A Modfied Real-Valued Feed-Forward Neural Network Low-Pass Equivalent Behavioral Model for Rf Power Amplfiers

This work proposes a modified Volterra-based series suitable for the low-pass equivalent behavioral modeling of radio frequency power amplifiers (RFPAs) for wireless communication systems. In a Volterra-based series, the instantaneous sample of the complex-valued output envelope is calculated by the sum of products that depend on the instantaneous and past (up to the memory length M) samples of the complex-valued input envelope. To comply with the constraints imposed by the bandpass behavior of RFPAs, the derivation of the proposed model starts from a general Volterra-based series given by the sum of contributions that include exactly one complex-valued information multiplied by a varying number (ranging from zero up to one less than the polynomial order truncation P) of real-valued amplitude components. A first reduction in the number of parameters is then performed by retaining only the one and two dimensional contributions. A second reduction in the number of parameters is finally achieved by introducing a third truncation factor S. In fact, if this additional truncation factor S is set equal to P −1, the proposed model contains all the two dimensional contributions. Moreover, when S is set equal to 0, the proposed model reduces to the largely adopted generalized memory polynomial (GMP) model. The proposed Volterra-based series retains the important property of being linear in its parameters and, in comparison with previous Volterra-based approaches, can provide a better compromise between number of parameters and modeling error. The proposed model is then compared with the GMP model in a scenario of same number of parameters. When applied to the modeling of input-output data obtained from a circuit-level description of a GaN HEMT Doherty PA excited by a LTE OFDMA signal, the proposed model reduces the normalized mean square error (NMSE) by up to 3.4 dB. Additionally, when applied to the modeling of input-output data measured on a GaN HEMT class AB PA excited by a WCDMA signal, the proposed model reduces the NMSE by up to 1.3 dB.

show abstract

A Modfied Real-Valued Feed-Forward Neural Network Low-Pass Equivalent Behavioral Model for Rf Power Amplfiers

Cited by 2 publications

References 16 publications

A behaviour model for power amplifiers based on sum of products between one-dimensional polynomials and finite impulse response filters

A behaviour model for power amplifiers based on sum of products between one-dimensional polynomials and finite impulse response filters

A Modified Two Dimensional Volterra-Based Series for the Low-Pass Equivalent Behavioral Modeling of Rf Power Amplifiers

Contact Info

Product

Resources

About