Nanoseconds to Microseconds

Counting Propagation Delay Down a Cable Run

Every metre of glass, copper or board laminate charges a signal a fixed toll in nanoseconds. Datasheets and field testers hand that toll over one metre at a time, while the budget you have to satisfy — a link latency target, a trading path, a clock distribution spec — is almost always written in microseconds. Dividing by a thousand is the small piece of arithmetic sitting between the two, and it gets done dozens of times while a route is being planned.

Conversion factor: 1 ns = 0.001 µs, so divide nanoseconds by 1 000. Single-mode fibre carries a pulse at roughly 4.9 ns per metre, so a 100 m riser costs about 489 ns — 0.489 µs, just under half a microsecond before a single transceiver has been powered up.

Why Each Medium Charges a Different Toll

Glass holds the pulse back

The group index of a standard single-mode core sits near 1.468, so light travels at about 68 % of its free-space speed — 4.9 ns per metre, which the industry rounds off to a convenient 5.

Copper is no quicker

Twisted pair has a nominal velocity of propagation around 0.65 to 0.70, landing on much the same 5 ns per metre. Cabling standards cap a 100 m channel at 555 ns for exactly that reason.

Boards carry their own bill

An FR-4 microstrip runs near 5.5 ps per millimetre — about 140 ps per inch — and a buried stripline is slower still at roughly 176 ps per inch, because all of its field sits in the laminate.

Air is the ceiling nothing beats

A radio hop crosses a metre in 3.336 ns, which is why a microwave path between two sites can undercut a fibre following the same map by a third of its flight time.

From a Length on the Drawing to a Delay Budget

The route length is usually the known quantity and the delay is what has to be proved, so the work runs in that order.

Multiply the run out and type the nanoseconds

A 250 m fibre path at 4.9 ns per metre comes to 1 225 ns. Type that in the left field and the microsecond value appears as you go. Spaces used to break up long figures are ignored, and a comma works in place of the decimal point.

Put the result beside the latency target

1.225 µs of pure flight time is easy to weigh against a one-way budget written in microseconds, and it shows immediately how much room is left for transceivers, forward error correction and switching.

Reverse it to size a run from a budget

The swap button (↔) turns the page into µs → ns, the direction you want when a spec allows 0.5 µs of propagation. That is 500 ns, and at 4.9 ns per metre it buys 102 m of fibre and nothing more.

Copy the digits into the budget sheet

Each field has its own copy control, handing over the bare number with no unit attached, so a row of route options assembles without editing every paste. Ctrl + C inside a field does the same thing.

Route length is not map length: a duct follows roads, climbs risers and carries slack coils at every splice enclosure, so the cable a signal really crosses is commonly 20 to 40 % longer than the straight line between two buildings. Delay budgets fail on that difference far more often than on the per-metre figure.

Propagation Delay by Medium and Run Length

Each row takes a transmission medium, its per-metre toll, a length that medium is genuinely used for, and what that run costs in both units. Velocity figures vary with cable construction, so treat them as planning values rather than acceptance criteria.

Medium	Delay per metre	Run length	Flight time (ns)	Flight time (µs)
FR-4 microstrip trace	5.5 ns/m	300 mm	1.65 ns	0.00165 µs
FR-4 stripline trace	6.9 ns/m	300 mm	2.07 ns	0.00207 µs
RG-6 coax, velocity factor 0.85	3.92 ns/m	30 m drop	117.7 ns	0.1177 µs
Cat 6 twisted pair	5.0 ns/m	100 m channel	500 ns	0.5 µs
Single-mode fibre, riser	4.895 ns/m	100 m	489.5 ns	0.4895 µs
Single-mode fibre, campus	4.895 ns/m	1 km	4 895 ns	4.895 µs
Single-mode fibre, amplifier span	4.895 ns/m	80 km	391 568 ns	391.568 µs
Free-space radio hop	3.336 ns/m	1 km	3 336 ns	3.336 µs

Read the rows top to bottom and the change of scale does the arguing for you. A long differential pair on a board is worth two nanoseconds, a patch panel run is worth half a microsecond, and one amplifier span between cities is worth as much as roughly two hundred thousand of those board traces laid end to end.

What Helps When You Are Adding Up a Path

Sweep a route without clearing anything

Both boxes stay editable and track each other, so you can walk a set of candidate lengths — 40 m, 250 m, 2 km — and watch where the total crosses out of nanoseconds.

Turn an allowance back into cable

The swap control flips the direction, so a microsecond allowance becomes a nanosecond figure you divide by the per-metre toll to reach an actual distance.

Picoseconds go in as decimals

Board-level skew is quoted in picoseconds and the smallest unit offered here is the nanosecond, so a 140 ps trace figure is entered as 0.14 and a 5 ps mismatch as 0.005.

Six-figure spans stay readable

Totals from a long-haul span are printed with spaced thousands, and the searchable dropdowns reach milliseconds for the moment a route grows past a few hundred kilometres.

Questions From the Cable Plant

What does a 100 m fibre run actually add to my latency?

About 489.5 ns one way, or 0.4895 µs, using 4.895 ns per metre for standard single-mode. Round trip it doubles to 979 ns. That is a real number but a small one: on the same path a pair of transceivers and one store-and-forward hop will usually cost several times more, so a 100 m difference in cable route is rarely where a latency complaint originates.

Why does a pulse crawl in glass when light in vacuum is so much quicker?

The core is a dense dielectric, and the quantity that governs a modulated pulse is the group index rather than the plain refractive index — near 1.468 at 1550 nm for standard fibre. Multiply the free-space 3.336 ns per metre by that ratio and you land on 4.895 ns per metre, an effective speed near 204 000 km per second instead of 299 792. Wavelength shifts it slightly, which is why precise timing work quotes the group index of the exact fibre type in service.

How far is one microsecond of delay in cable?

Roughly 204 m of single-mode fibre, since 1 000 ns divided by 4.895 gives 204.3. In free space that same microsecond stretches to 299.8 m. Scaled up, a kilometre of fibre costs 4.895 µs and a millisecond of one-way delay corresponds to about 204 km of glass — the rule of thumb behind every argument over where a matching engine ought to be hosted.

Does trace-length matching matter as much at 1 Gbps as it does at 10 Gbps?

It matters roughly ten times less, because the yardstick is the unit interval and that shrinks with the rate. One bit at 1 Gbps lasts 1 000 ps, so a 5 % skew allowance is 50 ps — around 9 mm of microstrip, which most layouts achieve without trying. At 10 Gbps the unit interval is 100 ps, the same 5 % becomes 5 ps, and that is under a millimetre of trace. Nothing about the laminate changed; the budget did.

Is the cable or the switch responsible for most of my delay?

Inside a building, the switch, by a wide margin. A store-and-forward device has to clock an entire frame in before it can send anything out, and a 1 518-byte frame at 1 Gbps takes 12.14 µs to serialise — still 1.21 µs at 10 Gbps. A 10 m patch cord contributes about 50 ns, or 0.05 µs, some two hundred times smaller. Cut-through forwarding exists precisely to remove that serialisation wait, and flight time only starts to dominate once the route leaves the campus.