Novel Multiband Linearization Technique for Closely-Spaced Dual-Band Signals of Wide Bandwidth Ying Liu* # , Patrick Roblin*, Hai Yu*, Shihai Shao # , and Youxi Tang # *Department of Electrical & Computer Engineering, OSU, Columbus, 43210, US A # National Key Lab of Science & Technology on Conunications, UESTC, Chengdu, 611731, China Abstract-In the concurrent dual-band digital predistortion (DPD), it is required that each band of interest be accurately measured, so that the in-band distortion can be modeled and suppressed. However when the two bands are closel y spaced, the inter-band modulation b y -products overlap with the cross modulation and the in-band intermodulation, thus leading to linearization performance degradation. On the other hand, b y directl y appl y ing the single band linearization for closel y -spaced dual-band signals, the one-band DPD approach will saturate in performance due to the wide bandwidth as well as the high nonlinear order and memor y depth required resulting in an ill-conditioning problem. To resolve the spectral overlaps taking place in closel y -spaced dual wide-band s y stem, (1) the in-band intermodulation, (2) the cross-modulation, and (3) the inter-band modulation b y -products are j ointl y extracted in the new proposed dual-band PA characterization scheme to y ield an improved frequenc y /time selective modeling. T he signals of interest extracted in each band using the proposed model are then used for dual-band DPD coefcient extraction. Improved modeling and linearization performances b y up to 5 dB are verifed with experiments performed on two 40-MHz LTE advanced signals 40 MHz apart. Index T erms-Digital predistortion, concurrent dual-band, memor y pol y nomial, power amplifer. I. INTRODUCTION Modern wireless communication systems are continuously developing in data rate and bandwidth to support more users and provide more data services, e.g., the Long Term Evolution (LTE)-advanced system are employing orthogonal fequency division multiplexing (OFDM) to achieve high peak data rate of 1 Gb/s in the downlink with a maximum instantaneous bandwidth of 100 MHz [1]. To further fully utilize the fre quency spectrum and support the signifcant increase of the modulated signal bandwidth, concurrent dual-band amplifers have been developed recently [2]. As the bandwidth increases, it poses stringent requirements on the power amplifer (PA) linearity and efciency. Digital predistortion (DPD) has proven to be one of the most cost effective methods to compensate for the PA nonlinearity and memory efect in order to suppress the spectral regrowth [3]. However, in concurrent dual-band transmitters, directly apply ing the conventional one-band DPD technique on the full band results in large transmitting bandwidth which requires analog to-digital converters (ADCs) and digital-to-analog converters (DACs) with prohibitively high sampling rates. Recently, several dual-band linearization methods have been proposed to compensate for the PA nonlinearities and memory effect in dual-band cases [4]-[7]. The authors in [4] pro posed a frequency-selective linearization technique based on PA characteristics measurement using a large-signal network analyzer (LSNA). It successfully removes in-band distortion and inter-modulation distortion up to the third order. Bassam et al. [5] proposed a 2D digital predistortion (2D-DPD) architecture, in which the radio fequency (RF) signals in each band are captured and digitized separately. The authors also proposed a multi-cell processing digital predistortion (MCP DPD) architecture [7], where the nonlinearities and memory effects of multi-carrier signals in each frequency band are compensated for independently of the signals in the other fequency bands. All these methods can signifcantly reduce the bandwidth and sampling rate requirement because the signal bands are widely located, thus the bands of interest and the corresponding distortions can be captured separately by means of fltering. However, when the transmitting signal bands are closely spaced, the in-band intermodulation, the cross-modulation, and the inter-band modulation by-products overlap with each other. In this case, the band of interest and its distortion cannot be simply isolated using fltering. This introduces an error which corrupts the DPD model extraction process and degrades the linearization performances. In this paper, we propose a novel multiband linearization technique to handle the dual-band signals with closely-spaced bands. A novel multiband PA model based on dual-band memory polynomials is proposed, which jointly extracts the in-band intermodulation, the cross-modulation, and the inter band modulation by-products. Using this new joint model estimation for the two bands of interest instead of separating them by fltering, the DPD coefcients for the adjacent bands can be extracted by using the indirect learning architecture. Experiments are performed on two 40-MHz LTE-advanced signals spaced by 40 MHz to demonstrate the effectiveness of this new method. II. MULTIBAND LINEARIZATION FOR DUAL-BAND SIGNAL Fig. 1. Block diagram of the proposed multiband DPD system. 978-1-4799-8275-2/15/$31.00 ©2015 IEEE