Transmitter Linearization Adaptable to Power-Varying Operation

Abstract

This paper presents the design of a power-scalable digital predistorter (DPD) for transmitter architectures. The target is to accomplish the joint compensation of impairments due to the I/Q modulator and nonlinearities associated with the power amplifier (PA), and procure a maintained linearization performance in a range of average working operation levels. The identification method for the linearizer parameters enriches the standard least-squares procedure with a synergistic integration with sparsity-based model pruning strategies. The method has been tested with a general complex-valued Volterra model applied to the linearization of two communications transmitters operating at 3.6 GHz. The linearizers designed for the two transmitters effectively provide the joint compensation of the nonlinear behavior. In addition to their good performance in terms of adjacent channel power ratio, the DPDs exhibit a wide range of power-varying adaptation.

Sparse Identification of Volterra Models for Power Amplifiers Without Pseudoinverse Computation

“…A DPD was calculated for each number of components, allowing to obtain the linearization normalized mean-squared error (NMSE), adjacent channel power ratio (ACPR), and error vector magnitude (EVM) as performance indicators. In addition, the Bayesian information criterion (BIC) was calculated over the model evolutions to indicate the optimum number of coefficients for each case [32]. Fig.…”

“…The residual is updated by subtracting the contribution of the last selected component (32) and the estimation of the model output is attained by adding the same contribution…”

Upgrading Behavioral Models for the Design of Digital Predistorters

“…Regressors are incorporated until the minimum of the BIC criterion is reached. The stopping indicator was defined by the minimum of the BIC written in terms of the normalized mean square error (NMSE) and a penalty term, as in [ 26 ], where is the number of regressors and is the NMSE corresponding to this number of regressors.…”

“…The lower plot of Figure 2 shows the normalized magnitude of the new estimated coefficients, and Figure 3 reveals the comparison of the error spectra for the s-FV(13,3) model and the upgraded FV model (blue traces). The spectrum of the error between and is also plotted to have a reference of the distortion generated by the PA. Once the normalized parameters were computed at an input level of 6 dBm, they were straightforwardly scaled to adapt the coefficients to other power levels and the corresponding NMSE were evaluated [ 26 ]. In the case of the reduced regressors set derived from the FV(13,3) set (a model with a shortfall in the regressors stock), there are eight active regressors.…”

On-Demand Real-Time Optimizable Dynamic Model Sizing for Digital Predistortion of Broadband RF Power Amplifiers

“…Power-adaptive DPD in [19] linearized PAs under different power levels by dynamically updating DPD coefficients with a coefficient interpolation block. In [20], the authors adopted a similar strategy but updated DPD coefficients with a theoretically derived scaling rule instead of optimized parameters. A neural network model was developed in [21] which explicitly measures and models PA characteristics under predetermined frequency, voltage and temperature conditions.…”