Wireless channel state and scene recognition method and system suitable for millimeter wave communication

Abstract

The present invention discloses a method and system for identifying wireless channel states and scenes, which belongs to the field of wireless communication technology, and in particular, is applicable to millimeter wave communications. The identification method establishes a multi-layer deep neural network; calculates channel statistical parameters and multipath cluster parameters, wherein the channel statistical parameters include signal receiving power, dynamic range, delay spread and Rice factor; the multipath cluster parameters include average cluster interval, cluster attenuation factor, first cluster multipath attenuation factor and average multipath attenuation factor; inputs the channel statistical parameters and multipath cluster parameters into the network, outputs the channel state and scene information to complete the training; uses the trained neural network to perform real-time identification of the channel state and scene. The identification method disclosed by the present invention overcomes the problem of low accuracy of traditional identification methods, does not require complex input parameters, effectively improves the real-time and accuracy of identification, and is applicable to fast time-varying applications such as high-speed railways and drones in millimeter wave communications, ensuring the safe and normal operation of millimeter wave communications.

Description

Technical Field

[0001] The present invention belongs to the technical field of wireless communications, and in particular relates to a method and system for identifying wireless channel states and scenes applicable to millimeter wave communications. Background Art

[0002] The construction of the fifth-generation mobile communication system (5G) is underway worldwide. It supports three major application scenarios: enhanced mobile broadband, massive device communications, and highly reliable and low-latency communications. It also provides solutions for vertical industry applications such as smart healthcare, smart factories, and intelligent transportation. The electromagnetic wave frequencies used in 5G system signal transmission primarily range from 2.6 GHz to 3.5 GHz, which are higher than those of 4G and 3G systems. Millimeter wave frequencies of 26 GHz and 39 GHz have been listed as candidate frequency bands for millimeter wave communications in my country.

[0003] Electromagnetic wave attenuation increases with frequency, and wireless signal attenuation in 5G systems is significantly higher than in 4G systems. Furthermore, 5G wireless channels, especially millimeter-wave channels, exhibit rapid time-varying characteristics, leading to significant differences in wireless signal transmission characteristics in different scenarios. Furthermore, because the diffraction capacity of electromagnetic waves decreases with frequency, the transmission characteristics of 5G wireless signals vary significantly under different channel conditions (such as line-of-sight and non-line-of-sight). With the continuous advancement of intelligent and information-based millimeter-wave communications, high-speed mobile services such as high-speed rail and drones will be widely used in millimeter-wave communication systems. my country's wireless communication needs are widespread and diverse in various scenarios. To ensure communication quality between base stations and high-speed rail, robots, and drones, differentiated channel models, physical layer algorithms, and network structures must be designed for different channel conditions and scenarios. Furthermore, accurate identification of the terminal device's state and scenario is required to match the optimal network configuration. Existing identification schemes mainly work on channel status or channel scenarios, and there is no solution that can simultaneously identify channel status and scenarios; and existing schemes all consider traditional channel statistical characteristics, such as received power, delay spread, Ricean factor, etc., but the channel statistical characteristics may be similar in similar scenarios, resulting in low accuracy of existing identification schemes, which in turn affects communication quality. Therefore, how to simultaneously identify channel status and scenarios, reduce the complexity of input parameters, and effectively improve the real-time and accuracy of identification has become a technical problem that needs to be solved urgently. Summary of the Invention

[0004] The present invention aims to provide a method for wireless channel status and scene recognition applicable to millimeter wave communication, characterized in that it includes the following steps:

[0005] Establish a multi-layer deep neural network consisting of 2 input layers, 2 output layers and 6 hidden layers;

[0006] Calculating channel statistical parameters, including signal received power, dynamic range, delay spread, and Ricean factor;

[0007] Use a multipath clustering algorithm to perform clustering and calculate multipath cluster parameters, wherein the multipath cluster parameters include: average cluster interval, cluster attenuation factor, first cluster multipath attenuation factor and average multipath attenuation factor;

[0008] Use channel statistics and multipath cluster parameters as input, and channel status and scene information as output to train a multi-layer deep neural network;

[0009] Using a trained multi-layer deep neural network, channel statistical parameters and multipath cluster parameters are input to perform real-time recognition of channel status and scene information.

[0010] The multi-layer deep neural network includes: a first input layer of N1 neurons, a first fully connected layer of 64 neurons, a first Dropout layer of 64 neurons, a second fully connected layer of 128 neurons, a scene output layer of 3 neurons, a second input layer of N2 neurons, a third fully connected layer of 64 neurons, a second Dropout layer of 64 neurons, a fourth fully connected layer of 32 neurons, and a channel state output layer of 2 neurons, wherein N1 and N2 are the number of input parameters of the first input layer and the second input layer respectively, and the number of neurons is set according to the number of input parameters; data passes through the first input layer, the first fully connected layer, the first Dropout layer, the second fully connected layer to the scene output layer in sequence, and the third fully connected layer merges the data of the scene output layer and the second input layer and then continues to pass through the second Dropout layer and the fourth fully connected layer in sequence to reach the channel state output layer.

[0011] The specific steps of calculating the channel statistical parameters are:

[0012] Calculate the delay power spectrum. Assume that the channel impulse response matrix at time m is h(m,τ). Then the delay power spectrum at time m is:

[0013] P(m,τ)=|h(m,τ)| 2 Where, τ is the time delay;

[0014] Calculate the noise power. The delay power spectrum includes both signal and noise. The noise power is obtained by averaging the last portion of the delay power spectrum that does not contain the signal. It is specifically defined as:

[0015]

[0016] Where T n 、T end The length selected for calculating the noise power and the total length of the delayed power spectrum;

[0017] Calculate the dynamic range and use the difference between the maximum power and noise in the delay power spectrum to calculate the dynamic range of the signal:

[0018]

[0019] Calculate the delay sample set of the non-noise signal and set the denoising threshold according to the dynamic range of the signal. Delay points greater than the denoising threshold will be extracted and regarded as signals. The delay sample set of the non-noise signal is defined as:

[0020]

[0021] where ξ τ,m is the maximum delay, and the delay points exceeding the maximum delay are regarded as noise. ρ is the denoising factor that varies with the environment. The signal received power is calculated by summing up all signal powers to obtain the signal received power at time m:

[0022]

[0023] where τ l is the signal delay point, is the set τ l Elements in

[0024] Calculate the delay spread to quantify the channel dispersion in the time domain. The calculation formula for delay spread is:

[0025]

[0026] Calculate the Rice factor, which is the ratio of the component with the largest power in the channel delay power spectrum to the sum of the powers of the other components, and is calculated by the following formula:

[0027]

[0028] The multipath clustering algorithm is a bubble clustering algorithm.

[0029] The specific steps of calculating the multipath cluster parameters are:

[0030] Introducing the SV channel model:

[0031]

[0032] where α rs and θ rs are the amplitude and phase of the sth multipath in the rth cluster, R m is the number of channel multipath clusters at time m, S r is the number of multipaths in the rth cluster, τ r,1 and τ r,sare the delays of the first and sth multipaths in the rth cluster that can be directly obtained from the clustering results, and j is an imaginary unit, that is,

[0033] Calculate the average cluster spacing to describe the abundance of scatterers in a scene. The more scatterers there are in the environment, the smaller the average cluster spacing. The calculation method of the average cluster spacing is as follows:

[0034]

[0035] Define the sth path in the rth cluster. Based on the assumption of the SV channel model, the received power of multipath decays exponentially with delay. The sth path in the rth cluster is modeled as:

[0036]

[0037] where 1 / Γ is the cluster attenuation factor, 1 / γ r is the multipath attenuation factor of the rth cluster.

[0038] The establishment of a multi-layer deep neural network including 2 input layers, 2 output layers and 6 hidden layers also includes:

[0039] Define the dataset for training:

[0040] D m ={(I 1,m ,I 2,m )|(O 1,m ,O 2,m )},m=1,2,...,M

[0041] Where M is the total number of data sets, I1 and I2 are the two input vectors of the two input layers of the multi-layer deep neural network, with lengths of IL1 and IL2 respectively; O1 and O2 are the two output vectors of the two output layers, with lengths of OL1 and OL2 respectively. Each element in the output vector is the probability of identifying a certain state or scene, and the state or scene of the recognition result is ultimately determined by the highest probability.

[0042] The rectified linear unit function is used in four fully connected layers and two dropout layers to provide positive input in the early stages of training to accelerate the training process;

[0043] The softmax activation function is used in the scene output layer and the channel state output layer to express the output value as the probability of identifying a certain class. The specific formula is expressed as follows:

[0044]

[0045] where z yis the output of the y-th neuron, and Y is the total number of output neurons;

[0046] The classification cross entropy function is set as the loss function, and the performance is evaluated by comparing the output value and the expected value. The loss function is defined as:

[0047]

[0048] Among them, m, x, y are data, scene and state category indexes respectively, p x,m and p y,m The probability of identifying it as the xth scene and the yth channel state, t x,m and t y,m The actual values ​​of the x-th scene and y-th state respectively;

[0049] The RMS propagation optimizer is used to adaptively update the learning rate to minimize the iterative training error.

[0050] The present invention also provides an identification system using the wireless channel state and scene identification method applicable to millimeter wave communication, characterized in that it includes:

[0051] Channel statistical parameter calculation module: extracts channel statistical parameters from channel impulse response data and outputs signal power, dynamic range, delay spread and Rice factor;

[0052] Multipath clustering module: uses a clustering algorithm to identify multipaths from the channel impulse response and divides the multipaths into different multipath clusters;

[0053] Multipath cluster parameter calculation module: extracts multipath cluster parameters from the clustering results and outputs average cluster interval, cluster attenuation factor, first cluster multipath attenuation factor and average multipath attenuation factor;

[0054] Multi-layer deep neural network module: uses channel statistical parameters, multipath cluster parameters, channel scenario and status information to train the multi-layer deep neural network. By inputting channel parameters, the trained multi-layer deep neural network can identify the channel scenario and status in real time;

[0055] The present invention also provides a computer-readable storage medium storing one or more programs, characterized in that: the one or more programs include instructions, which, when executed by a computing device, enable the computing device to execute the wireless channel state and scene recognition method suitable for millimeter wave communication.

[0056] The present invention further provides a computing device, characterized by comprising:

[0057] One or more processors, one or more memories, and one or more programs, wherein the one or more programs are stored in the one or more memories and are configured to be executed by the one or more processors, and the one or more programs include instructions for executing the wireless channel status and scene recognition method applicable to millimeter wave communication.

[0058] The beneficial effects of the present invention are:

[0059] This paper considers using channel statistics and multipath cluster parameters to train a multi-layer deep neural network to simultaneously identify channel states and scenarios in real time. This method can quickly and accurately identify channel states and scenarios, which is crucial for ensuring the robustness and reliability of communication systems and supporting the stable operation of millimeter-wave communications.

[0060] Unlike the existing technology, the present invention discloses a wireless channel state and scene recognition method suitable for millimeter wave communication, which can simultaneously identify the channel state and scene; selects a bubble clustering algorithm to avoid the computational complexity caused by the parameter extraction algorithm; adopts a simple network structure; adds a dropout layer to the neural network structure design to reduce overfitting and improve the robustness and reliability of recognition; uses the ReLu activation function in the fully connected layer and the dropout layer to accelerate the training process.

[0061] Unlike traditional methods that only consider channel statistics for scene and state recognition, this invention uses multipath clustering parameters for recognition. Wireless signals reflect and scatter off objects in space, forming multipath clusters. Multipath clustering parameters are directly related to the characteristics of objects in the environment. Therefore, using these parameters in the recognition process can effectively improve recognition accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] Figure 1 The present invention discloses a flow chart of a wireless channel state and scene recognition method applicable to millimeter wave communication;

[0063] Figure 2 This is a framework diagram of the wireless channel status and scene recognition system;

[0064] Figure 3 A diagram of the multi-layer deep neural network structure for wireless channel status and scene recognition;

[0065] Figure 4 (a) is a real picture of the substation channel measurement environment;

[0066] Figure 4 (b) is a real picture of the channel measurement environment in the waiting hall of the high-speed railway station;

[0067] Figure 4 (c) is a real picture of the indoor office channel measurement environment;

[0068] Figure 5 (a) is a distribution diagram of the recognition accuracy of the wireless channel state and scene recognition method applicable to millimeter wave communication disclosed in the present invention;

[0069] Figure 5 (b) is the recognition accuracy distribution diagram based on the back propagation neural network recognition method. DETAILED DESCRIPTION

[0070] The present invention provides a method for identifying wireless channel status and scenarios applicable to millimeter wave communications, which is further described in detail below with reference to the accompanying drawings.

[0071] Example 1:

[0072] like Figure 1 The embodiment of the present invention discloses a method for wireless channel status and scene recognition applicable to millimeter wave communication, including:

[0073] Establish a multi-layer deep neural network consisting of 2 input layers, 2 output layers and 6 hidden layers;

[0074] Calculating channel statistical parameters, including signal received power, dynamic range, delay spread, and Ricean factor;

[0075] Use a multipath clustering algorithm to perform clustering and calculate multipath cluster parameters, wherein the multipath cluster parameters include: average cluster interval, cluster attenuation factor, first cluster multipath attenuation factor and average multipath attenuation factor;

[0076] Use channel statistics and multipath cluster parameters as input, and channel status and scene information as output to train a multi-layer deep neural network;

[0077] Using a trained multi-layer deep neural network, channel statistical parameters and multipath cluster parameters are input to perform real-time recognition of channel status and scene information.

[0078] The specific implementation process is as follows:

[0079] First, if Figure 3 As shown in the figure, a multi-layer deep neural network including 2 input layers, 2 output layers and 6 hidden layers is established, wherein the 2 input layers input channel statistical parameters and multipath cluster parameters; the scene output layer outputs the recognized scene information, and the channel state output layer outputs the recognized channel state information. The scene output layer is connected to the second input layer, and the scene recognition result of the scene output layer is used as the input for subsequent channel state recognition.

[0080] The multi-layer deep neural network includes a first input layer of N1 neurons, a first fully connected layer of 64 neurons, a first Dropout layer of 64 neurons, a second fully connected layer of 128 neurons, a scene output layer of 3 neurons, a second input layer of N2 neurons, a third fully connected layer of 64 neurons, a second Dropout layer of 64 neurons, a fourth fully connected layer of 32 neurons, and a channel state output layer of 2 neurons, wherein N1 and N2 are the number of input parameters of the first input layer and the second input layer respectively, and the number of neurons is set according to the number of input parameters; data passes through the first input layer, the first fully connected layer, the first Dropout layer, the second fully connected layer to the scene output layer in sequence, and the third fully connected layer merges the data of the scene output layer and the second input layer and then continues to pass through the second Dropout layer and the fourth fully connected layer in sequence to reach the channel state output layer.

[0081] Taking into account the different correlations between channel statistical parameters and multipath cluster parameters and channel scenarios and states, it is necessary to determine the parameter selection of the two input layers according to actual conditions, and set the number of neurons according to the number of input parameters, that is, to determine the values ​​of N1 and N2.

[0082] Secondly, calculate the channel statistical parameters, which include: signal received power, dynamic range, delay spread and Rice factor. The specific steps of calculating the channel statistical parameters are:

[0083] Calculate the delay power spectrum. Assume that the channel impulse response matrix at time m is h(m,τ). Then the delay power spectrum at time m is:

[0084] P(m,τ)=|h(m,τ)| 2 Where, τ is the time delay;

[0085] Calculate the noise power. The delay power spectrum includes both signal and noise. The noise power is obtained by averaging the last portion of the delay power spectrum that does not contain the signal. It is specifically defined as:

[0086]

[0087] Where T n ,T end The length selected for calculating the noise power and the total length of the delayed power spectrum;

[0088] Calculate the dynamic range and use the difference between the maximum power and noise in the delay power spectrum to calculate the dynamic range of the signal:

[0089]

[0090] Calculate the signal and set the denoising threshold according to the dynamic range of the signal. The delay points greater than the denoising threshold will be extracted and regarded as signals.

[0091]

[0092] where ξ τ,m is the maximum delay, and the delay points exceeding the maximum delay are regarded as noise. ρ is the denoising factor that varies with the environment. The signal received power is calculated by summing up all signal powers to obtain the signal received power at time m:

[0093]

[0094] where τ l is the signal delay point, is the set τ l Elements in .

[0095] Calculate the delay spread to quantify the channel dispersion in the time domain. The calculation formula for delay spread is:

[0096]

[0097] Calculate the Rice factor, which is the ratio of the component with the largest power in the channel delay power spectrum to the sum of the powers of the other components, and is calculated by the following formula:

[0098]

[0099] Next, clustering is performed using a multipath clustering algorithm to calculate multipath cluster parameters, which include: average cluster interval, cluster attenuation factor, first cluster multipath attenuation factor, and average multipath attenuation factor.

[0100] In this embodiment, a clustering algorithm is used to divide the multipath in the wireless channel into multiple multipath clusters. The wireless channel state and scenario recognition method for millimeter-wave communications disclosed in this invention does not specifically limit the multipath clustering algorithm. Common algorithms such as K-means and multipath component distance can meet the requirements. However, the bubble clustering algorithm is recommended. It only uses the channel power delay profile to perform clustering, resulting in a simple process and greater ease of real-time state and scenario recognition in millimeter-wave communications.

[0101] In this embodiment, the specific steps of calculating the multipath cluster parameters are:

[0102] The SV channel model is introduced as a theoretical basis, and the SV channel model is defined as:

[0103]

[0104] where α rs and θ rs are the amplitude and phase of the sth multipath in the rth cluster, R m is the number of channel multipath clusters at time m, Sr is the number of multipaths in the rth cluster, τ r,1 and τ r,s are the delays of the first and sth multipaths in the rth cluster that can be directly obtained from the clustering results, and j is an imaginary unit, that is,

[0105] The average cluster spacing is used to describe the abundance of scatterers in a scene. The more scatterers there are in the environment, the smaller the average cluster spacing. The average cluster spacing is calculated as follows:

[0106]

[0107] Define the sth path in the rth cluster. Based on the assumption of the SV channel model, the received power of multipath decays exponentially with delay. The sth path in the rth cluster is modeled as:

[0108]

[0109] where 1 / Γ is the cluster attenuation factor, 1 / γ r is the multipath attenuation factor of the rth cluster. The cluster and multipath attenuation factor describe the speed of power attenuation and are used to quantify the abundance and intensity of scatterers in the scene. According to the SV channel model, a cluster attenuation factor and R can be fitted from the relationship between multipath power and delay in the clustering results of each delay power spectrum. m Multipath attenuation factors, for each channel sample, the number of multipath clusters varies, that is, R m The number will change, so only the first cluster multipath attenuation factor 1 / γ1 (that is, when r = 1) and the average multipath attenuation factor are considered in the identification. in R m The average value of the multipath attenuation factors is

[0110]

[0111] Again, channel statistical parameters and multipath cluster parameters are used as input, and channel status and scene information are used as output to train a multi-layer deep neural network.

[0112] Define the dataset for training:

[0113] D m ={(I 1,m ,I 2,m )|(O 1,m ,O 2,m )},m=1,2,...,M

[0114] Where M is the total number of data sets, I1 and I2 are the two input vectors of the two input layers of the multi-layer deep neural network, with lengths of IL1 and IL2 respectively; O1 and O2 are the two output vectors of the two output layers, with lengths of OL1 and OL2 respectively. Each element in the output vector is the probability of identifying a certain state or scene, and the state or scene of the recognition result is ultimately determined by the highest probability.

[0115] In this embodiment, the channel impulse response data in the data set is the original collected data and does not need to be preprocessed. There is no specific limit on the size of the data set, and the ratio of the training set to the test set is 7:3.

[0116] In the four fully connected layers, neurons are fully connected. These layers are used to explore the mapping relationship between input and output. To reduce overfitting in the neural network, a Dropout layer is added between the two fully connected layers. During each training batch of data, the Dropout layer randomly shuts down a certain number of neurons with a probability p. However, this probability should not exceed 0.2. Setting it too high can cause too many neurons to shut down, leading to large recognition errors.

[0117] Rectified linear unit (ReLu) is used in four fully connected layers and two Dropout layers to provide positive input in the early stages of training to accelerate the training process.

[0118] The softmax activation function is used in the scene output layer and the channel state output layer, which can map the output of the neuron to values ​​within [0,1], and the sum of these values ​​is 1, so these values ​​are regarded as the probability of identifying a certain class. The specific formula of the softmax function is expressed as:

[0119]

[0120] where z y is the output of the y-th neuron, and Y is the total number of output neurons.

[0121] During the training process, the classification cross entropy function is used as the loss function, and the performance is evaluated by comparing the output value and the expected value.

[0122]

[0123] Among them, m, x, and y are data, scene, and state category indexes respectively. x,m and p y,m The probability of being identified as the xth scene and the yth channel state respectively. x,m and t y,m The actual values ​​of the x-th scene and y-th state respectively.

[0124] Furthermore, in this implementation, the Root Mean Square Propagation (RMSProp) optimizer is used to adaptively update the learning rate to minimize the iterative training error.

[0125] Finally, a trained multi-layer deep neural network is used to input channel statistical parameters and multipath cluster parameters to perform real-time recognition of channel status and scenarios.

[0126] In order to verify the effectiveness of the wireless channel status and scene recognition method suitable for millimeter wave communication disclosed in the present invention, the recognition method based on a multi-layer deep neural network and the recognition method based on a back-propagation neural network proposed in the present invention were compared experimentally, and the specific recognition accuracy of substations, high-speed railway station waiting rooms, indoor scenes, and line-of-sight and non-line-of-sight channel states was analyzed. Figure 4 (a) Figure 4 (b) and Figure 4 (c) are the channel measurement environments of the substation, the waiting hall of the high-speed railway station and the indoor office. In the three environments, 1728, 1344 and 640 sets of line-of-sight state channel data and 384, 768 and 960 sets of non-line-of-sight state channel data were collected respectively. 70% of the data in each scene and each state are used to train the deep neural network, and the remaining 30% are used to verify the recognition results. In this experiment, six channel parameters, namely, average cluster interval, cluster attenuation factor, first cluster multipath attenuation factor, average multipath attenuation factor, delay spread and dynamic range, are used as inputs of input layer 1; six channel parameters, namely, signal receiving power, dynamic range, delay spread, Rice factor, cluster attenuation factor and first cluster multipath attenuation factor, are used as inputs of input layer 2. In the training of the deep neural network, the learning rate is selected as 10 -3 , the corresponding maximum number of iterations is 50; the momentum parameter of the RMSProp optimizer is set to 0.9, and the smoothing factor is set to 10 -8 ; The probabilities p of the two dropout layers are set to 0.04 and 0.02 respectively.

[0127] exist Figure 5 (a) and Figure 5 In (b), the x-axis represents the actual channel scene and state labels in the test dataset. Figure 5 (a) and Figure 5 The y-axes of (b) are the outputs of the recognition model based on multi-layer deep neural network and back propagation neural network respectively. Figure 4 It can be clearly found that the recognition method based on multi-layer deep neural network proposed in this invention, such as Figure 5 (a) and Figure 5As shown in Figure (b), in all channel scenarios and conditions, the recognition accuracy is higher than that of the recognition method based on backpropagation neural network. Furthermore, the present invention uses a simple bubble clustering algorithm to directly extract multipath parameters from the channel impulse response, reducing the computational complexity of channel feature extraction. The recognition system of the present invention can be embedded in various mobile communication terminals, and has good real-time recognition performance.

[0128] In summary, the present invention discloses a method for wireless channel status and scene recognition applicable to millimeter wave communication, which has the following advantages over the prior art:

[0129] 1. In terms of recognition speed:

[0130] Ability to identify channel status and scenarios simultaneously;

[0131] The network structure is simple, consisting only of fully connected layers and dropout layers;

[0132] The clustering method uses the bubble clustering algorithm to avoid the computational complexity caused by the parameter extraction algorithm;

[0133] Using ReLu activation function in fully connected layers and Dropout layers can speed up the training process;

[0134] 2. In terms of recognition accuracy:

[0135] Unlike traditional methods that only consider channel statistics for scene and state recognition, this invention uses multipath clustering parameters for recognition. Wireless signals reflect and scatter off objects in space, forming multipath clusters. Multipath clustering parameters are directly related to the characteristics of objects in the environment. Therefore, using these parameters in the recognition process can effectively improve recognition accuracy.

[0136] 3. In terms of recognition robustness and reliability:

[0137] Adding a dropout layer to the neural network structure design can reduce overfitting.

[0138] Example 2:

[0139] A second embodiment of the present invention discloses an identification system using the wireless channel state and scene identification method for millimeter wave communication disclosed in the present invention, comprising:

[0140] Channel statistical parameter calculation module: extracts channel statistical parameters from channel impulse response data and outputs signal power, dynamic range, delay spread and Rice factor;

[0141] Multipath clustering module: uses a clustering algorithm to identify multipaths from the channel impulse response and divides the multipaths into different multipath clusters;

[0142] Multipath cluster parameter calculation module: extracts multipath cluster parameters from the clustering results and outputs average cluster interval, cluster attenuation factor, first cluster multipath attenuation factor and average multipath attenuation factor;

[0143] Multi-layer deep neural network module: Use channel statistical parameters, multipath cluster parameters, channel scenarios and status information to train the multi-layer deep neural network. By inputting channel parameters, the trained multi-layer deep neural network can identify channel scenarios and status in real time.

[0144] After the receiver at the receiving end of any wireless communication device collects the signal, the channel impulse response data is input into the channel statistical parameter calculation module and the multipath clustering module respectively to extract the channel statistical parameters and perform multipath clustering; the multipath clustering results are input into the multipath cluster parameter calculation module to extract the multipath cluster parameters; finally, the outputs of the channel statistical parameter calculation module and the multipath cluster parameter calculation module are input into the multi-layer deep neural network module to output the corresponding channel scenario and status.

[0145] like Figure 2 As shown, the recognition system disclosed in this embodiment is divided into a training process and a real-time recognition process during its specific implementation:

[0146] During the training process, the channel parameters obtained from actual channel measurements and the corresponding channel states and scenarios are used to train a multi-layer deep neural network. The channel parameters include channel statistical parameters directly calculated from the channel impulse response and multipath cluster parameters obtained after clustering using a clustering algorithm.

[0147] The specific implementation steps of the training process are:

[0148] The transmitter sends a reference signal, and the receiver performs an autocorrelation operation after receiving the signal to obtain the channel impulse response and complete the channel measurement operation;

[0149] Inputting the channel impulse response data into a channel statistical parameter calculation module, extracting and outputting channel statistical parameters, wherein the channel statistical parameters include: output signal power, dynamic range, delay spread and Ricean factor;

[0150] Input the channel impulse response data into the multipath clustering module, use the clustering algorithm to identify the multipath from the channel impulse response and divide the multipath into different multipath clusters, and output the clustering results;

[0151] Input the clustering result into the multipath cluster parameter calculation module, extract and output the multipath cluster parameters from the clustering result, the multipath cluster parameters including: output average cluster interval, cluster attenuation factor, first cluster multipath attenuation factor and average multipath attenuation factor;

[0152] Channel statistical parameters, multipath cluster parameters, channel scenarios and state information are input into the multi-layer deep neural network module to train the multi-layer deep neural network.

[0153] During the real-time recognition process, a trained multi-layer deep neural network is deployed on a signal receiving device at the receiving end. Signals are received and a channel impulse response is obtained using a channel estimation method. The identified channel statistical parameters and multipath cluster parameters are then input into the multi-layer deep neural network to rapidly identify the channel state and scenario. In this embodiment, an existing channel estimation algorithm is used to perform channel estimation and obtain the channel impulse response. Compared to methods that obtain the channel impulse response by performing a cross-correlation operation on a decoded signal sequence and the received signal, the channel estimation algorithm used in this embodiment has the advantage of lower complexity.

[0154] The specific implementation steps of the real-time recognition process are:

[0155] Receive the signal and obtain the channel impulse response through the channel estimation method;

[0156] Inputting the channel impulse response data into a channel statistical parameter calculation module, extracting and outputting channel statistical parameters, wherein the channel statistical parameters include: output signal power, dynamic range, delay spread and Ricean factor;

[0157] Input the channel impulse response data into the multipath clustering module, use the clustering algorithm to identify the multipath from the channel impulse response and divide the multipath into different multipath clusters, and output the clustering results;

[0158] Input the clustering result into the multipath cluster parameter calculation module, extract and output the multipath cluster parameters from the clustering result, the multipath cluster parameters including: output average cluster interval, cluster attenuation factor, first cluster multipath attenuation factor and average multipath attenuation factor;

[0159] The channel statistical parameters and multipath cluster parameters are input into the multi-layer deep neural network module to perform real-time recognition of channel scenarios and states.

[0160] In this embodiment, the data processing flow of each module in the above-mentioned wireless channel status and scene recognition system applicable to millimeter wave communication is the same as the wireless channel status and scene recognition method applicable to millimeter wave communication disclosed in the first embodiment of the present invention, and will not be repeated here.

[0161] Example 3:

[0162] The third embodiment of the present invention further discloses a computer-readable storage medium storing one or more programs, wherein the one or more programs include instructions. When the instructions are executed by a computing device, the computing device executes the wireless channel state and scene recognition method suitable for millimeter wave communication disclosed in the first embodiment of the present invention.

[0163] Example 4:

[0164] The fourth embodiment of the present invention also discloses a computing device, including one or more processors, one or more memories, and one or more programs, wherein the one or more programs are stored in the one or more memories and are configured to be executed by the one or more processors, and the one or more programs include instructions for executing the wireless channel state and scene recognition method applicable to millimeter wave communication disclosed in the first embodiment of the present invention.

[0165] It will be understood by those skilled in the art that embodiments of the present invention may be provided as methods, systems, or computer program products. Thus, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0166] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as combinations of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowcharts and / or block diagrams. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0167] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0168] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0169] The above are merely embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention are included in the scope of the claims of the present invention to be approved.

Claims

1. A method for wireless channel status and scene recognition suitable for millimeter wave communication, characterized in that: The following steps are involved: Establish a multi-layer deep neural network consisting of 2 input layers, 2 output layers and 6 hidden layers; Calculating channel statistical parameters, including signal received power, dynamic range, delay spread, and Ricean factor; Use a multipath clustering algorithm to perform clustering and calculate multipath cluster parameters, wherein the multipath cluster parameters include: average cluster interval, cluster attenuation factor, first cluster multipath attenuation factor and average multipath attenuation factor; Use channel statistics and multipath cluster parameters as input, and channel status and scene information as output to train a multi-layer deep neural network; Using a trained multi-layer deep neural network, channel statistical parameters and multipath cluster parameters are input to perform real-time recognition of channel status and scene information; The specific steps of calculating the multipath cluster parameters are: Introducing the SV channel model: where α rs and θ rs are the amplitude and phase of the sth multipath in the rth cluster, Rm is the number of channel multipath clusters at time m, S r is the number of multipaths in the rth cluster, τ is the delay, τ r,1 and τ r,s are the delays of the first and sth multipaths in the rth cluster that can be directly obtained from the clustering results, and j is an imaginary unit, that is, Calculate the average cluster spacing, which is used to describe the abundance of scatterers in a scene. The more scatterers there are in the environment, the smaller the average cluster spacing. The average cluster spacing is calculated as follows: The multipath received power of the sth path in the rth cluster is defined as: where 1 / Γ is the cluster attenuation factor, 1 / γ r is the multipath attenuation factor of the rth cluster, and the delay power spectrum P(m,τ) is calculated by h(m,τ); The dynamic range is calculated by the maximum power and the noise power in the delay power spectrum.

2. The wireless channel state and scene recognition method applicable to millimeter wave communication according to claim 1, characterized in that: The multi-layer deep neural network includes: a first input layer of N1 neurons, a first fully connected layer of 64 neurons, a first Dropout layer of 64 neurons, a second fully connected layer of 128 neurons, a scene output layer of 3 neurons, a second input layer of N2 neurons, a third fully connected layer of 64 neurons, a second Dropout layer of 64 neurons, a fourth fully connected layer of 32 neurons, and a channel state output layer of 2 neurons, wherein N1 and N2 are the number of input parameters of the first input layer and the second input layer respectively, and the number of neurons is set according to the number of input parameters; data passes through the first input layer, the first fully connected layer, the first Dropout layer, the second fully connected layer to the scene output layer in sequence, and the third fully connected layer merges the data of the scene output layer and the second input layer and then continues to pass through the second Dropout layer and the fourth fully connected layer in sequence to reach the channel state output layer.

3. The wireless channel state and scene recognition method applicable to millimeter wave communication according to claim 1, characterized in that: The specific steps of calculating the channel statistical parameters are: Calculate the delay power spectrum. Assume that the channel impulse response matrix at time m is h(m,τ). Then the delay power spectrum at time m is: P(m,τ)=|h(m,τ)| 2 Where, τ is the time delay; Calculate the noise power. The delay power spectrum includes both signal and noise. The noise power is obtained by averaging the last portion of the delay power spectrum that does not contain the signal. It is specifically defined as: Where T n 、T end The length selected for calculating the noise power and the total length of the delayed power spectrum; Calculate the dynamic range and use the difference between the maximum power and the noise power in the delay power spectrum to calculate the dynamic range of the signal: Calculate the delay sample set of the non-noise signal and set the denoising threshold according to the dynamic range of the signal. Delay points greater than the denoising threshold will be extracted and regarded as signals. The delay sample set of the non-noise signal is defined as: where ξ τ,m is the maximum delay, and the delay points exceeding the maximum delay are regarded as noise. ρ is the denoising factor that changes with the environment. Calculate the signal received power by adding up all the signal received powers to get the signal received power at time m: where τ l is the signal delay point, is the set Elements in Calculate the delay spread to quantify the channel dispersion in the time domain. The calculation formula for delay spread is: Calculate the Rice factor, which is the ratio of the component with the largest power in the channel delay power spectrum to the sum of the powers of the other components, and is calculated by the following formula:

4. The wireless channel state and scene recognition method applicable to millimeter wave communication according to claim 1, characterized in that: The multipath clustering algorithm is a bubble clustering algorithm.

5. The wireless channel state and scene recognition method applicable to millimeter wave communication according to claim 2, characterized in that: The establishment of a multi-layer deep neural network including 2 input layers, 2 output layers and 6 hidden layers also includes: Define the dataset for training: D m ={(I 1,m ,I 2,m )|(O 1,m ,O 2,m )},m=1,2,...,M Where M is the total number of data sets, I1 and I2 are the two input vectors of the two input layers of the multi-layer deep neural network, with lengths of IL1 and IL2 respectively; O1 and O2 are the two output vectors of the two output layers, with lengths of OL1 and OL2 respectively. Each element in the output vector is the probability of identifying a certain state or scene, and the state or scene of the recognition result is ultimately determined by the highest probability. The rectified linear unit function is used in four fully connected layers and two dropout layers to provide positive input in the early stages of training to accelerate the training process; The softmax activation function is used in the scene output layer and the channel state output layer to express the output value as the probability of identifying a certain class. The specific formula is expressed as follows: where z y is the output of the y-th neuron, and Y is the total number of output neurons; The classification cross entropy function is set as the loss function, and the performance is evaluated by comparing the output value and the expected value. The loss function is defined as: Among them, m, x, y are data, scene and state category indexes respectively, p x,m and p y,m The probability of identifying it as the xth scene and the yth channel state, t x,m and t y,m The actual values ​​of the x-th scene and y-th state respectively; The RMS propagation optimizer is used to adaptively update the learning rate to minimize the iterative training error.

6. An identification system using the wireless channel state and scene identification method for millimeter wave communication according to any one of claims 1 to 5, characterized in that: include: Channel statistical parameter calculation module: extracts channel statistical parameters from channel impulse response data and outputs signal power, dynamic range, delay spread and Rice factor; Multipath clustering module: uses a multipath clustering algorithm to identify multipaths from the channel impulse response and divides the multipaths into different multipath clusters; Multipath cluster parameter calculation module: extracts multipath cluster parameters from the clustering results and outputs average cluster interval, cluster attenuation factor, first cluster multipath attenuation factor and average multipath attenuation factor; Multi-layer deep neural network module: Use channel statistical parameters, multipath cluster parameters, channel scenarios and status information to train the multi-layer deep neural network. By inputting channel statistical parameters and multipath cluster parameters, the trained multi-layer deep neural network can perform real-time recognition of channel scenarios and status.

7. A computer-readable storage medium storing one or more programs, characterized in that: The one or more programs include instructions that, when executed by a computing device, cause the computing device to perform any one of the methods according to claims 1 to 5 .

8. A computing device, characterized in that include: One or more processors, one or more memories, and one or more programs, wherein the one or more programs are stored in the one or more memories and configured to be executed by the one or more processors, the one or more programs comprising instructions for performing any of the methods according to claims 1 to 5.

Images