Research On Key Technologies Of Broadband Wireless Digital Predistortion

In order to satisfy the requirements of the fast developing wireless communication services, mobile communication systems have evolved from the first generation to the current fourth generation, and the bandwidth of associated transmit signal has also increased from 25 KHz to 100 MHz. Meanwhile, the peak to average power ratio(PAPR) of the transmit signal has also increased to a higher level. Thus, it poses more stringent requirement on the linearity of the radio frequency(RF) power amplifier with high efficiency and high output power level.To improve linearity of the power amplifier, this thesis focuses on core techniques of the broadband digital predistortion(DPD), and the contributions can be summarized as follows.Firstly, in this thesis, a method is developed to directly extract the reverse function from the transfer function of the power amplifier. Through model identification, a generalized memory polynomial(GMP) model is estimated to characterize the power amplifier behavior. By extracting the reverse function of the GMP model, the DPD function with an unknown DPD parameter is obtained. The DPD parameter is then accurately estimated by constructing a univariate polynomial from the GMP model and finding its roots. Thus, the parameter estimation error in conventional direct learning method is significantly reduced. This proposed method is demonstrated to outperform the conventional indirect learning DPD method by 3 d B and the direct learning DPD method by 2.2 d B in terms of the adjacent channel leakage ratio(ACLR) for LTE-A signals of bandwidth ranging from 20 MHz to 100 MHz.Secondly, this thesis proposes a wideband DPD technique to reduce the bandwidth of the DPD feedback channel. When the bandwidth of the DPD feedback channel is narrower than that of the amplified signal, the proposed method introduces bandwidth restriction to the conventional nonlinear model identification process. A digital filter is designed to apply frequency constraints on each nonlinear model basis and associated signals to ensure that the coefficients can be derived following the same identification process as the one without bandwidth limitation. The proposed technique is validated to significantly reduce the conventional feedback bandwidth by 72% with ACLR being lower than-45 d Bc for signals of bandwidth ranging from 20 MHz to 100 MHz. Specifically, for 100 MHz LTE-A signal, this technique can reduce the bandwidth from the conventional 500 MHz to 140 MHz.Finally, a novel DPD architecture is designed with an extendable dynamic range for the feedback channel. When the dynamic range of the feedback signal is beyond that of the ADC in the DPD feedback channel, the proposed architecture uses a RF cancellation chain to suppress the linear component of the feedback signal before entering the ADC to reduce its dynamic range. Subsequently, in the digital domain, the ADC samples are used to reconstruct the complete feedback signal of high dynamic range to estimate the DPD coefficients. Through finely tuning the RF cancellation and baseband reconstruction, the proposed architecture significantly reduces the requirement on the ADC dynamic range. For instance, for a 100 MHz LTE-A signal, the proposed architecture achieves the same ACLR performance with the dynamic range of the ADC being reduced by 26.3 d B compared to the conventional DPD method.In summary, this thesis studied the direct learning method, the technique to reduce the feedback bandwidth, and the architecture to reduce the ADC dynamic range of the broadband DPD. These results could be applied to the satellite communications, the broadband wireless local networks, and the next-generation mobile communication systems to improve the efficiency of the RF power amplifier, suppress the in-band nonlinear distortion, and reduce the spectral regrowth, particularly under the scenario of wide bandwidth and high PAPR.