Active IRS-assisted uplink IoT system channel estimation method and device based on CNN network
By adopting a CNN network-based channel estimation method in an active IRS-assisted IoT system, the problems of channel estimation error and calculation complexity are solved, and efficient and accurate channel estimation is achieved.
Patent Information
- Application Number
- CN202510010126.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-01-03
- Publication Date
- 2025-05-13
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
In an active IRS-assisted IoT system, how to effectively reduce channel estimation errors and reduce computational complexity, especially in resource-constrained environments.
The channel estimation method based on CNN network is used to train by building a system model, optimizing power allocation factors, generating training data sets and designing a convolutional neural network model, and finally used for channel estimation.
Lightweight and low computing complexity channel estimation is realized, which reduces the system's channel estimation error and improves the accuracy of channel estimation.
Smart Images
Figure CN119996120A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of channel estimation, and in particular to a method and device for channel estimation of an active IRS-assisted uplink IoT system based on a CNN network. Background Art
[0002] Intelligent Reflecting Surface (IRS) is a new type of wireless communication technology that uses large-scale reflective elements, such as tiny antennas or phase adjustment elements, to construct a dynamically controllable surface. These reflective elements can control the reflection, propagation and reception of wireless signals by changing their phase, amplitude or polarization state as needed. In wireless communication systems, intelligent reflective surfaces can be used as an infrastructure to enhance or optimize the propagation of wireless signals, and can also be used to implement advanced communication technologies such as directional propagation of wireless signals, beamforming, and multi-user diversity.
[0003] As a revolutionary paradigm for controlling wireless channels, passive IRS is expected to improve channel capacity, extend coverage, and save power in future wireless communication networks due to its high array gain, low cost, low power consumption, and negligible noise.
[0004] However, due to the "multiplicative fading" effect, traditional passive IRS can only achieve limited capacity gain in some direct link completely blocked or weak direct link scenarios. In order to overcome this basic physical limitation, active IRS came into being. Since active IRS has a built-in reflective amplifier, it amplifies the noise while amplifying the signal, and causes power consumption. How to allocate power between resource-constrained Internet of Things (IoT) devices and base stations (BS) so that the channel estimation error of active IRS-assisted IoT systems is minimized has become a major concern in academia and industry.
[0005] Traditional channel estimation methods mainly include the pilot-based least square (LS) method, the minimum mean square error (MMSE) method, and the compressed sensing (CS) method based on the sparsity of the millimeter wave channel. Due to the influence of noise, the error of the LS method is proportional to the square of the number of transmitting antennas; the MMSE method requires information about noise and channel correlation to achieve more accurate channel estimation, which leads to higher computational complexity; the CS method recovers the channel matrix by compressing and reconstructing the signal. However, when the iterative method is used to solve the nonlinear optimization problem in CS, there are problems of long processing time and high computational complexity. In order to further reduce the pilot overhead and complexity, deep learning is applied to IRS-assisted wireless networks for channel estimation. As the core architecture of channel estimation, CNN has powerful feature extraction capabilities, efficient end-to-end processing, nonlinear modeling capabilities and low complexity. These advantages make CNN perform well in channel estimation tasks and become one of the current research hotspots. Summary of the invention
[0006] The object of the present invention is to provide a lightweight, low-error and low-computational complexity CNN-based active IRS-assisted uplink IoT system channel estimation method and device.
[0007] The technical solution to achieve the purpose of the present invention is: a CNN network-based active IRS-assisted uplink IoT system channel estimation method, comprising the following steps:
[0008] Step 1: Build a system model of active IRS-assisted uplink IoT network;
[0009] Step 2: construct an optimization problem of minimizing the power allocation factor and the mean square error (MSE) of the channel estimation between the resource-constrained IoT device and the base station BS;
[0010] Step 3: Use the Cardan formula to solve the optimal power allocation factor between the IoT device and the BS. At this time, the channel estimation error of the system is the minimum channel estimation MSE.
[0011] Step 4: Based on the solved optimal power allocation factor, the known pilot signal and the active IRS phase shift matrix, a received signal data set is randomly generated, and the generated data set is segmented to obtain a training data set and a verification data set;
[0012] Step 5: Design a channel estimation network model based on a convolutional neural network (CNN), train the channel estimation neural network model using a training data set, and save the trained channel estimation neural network model;
[0013] Step 6: randomly regenerate a received cascade link signal data set, test the trained channel estimation neural network model, and use the tested channel estimation neural network model for channel estimation.
[0014] A CNN-based active IRS-assisted uplink IoT system channel estimation device, which is used to implement the CNN-based active IRS-assisted uplink IoT system channel estimation method, the device includes a system model building module, an optimization problem construction module, a power allocation factor calculation module, a data set generation module, a model training module and a model testing module, wherein:
[0015] System model building module, building the system model of active IRS-assisted uplink IoT network;
[0016] An optimization problem construction module is used to construct an optimization problem for minimizing the power allocation factor and the mean square error (MSE) of the channel estimation between resource-constrained IoT devices and base stations BS.
[0017] The power allocation factor calculation module uses the Cardan formula to solve the optimal power allocation factor between the IoT device and the BS. At this time, the channel estimation error of the system is the minimum channel estimation MSE;
[0018] The data set generation module randomly generates a received signal data set based on the solved optimal power allocation factor, the known pilot signal and the active IRS phase shift matrix, and divides the generated data set into a training data set and a verification data set;
[0019] Model training module, designs a channel estimation network model based on convolutional neural network CNN, uses training data sets to train the channel estimation neural network model, and saves the trained channel estimation neural network model;
[0020] The model testing module re-randomly generates a receiving cascade link signal data set to test the trained channel estimation neural network model. The channel estimation neural network model that passes the test is used for channel estimation.
[0021] A mobile terminal comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the active IRS-assisted uplink IoT system channel estimation method based on a CNN network is implemented.
[0022] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps in the CNN network-based active IRS-assisted uplink IoT system channel estimation method.
[0023] Compared with the prior art, the present invention has the following significant advantages: (1) by combining the convolution operation Conv and the rectified linear unit ReLU to extract the spatial features of the channel matrix, batch normalization BN is introduced between Conv and ReLU, thereby improving the stability of the network and accelerating the training speed; (2) by deploying the pooling layer to reduce the dimension of the features extracted from the previous layer, the parameters and computational complexity of the next layer are reduced, and the consumption of computing resources is significantly reduced while ensuring the performance of the model. In addition, the pooling layer can prevent overfitting and make the features more robust to noise, thereby enhancing the overall stability and generalization ability of the model; (3) the power allocation problem between resource-constrained IoT devices and BS is considered, and the optimal power allocation factor is obtained, which solves and reduces the channel estimation error of the system; (4) a direct and cascaded channel estimation network based on CNN is designed, which can obtain a lower channel estimation error, further reduce the pilot overhead, and improve the channel estimation accuracy. BRIEF DESCRIPTION OF THE DRAWINGS
[0024] Figure 1 The present invention is a flow chart of the active IRS-assisted uplink IoT system channel estimation method based on CNN network.
[0025] Figure 2 This is a system model diagram of the active IRS-assisted uplink IoT system in an embodiment of the present invention.
[0026] Figure 3 Schematic diagram of the structure of the pilot mode in an embodiment of the present invention.
[0027] Figure 4 Schematic diagram of the structure of a direct channel estimation network based on CNN in an embodiment of the present invention.
[0028] Figure 5 Schematic diagram of the structure of a CNN-based cascade channel estimation network in an embodiment of the present invention.
[0029] Figure 6 4 is a curve diagram showing the relationship between MSE and the optimal power allocation factor β in an embodiment of the present invention.
[0030] Figure 7 Graph showing the relationship between MSE and signal-to-noise ratio SNR in an embodiment of the present invention.
[0031] Figure 8 4 is a curve diagram showing the relationship between MSE and the number of BS antennas K in an embodiment of the present invention. DETAILED DESCRIPTION
[0032] The present invention provides a channel estimation method and device for an active IRS-assisted uplink IoT system based on a CNN network. First, a system model of an active IRS-assisted IoT is constructed; an optimization problem of minimizing the power allocation factor and the mean square error (MSE) of the channel estimation between a resource-constrained IoT device and a base station is constructed; the optimal power allocation factor is solved by using the Cardan formula, and the channel estimation error of the system is the minimum channel estimation error. In order to further reduce the pilot overhead, lightweight direct channel and cascade channel estimation networks are proposed based on convolutional neural networks. Based on the solved optimal power allocation factor, the known pilot signal and the active intelligent reflection surface phase shift matrix, 10,000 received direct link signal data sets and 10,000 received cascade link signal data sets are randomly generated, of which 90% are used as training data sets and 10% are used as verification data sets. Experimental results show that the MSE of the lightweight deep learning channel estimator designed by the present invention is lower than the MSE of the traditional least square (LS) method and the minimum mean square error (MMSE) method.
[0033] The present invention will be further described in detail below in conjunction with the accompanying drawings and specific embodiments.
[0034] like Figure 1 As shown, the present invention provides an active IRS-assisted uplink IoT system channel estimation method based on a CNN network, comprising the following steps:
[0035] Step 1: Build a system model of active IRS-assisted uplink IoT network, as follows:
[0036] Step 1.1, set the active IRS assisted uplink IoT network, the number of antennas of the base station BS is K, the IRS is equipped with N active reflective elements, and the IoT device is a single antenna user, such as Figure 2 As shown in the figure, it is assumed that the channels from IoT device to BS, IoT device to IRS, and IRS to BS are all Rayleigh fading channels; BS can estimate the channel characteristics between a large number of antennas through a small number of pilot signals in the uplink; based on channel reciprocity, BS can effectively predict or reconstruct the channel state information of the downlink;
[0037] Step 1.2: The transmission signal of the IoT device is expressed as:
[0038]
[0039] Where x is the transmitted signal with unit energy, P trepresents the total power of the IoT device and the active IRS, and β represents the power allocation factor. Then the received signal at the active IRS is:
[0040]
[0041] where Θ is the phase shift matrix of the active IRS, g represents the channel between the IoT device and the active IRS, and w i represents the additive white Gaussian noise at the active IRS and it obeys A complex Gaussian distribution, where 0 represents the mean, represents the IRS noise variance, I N represents the identity matrix of dimension N;
[0042] Since the system adopts an active IRS-assisted uplink IoT network, the received signal at the BS can be modeled as:
[0043]
[0044] Where h and F represent the channels between IoT devices and BS and IRS and BS respectively, w b represents the additive Gaussian white noise at the BS and it obeys A complex Gaussian distribution, where 0 represents the mean, represents the BS noise variance, I K Denote the identity matrix with dimension K; define G = diag(g), where diag represents the diagonal matrix with diagonal elements g; define Θ = diag(θ), where θ is the phase shift vector of the active IRS, then the cascade channel from the IoT device to the BS can be expressed as H biu =FG, therefore, equation (3) can be re-expressed as:
[0045]
[0046] Where W i =diag(w i ) indicates that the diagonal element is w i The diagonal matrix of ;
[0047] The phase shift vector θ for the active IRS can be written in normalized form, namely: Where ρ represents the amplification factor, represents the phase shift of regulation, so equation (4) can be re-expressed as:
[0048]
[0049] Correspondingly, the reflected power of the active IRS is:
[0050]
[0051] Where Ε{·} represents the expected operation, (·) H represents the conjugate transpose operation;
[0052] Through formula (6), we can get:
[0053]
[0054] Step 2: Construct an optimization problem for minimizing the power allocation factor and the mean square error (MSE) of the channel estimation between the resource-constrained IoT device and the base station BS, as follows:
[0055] Step 2.1: Figure 3 As shown, in order to establish the function of the channel estimation error with respect to the power allocation factor, a special pilot pattern is adopted, namely the Hadamard matrix is used as a special active IRS phase shift matrix. In four different pilot sequence periods, the active IRS phase shift matrix is sent in the order of V, V, -V, and -V, and the known pilot sequence of the IoT device is sent alternately in the order of x, x, -x, and -x.
[0056] Step 2.2: In the first pilot sequence period, we can get:
[0057]
[0058] where x=[x1,x2,...,x Q ] T is a pilot sequence consisting of Q symbols sent by the IoT device, [·] T represents the transposition operation; X = diag{x} represents a diagonal matrix with diagonal elements x; V is the phase shift matrix of the active IRS composed of Q phase shift vectors θ, that is, W b1 is the noise w at Q BSs b The noise matrix formed is W b1 =[w b1 ,w b2 ,…,w bQ ]; W i1 is the noise at the IRS during the first pilot sequence period;
[0059] Step 2.3: The received signals in the second, third and fourth pilot sequence periods are respectively expressed as:
[0060]
[0061] Where W i2 , W i3 , W i4 and W b2 , Wb3 , W b4 Represent the noise at the active IRS and the noise at the BS during the second, third, and fourth pilot sequence periods, respectively;
[0062] Step 2.4: According to the symmetry properties of equations (8) to (11), separate the direct signal and the cascade signal through algebraic operations, that is:
[0063]
[0064] in Only the direct channel h is included in Only the cascade channel H is included biu ;
[0065] Step 2.5: Multiply the right side of equation (12) by x * ,in(·) * represents taking the conjugate sign, and obtaining the LS estimator of the direct channel:
[0066]
[0067] Given E{x T x *}=N, then the MSE of the direct channel is estimated to be:
[0068]
[0069] where ||·|| F represents the F-norm;
[0070] Step 2.6: By performing quantization operations on both sides of equation (13), we can obtain:
[0071]
[0072] in vec(·) represents the orientation quantization operation, then the LS estimator of the cascade channel is:
[0073]
[0074] in(·) -1 Represents the matrix inversion operation;
[0075] The estimated MSE of the cascaded channel is:
[0076]
[0077] Step 2.7: Estimate the sum of the MSE of the direct channel and the cascade channel as:
[0078] ε=ε1+ε2 (19)
[0079] Then the optimization problem of minimizing the sum of the power allocation factor and the channel estimation MSE is:
[0080]
[0081] Step 3: Use the Cardan formula to solve the optimal power allocation factor between the IoT device and the BS. At this time, the channel estimation error of the system is the minimum channel estimation MSE, as follows:
[0082] Step 3.1: Convert the sum of the MSEs of the direct channel and the cascade channel into a function of the power allocation factor as follows:
[0083] Substituting formula (7) into formula (19), we can obtain:
[0084]
[0085] in
[0086]
[0087]
[0088] By solving the first-order derivative of β for equation (21), we can obtain:
[0089]
[0090] in
[0091]
[0092] Let (24) be equal to 0, then:
[0093] β 4 +d3β 3 +d2β 2 +d1β+d0=0 (26)
[0094] in
[0095]
[0096] Step 3.2: Calculate the optimal power allocation factor set S1 = {β1, β2, β3, β4}, where
[0097]
[0098] and where y * is any real solution of the following equation:
[0099] y 3 -d2y 2+(d3d1-4d0)y+(4d2d0-d1 2 -d3 2 d0)=0 (29)
[0100] When R≠0, there exists:
[0101]
[0102] When R = 0, there exists:
[0103]
[0104] Step 3.3: Get the power allocation factor by using the Cardan formula:
[0105]
[0106] in
[0107]
[0108] Step 4: Based on the solved optimal power allocation factor, the known pilot signal and the active IRS phase shift matrix, 10,000 received direct link signal data sets and 10,000 received cascade link signal data sets are randomly generated for channel estimation, and the generated data sets are divided, 90% of which are used as training data sets and 10% as verification data sets;
[0109] Step 5: Design a channel estimation neural network based on deep learning, use the training data set to train the channel estimation neural network, and save the trained channel estimation neural network, as follows:
[0110] Step 5.1: Design a direct channel estimation network based on convolutional neural network CNN. Figure 4 As shown, it contains 7 layers in total, namely three convolutional layers, three maximum pooling layers and one linear layer, as follows:
[0111] The spatial features of the channel matrix are extracted by combining the convolution Conv operation and the rectified linear unit ReLU. In order to improve the stability of the network and speed up the training, batch normalization BN is introduced between Conv and ReLU. Therefore, the first, third and fifth layers are all "Conv+BN+ReLU" operations;
[0112] The second, fourth, and sixth layers are maximum pooling layers, whose kernel size and stride are both set to 2. By deploying pooling layers to reduce the dimension of features extracted from the previous layer, the parameters and computational complexity of the next layer are reduced, which significantly reduces the consumption of computing resources while ensuring model performance. In addition, pooling layers can prevent overfitting and make features more robust to noise, thereby enhancing the overall stability and generalization ability of the model.
[0113] The seventh layer is a linear layer, whose output maps the real and imaginary parts of the estimated direct channel;
[0114] Step 5.2: Design a CNN-based cascade channel estimation network such as Figure 5 As shown, there are 10 layers in total, among which the first, third, fifth, seventh, eighth and ninth layers are all convolutional layers, the second, fourth and sixth layers are all maximum pooling layers, and the last layer is a linear layer;
[0115] Step 5.3: Deploy the designed direct channel estimation network and cascade channel estimation network on the BS side to obtain a channel estimation neural network based on deep learning. Use the training data set to train the channel estimation neural network and save the trained channel estimation neural network. The details are as follows:
[0116] The initial learning rate of the direct channel estimation network is set to 0.001, and the initial learning rate of the cascade channel estimation network is set to 0.0025. The channel estimation neural network is trained using the training data set, and the learning rate is decayed by half every 35 cycles. The training process contains a total of 200 cycles. After 200 iterations, the trained channel estimation neural network is saved, and the validation data set is used to verify the performance of the trained channel estimation neural network. During the training process, the AdamW optimization algorithm is used to update the weights of the model.
[0117] Step 6: Randomly regenerate 5000 received direct link signal data sets and 5000 received cascade link signal data sets to test the trained deep learning channel estimation neural network. Finally, it is found that the MSE of the deep learning-based channel estimation network designed by the present invention is smaller than the MSE of the traditional LS and MMSE methods.
[0118] The present invention also provides an active IRS-assisted uplink IoT system channel estimation device based on a CNN network, which is used to implement the active IRS-assisted uplink IoT system channel estimation method based on a CNN network. The device includes a system model construction module, an optimization problem construction module, a power allocation factor calculation module, a data set generation module, a model training module and a model testing module, wherein:
[0119] System model building module, building the system model of active IRS-assisted uplink IoT network;
[0120] An optimization problem construction module is used to construct an optimization problem for minimizing the power allocation factor and the mean square error (MSE) of the channel estimation between resource-constrained IoT devices and base stations BS.
[0121] The power allocation factor calculation module uses the Cardan formula to solve the optimal power allocation factor between the IoT device and the BS. At this time, the channel estimation error of the system is the minimum channel estimation MSE;
[0122] The data set generation module randomly generates a received signal data set based on the solved optimal power allocation factor, the known pilot signal and the active IRS phase shift matrix, and divides the generated data set into a training data set and a verification data set;
[0123] Model training module, designs a channel estimation network model based on convolutional neural network CNN, uses training data sets to train the channel estimation neural network model, and saves the trained channel estimation neural network model;
[0124] The model testing module re-randomly generates a receiving cascade link signal data set to test the trained channel estimation neural network model. The channel estimation neural network model that passes the test is used for channel estimation.
[0125] The present invention also provides a mobile terminal, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the active IRS-assisted uplink IoT system channel estimation method based on the CNN network is implemented.
[0126] The present invention also provides a computer-readable storage medium having a computer program stored thereon, and when the program is executed by a processor, the steps in the active IRS-assisted uplink IoT system channel estimation method based on a CNN network are implemented.
[0127] The present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.
[0128] Example
[0129] In this embodiment, an active IRS-assisted uplink IoT system with K=16 base station antennas and N=64 smart reflective surface elements is used for simulation experiments, and the MSE of the active IRS-assisted uplink IoT system channel estimation method based on the CNN network designed in the present invention is compared with the MSE of the traditional LS and MMSE methods.
[0130] Figure 6The simulation curves of the direct and cascade total channel estimation errors and the optimal power allocation factor are shown. It can be seen from the figure that when the total power between the IoT device and the base station is limited, there is an optimal power allocation factor. For example, when the number of base station antennas is 8 and the number of smart reflector elements is 64, the optimal power allocation factor is 0.5912.
[0131] Figure 7 The simulation curve between the channel estimation error and the signal-to-noise ratio is shown. It can be seen from the figure that as the signal-to-noise ratio increases, the channel estimation error gradually decreases. In addition, whether it is a direct channel or a cascade channel, the MSE of the active IRS-assisted uplink IoT system channel estimation method based on the CNN network designed in the present invention is much lower than the MSE of the traditional LS method and MMSE method.
[0132] Figure 8 The simulation curve between the channel estimation error and the number of base station antennas is shown. As can be seen from the figure, as the number of base station antennas increases, the channel estimation error gradually increases. This is because the size of the estimated channel increases with the increase in the number of base station antennas. In addition, the error of the active IRS-assisted uplink IoT system channel estimation method based on the CNN network designed by the present invention is smaller than that of the traditional LS method and MMSE method.
[0133] The above are only preferred embodiments of the present invention. It should be pointed out that, for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.
Claims
1. A CNN-based active IRS-assisted uplink IoT system channel estimation method, characterized in that: The following steps are involved: Step 1: Build a system model of active IRS-assisted uplink IoT network; Step 2: construct an optimization problem of minimizing the power allocation factor and the mean square error (MSE) of the channel estimation between the resource-constrained IoT device and the base station BS; Step 3: Use the Cardan formula to solve the optimal power allocation factor between the IoT device and the BS. At this time, the channel estimation error of the system is the minimum channel estimation MSE. Step 4: Based on the solved optimal power allocation factor, the known pilot signal and the active IRS phase shift matrix, a received signal data set is randomly generated, and the generated data set is segmented to obtain a training data set and a verification data set; Step 5: Design a channel estimation network model based on a convolutional neural network (CNN), train the channel estimation neural network model using a training data set, and save the trained channel estimation neural network model; Step 6: randomly regenerate a received cascade link signal data set, test the trained channel estimation neural network model, and use the tested channel estimation neural network model for channel estimation.
2. According to the CNN network-based active IRS-assisted uplink IoT system channel estimation method of claim 1, it is characterized in that: The system model for building an active IRS-assisted uplink IoT network described in step 1 is as follows: Step 1.1: In the active IRS-assisted uplink IoT network, the number of antennas of the base station BS is K, the IRS is equipped with N active reflective elements, and the IoT device is a single-antenna user; the channels from the IoT device to the BS, from the IoT device to the IRS, and from the IRS to the BS are all set to be Rayleigh fading channels; the BS estimates the channel characteristics between the antennas through the pilot signal in the uplink; based on the channel reciprocity, the BS predicts or reconstructs the channel state information of the downlink; Step 1.2: The transmitted signal s of the IoT device is expressed as: Where x is the transmitted signal with unit energy, P t represents the total power of the IoT device and the active IRS, and β represents the power allocation factor. Then the received signal at the active IRS is: where Θ is the phase shift matrix of the active IRS, g represents the channel between the IoT device and the active IRS, and w i represents the additive white Gaussian noise at the active IRS and obeys A complex Gaussian distribution, where 0 represents the mean, represents the IRS noise variance, I N represents the identity matrix of dimension N; Since the system adopts an active IRS-assisted uplink IoT network, the received signal at the BS is modeled as: Where h and F represent the channels between IoT devices and BS and IRS and BS respectively, w b represents the additive Gaussian white noise at the BS and it obeys A complex Gaussian distribution, where 0 represents the mean, represents the BS noise variance, I K Denote the identity matrix of dimension K; define G = diag(g), where diag represents a diagonal matrix with diagonal elements g; define Θ = diag(θ), where θ is the phase shift vector of the active IRS, then the cascade channel from the IoT device to the BS is represented as H biu =FG, therefore, equation (3) can be re-expressed as: Where W i =diag(w i ) indicates that the diagonal element is w i The diagonal matrix of ; The phase shift vector θ for the active IRS is written in normalized form, namely Where ρ represents the amplification factor, represents the phase shift of regulation, so equation (4) can be re-expressed as: Correspondingly, the reflected power of the active IRS is: Where Ε{·} represents the expected operation, (·) H represents the conjugate transpose operation; Through formula (6), we can get:
3. According to the CNN network-based active IRS-assisted uplink IoT system channel estimation method of claim 2, it is characterized in that: In step 2, the optimization problem of minimizing the power allocation factor and the channel estimation mean square error MSE between the resource-constrained IoT device and the base station BS is constructed as follows: Step 2.1: In order to establish the function of the channel estimation error with respect to the power allocation factor, the Hadamard matrix is used as the active IRS phase shift matrix. In four different pilot sequence periods, the active IRS phase shift matrix is sent in the order of V, V, -V, and -V, and the known pilot sequence of the IoT device is sent alternately in the order of x, x, -x, and -x. Step 2.2: In the first pilot sequence period, we get: where x=[x1,x2,...,x Q ] T is a pilot sequence consisting of Q symbols sent by the IoT device, [·] T represents the transposition operation; X = diag{x} represents a diagonal matrix with diagonal elements x; V is the phase shift matrix of the active IRS composed of Q phase shift vectors θ, that is, W b1 is the noise w at Q BSs b The noise matrix formed is W b1 =[w b1 ,w b2 ,...,w bQ ]; W i1 is the noise at the IRS during the first pilot sequence period; Step 2.3: The received signals in the second, third and fourth pilot sequence periods are respectively expressed as: Where W i2 , W i3 , W i4 and W b2 , W b3 , W b4 Represent the noise at the active IRS and the noise at the BS during the second, third, and fourth pilot sequence periods, respectively; Step 2.4: According to the symmetry properties of equations (8) to (11), the direct signal and the cascade signal are separated by algebraic operations, namely: in Only the direct channel h is included in Only the cascade channel H is included biu ; Step 2.5: Multiply the right side of equation (12) by x * ,in(·) * represents taking the conjugate sign, and obtaining the LS estimator of the direct channel: Given E{x T x * }=N, then the MSE of the direct channel is estimated to be: where ||·|| F represents the F-norm; Step 2.6: By simultaneously quantizing the left and right sides of equation (13), we can obtain: in vec(·) represents the orientation quantization operation, then the LS estimator of the cascade channel is: in(·) -1 Represents the matrix inversion operation; The estimated MSE of the cascaded channel is: Step 2.7: Estimate the sum of the MSE of the direct channel and the cascade channel as: ε=ε1+ε2 (19) Then the optimization problem of minimizing the sum of the power allocation factor and the channel estimation MSE is:
4. According to the CNN network-based active IRS-assisted uplink IoT system channel estimation method of claim 3, it is characterized in that: In step 3, the Cardan formula is used to solve the optimal power allocation factor between the IoT device and the BS. At this time, the channel estimation error of the system is the minimum channel estimation MSE, which is as follows: Step 3.1: Convert the sum of the MSEs of the direct channel and the cascade channel into a function of the power allocation factor, specifically: Substituting formula (7) into formula (19), we get: in By solving the first-order derivative of β for equation (21), we can obtain: in Let (24) be equal to 0, then: b 4 +d3b 3 +d2b 2 +d1β+d0=0 (26) in Step 3.2: Calculate the optimal power allocation factor set S1 = {β1, β2, β3, β4}, where and where y * is any real solution of the following equation: When R≠0, there exists: When R = 0, there exists: Step 3.3: Get the power allocation factor by using the Cardan formula: in 5. According to the CNN network-based active IRS-assisted uplink IoT system channel estimation method of claim 4, it is characterized in that: In step 4, the generated data set is split into 90% as a training data set and 10% as a validation data set.
6. According to the CNN network-based active IRS-assisted uplink IoT system channel estimation method of claim 5, it is characterized in that: In step 5, a channel estimation network model based on a convolutional neural network CNN is designed, the channel estimation neural network model is trained using a training data set, and the trained channel estimation neural network model is saved, as follows: Step 5.1, design a direct channel estimation network based on convolutional neural network CNN, which contains 7 layers, namely three convolutional layers, three maximum pooling layers and one linear layer; extract the spatial features of the channel matrix by combining convolution Conv operation and rectified linear unit ReLU; Batch normalization BN is introduced between Conv and ReLU, so the first, third and fifth layers are Conv+BN+ReLU operations; the second, fourth and sixth layers are maximum pooling layers, and the kernel size and stride are both set to 2; The seventh layer is a linear layer, which outputs the real and imaginary parts of the direct channel estimated by the mapping; Step 5.2, design a cascade channel estimation network based on convolutional neural network CNN, which contains 10 layers in total, among which the first, third, fifth, seventh, eighth and ninth layers are all convolutional layers, the second, fourth and sixth layers are all maximum pooling layers, and the last layer is a linear layer; Step 5.3: Deploy the designed direct channel estimation network and cascade channel estimation network on the BS side to obtain a channel estimation neural network based on deep learning, train the channel estimation neural network using the training data set, and save the trained channel estimation neural network.
7. The active IRS-assisted uplink IoT system channel estimation method based on CNN network according to claim 6 is characterized in that: In step 5.3, the channel estimation neural network is trained using the training data set, and the trained channel estimation neural network is saved, as follows: The initial learning rate of the direct channel estimation network is set to 0.001, and the initial learning rate of the cascade channel estimation network is set to 0.0025. The channel estimation neural network is trained using the training data set, and the learning rate is decayed by half every 35 cycles. The training process includes a total of 200 cycles. After 200 iteration cycles, the trained channel estimation neural network is saved, and the validation data set is used to verify the performance of the trained channel estimation neural network. During the training process, the AdamW optimization algorithm is used to update the weights of the model.
8. An active IRS-assisted uplink IoT system channel estimation device based on a CNN network, characterized in that: The device is used to implement the active IRS-assisted uplink IoT system channel estimation method based on a CNN network according to any one of claims 1 to 7, and the device includes a system model building module, an optimization problem construction module, a power allocation factor calculation module, a data set generation module, a model training module and a model testing module, wherein: System model building module, building the system model of active IRS-assisted uplink IoT network; An optimization problem construction module is used to construct an optimization problem for minimizing the power allocation factor and the mean square error (MSE) of the channel estimation between resource-constrained IoT devices and base stations BS. The power allocation factor calculation module uses the Cardan formula to solve the optimal power allocation factor between the IoT device and the BS. At this time, the channel estimation error of the system is the minimum channel estimation MSE; The data set generation module randomly generates a received signal data set based on the solved optimal power allocation factor, the known pilot signal and the active IRS phase shift matrix, and divides the generated data set into a training data set and a verification data set; Model training module, designs a channel estimation network model based on convolutional neural network CNN, uses training data sets to train the channel estimation neural network model, and saves the trained channel estimation neural network model; The model testing module re-randomly generates a receiving cascade link signal data set to test the trained channel estimation neural network model. The channel estimation neural network model that passes the test is used for channel estimation.
9. A mobile terminal comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the active IRS-assisted uplink IoT system channel estimation method based on the CNN network is implemented.
10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps in the active IRS-assisted uplink IoT system channel estimation method based on a CNN network are implemented.