Room Acoustics and the Wave Equation, Explained
After reading this you will know why a room rings the way it does, how the reverberation time RT60 is measured from an impulse response, and how to read the clarity, definition and mode plots the simulator produces.
What the simulator computes
Clap your hands in an empty tiled bathroom and the sound hangs for a second or more. Do the same in a bedroom full of soft furnishings and it dies almost at once. The difference is the impulse response: the pressure the room delivers to your ears when the source emits a single sharp pulse. Everything else follows from that one signal.
The tool solves the acoustic wave equation on your 2D floor plan. Air pressure p(x,y,t) obeys
Here c is the speed of sound, about 343 m/s in air at room temperature, and the two spatial terms measure how curved the pressure field is at each point. A pressure bump that is more curved than its neighbours accelerates outward. That is a wave spreading. The finite-difference time-domain (FDTD) scheme replaces each derivative with a difference between grid cells and steps the field forward in tiny time slices. Walls enter as boundary conditions that reflect some energy and absorb the rest.
When to use it, and when not
Use this for the questions a diffuse-field estimate cannot answer: where the dead spots sit, whether two parallel walls give you a flutter echo, how a curtain on one wall changes the tail. The FDTD result is exact wave physics up to the cutoff frequency shown in the results, usually in the 400 to 900 Hz range. That covers the room modes that colour bass and low mids, exactly the band where geometry matters most.
This is a 2D floor plan with the floor and ceiling folded in as a uniform loss (a "2.5D" model). It will not predict the true modal density of a real 3D room, and it treats each material with a single mid-frequency absorption coefficient. Read it as a teaching model of the horizontal sound field, not as a substitute for a measured room.
Do not use it for high-frequency detail, for scattering off small ornaments, or for concert-hall design decisions that turn on the ceiling shape. Above the cutoff the simulator synthesizes the tail from the measured energy decay rather than solving each wavefront.
Reverberation time and the Sabine formula
RT60 is the time for the sound level to fall by 60 decibels after the source stops, a factor of one million in energy. You rarely see a clean 60 dB of decay above the noise floor, so it is measured over a smaller span and extrapolated. The standard trick is the Schroeder integral: integrate the squared impulse response backward from the end to get a smooth decay curve, then fit a straight line to it.
For comparison the tool also shows Sabine's 1898 estimate:
Here V is the room volume in cubic metres and A is the total absorption in metric sabins, found by summing each surface area times its absorption coefficient \alpha. A perfectly reflective wall has \alpha = 0; an open window has \alpha = 1. The constant 0.161 is 24 \ln(10) / c with c = 343 m/s, which you can check: 24 \times 2.3026 / 343 = 0.1611.
Sabine assumes the sound field is diffuse: equal energy travelling in every direction, everywhere. Long corridors, curved focusing walls and rooms with all the absorption on one surface break that assumption, and there the FDTD result and the Sabine estimate will disagree. That disagreement is the interesting part.
Reproducing the demo room
Press Clap on the default Living room and work through the Sabine estimate by hand. The plan is a rectangle 5.5 m by 4.2 m with a 2.6 m ceiling: plaster walls at \alpha = 0.05 with a 2 m glass window at \alpha = 0.04, a 2 m curtain at \alpha = 0.50 hanging just in front of one wall, a wood floor at \alpha = 0.10, and a plaster ceiling at \alpha = 0.05.
- Volume: V = 5.5 \times 4.2 \times 2.6 = 60.1 cubic metres.
- Walls: the perimeter 2(5.5+4.2) = 19.4 m is
17.4m of plaster and2m of glass. Times the2.6m height: 17.4 \times 2.6 \times 0.05 = 2.26 sabins and 2 \times 2.6 \times 0.04 = 0.21 sabins. - Curtain: sound reaches both faces, so 2 \times 2 \times 2.6 \times 0.50 = 5.20 sabins.
- Floor: 23.1 \times 0.10 = 2.31 sabins. Ceiling: 23.1 \times 0.05 = 1.16 sabins.
- Total absorption A = 2.26 + 0.21 + 5.20 + 2.31 + 1.16 = 11.14 sabins.
- Sabine RT60: 0.161 \times 60.1 / 11.14 = 0.87 seconds.
The tool's own Sabine readout says 0.80 s: on its 5 cm grid the walls have thickness, which trims the air volume by about 5 %, and the curtain's ends add a little absorbing surface. The simulated RT60 comes out at 0.86 s. Now erase the curtain: A drops to 5.94 sabins and RT60 climbs to 0.161 \times 60.1 / 5.94 = 1.63 seconds. One soft surface roughly halves the reverberation.
Watch the decay build the number
The Schroeder curve is the clearest way to see RT60. The chart below plots the backward-integrated energy of the demo room in decibels against time. Fit a line to the straight middle section and its slope gives the decay rate. Over the fitted span the default room falls about 70 dB per second, so 60 dB takes 60 / 70 \approx 0.86 seconds, within a few percent of the Sabine hand calculation above.
If the Schroeder curve bends, with a fast early drop and a slower tail, the field is not diffuse. That is common when one wall carries most of the absorption. Report early decay time (EDT), the slope over the first 10 dB scaled to 60 dB, because that is closer to what your ear judges.
Reading clarity, definition and modes
RT60 is one number. The impulse response holds more. Two measures split the energy by arrival time, dividing at the direct sound.
- C80 (clarity)
- Ratio in decibels of energy in the first 80 ms to everything after. C_{80} = 10 \log_{10}(E_{0\text{-}80} / E_{80\text{-}\infty}). Positive values sound clear; music usually wants -2 to +2 dB.
- D50 (definition)
- Share of energy in the first 50 ms, D_{50} = E_{0\text{-}50} / E_{0\text{-}\infty}, as a fraction from 0 to 1. Speech wants above
0.50for good intelligibility.
The frequency response shows room modes: standing waves at frequencies where a half-wavelength fits between two walls. For a rigid pair of walls separated by length L, modes fall at
For the demo room's 5.5 m length, f_1 = 343 / (2 \times 5.5) = 31.2 Hz, then 62.4, 93.5 Hz, and so on. The 4.2 m width gives 343 / 8.4 = 40.8 Hz and its multiples. Each mode is a peak in the frequency response and a fixed pattern of loud and dead spots in the room. Move the listener onto a null and that note nearly vanishes.
Common mistakes
The first mistake is trusting Sabine in a room it was never meant for. In a 15 m corridor with soft ends and hard sides, sound bounces down the length long after the diffuse estimate says it should be gone. The FDTD tail will run longer than 0.161 V/A predicts, and the FDTD number is the trustworthy one.
The second is placing the listener at a mode null and concluding the room lacks bass. It does not; the seat does. Move it half a metre and the missing note returns.
The third is over-reading the high-frequency reverb tail. Above the cutoff the tail is synthesized from the energy decay, not solved wavefront by wavefront, so it captures the overall ring but not exact echo timing. The fourth is forgetting that each material uses one absorption number. A real curtain absorbs treble far better than bass, so a curtained room in the simulator will sound more uniformly dead than the same room in life.
Related tools
The same finite-difference idea, stepping a field forward on a grid, drives several other simulators here. The Wind Tunnel solves fluid flow rather than acoustic waves, and the Circuit Simulator steps voltages and currents through a network in time. If you like watching physics unfold on a grid, the Nuclear Reactor Simulator tracks neutron and heat balance the same way.
Frequently asked questions
Why is my measured RT60 different from the Sabine number?
Sabine assumes a diffuse field. When absorption is uneven or the room is long and thin, the real decay is not a single straight line, so the two disagree. The FDTD impulse response is the physical answer; treat Sabine as a quick sanity check, not the truth.
What counts as a good RT60?
It depends on use. A voice studio wants roughly 0.3 s, a living room 0.4 to 0.6 s, a classroom under 0.7 s, and a concert hall 1.8 to 2.2 s. Shorter is not always better: a dry room can sound lifeless.
Why does the simulation only go up to a few hundred hertz?
FDTD needs many grid cells per wavelength to stay accurate. A few-centimetre grid resolves waves down to a fraction of a metre, which sets the cutoff near 400 to 900 Hz. Above that the tool synthesizes the tail from the energy decay instead of solving each wave.
Can I use the impulse response WAV as a reverb?
Yes. Download it and load it into any convolution reverb plugin. Convolving a dry recording with the impulse response produces the sound as it would ring in that exact floor plan, within the model's limits.
What is a flutter echo and will I see it?
It is a repeating buzz between two hard parallel walls. In the simulator it shows up as regularly spaced spikes in the impulse response. Tilt or absorb one wall and the spikes fade.