Power penalty

Most impairments do not break a link; they make it need more light. The power penalty is that need made quantitative: the increase in received power, in dB, required to hold the target error ratio compared with the unimpaired case. Penalties are the currency that lets one budget line absorb physics as different as modulator bias, fiber dispersion, and connector reflections, because each is converted to the same unit and summed in the link budget.

The textbook example is finite extinction ratio. A transmitter that does not turn fully off wastes power on a pedestal under the signal, and if the receiver specification is referenced to average power, the penalty for extinction ratio $r$ (linear) is

which is 2.2 dB at 6 dB ER and grows quickly below 4 dB. The same physics referenced to OMA vanishes: the eye opening has not changed, only the floor under it, which is why modern PAM4 specifications quote OMA and let average power be a separate, laser-facing concern. Whether a penalty exists can depend on the reference convention, and misreading the convention misstates the link.

Chromatic dispersion enters as an eye-closure penalty: pulse spreading converts a fraction of each unit interval into intersymbol interference. Short-reach specifications keep this penalty implicit by bounding the tolerable spread; the practical accounting for datacenter links is the UI-spread view in the Datacenter Link Budget Explorer, where a few tenths of a dB of penalty corresponds to spread confined to a modest fraction of the symbol. Other standing entries in the ledger: jitter (sampling away from eye center), multipath interference from paired reflections, polarization-dependent loss, and relative intensity noise, each typically budgeted at a few tenths of a dB and each capable of much worse when out of specification.

The modern refinement is to measure a transmitter's aggregate penalty directly rather than summing terms: TDECQ passes the actual waveform through a standardized reference receiver and reports the equivalent power cost in one number. A specification then splits the budget cleanly: the transmitter may spend up to its TDECQ ceiling, the channel its loss and dispersion allocation, and the receiver sensitivity is specified with those spends already assumed. Penalties are not physically independent, and summing them in dB is itself an approximation that leans conservative; it survives because the alternative, a joint simulation of every impairment, is not something a datasheet can promise.

References: Agrawal, Fiber-Optic Communication Systems (4th ed.), Sec. 4.6 (power penalty mechanisms); IEEE Std 802.3 PMD clauses (penalty allocations). The TDECQ entry covers the standardized measured penalty, and the PAM4 Eye & TDECQ Explorer shows impairments becoming dB live.