On The Theory And Application Of Advanced Digital Predistortion

Power amplifier(PA)is the most critical component in wireless communication systems that produces nonlinearity.Digital Predistortion(DPD)has become a popular linearization technique for radio frequency front-end,due to the relative low cost and the flexibility in programming.The future trend of DPD development focuses on wideband scenario with low-cost architecture and low computational complexity.Concerning the fact that analog-to-digital converter(ADC)is one of the most expensive and energy-consuming components in the feedback path of a DPD system,the study is around reducing the resolution and sampling speed of ADCs.Low resolution forces down prices and power consumption and counteracts the so-called 'minimization',while re-ducing sampling speed reduces the acquisition bandwidth and thus push down the cost to counteract 'large-bandwidth'.Moreover,some other critical issues related to DPD implementation are also solved.Conventional DPD system depends upon acquisition bandwidth no less than five times the original signal bandwidth with high resolution.This prevents the conventional DPD technique from being applied to broadband system.This dissertation considers to reduce the ADC resolution to an extreme case of 1-bit,and derives the DPD algorithm with 1-bit acquisition theoretically.In the meantime,the existing issues in the proposed 1-bit DPD system are addressed one-by-one,including the iterative frequency domain based time alignment algorithm with 1-bit samples,the estimation of step size,and the overall system complexity analysis.The feasibility and linearization performance are validated in the experimental tests.A modified 1-bit DPD system is further proposed based on the preceding one,i.e.the forward modeling assisted 1-bit data acquisition based model extraction method,and it is proved that the modified one is also capable of linearizing the wideband PAs.In wideband applications,the feedback bandwidth of a DPD system can be limited,and the conventional model accuracy is also hard to increase.The spectral extrapola-tion method is introduced in the dissertation,which tends to recover the full band PA output signal iteratively using the band limited signal.This leads to a more accurate DPD model than the conventional approaches.The numerical stability of the spectral extrapolation method is analyzed afterwards,and a regularization algorithm is proposed to improve the numerical property.In terms of the cost of hardware implementation,a segmented fast Fourier transform method is designed to reduce the computational com-plexity,and thus the overall efficiency of the system is increased significantly.An improper-chosen normalization gain affects the linearization performance of DPD system,which is investigated in theory and simulations using indirect learning architecture.A one-step extraction of optimal normalization gain,which has strong and reasonable physical meaning,is proposed on the principle that the DPD block has a unity gain and does not vary the average power of the input signal.Besides,the I/Q imperfections in the feedback path of DPD system is analyzed,and an accurate inverse modeling approach is proposed to eliminate the effect of I/Q imbalance.This method is then extended to the scenario that both quadrature modulator and demodulator errors exist at the same time.The results of experimental test prove the feasibility of the theory.