Settings

The recorders, and when each heard it

RecorderxyHeard at (s)

The times only have to be on one clock: everything here is differences, so a clock wrong by a constant changes nothing at all. One recorder wrong relative to the others moves the answer and cannot be seen in the result, so reconcile the clocks first if they came off separate devices.

How fast it travelled, and how well you know the times

Sound in air is about 343 m/s at 20°C and changes by roughly 0.6 m/s per degree, so summer speed used on a winter night is two per cent out: four metres on a two hundred metre baseline, before any timing error at all. Radio is a different problem and takes its own speed. The timing figure is what the band below is computed from, so guessing it low is how a tool comes to promise a precision it does not have.

Where it came from

Squares are the recorders, the cross is the answer, and the faint cloud is every position consistent with your times at the accuracy you declared. The cloud is the point of the drawing: if it is a long smear rather than a blob, the array shape is doing that, not the arithmetic.

Do the times agree with each other

“Out by” is the gap between when a recorder actually heard it and when the answer says it should have. With more than three recorders these cannot all be zero, and a row that is out by far more than your stated accuracy means no single source produced all of these arrivals: a wrong position, a wrong sign, a mis-clocked recorder, or more than one event.

The working


    

Questions people ask about TDOA

Why do I get a band rather than a position?

Because a position on its own would be dishonest. Sound travels about a third of a metre in a millisecond, so a millisecond of timing error is a third of a metre of path difference, and the geometry multiplies that. The same error is worth about half a metre for a source inside a square of recorders and many times that when they are nearly in a line. The page perturbs every arrival time by the accuracy you declared, in every combination, re-solves each one, and shows you the spread. That spread is the answer; the cross in the middle of it is just where the spread is centred.

Why does it sometimes refuse to answer at all?

Two geometries genuinely have no solution. Recorders in a straight line cannot tell one side of the line from the other (four microphones along a street frontage is a line), and the equations are singular, so any answer would be an invention. And a source at the exact centre of a symmetric array is the same distance from every recorder, so all the arrival times are identical and nothing indicates which way to move. The second is a curiosity, since nothing real sits exactly on a centroid, but both get explained rather than silently producing nothing.

How many recorders do I need?

Three, for a position in two dimensions. There are three unknowns (two coordinates and the moment it happened, which nothing measured) and each recorder contributes one equation. A fourth and a fifth do not add unknowns, so they become checks: with three the times must fit perfectly by construction, and only with four or more can the page tell you the arrivals disagree with each other.

Does it matter that I do not know when it happened?

Not at all, and that is the elegance of the method. The emission time is solved for alongside the position rather than needing to be supplied. A practical consequence worth knowing: adding the same constant to every arrival time moves the answer not one millimetre, it only shifts the emission time by that constant. So the recorders must share a clock, but that clock can be arbitrarily wrong.

What is the array shape number?

How spread the recorders are across their own best-fit line, as a fraction of how spread they are along it. Zero is a perfect line and has no solution; near one is a square, which is as good as the geometry gets. It is shown next to the answer because it is the thing that decides how wide the band is, and a printed coordinate gives no hint of it at all.

Can I use this for radio rather than sound?

Yes: put the propagation speed in and the arithmetic is identical. Be aware of what that does to the timing, though: at the speed of light a microsecond is three hundred metres, so a position good to ten metres needs the arrival times to about thirty nanoseconds. That is a demanding requirement and it is the reason radio direction-finding uses specialised hardware rather than ordinary recorders.

Doing this often? OBSCURA Case keeps the analysis with the rest of the evidence, in the chronology and the case file. Everything is US$5 a month or US$48 a year.