Purpose
Chip batch variations and temperature drift can cause RF parameter dispersion. Calibration writes NV parameters into flash memory, ensuring that transmit power, IQ imbalance, and AGC curves fall within specifications. Cellular modules are often calibrated at multiple points according to band and channel.
Production line
Shielded box, programmable power supply, and automation scripts; yield and calibration time affect production capacity. Golden sample comparison prevents instrument drift.
Traceability
Serial number is linked to calibration log, enabling after-sales traceability.
Multi-band and temperature drift compensation
Cellular modules need to operate on multiple bands globally or regionally. Transmit power templates, EVM, spectrum templates, and receive sensitivity are all affected by crystal temperature drift, PA nonlinearity, and filter tolerances. Production line calibration usually collects data at several representative frequency points and target power levels, then fits or uses lookup tables to write into NV. For models supporting carrier aggregation or dual connectivity, it is also necessary to verify whether spurious emissions and intermodulation still meet certification limits under combined scenarios. Random checks in high and low temperature chambers can reveal batch device deviations; if necessary, broaden calibration points or tighten incoming material inspection standards to avoid edge cell access failures or network dropout complaints at customer sites.
Test instruments and data management
Vector signal sources, spectrum analyzers, and network analyzers need regular metrology and system-level calibration. Comparing against golden samples can detect early cable loss or port matching changes. The MES should retain the calibration curve version, instrument serial number, operator, and software script hash for each device to meet automotive and industrial customer traceability requirements. If OTA allows RF parameter fine-tuning, it must also prevent unauthorized modifications that lead to non-compliance; private key protection and signed upgrade packages are part of overall security. During capacity ramp-up, the calibration cycle and re-test rate directly impact the cost per unit, requiring a data-driven balance between yield, speed, and equipment investment.
Engineering Practice Supplement
For cross-team collaboration on "RF mass-production calibration": hardware provides impedance and power budget, RF provides spurious and sensitivity margins, embedded clarifies thread model and storage lifetime, and cloud clarifies SLA and throttling. For scenarios like NB-IoT, the business model must match wake-up cycles, packet length, and downlink capacity; for automotive and mobile scenarios, session continuity during handover and roaming must be considered. At project closeout, conduct a "red team" style review: define behavior for weak networks, power loss, certificate expiration, clock jumps, and full storage.
What It Means to Store a Power Template in Non-Volatile Parameters
Calibration on the line measures this module's RF deviation and writes it into non-volatile parameters, so that later transmit power, receive gain, and channel imbalance follow the target template. The deviation comes from the amplifier, the filter, the crystal, and board-level tolerance. Two chips of the same model do not land on top of each other by nature. Parameters are usually stored by band, channel, and power step, and channels in between are interpolated. Finishing calibration is not the same as passing certification. Inside the template, error vector, adjacent channel, and spurious still have to pass together. A frequency that was never written into the parameters transmits in the field with the wrong gain. The symptom is difficult access on a few channels and normal behavior on the rest.
The conditions of use are a shielded environment, a known cable loss, and a stable supply. If temperature is calibrated only at room temperature and high temperature is carried by margin, power walks out of the template as the temperature drifts. A model that supports multiple antennas or carrier combinations still needs a sample of intermodulation in the combined state after single-channel calibration passes. The parameter version has to be bound to the firmware version. If a firmware upgrade clears or shifts these parameters, the radio returns to an uncalibrated state while the version number looks new.
Path-Loss Drift and a Wrong Batch Call
The most common batch incident is a change in loss of a cable or a switch matrix while the instrument still uses last month's compensation. The whole lot is written with power that is too low or too high. If the gold sample does not pass through the station, the drift stays hidden until a customer complaint. The second is a retest that calibrates an already calibrated unit again, so the interpolation table is overwritten by the noise of the second pass. The third is to store only "pass" and not the curve. Service then cannot tell a corrupted parameter write from a mechanical revision. Without a script hash, an instrument id, and an operator record, two lines cannot be compared.
When over-the-air total power fails, first confirm whether port power is still inside the template; see radiated tests and OTA. A routing revision changes the reference plane the calibration assumed; see PCB RF routing. Old parameters cannot be dropped onto the new board.
Each unit's record should be able to reconstruct which frequencies and power steps were calibrated, the parameter version written, the path-loss compensation in use, the instrument id, and whether the result is a first pass or a retest. Before a retest overwrites parameters, read the old values out and save them, so the noise of a second pass does not wipe a table that already passed. Check path-loss compensation against the gold sample every day. If the gold sample drifts, stop the line. Do not keep building and rework the month at the end. If a firmware upgrade moves the parameter area, the upgrade script must check that key steps are still readable before and after the move. If the check fails, stay on the old firmware. When temperature is only sampled, a sample failure has to trace back to the crystal or amplifier lot, not fail one unit and release the rest. If port power is inside the template and the over-the-air result is well short, look at the structure and the antenna first. Do not keep raising the line's power target to make the air interface look better, or conducted spurious follows it out of limit. The power target may be adjusted only inside the template. Do not raise port power to make the over-the-air result look good.