Forward behavioral modeling of concurrent dual-band power amplifiers using extended real valued time delay neural networks

Abstract

The distortions induced by inter-modulations and cross-modulations in concurrent dual-band power amplifiers (PAs) are evidenced and characterized using multi-tone signals. Given the presence of cross-modulations, a comprehensive extended real-valued time-delay neural network (extended-RVTDNN) is proposed to model the nonlinear behavior of concurrent dual-band Pas. Two three-carrier WCDMA signals are applied to a dual-band Doherty PA prototype for modeling verification. The experimental results show that the proposed model approximates the PA with normalized mean square errors (NMSEs) of -38.48dB and -35.42dB in the lower and upper bands, respectively. Compared with the conventional single-band RVTDNN, this new method achieves an improvement in accuracy of more than 10dB.

A low-order nonlinear amplifier model with distributed delay terms

“…Among these * Correspondence: hayrettinyuzer@gmail.com published studies [4,[6][7][8][12][13][14] are baseband and [2,3,5,9,11] are passband modeling techniques. Mathematical definitions and performance comparisons of the models were also published for baseband [15][16][17][18] and passband modeling [18].…”

“…There are a number of behavioral modeling methods published in the literature, such as the Saleh model [2], general power series expansion [3][4][5], memoryless polynomial modeling [6], look-up-table modeling [7,8], Volterra series [9], Wiener-Hammerstein (3-box) [10,11], Volterra [12,13], and neural networks [14]. Among these * Correspondence: hayrettinyuzer@gmail.com published studies [4,[6][7][8][12][13][14] are baseband and [2,3,5,9,11] are passband modeling techniques. Mathematical definitions and performance comparisons of the models were also published for baseband [15][16][17][18] and passband modeling [18].…”

Behavioral model on transmitter power amplifier and IQ imbalance

“…The neural network has also been applied to model the behavior of power amplifier nonlinearity and memory effects, either in feed-forward structure [59][60][61][62][63]68] or in 39 feedback [64][65][66] structure. However, these models do not consider IQ imbalance.…”

“…typical Time-Delay Neural Network (TDNN) with 2 inner layers Time delay neural network (TDNN) is proposed to model PA behavior[59][60][61][62][63]. It is based on feed-forward neural network structure[60].…”

“…And ,( ) () il xn is the inner state of Neuron i and Layer l (inner stage l ). And l N denotes the total number of neurons of Layer l .In[62,63], a TDNN neural network is proposed to predistort the power amplifier, andthe performance of convergence rate is compared among various adaptive identification methods. The neural network is composed of 2 inner layers, with the activation function being tansig function.…”