An OFDM intelligent transmission method suitable for time-frequency dual-selective channels

Abstract

The present invention discloses an OFDM intelligent transmission method suitable for a time-frequency dual-selective channel. The method is for an OFDM system under a time-frequency dual-selective channel, and mainly includes the following steps: (1) generating a channel data set for training a neural network according to a time-frequency dual-selective channel model; (2) building a linear neural network, using the channel data generated in step (1) to train the network weights, and obtaining a transmitting and receiving equalization matrix for OFDM signal transmission, wherein the transmitting equalization matrix satisfies the transmitting end power constraint condition; (3) applying the pair of equalization matrices to the transmitting end and the receiving end of the OFDM system respectively; (4) the receiving end then uses the diagonal elements of the equivalent channel matrix to perform single subcarrier equalization, thereby restoring the transmission symbols on each subcarrier. The present invention can approach the optimal subcarrier joint linear equalization performance, and can significantly reduce the computational complexity of the optimal subcarrier joint linear equalization, which is conducive to engineering implementation.

Description

Technical Field

[0001] The present invention aims at an OFDM system under a time-frequency dual-selective channel and proposes an OFDM intelligent transmission method suitable for a time-frequency dual-selective channel, which significantly improves the performance of traditional single subcarrier equalization with lower calculation complexity. Background Art

[0002] For the dual-channel scenario of time and frequency selection, Doppler spread will cause mutual interference between OFDM subcarriers. If only traditional single subcarrier equalization is used at the receiving end, the system performance will be seriously degraded. To address this problem, multi-subcarrier joint equalization can be performed at the receiving end to eliminate the interference between subcarriers, such as the minimum mean square error (MMSE) equalization algorithm. Although this type of method can achieve excellent performance, the required computational complexity increases significantly as the number of OFDM subcarriers increases.

[0003] In recent years, the application of deep learning methods in the field of communications has been widely studied. Through the powerful optimization and fitting capabilities of neural networks, the performance of communication systems can be effectively improved. Compared with traditional methods, methods based on deep learning have the advantages of excellent performance, low online computational complexity, and strong robustness. Summary of the invention

[0004] In view of the defects of the above-mentioned prior art, the object of the present invention is to provide an OFDM intelligent transmission method suitable for time-frequency dual-selective channels, which can reduce the calculation complexity and approach the performance of MMSE equalization.

[0005] To achieve the above object, the present invention adopts the following technical solution:

[0006] An OFDM intelligent transmission method applicable to a time-frequency dual-selection channel comprises the following steps:

[0007] (1) Generate a channel data set for training a neural network based on a time-frequency dual-selection channel model;

[0008] (2) building a linear neural network, using the channel data generated in step (1) to train the network weights, and obtaining the transmit and receive equalization matrices for OFDM signal transmission, wherein the transmit equalization matrix satisfies the transmit power constraint;

[0009] (3) applying the pair of equalization matrices to the transmitting end and the receiving end of the OFDM system respectively;

[0010] (4) The receiving end then uses the diagonal elements of the equivalent channel matrix to perform single subcarrier equalization to restore the transmitted symbols on each subcarrier.

[0011] In the step (1), the channel data set is a set of channel matrices reflecting the fading characteristics of the dual-selection channel generated by the dual-selection channel model. where N is the number of subcarriers.

[0012] In step (2), the input of the linear neural network is the double-selection channel matrix H, and the output is the equivalent channel matrix H after the equalization matrix is ​​applied. eff =Q R HP T , where the transmit equalization matrix P T and the receiving equalization matrix Q R is a neural network trainable parameter and is obtained from the network weight after training is completed, and the emission equalization matrix P T Satisfy the transmitter power constraint, that is, in‖·‖ F represents the Frobenius norm, P max Indicates the maximum transmit power.

[0013] In step (2), the linear neural network is composed of two single-layer linear networks, and the weight of the single-layer linear network used for transmitter equalization is the transmitter equalization matrix P T The weight of the single-layer linear network used for receiving-end equalization is the receiving-end equalization matrix Q R , the output of the entire network is the equivalent channel matrix H eff =Q R HP T , participate in network training under unsupervised learning strategies.

[0014] In step (2), the following training objectives are used for network training:

[0015] The training objective is to minimize

[0016] in, represents the expectation of channel data, N is the number of subcarriers, |·| represents the complex modulus value, H k，k and H k，m Denote the equivalent channel matrix H eff The k-th diagonal element and the element in the k-th row and m-th column correspond to the channel gain on the k-th subcarrier of OFDM and the interference channel gain between the k-th subcarrier and the m-th subcarrier, respectively.

[0017] Beneficial effects: Compared with the prior art, the present invention has the following advantages:

[0018] 1. Compared with the traditional single subcarrier equalization method, the present invention achieves significant performance improvement with lower computational complexity.

[0019] 2. Compared with the multi-subcarrier joint MMSE equalization method, the present invention can achieve similar performance and significantly reduce the computational complexity.

[0020] 3. The neural network structure in the present invention is simple, which is conducive to engineering implementation. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 It is a flow chart of the method of the present invention.

[0022] Figure 2 It is a network structure diagram.

[0023] Figure 3 , Figure 4 It is the simulation experiment result diagram. DETAILED DESCRIPTION

[0024] The present invention is described in detail below with reference to the accompanying drawings and embodiments.

[0025] The typical application scenario of the present invention is single-user OFDM communication under a time-frequency dual-selective channel, and the design goal is to minimize the ratio of inter-subcarrier interference power to signal power, thereby eliminating the influence of inter-subcarrier interference.

[0026] like Figure 1 As shown, an OFDM intelligent transmission method applicable to a time-frequency dual-selection channel of the present invention comprises the following steps:

[0027] (1) Generate a channel data set for training a neural network based on a time-frequency dual-selection channel model;

[0028] The dual-selective channel model is a SISO channel model determined by the power delay spectrum under the 3GPP TDL-A model and the Doppler spectrum under the Clarke-Jakes model. The channel data set is a set of channel matrices generated by the above channel model that reflect the fading characteristics of the dual-selective channel. where N is the number of subcarriers.

[0029] (2) building a linear neural network, using the channel data generated in step (1) to train the network weights, and obtaining the transmit and receive equalization matrices for OFDM signal transmission, wherein the transmit equalization matrix satisfies the transmit power constraint;

[0030] The training objective is described as:

[0031] The training objective is to minimize

[0032] in, represents the expectation of channel data, N is the number of subcarriers, |·| represents the complex modulus value, H k，k and H k，m Denote the equivalent channel matrix H effThe k-th diagonal element and the element in the k-th row and m-th column correspond to the channel gain on the k-th subcarrier of OFDM and the interference channel gain between the k-th subcarrier and the m-th subcarrier, respectively.

[0033] The design scheme of the neural network structure is as follows: the frequency domain channel matrix H is used as input data, and the data is organized into a form in which the real part and the imaginary part are separated to facilitate network training; the linear neural network is composed of two single-layer linear networks, and the weight of the single-layer linear network used for transmitter equalization is the transmitter equalization matrix P T The weight of the single-layer linear network used for receiving-end equalization is the receiving-end equalization matrix Q R , the output of the entire network is the equivalent channel matrix H eff =Q R HP T , participate in network training under unsupervised learning strategy; after training, use the weights of the linear network to obtain the equilibrium matrix P T and Q R .

[0034] (3) applying the pair of equalization matrices to the transmitting end and the receiving end of the OFDM system respectively;

[0035] (4) The receiving end then uses the diagonal elements of the equivalent channel matrix to perform single subcarrier equalization to restore the transmitted symbols on each subcarrier.

[0036] In order to verify the technical effect of the present invention, a simulation experiment was conducted. Both layers of linear neural networks used 128 neurons, the corresponding subcarrier dimension was N=128, and the network weights were randomly initialized. The parameters involved in the simulation experiment are shown in the following table:

[0037] Table 1 Simulation experiment parameters

[0038] parameter Value Number of subcarriers 128 Subcarrier spacing 15kHz Number of symbols 1000 Modulation 16QAM Power Delay Spectrum TDL-A Delay Spread 100ns Doppler spectrum Jakes Normalized Doppler frequency deviation 0.04-0.2

[0039] In order to further illustrate the effect of the present invention, the present invention also simulates two comparative schemes of the prior art, including only the traditional single subcarrier equalization method and the multi-subcarrier joint MMSE equalization method, to compare the performance of the deep learning method.

[0040] Figure 3 The figure is a comparison result of the simulation experiment, where the horizontal axis is the signal-to-noise ratio (SNR) and the vertical axis is the bit error rate (BER). The simulation results show that the time-frequency dual-selective channel equalization method based on deep learning proposed in the present invention approaches the performance of multi-subcarrier joint MMSE equalization with lower complexity, and is significantly better than the commonly used single subcarrier equalization method.

[0041] Figure 4The comparison results of the simulation experiment are shown in Figure 1, where the horizontal axis is the normalized Doppler frequency deviation and the vertical axis is the bit error rate. The simulation results show that the time-frequency dual-selective channel equalization method based on deep learning proposed in the present invention has a certain robustness to Doppler spread.

[0042] 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. An OFDM intelligent transmission method suitable for time-frequency dual-selection channels, Features: The steps include: (1) Generate a channel data set for training a neural network based on a time-frequency dual-selection channel model; (2) building a linear neural network, using the channel data generated in step (1) to train the network weights, and obtaining the transmit and receive equalization matrices for OFDM signal transmission, wherein the transmit equalization matrix satisfies the transmit power constraint; (3) Apply the transmit and receive equalization matrices to the transmit and receive ends of the OFDM system respectively; (4) The receiving end then uses the diagonal elements of the equivalent channel matrix to perform single subcarrier equalization to restore the transmitted symbols on each subcarrier; In the step (1), the channel data set is a set of channel matrices reflecting the fading characteristics of the dual-selection channel generated by the dual-selection channel model. Composition, where N is the number of subcarriers; In the step (2), the input of the linear neural network is the double - selection channel matrix H, and the output is the equivalent channel matrix H after the action of the equalization matrix eff = Q R HP T , where the transmit equalization matrix P T and the receive equalization matrix Q R are trainable parameters of the neural network and are obtained from the network weights after training, and the transmit equalization matrix P T satisfies the transmit - end power constraint condition, that is where ‖·‖ F represents the Frobenius norm, and P max represents the maximum transmit power; In step (2), the linear neural network is composed of two single-layer linear networks, and the weight of the single-layer linear network used for transmitter equalization is the transmitter equalization matrix P T The weight of the single-layer linear network used for receiving-end equalization is the receiving-end equalization matrix Q R , the output of the entire network is the equivalent channel matrix H eff =Q R HP T , participate in network training under unsupervised learning strategy, and the double-selected channel matrix H is the input of the linear neural network; In step (2), the following training objectives are used for network training: The training objective is to minimize Among them, denotes taking the expectation of the channel data, N is the number of subcarriers, |·| represents the complex modulus, and H k,k and H k,m respectively denote the k-th diagonal element and the element in the k-th row and m-th column of the equivalent channel matrix H eff corresponding to the channel gain on the k-th subcarrier of OFDM and the interference channel gain between the k-th subcarrier and the m-th subcarrier, respectively.

Images