Quasi-Exact Inverse PA Model for Digital Predistorter Linearization Naveen Naraharisetti * , Patrick Roblin * , Christophe Quindroit * , Meenakshi Rawat * , Shahin Gheitanchi † * Ohio State University, Department of Electrical & Computer Engineering, Columbus, OH, 43210, USA † Altera Europe Limited, High Wycombe, Buckinghamshire, HP12 4XF, England Abstract— This paper reports the first experimental appli- cation of the recently reported quasi-exact inverse (QEI) for memory-polynomial or memory-spline models in the design of a digital predistorter (DPD) linearizing a power amplifier (PA). In comparison to indirect learning architecture, where the coefficients of the DPD are extracted by swapping the input and output variable in any PA model, the DPD extraction is performed from the PA model directly. One of the advantages of using this scheme is that the output noise of the PA is not included in the regression matrix, thus improving the performance. In this paper, B-splines are used to extract the PA model since the performance of the DPD depends on the accuracy of the PA model. The new DPD algorithm relies on an arbitrary number of memory delays as needed for the QEI of the PA model. The evaluation of the model’s performance is conducted on a real time application. A Long Term Evolution (LTE) signal of 10 MHz bandwidth is used to compare the performance with a memory polynomial (MP) DPD model used in indirect learning architecture. The measurement results demonstrate that there is a noticeable improvement in terms of Normalised Mean Square Error (NMSE) and Adjacent Channel Power Ratio(ACPR) when using the QEI model for DPD. Note that this is achieved without any iteration as in practical DPD systems. Better results are possible when the PA model represents the PA behavior more accurately. Index Terms—PA model, DPD , NMSE , ACPR , Quasi Exact, Direct Learning, indirect learning , amplifiers , linearization, FPGA. I. I NTRODUCTION Currently complex envelop techniques like Wideband Code Division Multiple Access(WCDMA) and Orthogonal Fre- quency Division Multiplexing(OFDM) signals are employed in high data rate transmissions for their spectral efficiency. However, these modulated schemes impose strict linearity requirements on the PA because of their non-constant envelope with high peak to average power ratio (PAPR). Due to the inherent nonlinearities of PA, the signal develops spectral regrowth in in-band, and intermodulation distortion (IMD) products in out-of-band. A linearizing technique is then needed in order to reduce the spectral regrowth while achieving a good power efficiency since the PA is operated near the saturation region. DPD is one of the commonly used linearizing tech- nique because of its robustness, moderate implementation cost and high accuracy. Most of the available predistorters (PDs) are based on indirect learning (IL) as shown in Fig. 2(a), wherein the inverse of the PA is modeled using a postdistorter inverse model and the coefficients are transferred to the PD. The two main drawbacks that affect the performance of this method are[1]: 1) When y is noisy due to measurement setup, IL requires to find an inversion of the noisy regression matrix. Due DPD DAC PA ADC DDC, Time Alignment, Coefficients Estimation ܮ ݔ Fig. 1. Block Diagram of Test Bench to this the adaptive algorithm converged to biased values. 2) The identified post distorter which is copied into the PD does not guarantee a good pre-inverse filter for the nonlinear device because of using commutative property for non-linear systems. In order to mitigate these issues a new scheme is developed in [2]. Initially an accurate PA model is estimated and then the DPD function is obtained by inverting the PA model. The DPD function is defined iteratively only for one memory delay. It takes multiple iterations for converging to the actual inverse model. This scheme is referred as direct learning (DL) and a comparison in performance with IL is reported in [3]. It is observed that DL algorithms achieve better performance in terms of NMSE, but with a few iterations. The PD based on IL model is estimated by using a least square (LS) method and it doesn’t need any iterative process whereas the PD developed in [2] is based on iterative process. In [4], the impact of noise on the identification process of PD in indirect learning architecture is studied and verified to contribute to the degradation in NMSE value. In this paper, a new DPD algorithm for PA model with an arbitrary number of memory delays is experimentally investi- gated for the first time. This DPD algorithm is based on the quasi-exact inverse of the PA model which achieves typically less than -84 dB NMSE in simulations when applied to the PA model itself [5]. The experimental verification of the model is performed on a testbench setup which closely resembles a real base station. Most of the DPDs in the current literature depend on vector signal generator (VSG) and vector signal analyzer (VSA) which exhibit very high performance and thus are not cost effective. The advantage of using a real system is that the non-idealities of the system can be accounted for and mitigated by the DPD algorithm. The testbench consists 978-1-4799-2935-1/13/$31.00 ©2013 IEEE