Stopping Sound Through a Partition: Mass, Gaps and Decoupling

The whole budget is a 33 mm square

Take an ordinary separating wall, 4.0 m long and 2.7 m high: 10.8 m² of partition. Ask it to hold 40 dB. The total open area across the whole of it — every unsealed slot, every cable notch nobody went back to, the clearance under the door, the gap the mullion was packed into but not sealed — has to stay under about 1,080 mm². That is a square 33 mm on a side, for the entire wall, once. Ask the same wall to hold 50 dB and the budget collapses to 108 mm², which is a square 10 mm on a side. Everything else on this page is spending decisions made against that one figure.

The arithmetic is short enough to do in your head. An aperture that is genuinely open passes essentially all the sound arriving at it, so once a fraction f of the wall area is open, at least that fraction of the incident energy is already through, and no construction on the remainder can lift the assembly above minus ten times the base-ten logarithm of f. One per cent open pins the wall at 20 dB. A tenth of a per cent pins it at 30. The other 99.9 per cent of the surface is not being ignored — it is simply not the term that decides the answer.

Treat the free-area figure as a design ceiling rather than a prediction, because the assumption underneath it is an idealisation. A narrow slot with depth to it is not a free opening: its transmission depends on frequency and on how deep the slot runs, and at frequencies where the slot column resonates it can pass more energy than its bare area suggests, while well away from that it passes less. What the arithmetic gives you is a limit you cannot argue your way past, in the speech range, using the one dimension anybody can actually measure on site.

So the budget gets spent before the build-up is chosen, not after. Write the target down, divide it by the wall area, convert the fraction into square millimetres, and then go through the drawing allocating that allowance: this much to the door threshold, this much to the service crossings, none at all to a transfer grille. If the allocation does not close, the answer is not a heavier wall. The sequencing of the seals themselves — what gets closed, at which hold point, and who owns the reseal when the last trade opens it again — is a site discipline covered in the companion guide on partition isolation, and it is a different job from this one.

Weak elements are weighted by area, then they refuse to average

A wall with different elements in it is combined on transmission coefficients weighted by area, not on ratings weighted by area, and the difference is the whole point. Convert each element's transmission loss to a coefficient, multiply by that element's area, sum, divide by the total area, convert back. Put an aperture the size of one flush mounting box — call it 0.0086 m² — into a 10.8 m² wall rated 55, and give that aperture a transmission loss of only 20 dB because it is a thin steel box with a skim over it. The wall reads 49.5. Five and a half decibels handed over to something the size of a postcard. Leave the same area genuinely open and the wall reads 31.

One qualification belongs on the drawing next to any number produced this way. The composite relationship is a statement about transmission loss in one-third octave bands, and STC is not a transmission loss — it is a contour fitted to sixteen of them. Running the area weighting directly on single-figure ratings is a screening calculation, good enough to rank options and to show a client where the money is going, and not good enough for a submittal. The defensible version takes ASTM E90 band data for each element, combines band by band, and re-rates the result to ASTM E413. Say which one you did.

Six decibels a doubling, and the doublings get expensive

Away from its resonances a single panel follows the mass law, commonly written for field incidence as transmission loss approximately equal to 20 log10 (m × f) − 47, with m the surface density in kg/m² and f the frequency in hertz. Two consequences fall straight out of the logarithm and they govern every argument about board schedules. Performance rises 6 dB for every doubling of surface density, and it rises 6 dB per octave, which is why the same wall that comfortably stops a conversation does nothing about a bass line.

Put board weights into that. A 15.9 mm Type X sheet runs somewhere around 10 to 11 kg/m², roughly 2.0 to 2.3 lb/ft², and lightweight board families are meaningfully under that — take the figure off the manufacturer's data sheet rather than from habit, because the two are not interchangeable once you are counting kilograms. A second layer takes one leaf from about 10 to about 20 kg/m². That is one doubling: about 6 dB. A third takes it from 20 to 30, worth 3.5 dB. A fourth takes 30 to 40, worth 2.5 dB. The layers get worse as they are added, and the fourth one weighs exactly as much as the first while buying less than half of it.

Underneath the mass law sits the thing that spoils it. Every panel has a critical frequency where the bending wave in the panel matches the airborne wave alongside it, the two couple efficiently, and the transmission loss falls into a dip before recovering above. That frequency drops as a panel gets thicker and stiffer, so a single thick board pulls its dip down toward the range that carries speech while two thinner boards keep theirs up out of the way. For 12.7 mm gypsum the dip sits in the region of 2.5 to 3 kHz. Two 12.7 mm layers and one 25 mm layer can be handed to you as the same surface density; acoustically they are not the same specification, and the assembly test report is the only place that difference is visible.

It is worth knowing how ASTM E413 treats such a dip when it turns sixteen bands into one number. The classification allows deficiencies below the reference contour up to a total of 32, with no single band deficient by more than 8. A narrow deep trough can therefore be absorbed and the rating still comes out where the specification wanted it, while a broad shallow shortfall across many bands costs more. That is not a flaw in the method; it is the method doing what it was built for. But it means an assembly can hold its rating and still be transparent at precisely the frequency somebody is complaining about, and no amount of re-reading the rating will show you that.

Two masses and a spring, resonating at 89 Hz

A double-leaf partition is not two walls in series, and treating it as though it were is where most of the disappointment comes from. It is a mass-spring-mass oscillator: the two boarded leaves are the masses and the trapped air in the cavity is the spring. Above the resonance of that system the pair does dramatically better than their combined weight would suggest, which is the entire reason anybody builds a cavity. At and below the resonance they do worse than one leaf of the same total weight, because the two leaves are moving as one and the spring is simply along for the ride.

The resonance frequency is commonly estimated as f0 ≈ 60 × √((m1 + m2) / (m1 × m2 × d)), with the surface densities in kg/m² and the cavity depth d in metres. Two 10 kg/m² leaves across a 90 mm cavity land at about 89 Hz. Deepen the same wall to 150 mm and it falls to about 69 Hz. Double the board on both faces, so each leaf is 20 kg/m², and 90 mm gives about 63 Hz while 150 mm gives about 49 Hz. Depth is the cheapest decibel available on the drawing and it is the one the floor plan usually refuses to hand over — which is a conversation to have with the architect at layout stage, not with the framer.

Then compare those frequencies with the band the wall is being rated across. ASTM E413 fits its contour from 125 Hz upward; the basic range in ISO 717-1 starts at 100 Hz. A resonance at 89 Hz is below both. The assembly is graded on the bands where it performs well and is not graded at all on the band where it has a hole, which is a fair description of why a party wall with an unimpeachable certificate still transmits a subwoofer through three rooms. This is also the reason ISO 717-1 carries the spectrum adaptation terms C and Ctr, the second of which weights a source spectrum with far more low-frequency energy in it, and the reason a specification written around Ctr is a genuinely harder specification than the same number written without it.

The door is seventeen per cent of the wall, and it moves

A 0.9 m by 2.1 m door leaf is 1.89 m², which is 17.5 per cent of the same 10.8 m² wall. Hang a door assembly rated 30 in a wall rated 55 and the composite reads 37.5. The wall has been thrown away, and it was thrown away on the door schedule, by someone who was pricing ironmongery. Specify the door assembly at 45 instead and the composite comes back to 50.9. Nothing about the wall changed in either case.

Then add the clearance. Ten millimetres across the 900 mm width is 0.009 m², which is eight hundredths of one per cent of the wall — and it drags the same 45-rated door assembly's composite down to 30.7. Even a three-millimetre clearance, on its own, caps the whole partition at 36 dB. The leaf's rating never gets an opportunity to matter. This is the sharpest illustration on the page of the opening argument: the number falls to whatever the weakest route allows, and the weakest route on most partitions is a gap somebody left for a reason that had nothing to do with acoustics.

So the door has to be specified as an assembly and bought as one. A rated door set is a leaf, a frame, perimeter seals, a threshold or drop seal, and the seal between the frame and the structure, tested together; a leaf bought on its own published figure and hung in somebody else's frame carries no rating at all, and there is no arithmetic that reconstructs one. The corollary belongs in the mechanical scope rather than the door schedule: ventilation transfer cannot be taken through an undercut on a partition with an acoustic requirement, so that air has to be found and attenuated somewhere else, and that has to be settled before the door schedule is issued rather than after.

One last thing separates a door from every other element on the wall: it is the only part with a moving mechanism whose performance depends on an adjustment. A drop seal is set against a specific finished floor level. Set it on a bare slab and it will either fail to deploy onto the finished floor or bind against it, and either way the sealing line the test report assumed does not exist. Re-check it after the floor finish is laid, and again at handover, and treat both checks as scheduled work rather than as a snagging item.

1. Take the door and glazed areas off separately, and hold their own ratings against them rather than netting them into the wall area.
2. Specify door sets by assembly rating, naming leaf, frame, seals and threshold as one item.
3. Confirm the transfer-air route on the mechanical drawings before issuing the door schedule.
4. Set the threshold clearance against the finished floor level, not the slab.
5. Re-adjust the drop seal once the floor finish is down, and record it.

Which number is the contract actually asking for

Two families of measurement are routinely confused on the same job, and the confusion is expensive because the obligation each creates is different. ASTM E90 measures a specimen mounted between two isolated chambers with everything around it deliberately suppressed, and ASTM E413 turns those bands into an STC. ASTM E336 measures two rooms in a finished building and reports field quantities that already contain whatever the building is doing around the wall. The ISO framework runs the same split: ISO 10140-2 in the laboratory, ISO 717-1 for the single-figure rating and its adaptation terms, ISO 16283-1 in the field. Specifying a laboratory rating and then testing to a field method is not a stricter version of the same thing; it is a different question, and the contract should say which one it is buying.

For the question that sits between those two — what will this actually measure, given everything else the building is made of — the named framework is ISO 12354-1, Building acoustics: estimation of acoustic performance of buildings from the performance of elements, Part 1, airborne sound insulation between rooms. It is the document to reach for when someone wants a predicted field figure rather than an element rating, and it is honest about the inputs it needs, which is usually the point at which a project discovers it does not have them.

Regulatory targets vary by jurisdiction and by adopted edition, and this is not a place to work from memory. In the United States the International Building Code's sound transmission provisions for Group R occupancies set the requirement for dwelling-unit separations and distinguish a laboratory-tested value from a lower field-tested one; confirm the figures against the edition your authority has actually adopted along with any local amendment. In England and Wales, Approved Document E, Resistance to the passage of sound, states its criterion as a field quantity that already includes the low-frequency Ctr adaptation term, with different minima for new build and for a material change of use. In Australia the National Construction Code likewise sets weighted ratings carrying a spectrum adaptation term. All three are asking a question that a laboratory STC on a submittal does not by itself answer.

The order the arithmetic implies is the order to work in, and it is close to the reverse of the order these decisions usually get made. Budget the open area against the target first, because that is a hard ceiling and nothing recovers from it. Take off the openings second and hold their own ratings against them, because they are weighted by area and they are where the improvement is cheap. Fix the cavity depth third, while the plan can still be argued about, because depth is what puts the resonance below the rated band. Choose the board schedule last, knowing that each layer is worth less than the one before it. Then hand the drawing over to the site sequence — the seals, the hold points, the reseal after the last trade — which the companion guide on partition isolation sets out in the order they are actually built, and where every one of these numbers is either kept or quietly lost.