Volterra kernel interpolation for system modeling and predistortion purposes

RF power amplifiers of wireless communication systems are usually driven deep into their nonlinear region to obtain a higher efficiency. The price paid for the increased efficiency is the distortion of the transmission signal caused by the inherent nonlinearity of the amplifier. To fulfill a given spectral mask for the transmission signal and to avoid an unacceptable high bit-error rate the amplifier must be linearized. One of the most powerful and efficient methods is digital baseband predistortion. The predistorter must be operated on a sampling frequency which is sufficiently high to be able to compensate the out-off-band spectral components caused by the amplifier. Also for system-level simulations of communication systems, amplifier behavior models (high sampling rate) are used to estimate the generated distortion. It is well known that nonlinear system identification can be accomplished on the Nyquist-rate regarding the input signal bandwidth, but however for several applications, systems operating on the output signal Nyquist-rate are required. In this paper we show how such systems can be obtained by a multi-dimensional kernel interpolation without the drawbacks of a high condition number for the kernel estimation and the demand for an expensive high sampling rate ADC.