Behavioral Modeling and Digital Predistortion of Wide- and Multi-Band Transmitter Systems
The demands for high data rates and ubiquitous/broadband wireless access necessitate the development of radio systems that deploy wide- and multi-band signals. These signals lead to high spectral utilization with negative trade-offs, such as rapidly varying envelopes and high peak-to-average power ratios (PAPRs). To deploy these types of signals, radio frequency (RF) transmitters face several challenges in power consumption, i.e., efficiency, and sources of distortions, i.e., nonlinearity. These challenges are most apparent in power amplifier (PAs) and degrade the overall performances of RF transmitters.
PA efficiency and linearity are characteristics that cannot be satisfied simultaneously. At high input power, PAs exhibits high efficiency; however, a PA in that region is inherently nonlinear. Conversely, at low input power, PAs are fairly linear at the expense of efficiency. With the deployment of wide- and multi-band signals, PAs exhibit strong static nonlinearity and memory effects at high input power. These effects lead to distortions, resulting in degradation of the error vector magnitude (EVM) and spectral regrowth. This regrowth creates adjacent channel interference and violates the emission requirements mandated by regulatory bodies. Digital predistortion (DPD) has been devised to mitigate the PA nonlinearity at high input power. Subsequently, DPD improves the achievable PA linearity versus power efficiency trade-off.
DPD incorporates an extra nonlinear function before the PA in order to preprocess the input signal. As a result, the cascaded system (DPD+PA) behaves linearly. DPD+PA linearity requires the DPD function to produce nonlinearities that have equal magnitude and are out phased compared to those generated by the PA. Thus, accurate PA behavioral modeling is essential for the development of DPD and is usually explored before DPD development.
In the context of wide- and multi-band signals, PA behavioral models/DPD schemes face several problems that have not been addressed appropriately in the literature. One particular problem is the exponential growth of the number of coefficients with nonlinearity order and memory depth. This growth leads to models' high complexity identification and implementation. Therefore, it restricts DPD performances in trading off linearity and efficiency. Moreover, for multi-band signals, the carrier frequencies' separation can be very large (in the order of hundreds of MHz). Consequently, the conventional single-input single-output DPD is not viable due to the unrealistic sampling rate required to cover a large range of frequencies. Finally, because a very high sampling rate is needed in the PA output observation path, the practicability of the DPD is rendered more complex and sometimes unfeasible when it is deploying wideband signals.
This thesis explores new schemes suitable for modeling and linearizing PA outputs when driven with wide- and multi-band signals. First, a new pruning technique is introduced to alleviate the complexity of DPD implementation. This pruning identifies the minimum set of dominant kernels needed in the Volterra series modeling scheme based on Wiener G-Functionals. The pruned Volterra series allows significant improvements of the numerical computation and stability of the DPD scheme. Second, new generalized memory polynomial (GMP) models for multi-band DPD are proposed. These multi-input multi-output GMP models show excellent potential in linearizing different nonlinear multi-band PAs. In addition, they involve cross terms and a reduced number of coefficients, which allows robustness against time delay misalignment between the multi-band signals. Finally, a new method to reduce the conventional high output sampling rate in the PA output observation path is proposed. Detailed analysis assesses the extent to which the output-sampling rate can be reduced through careful selection of the set of kernels used to represent the Volterra series DPD. This analysis results in a 2/5 reduction when the G-Functionals pruning technique is deployed. Extensive analysis and measurement validation are discussed and carefully analyzed to prove the validity of the different proposed solutions and to guarantee their generalizability.