← Interview Prep

Transceiver Power & dBm

Optical power, link budgets, and reading DDM/DOM diagnostics — the physical meaning, not the folklore.

The physical meaning of dBm

dBm isn't "some optics unit" — it's ordinary power in milliwatts, just expressed logarithmically:

P(dBm) = 10 × log₁₀( P(mW) / 1 mW )

So the reference point is 0 dBm = 1 mW. Anything below a milliwatt gives a negative number — which is why almost everything in optics carries a minus sign. That's not "bad," it just means powers on the order of microwatts.

It helps to memorize a few anchors — you never need to compute logarithms in your head:

The two-number rule: 3 dB is a factor of 2, 10 dB is a factor of 10. Combine them for anything: −13 dB = ÷10 ÷2 = a factor of 20.

Why a logarithm at all

Because losses along a path multiply, and a logarithm turns multiplication into addition. Fiber attenuates by 3×, a connector by another 1.2×, a splice by a little more — instead of multiplying fractions, you just add decibels:

RX(dBm) = TX(dBm) − loss(dB)

The whole link calculation collapses into one line of arithmetic. That's the entire point.

dB vs dBm — constantly confused

• dBm — absolute power (how much light). It applies to TX and RX.
• dB — a ratio, i.e. loss or gain. It applies to fiber, connectors, attenuators.

"A connector adds 0.5 dB" is correct. "A connector adds 0.5 dBm" is nonsense. Interviewers notice.

Power is set by the standard, not a rule of thumb

You'll hear claims like "the ideal TX is −6 dBm, range −1 to −7." That's a marketing generalization and it's wrong as a universal rule. Power is set rigidly by the standard for a specific transceiver: an 80 km ZR module has a normal TX of +3 dBm (eight times a milliwatt!), and −6 dBm would mean a dying laser. For an SR it's the opposite.

Exact numbers always come from the specific module's datasheet, but the relationship matters more: the longer the reach, the stronger the transmitter and the more sensitive the receiver.

Link budget: how to compute it

Budget = TX_min − RX_sensitivity

For 10GBASE-LR: −8.2 − (−14.4) = 6.2 dB — that's how much loss the path may absorb in the worst case.

What eats the budget:

Example: 8 km of single-mode with four cross-connects:

8 km × 0.35              = 2.8 dB
4 connector pairs × 0.5  = 2.0 dB
2 splices × 0.1          = 0.2 dB
──────────────────────────────
total loss                 5.0 dB
LR budget                  6.2 dB
margin                     1.2 dB   ← too little!

The design-minimum margin is 3 dB. You need it for laser aging, connector contamination during future work, temperature drift, and possible extra splices after a cable cut. In the example above the link will come up and run happily for years — and then one day drop after someone tidies the patch room.

Note the orders of magnitude: on short links the connectors eat more than the fiber itself. Five cross-connects inside a data center cost more than two kilometers of route.

Reading real diagnostics (DDM/DOM)

Modules with DDM/DOM report live telemetry:

Arista/Cisco:  show interfaces transceiver
Juniper:       show interfaces diagnostics optics
Linux:         ethtool -m eth0

What to look at and how to read it:

• RX power — the key number. Compare it to the computed TX_far_end − loss. A discrepancy of more than a couple of decibels means a path problem.
• TX power — if it's below normal for this module type, the laser is degrading.
• TX bias current — the most underrated parameter. The laser is power-stabilized: as it ages, the circuit raises the bias current to hold the output. So a rising bias with falling (or even normal) power = a module heading for failure. This lets you replace optics on a schedule instead of in an outage — directly valuable for trading, where swaps only happen in a maintenance window.
• Temperature and voltage — overheating degrades both power and sensitivity.
• Alarm/warning thresholds (high/low) are reported by the module itself — build alerting on those, not on absolute constants from a blog.

Diagnosis by RX value

About the −25/−30 dBm reading you'll see quoted as "interference and noise on the receiver": that's wrong. Optics are immune to electromagnetic pickup — that's the whole point of them. Such a value means exactly one thing: the photodiode sees no light and is reporting its noise floor. Look for a break, a dark far-end port, or swapped fibers.

Tie-in with latency

Since you're prepping for trading: fiber length affects not only the budget but also the delay — ≈ 5 µs/km. Two things worth saying out loud:

• Route selection is simultaneously an optical budget and microseconds. On arbitrage routes people accept more expensive optics with a bigger budget just to keep the route shorter.
• DWDM transponders and amplifiers buy you budget but add their own processing delay, and dispersion compensators built from spools of fiber are literally extra kilometers of glass — i.e. direct microseconds. Trading accounts for this explicitly and often drops compensation in favor of a shorter, unmodulated path.