machine running, receiver | 46 dB raw |
| Source room, machine running | 88 dB |
| Source room, loudspeaker only | 84 dB |
| Receiver, loudspeaker only | 39 dB raw |
| Receiver, machine running, one path sealed | 44.6 dB raw |
Correction, background, applied to the total. Margin is 14 dB. 10 log10 (104.6 - 103.2) = 45.8 dB. The correction moved it 0.2 dB and is carried anyway.
Correction, background, applied to the loudspeaker reading. Margin is only 7 dB, so this one is not optional. 10 log10 (103.9 - 103.2) = 38.0 dB. The correction moved it 1.0 dB, five times the previous one, in the direction that would otherwise have overstated the airborne share.
Correction, source-level normalization. Octave bands 125 through 4000 Hz agreed within 2 dB between machine and loudspeaker in the source room, so a single overall offset is acceptable here. The loudspeaker ran 4 dB below the machine, so the airborne share is 38.0 + 4 = 42.0 dB.
The split. Total energy 104.58 = 38,200. Airborne 104.20 = 15,800. Structure-borne = 38,200 - 15,800 = 22,400, which is 43.5 dB. Check: 15,800 + 22,400 = 38,200, back to 45.8 dB.
One path closed. The plant-room door undercut was temporarily sealed; the door is not rated, not in an exit route, and the only fuel-fired appliance in the room was already locked out. Corrected reading: 10 log10 (104.46 - 103.2) = 44.3 dB. The undercut's own contribution is 10 log10 (38,200 - 27,300) = 40.4 dB, and the remaining airborne paths are 10 log10 (15,800 - 10,900) = 36.9 dB. Check: 4,900 + 22,400 = 27,300, back to 44.3 dB.
The ceiling, printed. Remaining airborne energy 4,900 against structure-borne 22,400: 10 log10 (1 + 4,900 / 22,400) = 0.9 dB. Perfecting every remaining airborne path in the building buys nine tenths of a decibel. Running the same gate the other way, the structure-borne path is worth 10 log10 (1 + 22,400 / 4,900) = 7.5 dB.
What that decides. The undercut seal returned 1.5 dB, which is real and is below the roughly 3 dB most people report as a change in a steady broadband sound, so the tenant would not have called it fixed. The job is the mechanical connection between the machine and the structure, not the wall and not any further sealing. Nobody opened a wall.
Sibling-rule check, printed. Paths combined on energy rather than by subtracting decibels, per the flanking and small-hole cards: yes. Every level carries quantity, weighting, reference, bandwidth and time basis: yes. Background correction applied to every reported figure, declined nowhere, and its magnitude printed both times: yes. Gains quoted below their own ceilings, with no rounding taken in the flattering direction: yes.
Where this method breaks
A substitute source that is not omnidirectional. A directional loudspeaker aimed at the common wall loads that wall harder than the machine does and will overstate the airborne share. Aim it away from the partition and take a spatial average.
A machine whose airborne and structure-borne output are not independent. A duct or a pipe carries both, and switching the machine off removes both at once. Where services cross the boundary, the loudspeaker test isolates room air only, and the duct-borne path has to be closed separately.
A complaint that is tonal. Everything above runs on A-weighted overall levels, which is the right basis for a broadband complaint and the wrong one for a hum. Repeat lines 3 through 8 in the octave band that carries the tone. The band results are the ones that decide the fix; the overall number just tells you the test was clean.
Reading the surface instead of the path. A level that rises as you walk toward a ceiling or a column tells you that surface is radiating, not that it is the source. Confirm with a contact reading on the surface, and hand the tone-to-mechanical-cause step to the diagnostic family that owns it.
References
- 29 CFR 1910.95, Table G-16 (occupational noise, general industry); 29 CFR 1926.52 (construction); 29 CFR 1910.147 (control of hazardous energy, mechanical) and 29 CFR 1910.333(b)(2) (electrical work, carved out of 1910.147)
- 29 CFR 1910.1001 and 29 CFR 1926.1101 for presumed asbestos-containing thermal system insulation and surfacing material
- ASTM E336 (field measurement of airborne sound attenuation between rooms), in the edition the specification or adopted code cites; NFPA 80 and NFPA 101 in the editions adopted by the authority having jurisdiction
- See related: What Flanking Is and Why It Decides the Outcome; Why a Wall Never Performs Like Its Rating; How to Read a Spectrum Back to a Mechanical Cause; The Vibration Signature Reference