Complex Baseband Signal

The in-phase (I) and quadrature (Q) components are orthogonal carriers modulated by independent data streams and combined into a single signal. This orthogonality means the two components do not interfere with each other, effectively doubling spectral efficiency. In an ideal transmitter, the I and Q paths have exactly 90 degrees of phase separation and identical gain. Any deviation—known as IQ imbalance—creates an image signal that mirrors across the carrier frequency, degrading Error Vector Magnitude (EVM) and limiting predistortion performance.

The complex baseband signal is mathematically expressed as s(t) = I(t) + jQ(t), where j is the imaginary unit. The instantaneous envelope amplitude is √(I² + Q²), and the instantaneous phase is arctan(Q/I). This compact representation captures all modulation information—both amplitude and phase—without the high-frequency carrier term. It is the native domain for digital predistortion, where the predistorter applies a nonlinear correction directly to I(t) and Q(t) before upconversion to RF.

A complex baseband signal occupies a bandwidth from -B/2 to +B/2 around DC, where B is the total signal bandwidth. Unlike real-valued signals, complex signals can have asymmetric spectra—the positive and negative frequency components are independent. This property is exploited in carrier aggregation and concurrent multi-band systems. When a power amplifier introduces nonlinearity, the resulting spectral regrowth expands this bandwidth, requiring the predistorter to operate at a higher sampling rate to capture and cancel out-of-band distortion products.

The complex baseband signal's envelope amplitude fluctuates over time, and the ratio of its peak power to average power defines the PAPR. Modern wideband signals like OFDM exhibit high PAPR (often 10-12 dB), forcing power amplifiers to operate with significant back-off to avoid nonlinear saturation. This directly motivates crest factor reduction (CFR) algorithms, which clip and filter the complex baseband signal before predistortion to reduce PAPR while maintaining EVM and ACLR compliance.

Power amplifier nonlinearity is not static; it exhibits memory effects where the current output distortion depends on past input envelope values. In the complex baseband domain, these manifest as frequency-dependent AM/AM and AM/PM characteristics. Short-term memory arises from bias network impedance and matching circuits, while long-term memory stems from thermal dynamics and trapping effects in GaN transistors. Accurate behavioral models like the Generalized Memory Polynomial capture these dynamics by including delayed envelope terms in the predistorter structure.

To train a digital predistorter, the transmitted RF signal must be downconverted back to complex baseband through an observation receiver. This feedback path must have sufficient linearity and bandwidth to faithfully capture the PA's distortion products. Key impairments include ADC clipping from high PAPR signals, aliasing distortion from insufficient sampling rates, and IQ imbalance in the downconverter itself. Feedback path linearization is often a prerequisite step before DPD coefficient extraction to avoid learning a corrupted model.