Search

A channel estimation method for intelligence reflecting surface (IRS) assisted mine communication system based on self supervised learning (SSL) is proposed to address the problems of multipath fading, non line of sight communication, and difficulty in obtaining true labels caused by complex mine environments. The method builds an underground communication system model based on the Nakagami-g fading channel model and IRS signal transmission model, and solves the problems of multipath fading and non line of sight communication through IRS technology. Preliminary channel estimation is performed using the least squares (LS) algorithm, and then the channel estimation results are optimized using octave convolution (OCT) neural network under the SSL framework. OCT directly processes both high-frequency and low-frequency components, capturing both the rough features and subtle differences of the channel, providing comprehensive channel information, and thus more accurately estimating the channel state. The SSL algorithm uses received signals and their noisy versions as training data to improve the precision and efficiency of IRS assisted channel estimation through the intrinsic structure of unlabeled data, thereby reducing reliance on manual labeling. The simulation results show the following points. ① Introducing IRS technology can effectively reduce channel estimation errors. ② The loss value of OCT neural network is significantly lower than that of CNN, and the data fitting effect is better. OCT neural network has high computational efficiency and can improve the overall performance of channel estimation in communication systems. In environments with limited computing resources, OCT neural networks can maintain low parameter and memory usage. ③ The SSL algorithm can maintain a low normalized mean square error under all signal-to-noise ratio conditions, verifying its efficiency and robustness in channel estimation. ④ The channel estimation method for IRS assisted mine communication system based on SSL has good scalability and robustness in large-scale networks.

At present, most of the research on mine wireless signal detection only considers the ideal additive Gaussian white noise channel and Rayleigh fading channel. The signal detection has high bit error rate and complex network structure. In order to solve the above problems, a mine wireless signal detection method based on dual path network (DPN) is proposed. The method uses dual path network receiver (DPNR) to optimize the overall performance of the orthogonal frequency division multiplexing (OFDM) receiver and solve the problem of error accumulation in conventional receivers. Firstly, the residual (Res) block's shortcut is used to perform a convolution of shallow features, and the feature map after one convolution is added to the feature map after multiple convolutions. Secondly, the shallow layer of the Dense block is reused. The convolution calculation of the Dense block is performed to obtain the feature map after the convolution calculation. Finally, the feature maps of the two are fused into a new feature map, which extracts more features at the expense of less complexity, thereby improving detection performance. The experimental results show the following points. ① In OFDM systems, the bit error rate of DPNR is lower than that of conventional receivers. When the signal-to-noise ratio is 13, the bit error rate is zero. When the signal-to-noise ratio is greater than 7, the error rate of DPNR is reduced by more than one order of magnitude compared to conventional receivers in mine environments. When the signal-to-noise ratio is greater than 11, the bit error rate of DPNR is more than one order of magnitude lower than that of conventional receivers under additive Gaussian white noise. ② In the multi-carrier/offset orthogonal amplitude modulation of communication system filter banks, the error rate of DPNR is reduced by more than two orders of magnitude compared to conventional receivers, indicating its good robustness. ③ As the signal-to-noise ratio increases, the bit error rate of DPNR and residual neural network (ResNet) receivers is lower than that of densely connected convolutional networks (DenseNet) receivers. The bit error rate of DPNR is lower in the final stage when the signal-to-noise ratio is greater than 13. ④ At higher signal-to-noise ratios, the bit error rate of DPNR is much lower than that of deep receivers. When the signal-to-noise ratio is greater than 8, the bit error rate of DPNR is reduced by more than one order of magnitude compared to deep receivers.

Due to space limitations, multi frequency multiple-input multiple-output (MIMO) antennas have strong coupling problems caused by small unit spacing. In order to solve the above problems, a mine high isolation tri-band MIMO antenna has been designed. By loading two L-shaped branches at both ends of a rectangular branch to form a trident monopole antenna, the antenna has tri-band features. Two trident monopole antenna units are placed symmetrically. A T-shaped branch is loaded on the metal floor between the two units. The opposite current generated by parasitic branches is used to offset the coupling current without branches. Two symmetrical rectangular slots are etched to suppress the mutual coupling caused by surface waves on the floor by changing the current distribution on the floor. The high isolation of the antenna throughout the entire band is achieved. The simulation results show that the antenna operates in frequency bands of 1.85-2.70, 3.24-3.99, 4.65-5.80 GHz, can effectively covering coal mines' underground WiMAX/WiFi/4G/5G NR operating band. The isolation of the antenna in three bands is greater than 20, 22, 22 dB, respectively. It is 11, 9, 10 dB higher than the isolation of the antenna before decoupling; The envelope correlation coefficient is less than 0.2, indicating good diversity performance. The antenna has stable gain variation within the operating band and good omnidirectional radiation features. This antenna has the advantages of simple and compact structure, easy processing, and low profile. It has a wide range of application scenarios in wireless communication in coal mines.

With the application of 5G technology in the coal mine, the signal interference between multiple systems becomes more and more intensive. This seriously affects the quality of data, voice and image communication. The coexistence of multiple systems in coal mine becomes increasingly prominent. To solve this problem, a multi-band microstrip antenna for mine is designed, which can work in WiMAX/WiFi/4G/5G NR band at the same time. Based on a planar monopole antenna, the antenna can work in multiple frequency bands in the mode of loading two L-shaped branches and loading an inverted L-shaped branch on the floor. The simulation results show that the antenna's middle, right and left branches produce 2.4, 3.5 GHz and 4.8 GHz resonance points respectively. The inverted L-shaped branch loaded on the floor provides 1.9 GHz resonance point. The antenna can work in three frequency bands, which are 1.88-2.73, 3.26-3.79 GHz and 4.7-5.9 GHz respectively. The antenna can effectively cover all the operating frequency bands of WiMAX/WiFi/4G/5G NR in coal mines. According to the antenna's peak gain and normalized pattern, the antenna has good gain performance and overall radiation performance in the required operating frequency band.

To address the common issues in traditional convolutional neural network (CNN)-based spectrum sensing algorithms in cognitive radio networks (CRNs), including inadequate signal feature representation, inefficient utilization of feature map information, and limited feature extraction capabilities due to shallow network structures, this paper proposes a spectrum sensing algorithm based on a short-time Fourier transform (STFT) and residual attention dense network (RADN). Specifically, the RADN model improves the basic residual block and introduces the convolutional block attention module (CBAM), combining residual connections and dense connections to form a powerful deep feature extraction structure known as residual in dense (RID). This significantly enhances the network's feature extraction capabilities. By performing STFT on the received signals and normalizing them, the signals are converted into time–frequency spectrograms as network inputs, better capturing signal features. The RADN is trained to extract abstract features from the time–frequency images, and the trained RADN serves as the final classifier for spectrum sensing. Experimental results demonstrate that the STFT-RADN spectrum sensing method significantly improves performance under low signal-to-noise ratio (SNR) conditions compared to traditional deep-learning-based methods. This method not only adapts to various modulation schemes but also exhibits high detection probability and strong robustness. [ABSTRACT FROM AUTHOR]

The thinned and sparse beamforming for semicircular FDAs were investigated, where the excitation amplitudes were also considered in thinned semicircular FDAs, and only the elements' positions were incorporated into the sparse semicircular FDA. Firstly, the transmit–receive model was introduced to handle the inherent time-varying issue of FDA, followed by the thinned and sparse implementations successively. Note that three types of non-linearly varying frequency offsets (FO), i.e., log-FO, sin-FO, and tanh-FO, were adopted during the investigations. Under the same assumption that 50% of the elements should be saved, the sidelobe levels (SLLs) of the thinned semicircular -FDAs were reduced by 5.8 dB, 4.4 dB, and 4.4 dB, and the widths of the mainlobes were all widened by 3° in their angle dimension. Compared with the thinned semicircular FDAs, the phenomenon of mainlobe widening was alleviated in the sparse semicircular FDAs where the SLLs were reduced by 2.2 dB, 3.7 dB and 3.5 dB, and the mainlobes' widths in the angle dimension were widened by 1°, 0° and 1°, respectively. It should be highlighted that the sparse semicircular FDA with sin-FO did not broaden the mainlobe in the angle dimension. Therefore, it can be concluded that a sparse semicircular FDA is superior over a thinned semicircular FDA, since it can reduce the same cost with a higher array resolution. [ABSTRACT FROM AUTHOR]

In the process of using optical fiber frequency shift value monitored by optical fiber sensor to characterize the strata behaviors law, the data collected by the sensor is missing, and the strata behaviors law can not be accurately predicted. In order to solve this problem, taking Qianqiu Coal Mine as the engineering background, under the premise of partial data loss of the lower half of the optical fiber, two prediction models, GRU (Gated Recurrent Unit) and LSTM (Long Short-Term Memory), are introduced to compare and predict the missing optical fiber frequency shift value. The convergence speed of the GRU model is better than that of the LSTM model, which shows that the missing value processing method based on the GRU model is better. The original and complete optical fiber frequency shift value is converted into the average optical fiber frequency shift change which can characterize the strata behaviors position, and the XGBoost (eXtreme Gradient Boosting) model and the BP neural network model are introduced for comparative prediction. The XGBoost model can predict all the 'peak' positions in the test set accurately. However, the BP neural network model can only predict two 'peak' positions, which shows that the prediction effect of the XGBoost model is better than that of the BP neural network model. The predicted optical fiber frequency shift missing value is replaced to the missing position to form 'complete' optical fiber frequency shift value data. The data is converted into the average optical fiber frequency shift change and then the XGBoost model is used for prediction. The results show that both the LSTM model and the GRU model can predict the data of the lower half of the optical fiber accurately, and the GRU model has higher accuracy than the LSTM model. The XGBoost model can predict the periodic pressure in the test set accurately. After the missing data predicted by the GRU model is integrated into the missing position, the XGBoost model can still predict the strata behaviors effectively.

In order to solve the problems of the complex modeling process, high computational complexity, and low prediction accuracy of the existing field strength prediction models, a mine roadway field strength prediction model based on improved convolutional neural network(CNN) is proposed. By analyzing the influence factors of electromagnetic wave transmission in large-scale fading channels in mines, using antenna operating frequency, roadway cross-sectional dimensions, roadway wall roughness, roadway wall inclination, roadway wall relative permittivity and transceiver distance as model inputs, using the electromagnetic wave propagation path loss as model outputs, the model is able to predict the changes of the roadway field strength. The improved CNN adds batch normalization layer after each convolutional layer to replace the original pooling layer so as to avoid the loss of data characteristics due to down-sampling of the pooling layer, to keep the output of each convolutional layer similarly distributed, to improve the network generalization capacity and to speed up the network convergence. The simulation results show that compared with the field strength prediction models based on CNN, BP neural network and support vector machine, the model has high consistency between the predicted value and the actual value, has stronger robustness, and improves the accuracy of mine roadway field strength prediction effectively.

When the traditional adaptive OFDM (Orthogonal Frequency Division Multiplexing) modulation technology based on fixed signal-to-noise ratio threshold is applied to complex mine channels, the feedback channel state cannot completely match the actual channel state, resulting in high bit error rate and low throughput. In order to solve the above problem, a Q-learning algorithm based mine adaptive OFDM modulation method is proposed and applied to the mine adaptive OFDM modulation system. The system is composed of a transmitter, a mine wireless channel and a receiver. The transmitter is a sensor-equipped mine cart, which can move freely in a narrow roadway. The transmitter uses Q-learning algorithm to update the state-action value function continuously in the dynamic interaction with the mine wireless channel. And the transmitter uses a greedy strategy to select the modulation method according to the updated state-action value function to approximate the optimal adaptive modulation strategy so as to reduce the system BER and improve the communication throughput. The performance of two mine adaptive OFDM modulation systems based on SARSA algorithm and fixed signal-to-noise ratio threshold is compared. The result shows that the average BER of the adaptive OFDM modulation system based on Q-learning algorithm are 1.1×10-3，2.1×10-3, and the total throughput are 3,115 bit, 2,719 bit respectively in the uniform and non-uniform movement states of mine cart. These results are better than the adaptive OFDM modulation system based on SARSA algorithm and fixed signal-to-noise ratio threshold. And the convergence speed of Q-learning algorithm in the system is better than that of SARSA algorithm.

In view of complex and heterogeneous wireless environment of mine, an underground signal recognition method based on higher-order cumulants and DNN model was proposed to realize automatic modulation recognition of underground digital signals of BPSK, QPSK, 8PSK, 2FSK, 4FSK, 8FSK, 32QAM, 64QAM, OFDM. Theoretical values of high-order cumulants of the 9 kinds of digital signals were obtained by analysis, and the signal identification was improved by Fourier transform. The influence of underground small-scale fading channels on high-order cumulants were analyzed, high-order cumulants calculation expression of the signal after passing through the underground channel was derived, and signal recognition was realized using characteristic parameters constructed according high-order cumulants to train DNN model. The simulation analysis results show that the method has excellent modulation recognition performance in mine Nakagami-m fading channel, average correct recognition rate is more than 89.2% when the signal-to-noise ratio is -5 dB, and the average correct recognition rate is 100% when the signal-to-noise ratio is 5 dB or more. The method provides a new idea for signal recognition and detection in special and complex environments.

In order to study propagation loss law of electric wave and improve prediction accuracy of field intensity coverage in mine tunnel, least square support vector machine (LS-SVM) method optimized by genetic algorithm was used to forecast underground field intensity in mine tunnel. Firstly, simulated field intensity data was generated by computer software and divided into training set and testing set. Then the LS-SVM machine method was used to study training set, genetic algorithm was used to optimize parameters of LS-SVM, and testing set was used to verify performance of the method. Finally the LS-SVM method optimized by genetic algorithm was used to forecast underground field intensity in mine tunnel. The simulation and experiment results prove that the LS-SVM optimized by genetic algorithm can effectively improve prediction accuracy of field intensity in mine tunnel, and achieve good prediction effect.

For problem of low accuracy of current prediction of field intensity in mine tunnel, a prediction model based on the least squares support vector machine method was proposed to predict the field intensity in mine tunnel by taking measured data of a tunnel as training sample. Influence of training set construction and parameters selection on prediction effect were analyzed in details. The simulation results show that the LS-SVM prediction model has higher prediction accuracy than dual-slope model and logarithmic correction model.

In view of problem that fault of underground power supply equipment leads great hidden danger to coal mine safety production, an intelligent fault detection and analysis system of coal mine power based on TD-SCDMA technology was proposed. The system collects each segment bus voltage, each branch current, power, state of breaker and fault information in real-time through detection and collection system of power supply, then transmits running state, running parameters, fault information of coal mine power supply and distribution system to intelligent power fault analysis system through TD-SCDMA/industrial Ethernet support system, so as to complete information processing and analysis and fault diagnosis and processing of coal mine power supply and distribution system. The application result shows that the system improves reliability of power supply, and provides guarantee for coal mine safety production.

For development demands and characteristics of informatization of coal industry, the paper analyzed and compared private network solution and rent network solution mainly from adaptability, disaster tolerance, scheduling functions, security, voice quality and other aspects, and got conclusions that private network solution had significant advantages. It analyzed advantages of TD-SCDMA network solution specifically, and indicated that private network solution can solve current communication problems of coal industry more effectively and meet the needs of coal mine safe production and communication development.

The paper gave construction target of intelligent management platform of coal mine safety production, introduced general design and realization of the platform. It shows that the platform realizes fine management of coal mine real timly and dynamically based on the practical application example of the platform in a coal mine.