Digital Pre-Distortion Linearization Technologies For Concurrent Multi-band Multi-mode RF Power Amplifiers

Concurrent multi-band wireless technology is the key technology of the future multi-bandfusion architecture wireless systems, which can greatly reduce the size of systems and energyconsumption, effectively use the existing discontinuous spectrum resource, and achieve seamlesssystem upgrade. A concurrent multi-band RF power amplifier (CMBRFPA) is a core componentof multi-band fusion architecture in future wireless communication systems, which has the largestenergy consumption with the strongest nonlinearity. Therefore, super-linear and high efficiencyCMBRFPA is the key factor to achieve small size, low power consumption andgreencommunication for the future multi-band fusion architecture wireless system, achieving long-termsustainable development. Therefore, this thesis will firstly do research on dynamic nonlinearmodeling and digital pre-distortion technology of single-band radio frequency power amplifier(RFPA) for broadband application. Then theoretical analysis and experimental research will bepresented for concurrent multiband RFPA’s nonlinear characteristics. Finally, the thesis will studythe nonlinear compensation problem of a concurrent multi-band RFPA, especially the dynamicnonlinear modeling and baseband digital pre-distortion linearization technology of the dual-bandRFPA. The main contents of this thesis is divided into two parts as following:First part: the dynamic nonlinear behavior model of a single-band RFPA for broadbandapplication is studied. The augmented general fractional order memory polynomial(AGFMP) model is proposed. The experimental results show that the numbers of thecoefficients of a AGFMP are30%less than those of a general memory polynomial(GMP), but modeling and linearization ability of the AGFMP is just reduced by lessthan1dB, comparing to that of a GMP. In order to reduce the order of the polynomialmodel, the augmented parallel two-box general memory polynomial model is designed,which avoids the problem of twice solving model coefficients needed by thetraditional two-box model. The experimental results demonstrate that it leads tonearly0.5dB improvement in terms of normalized mean square error (NMSE)compared to that of the memory polynomial (MP) model, while the modelcoefficients decrease50%for modeling the strongly nonlinearity of a broadbandDoherty PA. In order to simplify the complex structure of the neural network (NN) andimprove the convergence speed, a real-valued time-delay radial basis function NNmodel is presented in this thesis. The training time of this NN is lower than the real-value time-delay NN model based on BP, but it cannot accurately describe the strongnonlinearity of the power amplifiers. In order to solve the problem of strong memoryeffect and nonlinearity linearization for the broadband Doherty PA, the augmentedradial basis functions NN model is proposed, which can compensate for the strongnonlinearity of a broadband Doherty PA more accurate than a traditional NN model,and has the advantages of simple structure and fast training speed.Second part:the nonlinear modeling and linearization method of concurrent dual-band RFPA isinvestigated in-depth. First of all, the nonlinearity of concurrent dual-band RFPA (CDB-RFPA) isclassified. The importance of the sub-band interval for concurrent three-band RFPA has also beenanalyzed. Then the sub-band separation baseband signal extraction technology is proposed for theconcurrent multi-band application. The sampling frequency of analog-to-digital converters (ADCs)for the proposed extraction method depends only on the maximum sub-band signal bandwidth. Theinfluence of the synchronicity between the low frequency and the high frequency sub-band to thelinearization is studied after that. The importance of two sub-band synchronization is confirmedwith measurement results. At the same time, a new synchronous method for dual-band signals isproposed, and the subcarrier polymerizable dual-band RF signal has been obtained. A dual branchpre-distortion method is proposed for restraining the nonlinearity of a special sub-carrieraggregation dual-band RFPA (DB-RFPA),. The sub-band separation sub-carrier aggregationdigital pre-distortion method is presented for linearizing the universal sub-carrier aggregation DB-RFPA. The above two methods can effectively suppress the spectral re-growth in the carrierfrequency of two sub-bands for sub-carrier aggregation DB-RFPA while the sampling rate ofADCs is significantly reduced. However, the requirements for the rate of digital-to-analogconverters (DACs) are not lowered by these two methods. Therefore, the sub-band separationconcurrent dual-band digital pre-distortion method is proposed in this thesis. The proposed methodgreatly reduces the sampling rate of ADCs and DACs, and the conversion rate depends only on themaximum sub-band signals bandwidth, which is independent of the interval of two sub-bands. Butthis method inevitably improves the complexity in digital processing. Finally, the sub-bandseparation mutual coupling modeling method is introduced to avoid the complicated digitalprocessing. And the modeling capability forthe nonlinearity of a CDB-RFPA is demonstrated withthe experimental results.