Internal & external calibration

A photodiode does not output dBm and a thermistor does not output °C. Somewhere the raw sensor reading has to be turned into an engineering value — and SFF-8472 allows that to happen either inside the module or on the host, using constants the module provides. Getting this wrong is the usual reason a switch displays absurd DDM values for a module that works perfectly.

Two calibration modes

Mode	Declared by	Who converts	What the monitor bytes contain
Internal	A0h byte 92 bit 5	the module's microcontroller	final values in standard units (1/256 °C, 100 µV, 2 µA, 0.1 µW)
External	A0h byte 92 bit 4	the host	raw A/D counts; host applies constants from A2h 56–95

Almost every modern module is internally calibrated. Externally calibrated modules are older or cost-optimised designs; their values read as garbage on hosts that skip the conversion.

External calibration constants — A2h bytes 56–95

Bytes	Constant	Format	Applies to
56–75	Rx_PWR(4) … Rx_PWR(0)	five IEEE-754 single-precision floats	Rx power (4th-order polynomial)
76–77	Tx_I slope	unsigned fixed-point 8.8	Tx bias
78–79	Tx_I offset	signed 16-bit	Tx bias
80–81	Tx_PWR slope	unsigned 8.8	Tx power
82–83	Tx_PWR offset	signed 16-bit	Tx power
84–85	T slope	unsigned 8.8	temperature
86–87	T offset	signed 16-bit	temperature
88–89	V slope	unsigned 8.8	Vcc
90–91	V offset	signed 16-bit	Vcc
92–94	reserved	—	—
95	CC_DMI	checksum of bytes 0–94	—

The conversion formulas

Rx_PWR (µW)      = Rx4·raw⁴ + Rx3·raw³ + Rx2·raw² + Rx1·raw + Rx0
Tx_I   (µA)      = Tx_I_slope · raw + Tx_I_offset
Tx_PWR (0.1 µW)  = Tx_PWR_slope · raw + Tx_PWR_offset
T      (1/256 °C)= T_slope · raw + T_offset
V      (100 µV)  = V_slope · raw + V_offset

For an internally calibrated module the constants are ignored; many vendors still fill them with identity values (slope 1.0, offset 0, Rx polynomial 0,0,0,1,0) so that a host applying them anyway gets the right answer.

Average vs OMA

A0h byte 92 bit 3 says how Rx power is expressed: average power (1) or OMA — optical modulation amplitude (0). Nearly all modules report average power; OMA appears on some multi-mode Ethernet optics where the standard specifies sensitivity that way. Comparing an OMA figure with an average-power sensitivity spec misleads by up to ~3 dB.

Symptoms of a calibration problem

What you see	Likely cause
Temperature of 0 °C or −128 °C, Vcc 0.00 V	host reads raw counts from an externally calibrated module without converting
Rx power impossibly high (+10 dBm) or always −40 dBm	wrong calibration mode assumed, or zeroed constants
Values differ wildly between two switch models with the same module	one host honours byte 92, the other assumes internal
Tx power identical on every module of a batch	not calibration — a constant Tx (DDM levels)

In CodingBox

CodingBox reads byte 92, applies external calibration constants when the module declares them, and shows both raw and converted values, so a module that a switch displays wrongly can be judged on the bench. The constants block 56–95 is editable in the EEPROM editor with CC_DMI recalculated automatically — useful for repairing modules whose calibration area was overwritten.