I/Q Calibration

Calibration strategies are categorized by their execution environment and persistence model.

• Factory Calibration: Performed during manufacturing using precision test equipment (vector signal analyzers, signal generators). Coefficients are stored in non-volatile memory and remain static for the device lifetime. This corrects for process variation and component tolerance.
• Field Calibration: Executed at device startup or periodically during operation. Leverages an on-chip observation receiver to capture the transmitted signal and adapt coefficients to compensate for temperature drift and aging.
• Hybrid Approach: Factory calibration establishes a baseline, while field calibration tracks dynamic variations using adaptive algorithms like LMS (Least Mean Squares).

I/Q imbalance is mathematically modeled as a widely-linear system, where the impaired output is a linear combination of the ideal signal and its complex conjugate.

• The relationship is defined by the I/Q Mismatch Matrix: a 2x2 transformation that maps the ideal I/Q vector to the impaired vector.
• The key parameter is the complex mismatch coefficient (K), representing the ratio of the image-producing path gain to the desired signal path gain.
• Correction applies the inverse matrix, effectively performing a complex scaling and conjugate cancellation to restore signal circularity and suppress the image sideband.

Modern wideband signals (e.g., 100 MHz 5G NR carriers) expose frequency-dependent I/Q imbalance caused by mismatched anti-aliasing filters, DAC roll-off, and PCB trace length differences.

• Frequency-Independent Correction: A single complex scalar multiplication corrects static gain and phase errors across a narrow bandwidth.
• Frequency-Dependent Correction: Requires a complex FIR filter (I/Q Mismatch Filter) to equalize the amplitude and phase ripple across the entire signal bandwidth.
• Calibration algorithms must estimate the full impulse response of the mismatch path, often using multi-tone or wideband chirp test signals during the characterization phase.

Advanced calibration algorithms can extract imbalance parameters without dedicated training sequences, operating directly on the modulated signal's statistical properties.

• Circularity-Based Estimation: A properly balanced QAM signal has circular symmetry in the complex plane. I/Q imbalance destroys this properness. Algorithms minimize the signal's non-circularity (variance of the real and imaginary parts) to iteratively solve for the mismatch coefficients.
• Spectral Symmetry: The unwanted image sideband violates spectral symmetry. Blind algorithms minimize the power in the image frequency band relative to the desired signal band.
• These techniques enable continuous background calibration without interrupting data transmission.

A critical sub-function of I/Q calibration is the suppression of carrier feedthrough caused by DC offsets at the modulator input.

• Source: Local oscillator self-mixing and component mismatch introduce a static DC voltage on the I and Q baseband paths, which upconverts directly to the carrier frequency.
• Calibration Loop: The observation receiver measures the LO leakage power at the transmitter output. A successive approximation or gradient descent algorithm adjusts compensating DC offset values applied to the baseband DACs.
• Performance Metric: Effective calibration achieves an LO suppression of -50 dBc or better, preventing the carrier tone from violating spectral emission masks and wasting transmit power.

The accuracy of coefficient extraction depends heavily on the design of the calibration stimulus signal.

• Single-Tone Test: A continuous wave (CW) tone generates a known image tone. Simple and fast, but only characterizes frequency-independent imbalance at one frequency.
• Multi-Tone Test: A set of discrete tones across the band allows estimation of frequency-dependent gain and phase ripple at specific frequency bins.
• Wideband Modulated Signal: Using the actual modulation format (e.g., OFDM) with a known data sequence provides the most realistic characterization, capturing the effects of peak-to-average power ratio and signal statistics on the impairment model.