Morse Telegram Intelligent Transceiver System

Today,with various new communication technologies emerging in an endless stream,Morse Communication,which has been born for more than 180 years,still plays an important role in some special fields.In recent years,some researchers have implemented FPGA-based Morse wireless communication systems,and some scholars have extended and applied algorithms such as machine learning and deep learning to the decoding of Morse telegrams.However,in the real environment,the sender usually uses a higher code rate to send,and it is accompanied by time jitter,etc.The signal received by the receiver is usually mixed with a variety of noises,which interferes with the sending and receiving of Morse telegrams.As a result,Morse communication still relies on professional telegraphers.In order to solve this situation,this thesis designs and implements a Web-based Morse telegram automatic sending and receiving system.The main tasks are as follows:This thesis uses the Kmeans algorithm and the CRNN model to decode Morse audio,and tests the performance of the two implementations.The Kmeans algorithm distinguishes high and low levels by thresholding,then clusters the duration of the two states of morse dot and dash,calculates the weighted average to get the frame length of the morse signal,and thus divides the five states of the morse code to achieve decoding.The CRNN decoding model needs to be trained using the dataset generated by the authors,with a convolutional layer for feature extraction,a recurrent layer for feature prediction,and a transcription layer for decoding.The final test shows that when the frame length is fixed and the signal-to-noise ratio is higher than-2dB,the decoding accuracy rate of the Kmeans algorithm is higher than 98%,and the time consumption is very short.The CRNN model takes a long time to decode,but the correct rate is still higher than 98%when decoding the audio whose frame length is not fixed and the signal-to-noise ratio is higher than-6dB.This thesis implements real-time decoding of Morse audio streams by means of a front-end recording control module and a back-end sliding window module.The front-end controls the browser to send the recording data to the back-end,and the back-end converts the recording data into a time-frequency graph and caches it in the sliding window,and then sends the time-frequency graph in the sliding window to the CRNN model,and extracts the effective parts of the prediction sequence for splicing into the total predicted sequence,and transcribe the total predicted sequence to obtain real-time decoding results.In addition,the decoding of Morse audio streams in real time is finally achieved according to the header and tailer.Based on the file decoding module,the real-time receiving and decoding module,and the modules of encoding and decoding,sending,and personal center,a Web-based Morse telegram intelligent copying and sending system is realized.After testing,the system has realized the automatic sending and receiving of Morse telegrams.