ROADM Power Equalization and APC Explained

In a DWDM network the difference between a channel that runs for years and one that steadily degrades is usually per-channel power, not the fibre. Amplifiers apply gain to the whole band, so adding or dropping a wavelength changes the input power every surviving channel sees, and small errors accumulate across ROADM hops into OSNR loss and bit errors. This article explains what automatic power control (APC) does at the amplifier and node level, how ROADM power equalization keeps the band flat, and which measurements to check when channels drift.

Why Power Drifts in the First Place

• Channel count changes: an EDFA with constant gain sees total input power move when wavelengths are added or dropped, and a fast transient-suppression loop must rebalance output power to keep per-channel power constant.
• Span loss changes: ageing fibre, added splices, dirty connectors and re-patched jumpers all shift the loss a span presents, so a gain setpoint calculated at turn-up becomes wrong.
• Component ageing: WSS and VOA insertion losses drift, and the node's expected power values stop matching the photodiode readings.

None of these faults announce themselves. The link keeps carrying traffic, but the OSNR margin that was 4 dB at commissioning quietly becomes 1.5 dB, and the first hot day or the first fibre bend pushes pre-FEC BER past the FEC threshold. Power control is the discipline that keeps that margin intact.

What APC Does

At the amplifier card level, the amplifier calculates its gain setpoint from the expected per-channel power after the first channel is provisioned, then holds that setpoint - the value is not continuously re-optimised, and it is recalculated when the provisioned channel count returns to zero. At node and network level, APC compares measured power with expected power and corrects gain or VOA setpoints to compensate for span loss. Expected power is derived from the provisioned per-channel power, the channel distribution (express, add and drop channels in the node) and an ASE estimate.

! typical verification views on a DWDM NMS/CLI
show amplifier gain
show channel power
show roadm equalization
show span loss

The important distinction is timescale. The card-level loop reacts in microseconds to protect surviving channels from a power excursion. The node-level and network-level APC is a slower, supervisory loop: it may take seconds to minutes to converge and it is correcting for a physical change (span loss drifted) rather than for a transient. Confusing the two leads to people disabling transient suppression to "fix" a level problem, which is the wrong lever entirely.

Amplifier Control Modes: APC, AGC and ACC

Every optical amplifier exposes at least three control modes, and the mode in use determines which knob the node controller can turn:

• ACC (automatic current control): pump current is held fixed, so gain floats with input power and the output power follows whatever the input does. Useful for testing and for the very first turn-up before the channel plan is known.
• AGC (automatic gain control): gain is held at a setpoint, so output power tracks input power plus gain. This is the normal operating mode on a span amplifier because the gain setpoint maps directly to the span loss it must compensate. It is also the mode that produces per-channel power excursions when channels are added, unless a fast transient loop overlays it.
• APC (automatic power control): total output power is held constant by adjusting gain. Adding a channel therefore reduces the gain, which correctly keeps using the last few dB of pump headroom instead of driving the amplifier into saturation.

In practice a modern line amplifier runs AGC for steady-state accuracy with a transient-suppression function that behaves like APC for the few microseconds around a channel event. That combination is why a well-configured 40-channel span survives a 20-channel add without the surviving channels losing more than a fraction of a dB.

Target Power, Per-Channel Power and the Arithmetic Nobody Does

Per-channel power is a design number, not a measurement. A typical coherent 100G or 200G channel is provisioned at around -2 to +1 dBm at the amplifier output, and the total output power the amplifier must deliver follows directly from the channel count:

Total output power (dBm) = per-channel power (dBm) + 10*log10(N channels)

Example: 0 dBm per channel
  1 channel   ->  0.0 dBm
  40 channels -> 16.0 dBm
  80 channels -> 19.0 dBm
  96 channels -> 19.8 dBm
  10 channels removed from 96 -> 19.3 dBm

This is the single most useful equation in the whole topic. It tells you why a 20 dBm amplifier runs out of headroom at 96 channels of 0 dBm, why the last few channel adds barely move the total, and what the amplifier output should read when you suspect the photodiode. If the measured total output is 3 dB higher than the arithmetic predicts, either the channel count is wrong or the amplifier has silently moved into a different control mode.

ROADM Power Equalization

An equalization sweep measures each channel and adjusts the attenuator or WSS setting so all channels leave the node at the same target power. Two symptoms point at unequalised channels: one channel shows a significantly lower OSNR than its neighbours, or the pre-FEC BER of a single wavelength degrades while the rest of the band is stable. Both are node-level problems, not amplifier problems - raising the band's gain makes the offending channel worse and pushes good channels into non-linear territory.

Equalization is per hop, and errors compound. A 1 dB uncompensated error at one node becomes a 4 dB error after four hops if nothing corrects it, and OSNR contributions from each amplifier then diverge. This is why the equalization state of every ROADM in the path - not just the one nearest the reporting channel - belongs in the incident notes.

How an Equalization Sweep Actually Works

1. The node controller knows the channel plan from the provisioned cross-connects and the measured optical spectrum from the per-degree monitor ports.
2. It measures each channel's power at the WSS input and output and compares it with the target.
3. It applies a per-channel attenuation correction at the WSS (or at a VOA in older designs) so the launch power out of each degree converges on the target.
4. It re-measures and iterates until the residual error is inside the platform's tolerance, typically plus or minus 0.5 dB.
5. It stores the resulting attenuation map. If that map later diverges from the current spectrum by more than a set threshold, most platforms raise a power-degradation alarm.

A sweep is a maintenance action, not a background task on most equipment. Running one during peak traffic causes a short but real power perturbation on every channel in the node, which is why change windows exist. On agile platforms the sweep can be done per degree, so an add/drop degree can be re-equalised without touching an express degree that carries 80 protected channels.

Transient Suppression When Wavelengths Change

The worst-case event in a DWDM network is not a fibre cut - it is a channel cut. When one of ninety-six wavelengths disappears, the amplifier suddenly has more gain available per surviving channel, and without suppression the survivors can see several dB of overshoot for milliseconds. That overshoot temporarily pushes channels into the non-linear regime, degrades OSNR and, in extreme cases, blinds a receiver long enough to trigger a protection switch.

! what to look at on the amplifier and node
show amplifier transient-suppression status
show amplifier power excursion log
show channel power history <channel>
show roadm degree <n> spectrum

Fast transient suppression in modern amplifier hardware responds in microseconds and holds the excursion to a fraction of a dB. Two things defeat it: an amplifier running without enough pump headroom (the loop cannot move gain fast enough because there is nothing left to reduce), and a network-level APC setpoint that fights the fast loop by trying to hold total power constant on a slow timescale. Set the pump headroom target generously on spans that will carry many protected channels.

Span Loss Verification

1. Measure span loss at turn-up and record it per span; include connector and patch-panel losses honestly.
2. Re-measure after any fibre work and compare with the recorded baseline.
3. If measured loss exceeds the design, fix the physical layer first: cleaning connectors resolves a surprising share of "amplifier" tickets.
4. Confirm the gain setpoint still matches span loss, and use APC corrections rather than manual offsets where the platform supports them.

A useful sanity check is that the amplifier's measured gain should equal the span loss it compensates, plus or minus the amplifier's own design margin. If a span with a recorded 22 dB loss shows a 22 dB gain setpoint but the photodiodes disagree by 3 dB, the disagreement is either a measurement problem (a dirty connector at the monitor port) or an attenuation element somewhere in the path that was never documented. Both are worth an hour of a maintenance window, because undocumented attenuation is what turns every future capacity upgrade into a fight with the power budget.

Symptom to Cause: A Short Table

Pre-FEC BER and OSNR Are the Referee

Power and equalization are means to an end, and the end is pre-FEC BER with margin to the FEC threshold. A span can measure perfectly on every power meter and still fail because the noise figure is poor, the filter shape is wrong at the channel's edge, or the channel has been placed on a grid point that another channel's skirt is clipping. When the power readings are clean and BER is not, the investigation moves to the coherent layer - the detailed method is in coherent optics pre-FEC BER and OSNR checks.

Turn-Up and Commissioning Order

Equalise before you judge performance. Commissioning order matters: verify amplifier gain and span loss, equalise channels at each ROADM node, then measure OSNR and pre-FEC BER. Measuring OSNR before equalisation produces numbers that cannot be compared with the design budget and leads to pointless amplifier tweaking. Keep the commissioning record - the per-span loss, per-channel power target and measured OSNR - as the baseline for every future incident.

The commissioning record is the artefact that makes every later incident tractable. A test report that lists span loss, per-channel launch power, equalization residuals and measured OSNR per node turns a four-hour fault hunt into a ten-minute comparison. The wider checklist is the DWDM system commissioning checklist linked at the end of this article.

Common Mistakes That Cause Repeat Tickets

• Raising band gain to fix one low channel. This pushes good channels toward non-linearity and rarely helps the weak channel, which is usually an equalization or filter problem.
• Turning off transient suppression during a maintenance window and forgetting to re-enable it. The next unprotected channel event produces an OSNR hit nobody can explain.
• Treating the amplifier setpoint as a measurement. The setpoint is what the node intends; the photodiode is what actually happened. Compare them.
• Equalising once at turn-up and never again after a fibre repair. Every splice changes the loss a span presents, which invalidates the gain setpoint that was correct last year.
• Mixing per-channel target powers across a path. Pre-emphasis is a deliberate design choice; accidentally running one node 2 dB hot makes the OSNR contribution budget meaningless.

Frequently Asked Questions

Should APC run continuously? On most platforms the fast transient loop is always on and the supervisory APC runs on a schedule or on trigger. Run supervising APC on trigger plus a periodic sweep; continuous correction chasing measurement noise produces more alarms than it clears.

Is a flat spectrum the goal? Equal per-channel power at each amplifier output is the usual goal, but a deliberate tilt of a fraction of a dB across the band is sometimes used to pre-compensate for a span with wavelength-dependent loss. Document any tilt as part of the design, not as an accident.

How much OSNR margin is enough? A common working target is 3 to 4 dB of margin above the FEC threshold at commissioning. Below 2 dB, the link will pass traffic today and fail the first time the plant ages, the temperature drifts, or a channel is moved.

Related reading: WDM optical transmission basics, DWDM system commissioning checklist, DWDM system components (mux, EDFA and demux) and Optical transceiver power: dBm RX/TX thresholds.

原文链接：Cisco ONS 15454 DWDM Network Reference: automatic power control