A DNN-based OTFS time-frequency domain situation awareness and channel estimation method

Abstract

A DNN-based OTFS time-frequency domain situation awareness and channel estimation method models the frequency domain selective fading channel in wireless communication as an autoregressive process, and then converts the estimation problem into an autoregressive coefficient estimation process. Training data is obtained through processes such as discrete sampling of the channel impulse response function, DFT transformation, feature matrix representation, and channel prior information generation. The preprocessing of the training data is completed, and a data and model driven neural network is designed. The neural network is trained with a large amount of prior channel data. The optimal frequency domain correlation parameter estimation under the minimum mean square error of the network is solved by offline training and online estimation, and then the corresponding channel response is obtained. The algorithm can achieve lower bit error rate and complexity under the same signal-to-noise ratio.

Description

Technical Field

[0001] This invention belongs to the field of wireless communication, specifically relating to a DNN-based OTFS time-frequency domain situational awareness and channel estimation method, which can be applied to the future development of wireless communication technology and to ensure communication transmission for highly dynamic targets. Background Technology

[0002] With the development of wireless communication technology, the demand for more efficient, reliable, and higher-quality communication has driven the development of Orthogonal Frequency Division Multiplexing (OFDM). This technology has achieved high-speed, low-latency wireless communication to a certain extent, while also possessing good anti-multipath interference capabilities to ensure reliable data transmission. However, in high-speed mobile scenarios, the multipath effect of OFDM technology not only reduces the system's bit error rate but also increases the complexity of the receiver. Therefore, based on the requirements of 6G communication, Orthogonal Time Frequency Space (OTFS) technology has been proposed. It can achieve highly robust communication in environments with high multipath and Doppler effects, significantly reducing signal distortion and inter-symbol interference while reducing the complexity of the communication system, exhibiting a superior performance-complexity tradeoff.

[0003] However, OTFS technology still faces many problems that need to be solved. Channel estimation and pilot design in OTFS systems are particularly challenging, for example, high pilot overhead, high requirements for the signal-to-noise ratio of pilot signals, and high computational complexity. The sparsity characteristics and time correlation characteristics of the time-delay-Doppler domain channel are not fully utilized in current channel estimation schemes, and traditional OTFS channel estimation methods have low spectral efficiency, making them unsuitable for communication scenarios with high Doppler frequency offsets, such as low-Earth orbit satellites. Therefore, research on channel estimation and signal detection schemes for satellite OTFS communication systems is essential. This invention proposes a DNN-based OTFS time-frequency situational awareness and estimation method. By designing a data- and model-driven neural network, and iteratively training the network multiple times to solve for the optimal frequency domain correlation coefficient under the minimum mean square error, the corresponding channel response is calculated, thereby achieving better channel estimation performance.

[0004] Based on publicly available literature on OTFS channel estimation from both domestic and international sources, the existing research findings mainly focus on the following aspects: Literature such as "Pilot Sequence Design and Channel Estimation Based on Orthogonal Time-Frequency-Space (OTFS) Systems" proposes a channel estimation algorithm based on the autocorrelation characteristics of pilot sequences. The performance of this algorithm is related to the length of the PN sequence; only a longer PN sequence can guarantee high channel estimation accuracy. However, a longer sequence length leads to higher computational complexity, and this scheme is not suitable for communication scenarios where channel states change rapidly. The literature, such as "A novel channel estimation scheme for OTFS", proposes a threshold-based estimation algorithm. This algorithm mainly optimizes channel estimation performance by setting one or more thresholds, aiming to handle channel estimation problems under low signal-to-noise ratio conditions. It sets up Doppler domain transmission pulse pilots at the transmitter and performs pulse detection at the receiver based on the threshold method. The time delay-Doppler domain channel is estimated by using the Doppler domain offset, pulse amplitude and number. However, this method has large pilot overhead and low spectrum utilization. The literature, such as "Research on OTFS channel estimation algorithm in high-speed mobile communication system", proposes an estimation method based on compressed sensing. It uses the compressed sensing principle to estimate the time delay, Doppler frequency shift and channel gain of each path, and then determines the size of the pilot transmission matrix to improve the accuracy of channel estimation and diversity order. However, this method only considers integer Doppler frequency shift and does not consider the impact of fractional Doppler frequency shift on the OTFS channel estimation performance.

[0005] With the rapid development of deep learning, related technologies have been applied to communication fields such as modulation recognition and channel equalization. The "OTFS channel estimation based on model-driven deep learning" uses a model-driven deep learning LDAMP algorithm to estimate the OTFS time-delay Doppler channel. The "OTFS system channel estimation method based on CNN in satellite-ground scenario" uses a convolutional neural network to process the OTFS system in an end-to-end manner. The "OTFS system signal detection method based on DNN" uses a deep learning method to estimate CSI and directly recover the transmitted symbols. However, these methods have poor performance in frequency-selective fading environments and directly estimate the received signal, which to some extent reduces transmission efficiency and increases complexity. At present, there is no OTFS time-frequency domain channel estimation algorithm that balances high performance and low complexity to achieve highly reliable communication transmission. Therefore, the method proposed in this invention has obvious novel features. Summary of the Invention

[0006] This invention proposes a DNN-based OTFS time-frequency domain situational awareness and channel estimation method. This method models the frequency-selective fading channel in wireless communication as an autoregressive process, and then transforms the estimation problem into an autoregressive coefficient estimation process. A data- and model-driven neural network is designed, and the neural network is trained with a large amount of prior channel data. The optimal parameter estimation under the minimum mean square error of the network is solved through multiple iterations of training, thereby achieving better estimation performance.

[0007] The technical solution of this invention is: a DNN-based OTFS time-frequency domain situational awareness and channel estimation method, comprising the following steps:

[0008] 1) Obtain preprocessed training data through processes such as discretization sampling, DFT transformation, feature matrix representation, and generation of channel prior information based on the channel impulse response function;

[0009] 2) Design a neural network model and obtain the optimal frequency domain correlation coefficient based on offline training;

[0010] 3) The channel response is estimated using the LS algorithm, and the optimal frequency domain channel response is obtained based on online estimation;

[0011] 4) The received signal sequence is obtained by using the least squares method. After demodulation and decoding, it is compared with the information sequence of the transmitting end to obtain the bit error rate curve.

[0012] Furthermore, in step 1), training data is acquired and preprocessed according to the OTFS system model;

[0013] a) Taking an OTFS symbol as an example, assuming the channel response remains unchanged within the symbol, the channel impulse response is expressed with a sampling interval T. s Discretize into g i (n) is represented as

[0014]

[0015] In the above formula, α l Let n be the average gain of the channel on the l-th path; neglecting inter-symbol interference in the system, then n l Let n be the channel delay on the l-th path, satisfying n l =[τ l / T s ].

[0016] b) Assume that an OTFS symbol has N subcarriers, N p After uniformly inserting pilot subcarriers, performing an N-point DFT transform on the discrete impulse response of the system yields the channel frequency domain response on the k-th subcarrier as follows:

[0017]

[0018] c) The channel response vector on each pilot subcarrier within one OTFS symbol time can be obtained as follows:

[0019] h = [h(1), ..., h(N)] p )] T

[0020] d) Based on the channel response vector, first obtain the eigenvalues ​​D of h. N×1 and the corresponding column eigenvector matrix U N×N The eigenvalues ​​are sorted in descending order to obtain D' N×1 The corresponding feature vectors are arranged by index to obtain U' N×N Taking the first cp eigenvalues, the representation matrix is ​​obtained as follows:

[0021] if k≤cp delta(k)=D'(k) / (D'(k)+β / Noise);

[0022] else delta(k) = 0;

[0023] e) Generate channel prior information using the channel representation matrix as training data for the DNN, as shown in the following expression:

[0024] Ch_feature=U'*diag(delta)*U' T

[0025] Furthermore, step 2) involves obtaining channel estimation parameters based on a neural network training model.

[0026] A deep neural network (DNN) is a complex artificial neural network structure composed of multiple layers of neurons. Each neuron is responsible for receiving input, processing it, and generating output. This situational awareness and estimation process mainly consists of two parts: offline training and online prediction.

[0027] During the offline training phase, the CSI at the pilot signal and the CSI at one OTFS symbol are used as training data. The CSI at the pilot signal is used as the input data of the DNN, and the CSI at one OTFS symbol is used as the label data to train the DNN.

[0028] This neural network mainly consists of an input layer, a hidden layer, and an output layer.

[0029] 1) Use the CFR vector at the pilot obtained by LS estimation as the input data, i.e.

[0030]

[0031] 2) Preprocess the input data by extracting the real and imaginary parts of each CFR vector component and concatenating them together to obtain the actual input data.

[0032] 3) The nonlinear transformation of the weighted sum of all data from the previous layer is used as the output of the hidden layer's neurons. The transformation expression is:

[0033]

[0034] Among them, l l,i w l Let b and b be the output, weights, and biases of the i-th neuron in the l-th hidden layer, respectively. T () represents the activation function, primarily the Tanh activation function, with the following expression:

[0035]

[0036] 4) The transformation of a neural network can be expressed by the following cascaded mathematical expression:

[0037]

[0038] Where L represents the number of layers in the neural network, and θ represents all the parameters involved in the network. This represents the result obtained from channel estimation.

[0039] 5) Use the Y and channel estimation results output by the OTFS communication system. The received sequence r is obtained based on the least squares (LS) channel estimation algorithm. n :

[0040]

[0041] Furthermore, in step 2), the frequency domain correlation coefficient tmps is obtained through offline training of a DNN neural network. n ;

[0042] The DNN neural network structure is designed with k hidden layers and 1 output layer. Assuming the input is Ptrain, the training phase of the neural network can be represented as follows:

[0043] tmps1 = Ptrain * W1 + B1

[0044] tmps i =tmps i-1 *W i-1 +BI i-1

[0045] Where i = 2, 3, 4, ..., n, the loss during training is obtained as error = Ptrain - tmps. n Used to update W i and B.I. i With equal weights and a learning rate Lr, the expression is as follows:

[0046] W i =W i +Lr*error

[0047] BI i =BI i +Lr*error

[0048] The iteration count is set to epochs, and the final output layer expression is as follows:

[0049] WO = WO + Lr * error;

[0050] BO = BO + Lr * error;

[0051] Therefore, the frequency domain correlation coefficient can be calculated as tmps n =tmps n-1 *WO+BO.

[0052] Furthermore, in step 3) the online estimation stage, the frequency domain channel response H is obtained using the channel response and the frequency domain correlation coefficient. tmp .

[0053] 1) The initial estimate of H can be obtained using the pilot sequence and LS. in =r_pilot / Data_pilot, to obtain the channel response H in ;

[0054] 2) Using the frequency domain correlation coefficient tmps6 obtained through deep learning, the channel response H is calculated. tmp =H in *tmps6.

[0055] 3) The received signal sequence r is obtained by least squares (LS) estimation. n for

[0056] r n =Y*H tmp

[0057] For the received signal r n QPSK demodulation and Turbo decoding are performed to obtain the received signal sequence R. n By comparing the information sequences of the receiver and transmitter, the bit error rate curve and algorithm running time under this situational awareness and estimation algorithm are obtained.

[0058] The advantages of this invention compared to the prior art are:

[0059] (1) This technology proposes an OTFS time-frequency domain situational awareness and channel estimation method based on DNN. This method obtains training data through processes such as discretization sampling based on the channel impulse response function, DFT transformation, feature matrix representation, and generation of channel prior information, rather than directly using pilot data or data after simple normalization as training data. This method can ensure good channel estimation performance with low complexity.

[0060] (2) The proposed DNN-based OTFS time-frequency domain situational awareness and channel estimation method is mainly divided into offline training and online estimation processes. In the offline training stage, channel data obtained by pilot sequence processing is used as training data. In the online estimation stage, the channel response is obtained through the trained channel estimation model. Finally, the least squares estimation algorithm is used to obtain the received sequence signal instead of directly estimating the received signal. This technology can achieve lower bit error rate and complexity in communication under the same signal-to-noise ratio conditions. Attached Figure Description

[0061] Figure 1 This is a flowchart of the present invention.

[0062] Figure 2 It is a neural network channel estimation process.

[0063] Figure 3 This invention compares the information sequences of the receiver and transmitter of LS estimation, MMSE estimation, LMMSE estimation, and DNN-based time-frequency domain situational awareness and estimation algorithms to obtain bit error rate curves under different channel estimation algorithms. Detailed Implementation

[0064] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments.

[0065] (1) Based on the constructed OTFS system model, acquire training data and perform preprocessing simultaneously;

[0066] a) Taking an OTFS symbol as an example, assuming the channel response remains unchanged within the symbol, the channel impulse response is expressed with a sampling interval T. s Discretize into g i (n) is represented as

[0067]

[0068] In the above formula, α l Let n be the average gain of the channel on the l-th path; neglecting inter-symbol interference in the system, then n l Let n be the channel delay on the l-th path, satisfying n l =[τ l / T s ].

[0069] b) Assume that the number of subcarriers N in an OTFS symbol is 64, and the number of pilot subcarriers N p Setting it to 4, after uniformly inserting pilot signals, performing an N-point DFT transform on the discrete impulse response of the system yields the channel frequency domain response on the k-th subcarrier.

[0070]

[0071] c) The channel response vector on each pilot subcarrier within one OTFS symbol time can be obtained as follows:

[0072] h = [h(1), ..., h(N)] p )] T

[0073] d) Based on the channel response vector, first obtain the eigenvalues ​​D of h. N×1 and the corresponding column eigenvector matrix U N×N The eigenvalues ​​are sorted in descending order to obtain D' N×1 The corresponding feature vectors are arranged by index to obtain U' N×N Taking the first cp eigenvalues, the representation matrix is ​​obtained as follows:

[0074] if k≤cp delta(k)=D'(k) / (D'(k)+β / Noise);

[0075] else delta(k) = 0;

[0076] e) Generate channel prior information using the channel representation matrix as training data for the DNN, as shown in the following expression:

[0077] Ch_feature=U'*diag(delta)*U' T

[0078] (2) The frequency domain correlation coefficient tmps6 was obtained by training a DNN neural network;

[0079] The DNN neural network structure is designed with 4 hidden layers and 1 output layer. Assuming the input training data is Ptrain2, the training phase of the neural network can be represented as follows:

[0080] tmps1 = Ptrain2 * W1 + B1

[0081] tmps i =tmps i-1 *W i-1 +BI i-1

[0082] Where i = 2, 3, 4, 5, 6, the loss during training, error = Ptrain2 - tmps6, is obtained and used to update W. i and B.I. i With equal weights and a learning rate Lr = 0.005, the expression is as follows:

[0083] W i =W i +Lr*error

[0084] BI i =BI i +Lr*error

[0085] The iteration count is set to epoch = 2000, and the final output layer expression is as follows:

[0086] WO = WO + Lr * error;

[0087] BO = BO + Lr * error;

[0088] Therefore, the frequency domain correlation coefficient can be calculated as tmps6 = tmps5 * WO + BO.

[0089] In step 3) the online estimation stage, the frequency domain channel response H is obtained using the channel response and the frequency domain correlation coefficient. tmp .

[0090] 1) The initial estimate of H can be obtained using the pilot sequence and LS. in =r_pilot / Data_pilot, to obtain the channel response H in ;

[0091] 2) Using the frequency domain correlation coefficient tmps6 obtained through deep learning, the channel response H is calculated. tmp =H in *tmps6.

[0092] 3) The received signal sequence r is obtained by least squares (LS) estimation. n for

[0093] r n =Y*H tmp

[0094] This invention is mainly verified using simulation data, and all steps and conclusions have been verified to be correct on Matlab.

[0095] By comparing the information sequences of the receiver and transmitter from LS estimation, MMSE estimation, LMMSE estimation, and the DNN-based time-frequency domain situational awareness and estimation algorithm, the bit error rate curves under different channel estimation algorithms are obtained. (See [link to relevant documentation]). Figure 3 .

[0096] Table 1 Comparison of running times of different algorithms

[0097] algorithm LS channel estimation MMSE channel estimation IMMSE Channel Estimation DNN-based channel estimation Runtime / s 176.796548 1172.333143 968.738319 930.13672

[0098] Bit error rate (BER) reflects the performance of channel estimation; a lower BER indicates a more accurate estimated received sequence and better channel estimation algorithm performance. Algorithm running time, to some extent, reflects algorithm complexity; a shorter running time indicates lower algorithm complexity.

[0099] Figure 3 The performance of LS estimation, MMSE estimation, IMMSE estimation, and DNN-based time-frequency domain situational awareness and estimation algorithms were compared with Table 1. It can be seen that although the LS estimation algorithm has low computational complexity, its estimation performance is poor, resulting in reduced anti-interference capability. The MMSE and IMMSE estimation algorithms have good performance, but they require prior channel information and have high computational complexity. The DNN-based time-frequency domain situational awareness and estimation algorithm has performance between the MMSE and IMMSE algorithms, but its computational complexity is significantly reduced; when SNR ≥ 6dB, the bit error rate can reach 10%. -4 It strikes a good balance between algorithm complexity and estimation performance, and has good anti-interference ability.

[0100] The contents not described in detail in this specification are common knowledge to those skilled in the art.

Claims

1. A time-frequency domain situational awareness and channel estimation method based on DNN-based OTFS, characterized in that, include: 1) Obtain preprocessed training data through discretization sampling, DFT transformation, feature matrix representation, and channel prior information generation based on the channel impulse response function; 2) Design a neural network model and obtain the optimal frequency domain correlation coefficient based on offline training; 3) The channel response is estimated using the LS algorithm, and the optimal frequency domain channel response is obtained based on online estimation; 4) The received signal sequence is obtained using the least squares method, demodulated and decoded, and then compared with the information sequence at the transmitting end to obtain the bit error rate curve; Step 1) involves acquiring training data and preprocessing it based on the OTFS system model. The specific process is as follows: a) Divide the channel impulse response into sampling intervals Discretize into g l (n) is represented as: g l (n) represents the channel impulse response at different sampling times; For the first The average gain of the channel along the path; ignoring inter-symbol interference in the system, then For the first The channel delay on each path satisfies ; b) Assume that a OTFS symbol contains a total of Subcarriers, After uniformly inserting the pilot subcarriers, perform an N-point DFT transform on the discrete impulse response of the system to obtain the nth pilot subcarrier. The channel frequency domain response on each subcarrier is: c) The channel response vectors on each pilot subcarrier within one OTFS symbol time are obtained as follows: d) Based on the channel response vector, first obtain eigenvalues and the corresponding column eigenvector matrix , eigenvalue Sort in descending order to get The corresponding column feature vector Arranged by index Take the front With eigenvalues, the representation matrix is ​​obtained as follows: e) Generate channel prior information using the channel representation matrix as training data for the DNN, as shown in the following expression: ； Step 2) uses a DNN to design a neural network for offline training to obtain the frequency domain correlation coefficient. ; The DNN neural network structure is designed as follows: k hidden layers and 1 output layer. Assume the input... The training phase of a neural network can be represented as follows: in Loss during training Used for updating and Equal weights, set learning rate The expression is as follows: The number of iterations is set to The final output layer expression is as follows: Therefore, the frequency domain correlation coefficient can be calculated as follows: .

2. The DNN-based OTFS time-frequency domain situational awareness and channel estimation method according to claim 1, characterized in that, In step 3) the online estimation stage, the frequency domain channel response is obtained using the channel response and the frequency domain correlation coefficient. The calculation method is as follows: 1) Initial estimation using LS can be obtained from the pilot sequence. The channel response is obtained. ; 2) Frequency domain correlation coefficient obtained using deep learning The channel response is calculated. ; 3) Obtain the received signal sequence using least squares (LS) estimation. for: 。

Images