[Objective]Power Line Carrier(PLC)communication adopts the frame burst transmission mode.Due to the carrier frequency offset between transceivers,various noise and time-varying characteristics of PLC channel and the system has no ded-icated reference signal.The traditional channel estimation has no tracking and prediction ability for the channel,which leads to the deterioration of the PLC system performance.[Methods]Aiming at the existing problems,this paper proposes a Denoising Long Short Term Memory(DnLSTM)neural network based on Long Short Term Memory(LSTM)neural network and De-noising Convolutional Neural Network(DnCNN),which is used for PLC channel estimation.First,offline training is per-formed on DnLSTM and the parameters are saved after training.Then the trained parameters are deployed in PLC system.Af-ter loading parameters,online prediction is performed to obtain the predicted PLC system channel estimation.In the simulation of PLC system,this paper uses Least Squares(LS)algorithm,Minimum Mean Square Error(MMSE)algorithm and DnL-STM to estimate the channel response,and gives the simulation results under the conditions of Additive White Gaussian Noise(AWGN),combined noise,impulsive noise and colored noise.Meanwhile,simulations for different number of preamble sym-bols for channel estimation are performed.[Results]The results show that there is a relationship between the accuracy of DnL-STM channel estimation and the number of preamble symbols.Using four preamble symbols for channel estimation,its esti-mation accuracy is better than LS and close to MMSE algorithm.DnLSTM has a good ability to resist carrier frequency offset and channel time-varying.When the number of preamble symbols for channel estimation increases,PLC system performance with low Singnal to Noise Ratio(SNR)gets better and PLC system performance is similar with high SNR.[Conclusion]Ac-cording to simulation results above,it can be concluded that DnLSTM,which is based on DnCNN and LSTM,can predict PLC system channel response with frequency offset very well and it can track the varying PLC system channel response in real time.
维普
万方数据