Multiband Linearization Technique for Broadband Signal With Multiple Closely Spaced Bands

Abstract

Under the scenario of noncontiguous carrier aggregation with closely spaced multiple bands, the interband modulation products after power amplifying will overlap with the bands of interest. This leads to linearization performance degradation for the conventional three-band digital predistortion (DPD) method which only accounts for in-band intermodulation (IMD) and cross modulation. To resolve the spectral overlaps, a multiband linearization technique combining the direct power amplifier (PA) model extraction and the indirect DPD learning strategies is proposed. A multidimensional memory polynomial (MD-MP) model is first presented to jointly extract: 1) the in-band IMD; 2) the cross modulation; and 3) the interband modulation products from the observations for the multiple bands of interest to yield an improved frequency/time selective modeling. The extracted MD-MP models are then used in the indirect learning structure instead of the physically filtered observations to estimate the multiband DPD coefficients, given the fact that the observations spectrally overlap with each other which would result in interference for DPD coefficients estimation with direct filtering. Moreover, a novel MD cubic spline (MD-CS) basis is applied for the first time to the considered three-band scenario while the principal component analysis is used to alleviate any ill-conditioning problem during the model extraction. Improved PA modeling and DPD linearization performances are verified through experiments performed on three groups of wide-bandwidth noncontiguous aggregated signals covering a variety of application scenarios.