Research On OFDM And Deep Learning Techniques For LEO Satellite-to-ground High-speed Systems

Numerous satellites are running at the low Earth orbit(LEO)for communications,meteorology,land resources and scientific exploration,etc.LEO satellites are playing a significant role in the exploration of science and improving the quality of people's daily life.Nowadays,following the high-precision,multi-function,and multi-task trend of space-borne exploration techniques,the demand of the high-speed communication system is increasing.Traditional LEO satellite-to-ground communication systems have been challenging to meet the requirement of future LEO satellites.X-band LEO satellite-to-ground data transmission systems are now very developed.The development of new higher frequency communication systems,such as Ku/Ka-band-based LEO satellite-to-ground data transmission systems,can only temporarily relieve the tension of spectrum resources.Therefore,making full use of limited spectrum resources is more important.Based on the characteristics of the X-band LEO satellite-to-ground data transmission system,this dissertation improves the existing orthogonal frequency division multiplexing(OFDM)technique,which is known for its high spectral utilization.At the meantime,deep learning technologies are introduced to improve the performance of the entire communication system for further improving the performance of LEO satellite-to-ground data transmission.This dissertation mainly solves two disadvantages of the OFDM technique,which are the high side-lobe and high peak-to-average power ratio of the OFDM signal.The main contributions of this dissertation are as follows:1.The high side-lobe of traditional OFDM signals leads to a significant decrease in spectral efficiency,which is one of the reasons that limit the development of the OFDM technique in the aerospace field.Therefore,this dissertation proposes a new transmission signal based on the OFDM technique,which is hybrid modulation based filtered orthogonal frequency division multiplexing(HMF-OFDM),and hence the high side-lobe of the OFDM signal could be handled.The spectral efficiency,which is calculated in-30 d B minimum power spectral density bandwidth,of HMF-OFDM signals is 1.81 times higher than that of traditional OFDM signals.Meanwhile,a hybrid modulation method is adopted to avoid the interference of the filter to the sub-carriers of OFDM signals,and then the utilization of spectral resources could be maximized.2.According to a large carrier frequency offset caused by LEO satellite-to-ground data transmission link,a sliding differential correlation algorithm,which is a symbol timing synchronization algorithm based on variable-length training sequence,is proposed.Traditional data-aided timing synchronization algorithms feature a long duration of the training sequence,poor robustness against noise,and sensitivity to carrier frequency offsets.Compared with traditional algorithms,the proposed algorithm can flexibly adjust the duration of the training sequence by pre-evaluating the communication channel so as to maximize the duration of data frames to improve the data transmission efficiency.In addition,the proposed algorithm is significantly robust to additive white Gaussian noise and carrier frequency offsets.Simulation results show that the carrier frequency offsets robustness of the proposed algorithm exceeds the possible carrier frequency offset caused by the LEO satellite-to-ground data transmission link.3.To handle the nonlinear distortion from the LEO satellite-to-ground data transmission system,this dissertation proposes a deep neural network(DNN)based digital signal recovery(DSR)technique.The proposed technique captures the prior knowledge,such as nonlinearities of radio frequency power amplifiers(RF-PAs),of space-borne transmitters to improve the quality of the signal received at round stations by modeling the prior knowledge using DNNs.Benefiting from the noise and power variation robustness of the DNN,the proposed technique can handle the large-scale power variation of the received signal and additive white Gaussian noise introduced by the LEO satellite-to-ground communication channel.The proposed technique can correct high signal distortions caused by the space-borne RF-PA at the ground station and hence allows the space-borne RF-PA to work close to its saturation region.This way can indirectly handle the high peak-to-average power ratio of the OFDM signal,leading to a high power efficiency of the LEO satellites.Experimental results show that the proposed technique could improve the power-added efficiency of the RF-PA from 32.6% to45% with the same error vector magnitude(EVM),compared with the traditional power back-off technique.The proposed technique could work stably with a received signal power variation of 12 d B,while it is a great challenge for the conventional memory polynomial-based DSR technique.4.Based on the DSR technique,this dissertation proposes a DNN based filtered OFDM receiver.In the proposed design,DNN can learns not only the nonlinearity of the space-borne RF-PA,but also the modulation schemes of the received signal.This work integrates the recovery of the nonlinear distortion and the digital demodulation into one model.Compared with existing DNN based receivers,this technique concurrently considers the features of the hardware and communication channel.Simulation results show that the proposed receiver can accommodate at least five types of filtered OFDM modulations(QPSK,8PSK,8QAM,16 QAM,and 16APSK)and two typical working states of the RF-PA(class AB and class B).The proposed technique has the potential to improve the performance of on-orbit LEO satellite communication systems with minimal modifications to LEO satellites.It also provides a feasible method for future LEO satellite-to-ground data transmission systems.This dissertation develops the physical layer of a communication system and introduces deep learning techniques to improve the overall performance of X-band LEO satellite-to-ground data transmission systems.Experiments and simulations are made for verifying the proposed techniques,which could provide more feasible solutions for future LEO satellite-to-ground data transmission systems.