The Detection Algorithms For The Faster-than-Nyquist System Under Time-varing Multipath Channel

In the future,the demand for diversified application scenarios with space-air-ground integrated network is constantly stimulating people to move forward to the next generation of mobile communications,and 6G poses a challenge to the spectrum efficiency to reach more than twice the spectrum efficiency of 5G.The Nyquist system,which is now widely used,is not conducive to the improvement of system capacity and spectral efficiency.By transmitting a smaller time interval and/or a narrower frequency interval,the Faster-than-Nyquist technology can transmit more data on a limited time-frequency resource,which is expected to greatly improve spectral efficiency.The time-frequency compression makes the Faster-than-Nyquist system having to solve the inherent interference problem.This thesis mainly studies the spectral efficiency of time-domain Faster-than-Nyquist systems and frequency-domain Faster-than-Nyquist systems and the bit error rate performance of multiple equalization detection algorithms.In this thesis,a new time-domain Faster-than-Nyquist implementation method is also studied,which can reduce the difficulty of receiver equalization,making the Faster-than-Nyquist technology one of the candidate technologies for 6G.The main work and innovations in this thesis include the following four aspects:1.For the time-domain non-causal Faster-than-Nyquist system,an equalization detection scheme based on bidirectional long short-term memory neural network is proposed.In view of the non-causal characteristics of the system’s current symbol equilibrium calculation related to several symbols before and after,the two-directional long short-term memory network can jointly decide the label of the current moment symbol by extracting the historical moment and future moment information,and determine the corresponding bit information through the unified mapping law of the real and imaginary parts,so as to complete the purpose of detection.The detection complexity of the proposed scheme has a linear relationship with the length of the received time series,which has more advantages than the existing benchmark BCJR algorithms in terms of complexity and time-consuming.In terms of detection performance,the proposed scheme has almost no performance loss from the existing benchmark BCJR algorithm when the system compression degree is small,and the detection performance gap with the benchmark BCJR algorithm can be improved by increasing the number of hidden layers in the case of large system compression.2.For the time-domain causal Faster-than-Nyquist system,an equalization detection scheme based on temporal convolutional neural network is proposed,Based on the fact that neural networks do not require accurate signal-to-noise ratio knowledge,this thesis abandons the matching and receiving step in the signal transmission process,and transforms the non-causal Faster-than-Nyquist transmission into causal Faster-than-Nyquist transmission,and at the same time,the temporal convolutional neural network is not only much lower than the bidirectional long short-term memory network in terms of complexity and time-consuming,but also has a more powerful classification accuracy advantage in the causal time series.The simulation results show that the detection performance of the proposed scheme is close to the bit error performance of the orthogonal Nyquist system under the AWGN channel,while the benchmark optimal algorithm and the bidirectional long short-term memory network equalization scheme have certain bit error performance losses with the orthogonal system.3.For the modeling and simulation of the Faster-than-Nyquist technology in fading channel,this thesis places the time-domain Faster-than-Nyquist system in a 3-path multipath channel known by the tap,and compares and evaluates the above two neural network detection algorithms with the existing optimal algorithms.The simulation results show that the detection algorithm based on bidirectional long short-term memory neural network is still a very potential solution for non-causal Faster-than-Nyquist signal detection,while the detection algorithm based on temporal convolutional neural network still performs best in multipath channel.For the frequency-domain Faster-than-Nyquist technology,this thesis applies it to the physical downlink shared channel platform and compares and evaluates it with the existing orthogonal frequency-division multiplexing technology,analyzes the defects of the current equalization algorithm,and verifies the essential characteristics of non-orthogonal transmission to obtain bandwidth efficiency improvement by sacrificing bit error performance.4.For the overlapping multiplexing characteristics of time-domain Faster-than-Nyquist signal,this thesis proposes a new implementation method,which designs the forming function of each stream through the principle of Schmitt orthogonalization,and converts the transmission mode of interlayer interference and intralayer orthogonal of the existing Faster-than-Nyquist system into a new transmission mode of interlayer orthogonal and intralayer self-disturbance.Through the waveform design,this new implementation method can not only ensure that the spectrum utilization of the system is consistent with the existing implementation,but also reduce the equalization complexity of the receiver,achieve the coexistence of low complexity and spectral efficiency lossless,which is conducive to the commercial use of Faster-than-Nyquist technology in the future.