Frequency-Dependent I/Q Imbalance

Unlike frequency-independent imbalance corrected by a single complex coefficient, frequency-dependent mismatch is modeled as a widely-linear filter. The impaired output is the sum of a linear convolution with the desired signal and a convolution with its complex conjugate. This requires a complex FIR filter for compensation, where each tap addresses a specific frequency region of the mismatch profile.

The frequency selectivity originates from physical analog imperfections:

• Mismatched anti-aliasing filters: Different cutoff frequencies or ripple in I and Q low-pass filters
• Unequal trace lengths: PCB routing differences causing frequency-dependent phase skew
• DAC bandwidth mismatch: Differing sin(x)/x roll-off characteristics between I and Q digital-to-analog converters
• Amplifier gain ripple: Non-flat frequency response in I or Q baseband amplifiers

Correction requires an adaptive complex equalizer that implements the inverse of the widely-linear system. The filter structure typically uses a 2x2 MIMO architecture with four real filters or a single complex FIR filter with conjugate taps. Coefficients are estimated using algorithms like Least Mean Squares (LMS) or Recursive Least Squares (RLS) operating on the circularity property of proper complex signals.

For narrowband signals, frequency-dependent imbalance may appear static. However, in 5G NR and WiFi 6 systems with 100+ MHz bandwidths, the variation becomes severe:

• EVM degradation varies across subcarriers in OFDM
• Image suppression is frequency-selective, with poor rejection at band edges
• Spectral regrowth becomes asymmetric, complicating ACLR compliance
• Higher-order modulations (256-QAM, 1024-QAM) are particularly vulnerable

A critical subset of frequency-dependent imbalance is I/Q skew—a relative time delay between I and Q sampling instants. This manifests as a linear phase distortion across frequency, equivalent to a frequency-dependent phase imbalance. Skew of even a few picoseconds can severely degrade Error Vector Magnitude (EVM) in multi-GHz bandwidth systems. Correction requires fractional-delay interpolation filters.

Frequency-dependent parameters are often estimated blindly without training sequences, exploiting the statistical property of circularity (properness). A properly modulated complex signal has zero complementary autocorrelation. Any non-zero complementary autocorrelation indicates imbalance. Algorithms like Widely-Linear Bussgang or spectral circularity-based methods extract the mismatch filter coefficients directly from the transmitted signal's second-order statistics.