Research On Channel Estimation And Signal Detection Based On Deep Learning In DM-OFDM-IM System

After decades of development,wireless mobile technology has entered the fifth generation mobile communication system(5G)era.Different from the previous 1G to 4G communication systems,5G should not only solve the communication problems between people,but also realize the information interaction between people and things,the expansion of communication scenarios,and more diverse and differentiated service requirements such as AR and VR,which require the further improvement of 5G data transmission rate.Orthogonal frequency division multiplexing(OFDM)technology,based on its advantages of high spectrum efficiency,strong anti inter symbol interference and simple structure,has become the core air port technology of 4G,and it is recognized that it can still play an important role in 5G system.However,facing the requirements of 5G higher energy and spectrum efficiency,OFDM technology also needs to be further up-graded.Dual mode OFDM(DM-OFDM)technology is combined with index modulation to improve the throughput of traditional index modulation based OFDM system.Specifically:in the DM-OFDM system,the subcarriers are divided into several sub-blocks,all subcarriers are divided into two groups in each subblock,which are modulated by two distingu ishable constellations,respectively.Therefore,information bits can be transmitted not only by traditional constellation,but by index related to the activated subcarriers and the other constellation.Under the given throughput,the DM-OFDM system has better BER performance and the same or lower computational complexity than other OFDM systems using index modulation.In this context,orthogonal frequency division multiplexing index modulation(OFDM-IM)based on index modulation is proposed by the industry.Different from traditional OFDM,OFDM-IM only activates some subcarriers to transmit modulated signals.The sequence number of the activated subcarrier is determined by some information bits to be transmitted.This part of information can also be obtained by the decision of activating the subcarrier index at the receiving end.Therefore,OFDM-IM can effectively improve the system energy efficiency and anti frequency offset reliability by silencing some subcarriers.However,this advantage is obtained at the loss of certain spectral efficiency.Most of the existing detectors use the channel state information(CSI)as the known condition to detect the bit error rate.This paper combines the channel estimation with the neural network detector to detect the bit error rate of the OFDM index modulation system given that the channel state information is unknown.On the basis of traditional channel estimation methods,this paper also proposes a channel estimation method based on deep learning.Compared with the traditional least squares(LS)channel estimation algorithm,the channel estimation method based on deep learning can get better mean square error(MSE).In order to further improve the spectral efficiency of the system,another idea using index modulation concept,namely dual mode orthogonal frequency division multiplexing index modulation(DM OFDM IM),is proposed by the industry.Similar to OFDM IM,the transmission information sequence is used to select some subcarrier index in DM-OFDM-IM system.The difference is that the selected and bit selected subcarriers send signals,but only in two independent modulation symbol sets.Therefore,in DM-OFDM-IM,information bits can be transmitted to the receiver through two modulation symbol sets and subcarrier index.So DM-OFDM-IM can obviously get better throughput performance.However,in order to obtain the transmitted information bits,it is necessary to judge the subcarrier selection index and the two symbol sets at the receiver of DM-OFDM-IM at the same time.Although the traditional maximum likelihood(ML)detector can obtain the optimal detection performance,it is very complex and even difficult to realize when the modulation symbol set is large and the number of subcarriers is large.For the design of receiving detector for OFDM-IM,this paper attempts to use deep learning method to construct a signal detector pair based on deep neural network(DNN).The purpose is to obtain better BER performance and reduce the complexity of detector effectively.The DNN detector uses a deep neural network with multiple full connection layers to recover the bit information in the DM-OFDM-IM system.In order to improve the performance of the detector,before the received signal is sent to the neural network,the received signal of the dual-mode orthogonal frequency division multiplexing index modulation system is preprocessed,and then the preprocessed data set is used to train the detector.Finally,the trained model is used for real-time detection of the dual-mode orthogonal frequency division multiplexing index modulation system.The complexity of on-line detection of DNN detector is obviously lower than that of ML detector,which can effectively improve the detection efficiency and obtain the bit error rate similar to that of ML detector.The simulation results show that the DNN detector can achieve a bit error rate performance similar to ml with very low running time.On this basis,we also consider the combination of channel estimation and signal detection.Existing detectors,including the DNN detection method designed in this paper,need to rely on channel state information(CSI).In practical applications,CSI is often obtained by inserting pilot into the transmission sequence and channel estimation.However,due to the existence of estimation error,the robustness of the ideal signal detection model in practical application is poor.In order to solve this practical problem,this paper combines channel estimation with deep neural network,and considers the signal detection method of OFDM-IM under the condition of channel estimation error.Based on the traditional channel estimation method,this paper uses a channel estimation method based on deep learning.Compared with the traditional least square(LS)channel estimation algorithm,the channel estimation method based on deep learning can get better mean square error(MSE).