Accuracy and limits of DDM

DDM numbers look precise — two decimals of dBm, hundredths of a volt — but the specification behind them promises far less. Knowing what a monitor physically measures, how accurate it is required to be and where it stops working keeps you from chasing a "1 dB drop" that is inside the noise, and from trusting a "normal" reading that a cheap module simply made up.

What the specifications guarantee

Monitor	Required accuracy (SFF-8472 / SFF-8636, internally calibrated)	Resolution (LSB)	Over what range
Temperature	±3 °C	1/256 °C	module's operating range
Supply voltage	±3 % of nominal	100 µV	0 … 6.55 V
Tx bias	±10 %	2 µA	0 … 131 mA
Tx power	±3 dB	0.1 µW	vendor-specified range, typically −8 … +4 dBm
Rx power	±3 dB	0.1 µW	vendor-specified range, typically −20 … 0 dBm or −30 … −7 dBm for APD

Many vendors do better — ±1.5–2 dB is common — but only the datasheet says so. The ±3 dB means two modules on the same fibre may legitimately report Rx values 6 dB apart from each other's Tx minus the real loss. CMIS modules inherit the same monitor formats; their advertising pages may state tighter figures.

What is actually measured

Reported as	Physically	Consequence
Tx power	current of a monitor photodiode catching a fraction of the laser output (back facet or tap)	an estimate scaled at the factory; drifts with temperature and coupling; on some modules replaced by a constant — DDM levels
Rx power	average photocurrent of the receiver photodiode (before or after the TIA)	true average power; independent of data content for NRZ, but responsivity depends on wavelength — a receiver calibrated at 1310 nm mis-reads a 1550 nm signal by ~1 dB
Tx bias	the current the driver sets	usually accurate; on EML modules only the laser section, not the modulator
Temperature	a sensor on the module PCB or in the controller	a few °C above case temperature; not the laser temperature (that is the optional laser-temperature monitor)
Vcc	supply measured inside the module	slightly below the host rail because of the connector and filter

Average vs OMA

Specifications express sensitivity in OMA (optical modulation amplitude); DDM reports average power. For an NRZ signal with typical extinction ratio the two differ by about 2–3 dB (average is lower than OMA + 3 dB… the exact offset depends on extinction ratio). PAM4 makes it worse: average power says little about the eye. A module can therefore be "in range" by average power and still fail sensitivity. Byte 92 bit 3 of SFF-8472 tells which kind of number the module reports (Calibration).

Floors and ceilings

Rx floor — below about −30 dBm (PIN) or −35 dBm (APD) the photocurrent is in the noise; the raw value hits 0 and displays as −40 dBm (10·log10 of 0.1 µW/1 mW). A reading of −40 means "no measurable light", not "−40 dBm of light".
Rx ceiling — above the monitor's range the value saturates; overload of the receiver (errors) may begin before or after the monitor saturates.
Tx — meaningful only while the laser is on; with TX_DISABLE asserted it should fall to the floor. A Tx that stays at its nominal value with Tx disabled is a constant.

Timing

Modules refresh monitors internally several times per second; hosts poll every few seconds; NMS every minutes. A brief event (a flap of 200 ms) may never appear in DDM.
Right after insertion the data-not-ready bit is set (SFF-8472 A2h 110.0, SFF-8636 byte 2.0); values read then are garbage.
Flags are latched: an alarm you see may be from an event minutes ago that has since recovered (Thresholds & alarms).

Where DDM is absent or misleading

Module type	What you get	Note
Passive DAC	nothing — no A2h device	identity only
1000BASE-T / 10GBASE-T SFP	often temperature only, or none; Tx/Rx fields zero or constant	a copper PHY has no optics to measure
AOC	full DDM at both ends	levels are internal to the cable; a low Rx inside an AOC means a failing cable, not a dirty connector
PON OLT module	Rx is a burst-mode average over all ONUs	per-ONU power comes from the OLT's ranging, not from DDM — PON problems
PON ONU stick	correct Rx; Tx meaningful only while transmitting bursts	Tx may read low or zero when idle
Tunable DWDM	Tx/Rx fine; adds laser temperature, TEC current, wavelength/frequency monitors	Tunable maps
Externally calibrated modules	raw values until constants are applied	absurd numbers if the host or tool ignores byte 92 — Calibration
Low-cost or cloned modules	Tx and sometimes bias are constants; Rx usually real	compare against Tx disable and across a batch
Media converters / DDM through a PHY	some devices synthesise values	trust only what the module itself reports on the bench

Practical rules

Treat differences under 1 dB as noise; act on 2–3 dB changes from a baseline (Monitoring).
Compare Rx on both ends and Tx vs Rx across the link, not single numbers.
Verify a suspicious Tx with an external power meter once; if the module lies, note it in the code database and rely on the far end's Rx.
Do not use DDM to certify a fibre plant — that is an OTDR/light-source job. DDM is a built-in health indicator, not a test instrument.
When the numbers are absurd (0 °C, +10 dBm, 0.00 V), suspect calibration, not physics.

Expected windows to compare against: Typical values.