Adaptive LUT

The defining characteristic of an adaptive LUT is its integration within a closed-loop control system. A portion of the power amplifier (PA) output is coupled, downconverted, and digitized through an observation receiver. An error signal is computed by comparing this feedback with the original baseband input. This error drives an adaptation algorithm that iteratively updates LUT coefficients to minimize the difference, allowing the system to self-correct for PA aging, thermal drift, and supply voltage variations without manual recalibration.

Adaptive LUTs rely on iterative algorithms to update entries in real-time. Common approaches include:

• Least Mean Squares (LMS): A computationally simple, stochastic gradient descent method that updates coefficients proportionally to the instantaneous error.
• Recursive Least Squares (RLS): Offers faster convergence than LMS at the cost of higher computational complexity, suitable for rapidly changing signal conditions.
• Secant Method: A root-finding approach that directly computes the complex gain correction needed to drive the error to zero for each LUT bin. The choice of algorithm balances convergence speed against steady-state noise and hardware resource utilization.

To ensure seamless, glitch-free predistortion during coefficient updates, adaptive LUTs are typically implemented using a dual-buffer memory structure known as ping-pong buffering. While one LUT (the 'active' table) is used for real-time predistortion on the transmit path, the adaptation engine writes updated coefficients to the second LUT (the 'shadow' table) in the background. Once the update cycle is complete, a multiplexer switch instantly swaps the roles of the two tables. This prevents the predistorter from ever applying a partially updated, potentially unstable set of coefficients.

The adaptation rate governs how aggressively LUT coefficients are modified in response to the error signal. This is typically controlled by a step size parameter (μ) in gradient-based algorithms. A large step size enables rapid tracking of fast-changing PA nonlinearities but introduces steady-state jitter that can degrade ACLR. A small step size yields smooth, low-noise coefficients but may lag behind thermal transients. Advanced implementations use variable step size techniques, adapting μ based on error magnitude or signal statistics to optimize the trade-off between tracking agility and correction accuracy.

A primary motivation for adaptive LUTs is their ability to compensate for long-term parametric drift in power amplifiers. GaN and LDMOS PAs exhibit gain and phase variations as junction temperature changes during operation. Over months of deployment, device aging further shifts the AM-AM and AM-PM characteristics. An adaptive LUT continuously tracks these slow changes by observing the persistent error residual. Some implementations incorporate a dedicated temperature sensor to provide a feed-forward bias to the LUT indexing, combining open-loop temperature compensation with closed-loop adaptation for enhanced stability.

Ensuring stable adaptation is critical to prevent the predistorter from diverging and damaging the PA or violating spectral emission masks. Adaptive LUT controllers implement convergence monitoring by tracking the normalized mean squared error (NMSE) between the desired and observed signals. If the error fails to decrease or begins to oscillate, the adaptation engine can freeze updates, revert to a previously known-good coefficient set, or reduce the step size. Leakage factors are often added to LMS updates to prevent coefficient drift in low-excitation LUT bins, ensuring long-term numerical stability.