Abstract

Wireless communications play an extremely important role in our daily life. The ubiquitous cellphones and smartphones can provide not only voice service but also data service when they are connected to cellular network. With the enormous amount of mobile applications available to smartphones, people can listen to music, watch movies, check emails, update and share information on social media anytime and anywhere on their smartphones. All of these activities are possible because of the fast development of the wireless communication technology. Nowadays many people are enjoying the entertainments brought by 4G technology. In the meantime, companies and engineers in telecommunication industry have already been developing the 5G technology.
To support the high data transmission rate of 4G or 5G technology, a complex and robust infrastructure is essential. Radio-over-Fiber (RoF) technology is a good solution to build such an infrastructure. RoF transmission can provide extremely high capacity with very low attenuation. Besides, RoF transmission system is much cheaper compared to the traditional way of transmission such as microwave coaxial cable or digital optical fiber transmission. However, like other analog systems, RoF system suffers from signal distortion induced by nonlinearities.
RF power amplifier is needed to transmit the signal from a base station to the handsets because the attenuation of RF signal in the air is very high. Although many new types of RF power amplifiers have been designed to improve the power efficiency and the linearity, nonlinearity is still a big problem.
To reduce the distortion of an RoF system, as well as that induced by an RF power amplifier, a new digital predistortion (DPD) technique, envelope-assisted RF digital predistortion, is presented in this thesis. This DPD model is evolved from the memory polynomial, and composed of an RF memory polynomial and a baseband memory polynomial. Therefore, this model takes both the RF signal and the baseband envelope as inputs to realize the digital predistortion. The RF polynomial is aimed to suppress the in-band nonlinearity, out-of-band nonlinearity and short-term memory effect, while the baseband polynomial is aimed to weaken the in-band nonlinearity and long-term memory effect. The indirect learning architecture is adopted and the least squares method is used to calculate the coefficients of the model.
The validity of the proposed DPD has been proved by simulation and experiment. In the simulation system, an RoF model is built which can induce both short-term and long-term memory effects, as well as the in-band and out-of-band nonlinearities. The optimal RF and baseband memory depths of the proposed DPD adopted in the simulation system have been found in terms of error vector magnitude (EVM) improvement and complexity. After the establishment of the memory depths, three simulation cases have been run to evaluate the performance of the proposed DPD. The first case is a normal two-band test, where two 20 MHz LTE signals are located at 800 MHz and 900 MHz. Simulation results show that EVM is improved by 14.6 dB, adjacent channel power ratio (ACPR) is suppressed by 13.5 dB, and third order intermodulation distortion (IMD3) is weakened by 19.5 dB. In the second case, two LTE signals are closely located at 800 MHz and 840 MHz. Simulation results show that EVM is improved by 16.3 dB, ACPR is suppressed by 15.5 dB, and IMD3 is weakened by 27.5 dB. In the third case, three LTE signals are located at 800 MHz, 850 MHz and 900 MHz. EVM is improved by 15.7 dB, ACPR is suppressed by 15.2 dB and IMD3 is weakened by 14.1 dB. The performance of the proposed DPD has also been compared to that of a baseband 2D DPD. Comparison results show that 2D DPD is only better at improving the EVM and ACPR in the first case, and the proposed DPD is superior to the 2D DPD on all the other aspects. Especially, the proposed DPD is not limited by the number of signal bands, while the 2D DPD is limited to two-band scenarios.
Experimentation has been conducted to further prove the effectiveness of the proposed DPD in a real RoF system. A similar test flow is adopted as in the simulation work. Firstly the optimal RF and baseband memory depths have been found. Then three test cases have been done. In the first case, EVM is improved by 6.9 dB, ACPR is suppressed by 14.8 dB, and IMD3 is weakened by 12.2 dB. In the second case, EVM is improved by 4.8 dB, ACPR is suppressed by 8.6 dB, and IMD3 is weakened by 9.5 dB. In the third case, EVM is improved by 8.9 dB, ACPR is suppressed by 16 dB, and IMD3 is weakened by 13.6 dB. Comparison with the 2D DPD shows the same result as in the simulation that the 2D DPD is only better at improving the EVM and ACPR in the first case. Influence of input power on EVM improvement has been studied. Experiment results show that the EVM improvement can be optimized at specific input power level. Sampling bandwidth has been reduced by down-converting the RF signal and sampling it at an intermediate frequency. Therefore, the implementation cost of the proposed DPD could be greatly reduced. In the two-band test, experiment results show that the performance of the proposed DPD with down conversion is even better than that of without down conversion. Influence of sampling bandwidth on EVM improvement has also been studied and the experiment results show that after down conversion, higher sampling bandwidth does not lead to significant EVM improvement.