Embedded Calibration of Integrated RF Frontend Blocks

Parameter variability due to process, voltage, and temperature variations (PVT) in the contemporary submicron CMOS chips makes RF circuit design risky unless proper calibration procedures are applied. They usually require careful on-chip measurements by RF sensors that can be used to compensate for the possible impairments in specs such as gain or gain mismatch which is critical e.g. in IQ setups or multi-path MIMO systems. As the sensors are exposed to the same variations as the actual RF circuitry they need to be calibrated first. In this paper the focus is on a calibration technique suitable for embedded RF sensors. This technique is based on DC measurements and statistical multivariate regression which enable mapping of the sensor DC characteristics into RF characteristics. The external RF measurements are restricted to the statistical training process preceding the actual calibration. The results obtained by simulation demonstrate reduction of output signal variations among sensors from 50% to 5% for moderate signal level and from 25% to 2.5% at full range.