An Efficient Scheme for Nonlinear Modeling and Predistortion in Mixed-Signal Systems

Abstract

Abstract-A novel identification and predistortion scheme of weakly nonlinear systems for mixed-signal devices, which takes into account practical implementation aspects, is presented. It is well known that for the identification of weakly nonlinear systems, despite the spectral regrowth, it suffices to sample the input-output (I/O) data of the system at the Nyquist rate of the input signal. Many applications such as linearization and mixed-signal simulations require system models at a higher sampling rate than Nyquist. Up to now, the construction of such high-rate models has been done by oversampling the corresponding I/O data. This leads to high computational complexity, ill posedness of the estimation, and high demand on the analog-to-digital-converter sampling rate for the implementation. This brief discusses an efficient way to obtain high-rate models and predistorters from low-rate models and shows the validity of the proposed scheme for a very-high-speed-digital-subscriber-line power amplifier, where an adjacent channel power supression of 20 dB is achieved.

A Local Nonlinear Model for the Approximation and Identification of a Class of Systems

“…2 can be thought of consisting of three parts, namely the central block that realizes a general quadratic vector function of N = 2 arguments and the two adjacent parts comprising the diagonal linear transfer functionT(s). Considering the general expression for the higher-order kernels (8), reveals that its first summand is analogous to the above discussed expressions in (12) for higher orders. For example the thirdorder kernel would involve a cubic vector function of N = 2 arguments and the same adjacent transfer functionsT(s).…”

“…Although, this condition seems to be ad-hoc at first, it actually corresponds to a situation encountered in practice. For instance in the generation of macromodels for large-scale weakly-nonlinear circuits, such as transceiver front-ends [8] the linear part of the system can accurately be extracted from a small-signal analysis. In the absence of a priori knowledge about the linear characteristics a sequential procedure can be applied, where first the linear part is estimated [9], [10], [11] (possibly with a smaller amplitude than nominal) and second the function g(ϑ) is estimated.…”

Energy-Efficient Distributed Estimation by Utilizing a Nonlinear Amplifier

“…This requires oversampling the transmitted data by twice the order of the highest nonlinear term [51]. This oversampling increases power requirements in the baseband analog converters and the digital signal processor of a conventional in-phase/quadrature modulated system.…”

Sampling Rate Reduction for Digital Predistortion of Broadband RF Power Amplifiers

“…In this case, the model extraction algorithms need to apply spectral extrapolation to the loss function to achieve full-band distortion cancellation. Another method is to remove the anti-aliasing filter before sampling [15]- [18]. The aliased feedback signal is then used together with specially processed input signals to extract model coefficients.…”