Altera原版文章：数字预矫正(DPD)3.ppt

Digital Predistortion using Polynomial Approach Agenda Introduction Indirect Learning Architecture Forward Path Feedback Path Block diagram of complete solution Example Resource Estimate Numerical Accuracy Summary Introduction: Predistortion DPD - Implementation Choices Two basic approaches Look-Up-Table (LUT) contains points on the transfer function Polynomial representation of the transfer function Agenda Introduction Indirect Learning Architecture Forward Path Feedback Path Block diagram of complete solution Example Resource Estimate Numerical Accuracy Summary Indirect Learning Architecture Indirect Learning Architecture Feedback path (predistorter training block A) can work offline A block of [z(n), y(n)], i.e., [input to PA, output from PA] can be stored in a look up table The predistorter training block derives the inverse response of the PA by performing polynomial curve fitting on the data in the LUT (produces polynomial coefficients by solving the least squares problem) The polynomial coefficients are computed such that e(n)=0, then y(n)= x(n) The generated coefficients in the feedback path are used by the predistorter in the forward path until the next update by the training block Agenda Introduction Indirect Learning Architecture Forward Path Feedback Path Block diagram of complete solution Example Resource Estimate Numerical Accuracy Summary Predistorter Model Assume Predistorter is modelled by memory polynomial model[1] [eq 1] z(n) = c10 . x(n) + c30 . x(n) . |x(n)|2 + c50 . x(n) . |x(n)|4 +… c(2K+1)0 . x(n) . |x(n)|2K + c11 . x(n-1) + c31 . x(n-1) . |x(n-1)|2 + c51 . x(n-1) . |x(n-1)|4 +… c(2K+1)1 . x(n-1) . |x(n-1)|2K + c12 . x(n-2) + c32 . x(n-2) . |x(n-2)|2 + c52 . x(n-2) . |x(n-2)|4 +… c(2K+1)2 . x(n-2) . |x(n-2)|2K +... c1Q . x(n-Q) + c3Q . x(n-Q) . |x(n-Q)|2 + c5Q . x(n-Q) . |x(n-Q)|4 +… c(2K+1)Q