The effectiveness of DPD systems in canceling the distortions present at the output of PAs and transmitters extensively depends on the match between the predistorter’s nonlinear characteristics and that of the DUT to be linearized. Since the nonlinearity exhibited by the DUT varies with time due to changes in the drive signal, aging, or drifts, it is essential to continuously update the predistortion function to maintain the linear operation of the system made of the predistorter and the DUT. Adaptive digital predistorters can be implemented either in closed loop or open loop configuration. This classification depends on the location of the predistortion function with respect to the adaptation loop.

Closed Loop Adaptive Digital Predistorters

In closed loop adaptive digital predistorters, the predistortion function is located inside the adaptation loop used to update the predistortion function coefficients.

The functional block diagram of closed loop DPD systems is depicted in Figure 9.4. The signal at the input of the digital predistorter (x_{in}__{DPD}(n)) and that at the output of

Figure 9.4 Closed loop adaptive digital predistortion system

the PA (x_{out}__{PA} (n)) are used to compute the error signal of the closed loop DPD system

(ecL_DPD(n)) defined by:

where G_{LPA} is the gain of the linearized PA.

The adaptive algorithm is then used to minimize this error signal and ensure that the amplifier’s output signal is a scaled replica of the predistorter’s input signal.

This concept is also known as “model the reference adaptive system” (MRAS) in control theory. Closed loop DPD systems employ the direct learning technique to identify the predistorter’s coefficients. The direct learning refers to the method used to update the predistorter’s coefficients by considering the input and output signals of the linearized PA made of the cascade of the predistorter and the amplifier [1, 2]. Closed loop predistorters usually exhibit slow convergence and high computational complexity since there is no direct relation between the error signal and the predistorter’s coefficients. They are also prone to divergence if the PA is driven into saturation, which will cause the adaptive algorithm to repeatedly and unsuccessfully try to increase the output power of the predistorter to correct for the uncorrectable saturation induced distortions.