In-Phase/Quadrature (IQ) Data

IQ data represents a bandpass signal as a complex lowpass equivalent, where the I component is the real part and the Q component is the imaginary part. This eliminates the carrier frequency from analysis, dramatically simplifying the mathematical modeling of power amplifier nonlinearity. The instantaneous amplitude is calculated as sqrt(I² + Q²), while the instantaneous phase is arctan(Q/I). This compact representation allows DPD algorithms to operate at baseband sample rates rather than RF frequencies.

The I and Q channels are mathematically orthogonal, meaning they can carry independent information streams without mutual interference. In a physical transmitter, this orthogonality is maintained by mixing the I component with a local oscillator signal and the Q component with the same oscillator shifted by 90 degrees. Any IQ imbalance—amplitude mismatch or phase deviation from perfect quadrature—destroys this orthogonality and creates unwanted image signals that degrade DPD performance.

Digital predistortion operates directly on complex IQ samples because nonlinear distortion affects both amplitude (AM-AM) and phase (AM-PM) simultaneously. The predistorter applies a complex gain correction factor to each IQ sample based on the instantaneous input magnitude. This correction is itself a complex value, modifying both I and Q components to pre-compensate for the power amplifier's nonlinear transfer characteristic before the signal reaches the RF stage.

To capture and correct nonlinear distortion, the IQ sample rate must be 3 to 5 times the signal bandwidth to accommodate the expanded bandwidth of intermodulation products. For a 100 MHz 5G NR carrier, this demands IQ sample rates of 300-500 MHz or higher. This wideband IQ capture is critical for accurate behavioral modeling, as truncating the observation bandwidth removes distortion products that fold back into the band of interest through aliasing.

IQ data streams inherently preserve the temporal history of the signal, which is essential for modeling power amplifier memory effects. By storing past IQ samples in tapped delay lines, memory polynomial models can correlate current distortion with previous signal states. The complex nature of IQ data means both the magnitude and phase history influence the PA's current behavior, capturing phenomena like thermal memory and bias circuit relaxation that simpler magnitude-only models miss.

DPD coefficient extraction demands high-precision IQ data to resolve the subtle distortion components that are 30-50 dB below the main signal. Typically, 16-bit or higher ADC resolution is required for the feedback observation path. Quantization noise in the IQ samples directly limits the achievable linearization depth. Modern FPGA-based DPD implementations often use 18-bit or 32-bit fixed-point arithmetic to preserve the dynamic range needed for accurate predistorter synthesis.