2D Look-Up Table (2D-LUT)

Unlike a 1D-LUT that maps a single signal magnitude to a correction value, a 2D-LUT uses a pair of instantaneous envelope magnitudes—(|x₁|, |x₂|)—from two concurrent baseband signals to form a unique address. This 2D address directly indexes a pre-computed complex gain correction value, eliminating the need for real-time polynomial evaluation and drastically reducing computational latency in the feedback path.

The 2D-LUT inherently captures cross-modulation and intermodulation distortion (IMD) products generated by the interaction of two carrier signals within a power amplifier. By indexing on both magnitudes simultaneously, a single table entry stores the predistortion coefficient that compensates for the instantaneous nonlinear mixing, including cross-band memory effects that simpler parallel 1D-LUT architectures fail to address.

To mitigate the exponential memory growth of a full 2D table, implementations employ bilinear interpolation between adjacent stored points. The 2D address space is uniformly quantized into a grid, and the final correction value is computed by interpolating between the four nearest neighbors. This technique allows a relatively small table (e.g., 64x64 or 128x128 entries) to approximate a continuous 2D predistortion function with high fidelity.

In an adaptive 2D-LUT DPD system, the table entries are continuously updated using a direct learning architecture (DLA) . The error between the desired linear output and the actual PA output is used to compute a correction delta for the specific 2D address bin. A recursive update algorithm, often employing least mean squares (LMS) or a normalized variant, adjusts the stored complex gain to minimize the residual distortion in real-time.

The 2D-LUT is highly amenable to FPGA implementation using dual-port Block RAM (BRAM). The two magnitude signals serve as the row and column addresses for a memory block. A single clock cycle read operation retrieves the complex correction factor, which is then multiplied with the composite input signal. This single-cycle latency is critical for wideband applications where the DPD must operate at high sample rates with minimal processing delay.

The instantaneous magnitudes |x₁(n)| and |x₂(n)| are computed using a CORDIC algorithm or a coordinate rotation block to extract the envelope. These values are then quantized to M and N bits respectively to form the integer row and column addresses for the LUT. The choice of quantization depth (e.g., 6-bit for 64 levels) represents a direct trade-off between linearization accuracy and memory footprint, with finer quantization capturing more subtle nonlinear behavior.

The predistortion model that directly utilizes the 2D-LUT indexing structure. It synthesizes a correction signal based on a two-dimensional function of the instantaneous magnitudes of two concurrent baseband signals. Unlike 1D approaches, 2D-DPD explicitly accounts for the cross-modulation products generated when both carriers interact within the power amplifier's nonlinear region. The 2D-LUT serves as the most hardware-efficient implementation of this model, mapping the dual-envelope address directly to complex gain correction values without requiring real-time polynomial computation.

Nonlinear interference products generated by the interaction of multiple carrier signals within a shared power amplifier. These products fall directly on top of or adjacent to the desired transmit bands and cannot be filtered without damaging the wanted signals. Key types include:

• Intermodulation Distortion (IMD): Unwanted frequency components from nonlinear mixing
• Cross-Modulation: Envelope transfer from a strong interferer onto a desired signal The 2D-LUT addresses these by indexing correction values based on the instantaneous envelope of both signals simultaneously, enabling cancellation of distortion that depends on the joint signal state.

A behavioral model that extends the standard memory polynomial into two dimensions by incorporating cross-terms dependent on the envelope magnitudes of both concurrent bands. While the 2D-LUT provides a non-parametric look-up approach, the 2D-MMP offers a parametric polynomial formulation that explicitly models:

• In-band memory effects for each carrier
• Cross-band memory effects where past envelope values in one band influence current distortion in another The 2D-MMP coefficients can be extracted offline and used to populate or validate the 2D-LUT contents during system calibration.

A closed-loop adaptation method for updating 2D-LUT contents during live operation. The architecture works by:

• Placing a post-distorter model in the feedback path, identified from the attenuated PA output
• Copying the trained post-distorter coefficients to the predistorter LUT in the forward path This approach avoids the need for a direct PA inverse model and handles the non-commutative nature of nonlinear systems. For 2D-LUT implementations, MB-ILA enables adaptive table updates that track changes due to thermal drift, aging, and channel reconfiguration without interrupting transmission.

A signal conditioning technique applied upstream of the 2D-LUT that jointly reduces the peak-to-average power ratio (PAPR) of the composite multi-band signal. MB-CFR is critical because:

• Excessive PAPR drives the PA deeper into saturation, expanding the nonlinear region the 2D-LUT must linearize
• Joint CFR prevents re-growth of peaks after individual carrier clipping
• Reduced PAPR allows the 2D-LUT to operate within a smaller dynamic range, improving correction accuracy The 2D-LUT and MB-CFR form a co-designed transmit chain where CFR limits the signal envelope extremes and the LUT handles the remaining compression characteristics.

The mathematical foundation from which 2D-LUT indexing strategies are derived. This model analytically describes the baseband nonlinear behavior in a dual-band transmitter by pruning the full passband Volterra series to retain only terms that fall within the two transmit bands. The resulting formulation reveals that the distortion in each band depends on a two-dimensional basis function of the envelope magnitudes of both signals—the exact relationship that the 2D-LUT captures through its dual-magnitude addressing scheme. Understanding this derivation is essential for determining optimal LUT address quantization and interpolation strategies.