In this paper, a new linearization algorithm of Power Amplifier, based on Kalman filtering theory is proposed for obtaining fast convergence of the adaptive digital predistortion. The proposed method uses the real-time digital processing of baseband signals to compensate the nonlinearities and memory effects in radio-frequency Power Amplifier. To reduce the complexity of computing in classical Kalman Filtering, a sliding timewindow has been inserted which combines off-line measurement and on-line parameter estimation with high sampling time to track the changes in the PA characteristics. We evaluated the performance of the proposed linearization scheme through simulation and experiments. Using digital signal processing, experimental results with commercial power amplifier are presented for multicarrier signals to demonstrate the effectiveness of this new approach.

International audienceThis paper presents a new time-domain model of radio frequency power amplifiers (RF PA) for wideband applications. The proposed model identification requires a set of characterization procedures of a memory polynomial model with delay taps for a two-tone test. This first characterized model is described using amplitude AM/AM and phase AM/PM characteristics including delay and gain parameters, which is able to reproduce distortion sideband asymmetries. A two-tone signal with varying tone spacings is used to find the frequency response of parameters on each nonlinearity order. These response are used to estimate equivalent digital filters giving the final structure of a wideband model. The measurement tests on a Class AB-$1.85\,$GHz RF PA from Advanced semiconductor business with various multitone signals show that the proposed architecture describes PA nonlinearities and memory effects with a high level of accurac

International audienceCorrect modeling of radio frequency power amplifiers (RF PA) is difficult, as nonlinear characteristics are not only statics but also dynamics and strongly depend on input signal frequencies. Recent studies in PA behavioral model have shown the interest of structures based on memory polynomial functions with sparse delay taps between instantaneous and previous input signal. The extension of such structures is proposed in this work to achieve a measurement based wideband model of PA for digital wireless communication systems. The model uses Infinite Impulse Response filters which describe the frequency response of each nonlinearity order. The characterization of this new model structure includes direct calculation of polynomial parameters and delay taps with two-tone test for different frequencies and parameter estimation of equivalents filters in frequency domain using least mean square algorithm and Instrumental Variable method. The obtained architecture improves memory effect description and provides computational efficiency suitable for dynamic time-domain PA studies. This new approach is validated using a 1.85GHz RF PA from Advanced semiconductor business with various modulations such as two-tone test, WCDMA and WiMax signals