Channel prediction method and device suitable for non-ground network

Abstract

The invention discloses a channel prediction method and device suitable for a non-ground network, and relates to the technical field of artificial intelligence, and the method comprises the steps: S1, user equipment located on the ground carries out uplink communication with a gNodeB located on an LEO satellite; s2, the CP-OFDM waveform carrying the resource grid is sent by the user equipment and is propagated through a frequency selective channel; s3, at a receiving end, removing the cyclic prefix from the signal, and performing demultiplexing on the received symbol; and S4, for the estimation value of the CFR matrix, applying least square channel estimation on the received pilot symbol, and performing interpolation in the time slot duration. According to the channel prediction method and device suitable for the non-ground network, channel estimation can be not affected by removing pilot frequency every other OFDM time slot, the uplink peak throughput can be remarkably improved regardless of the data modulation order in most cases, and the performance is hardly affected.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and specifically provides a channel prediction method and device applicable to non-terrestrial networks. Background Art

[0002] Channel prediction is a method of obtaining information about the propagation channel state when the channel state cannot be actually estimated. In a communication system, obtaining perfect CSI is crucial for the communication system. If the transmitted CSI becomes potentially outdated due to channel fluctuations, then channel prediction becomes important, and the predicted CSI needs to be sent to the transmitter to improve the performance of the MIMO system. Channel prediction helps improve the performance of various operations performed at the transmitter or receiver, such as adaptive coding and modulation, decoding processing, channel equalization, and antenna beamforming.

[0003] Such applications are suitable for non-terrestrial networks (NTN), which is one of the 6G wireless communication standards. The typical characteristics are dynamic channel conditions, including time-varying delay and Doppler frequency shift. Channel prediction methods are usually used to cope with the channel aging effect, that is, due to changes in propagation conditions, such as the movement of LEO satellites, etc., the accuracy of the estimated channel information decreases, hindering the performance of the satellite communication system.

[0004] For general research, most of the work on channel prediction uses MSE as an indicator, while ignoring the impact on key communication indicators. In this case, for the 6G NTN system, orthogonal frequency division multiplexing OFDM is used as the waveform, and a convolutional-based coding and decoding neural network is used to predict the channel frequency response matrix of the upcoming OFDM from the current estimated value. This case provides a lightweight channel prediction method based on CNN and LSTM, which uses a convolutional-based encoder-decoder structure and predicts hidden features through an LSTM layer. Summary of the Invention

[0005] The purpose of the present invention is to provide a channel prediction method and device applicable to non-terrestrial networks to solve the problems raised in the above background art.

[0006] To achieve the above purpose, the present invention provides the following technical solutions:

[0007] In the first aspect, a channel prediction method applicable to non-terrestrial networks is provided, including the following steps:

[0008] S1. The user equipment located on the ground communicates with the gNodeB on the LEO satellite, and the transmission link of the equipment uses a low-density parity-check (LDPC) channel encoder, a bit interleaver, an M-ary quadrature amplitude modulation mapper (M-QAM), and a CP-OFDM multiplexer;

[0009] S2. The CP-OFDM waveform carrying the resource grid is sent by the user equipment and propagates through a frequency-selective channel, which is modeled as the channel impulse response of a TDL with [number of paths] paths:

[0010]

[0011] wherein, ∈D~N(0,σ D ) is the residual frequency synchronization error component, which is a Gaussian distribution N with an expectation of 0 and a standard deviation of, h n (t) and τ n represent the value and delay of the nth tap, where τ 1 = 0, and δ(·) represents the Dirac function;

[0012] S3. At the receiving end, the additive white Gaussian noise (AWGN) interferes with the waveform, so the cyclic prefix is removed from the signal and the received symbols are demultiplexed;

[0013] S4. For the estimated value of the CFR matrix Apply the least squares channel estimation to the received pilot symbols, and the received pilot symbols can be expressed as:

[0014]

[0015] The estimated value calculated in this way is interpolated within the slot duration to cover the time span of the resource grid.

[0016] Furthermore, in step S1, the user equipment is equipped with a 6G transceiver and has different degrees of mobility, and the equipment can pre-compensate for the Doppler frequency shift and track the Doppler rate change caused by the mobility of the considered NTN node, and only the residual frequency synchronization error that affects the received cyclic prefix orthogonal frequency division multiplexing (CP-OFDM) waveform exists.

[0017] Furthermore, in step S2, for a user equipment, N sc -dimensional subcarriers are used for uplink transmission, and the number of subcarriers is N sym , and the dimension of the entire transmission-side resource grid X is (N sc ×N sym ).

[0018] Further, in step S2, in one time slot, M π QAM pilot symbols are inserted into each subcarrier, and the position index is denoted in the resource grid of OFDM as indicating that the number of pilot symbols of OFDM in each time slot is 7;

[0019] And every other time slot, all pilot symbols are removed to reduce the pilot overhead, that is, half of the time slots only contain data symbols, so there is a limitation.

[0020] Further, in step S3, the CFR matrix is represented by H = FFT[h(t,τ)], where FFT represents the fast Fourier transform, and the OFDM resource grid at the receiving end can be represented as:

[0021] Y = H⊙X + W;

[0022] In the formula, W represents the influence of AWGN on the OFDM grid, and N 0 represents the noise power spectral density, and ⊙ is the Hadamard multiplication operator of the matrix.

[0023] Further, in step S4, for the selected system model, the interpolation not only covers the data symbol positions in the first time slot, but also covers all symbol positions in the second time slot, and there are no pilots in the second time slot. Channel prediction is used to improve the accuracy of channel estimation;

[0024] The execution logic of step S4: First, use the obtained channel estimation value to equalize and demap the data symbols in the first time slot; then remap the obtained data bits to M-QAM symbols, and these symbols are used to perform the least squares method on the entire first time slot, so as to form an overall data and pilot estimation scheme.

[0025] Further, the demapping error will lead to inaccurate quantization of the estimated channel matrix. For example, an error on a 4-QAM data symbol may cause a phase shift of the corresponding channel estimation. Specifically, a channel predictor is used to identify and equalize such errors, and to predict the channel;

[0026] After the entire resource grid is equalized through the estimated or predicted channel coefficients, the data symbols are demapped, and then the obtained data is decoded.

[0027] Further, the main structure of the channel predictor includes three parts: an encoding layer, a prediction layer, and a decoding layer. Among them, Flatten is a custom layer that flattens the tensor on the frequency axis, with each frequency as a dimension of the tensor. TimeMirror is a custom layer that mirrors the generated tensor on the time axis and arranges it by time to ensure that the first prediction result depends on the latest sample rather than the subsequent samples, thereby improving the prediction accuracy. The Relu layer is an activation function with the formula f(x) = max(0, x). The BatchNorm layer is a normalization operation, and the units of the LSTM are configured as 32.

[0028] In a second aspect, a channel prediction device applicable to non-terrestrial networks is provided, which is applied to the above-mentioned channel prediction method applicable to non-terrestrial networks. The device includes: a memory, a processor, and computer program instructions stored on the memory and executable on the processor. When the processor executes the computer program instructions, the above-mentioned channel prediction method applicable to non-terrestrial networks is implemented.

[0029] In a third aspect, a computer-readable storage medium is provided, which stores a computer program that can be loaded and executed by a processor to implement the above-mentioned channel prediction method applicable to non-terrestrial networks.

[0030] The present invention provides a channel prediction method and device applicable to non-terrestrial networks, having the following beneficial effects:

[0031] The present invention can achieve removing pilots every other OFDM time slot without affecting channel estimation. In most cases, regardless of the data modulation order, it can significantly improve the uplink peak throughput, and the performance is hardly affected. When the speed is limited within a certain range, the selected neural network is very stable in the performance between the training set and the test set user equipment, and shows good scalability. If it is implemented by hardware in the future and extended to multiple input-output OFDM time slots, it is expected to obtain more accurate prediction performance and has the potential to further reduce pilot signals. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 It is a schematic diagram of the overall solution of a channel prediction method applicable to non-terrestrial networks of the present invention;

[0033] Figure 2 It is a schematic diagram of the main structure of the channel predictor of a channel prediction method applicable to non-terrestrial networks of the present invention;

[0034] Figure 3 It is a parameter configuration diagram of the convolutional layer of a channel prediction method applicable to non-terrestrial networks of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0035] The following further describes in detail the implementation manners of the present invention in conjunction with the accompanying drawings and embodiments. The following embodiments are used to illustrate the present invention, but cannot be used to limit the scope of the present invention.

[0036] As Figures 1 - 3 shown, a channel prediction method applicable to non-terrestrial networks includes the following steps:

[0037] S1. The user equipment located on the ground communicates with the gNodeB on the LEO satellite through the uplink, and the transmission link of the equipment uses a low-density parity-check (LDPC) channel encoder, a bit interleaver, an M-ary quadrature amplitude modulation mapper (M-QAM), and a CP-OFDM multiplexer.

[0038] In this embodiment, the user equipment is equipped with a 6G transceiver and has different degrees of mobility. The equipment can pre-compensate for the Doppler frequency shift and track the Doppler rate change caused by the mobility of the considered NTN node. There is only a residual frequency synchronization error that affects the received cyclic prefix orthogonal frequency division multiplexing (CP-OFDM) waveform.

[0039] S2. The N sc -dimensional subcarriers of the user equipment are used for uplink transmission, and the number of subcarriers is N sym , and the dimension of the entire transmission-side resource grid X is (N sc × N sym ); in one time slot, M π QAM pilot symbols are inserted into each subcarrier, and the position index is denoted as in the resource grid of OFDM It means that the number of pilot symbols of OFDM in each time slot is 7. In this embodiment, all pilot symbols are removed every other time slot to reduce the pilot overhead, that is, half of the time slots only contain data symbols, so there is a limitation. The CP-OFDM waveform carrying the resource grid is sent by the user equipment and propagates through a frequency-selective channel, and the channel is modeled as the channel impulse response of a TDL with paths:

[0040]

[0041] In the formula, ∈ D ∼ N(0, σ D ) is the residual frequency synchronization error component, which is a Gaussian distribution N with an expectation of 0 and a standard deviation of, h n (t) and τ n represent the value and delay of the nth tap, where τ 1 = 0, and δ(·) represents the Dirac function (for the specific function, refer to the following website:

[0042] https: / / baike.baidu.com / item / %E7%8B%84%E6%8B%89%E5%85%8B%E5%B8%83%E4%B8%8E%E6%95%B0 / 5760582).

[0043] S3. At the receiving end, additive white Gaussian noise (AWGN) interferes with the waveform. Therefore, the cyclic prefix is removed from the signal, and the received symbols are demultiplexed. Let H = FFT[h(t,τ)] represent the CFR matrix, where FFT represents the fast Fourier transform. The OFDM resource grid at the receiving end can be expressed as:

[0044] Y = H⊙X + W;

[0045] In the formula, W represents the influence of AWGN on the OFDM grid, and N 0 represents the noise power spectral density, and ⊙ is the Hadamard multiplication operator of the matrix.

[0046] S4. For the estimated value of the CFR matrix Apply least squares channel estimation to the received pilot symbols. The received pilot symbols can be expressed as:

[0047]

[0048] The estimated value calculated in this way is interpolated within the slot duration to cover the time span of the resource grid. For the selected system model, the interpolation covers not only the data symbol positions in the first slot but also all the symbol positions in the second slot, where there are no pilots. Here, we use channel prediction to improve the accuracy of channel estimation.

[0049] Generally speaking, first, use the obtained channel estimation value to equalize and demap the data symbols in the first slot; then remap the obtained data bits to M-QAM symbols, and these symbols are used to perform the least squares method on the entire first slot, thus forming an overall data and pilot estimation scheme.

[0050] Demapping errors will lead to inaccurate quantization of the estimated channel matrix. For example, errors in 4-QAM data symbols may cause phase shifts in the corresponding channel estimation. Therefore, the channel predictor not only has to predict the channel but also identify and equalize such errors. After the entire resource grid is equalized through the estimated or predicted channel coefficients, the data symbols are demapped, and then the obtained data is decoded.

[0051] The main structure of the channel predictor adopted in this embodiment is as Figure 2As shown, it mainly includes three parts: an encoding layer, a prediction layer, and a decoding layer;

[0052] Among them, Flatten is a custom layer that flattens the tensor on the frequency axis, with each frequency as a dimension of the tensor. TimeMirror is a custom layer that mirrors the generated tensor on the time axis and arranges it by time to ensure that the first prediction result depends on the latest sample rather than the subsequent samples, thereby improving the prediction accuracy. The Relu layer is an activation function, and the BatchNorm layer is a normalization operation. The main formula of BatchNorm can be referred to at https: / / zhuanlan.zhihu.com / p / 168791054. The units of LSTM are configured as 32, and the main parameter configuration of the convolutional layer is as Figure 3 shown;

[0053] The main formula of the Relu activation function is:

[0054] f(x) = max(0, x);

[0055] Figure 1 In, the input of the channel prediction structure will first be normalized, and then the real part and the imaginary part need to be taken as a dimension respectively and merged into a tensor, and the formed tensor dimension is Adopting such an expression method makes the training performance similar whether using Cartesian or polar coordinates.

[0056] Figure 1 In, the output of the channel prediction structure The output tensor structure is the same as and is also The real part and the imaginary part are taken as a dimension respectively.

[0057] The loss function of the network adopts the mean squared error function, and the optimization function adopts Adam, which are both commonly used in machine learning and will not be described additionally.

[0058] The equalizer used for equalization adopts a feed-forward equalizer (FFE) (specific reference: https: / / zhuanlan.zhihu.com / p / 652650099).

[0059] A channel prediction device applicable to non-terrestrial networks, which is applied to the above-mentioned channel prediction method applicable to non-terrestrial networks, and is characterized in that the device includes: a memory, a processor, and computer program instructions stored on the memory and executable on the processor, and when the processor executes the computer program instructions, it implements the above-mentioned channel prediction method applicable to non-terrestrial networks.

[0060] A computer-readable storage medium, characterized in that it stores a computer program that can be loaded and executed by a processor to perform the channel prediction method applicable to non-terrestrial networks as described above.

[0061] The embodiments of the present invention are given for purposes of illustration and description, and are not exhaustive or limit the present invention to the disclosed form. Many modifications and variations are obvious to those of ordinary skill in the art. The embodiments are chosen and described in order to better illustrate the principles and practical applications of the present invention, and to enable those of ordinary skill in the art to understand the present invention and design various embodiments with various modifications suitable for a particular purpose.

Claims

1. A channel prediction method applicable to non-terrestrial networks, characterized in that: The following steps are involved: S1, the user equipment located on the ground performs uplink communication with the gNodeB located on the LEO satellite, and the transmission link of the equipment adopts a low-density parity check channel encoder, a bit interleaver, an M-ary orthogonal amplitude modulation mapper and a CP-OFDM multiplexer; S2. The CP-OFDM waveform carrying the resource grid is sent by the user equipment and propagates through a frequency selective channel, which is modeled as a channel impulse response of a TDL with 1 path: In the formula, ∈D~N(0,σ D ) is the residual frequency synchronization error component, which is a Gaussian distribution with an expected value of 0 and a standard deviation of N,h n (t) and τ n represents the value and delay of the nth tap, where τ1 = 0 and δ(·) represents the Dirac function; S3. At the receiving end, the cyclic prefix is ​​removed from the signal and the received symbols are demultiplexed; S4. Estimated values ​​for the CFR matrix Applying least squares channel estimation on the received pilot symbols, the received pilot symbols It is expressed as: The estimates so calculated are interpolated over the slot duration to cover the time span of the resource grid.

2. A channel prediction method applicable to non-terrestrial networks according to claim 1, characterized in that: In step S1, the user equipment is equipped with a 6G transceiver and has different degrees of mobility, and the equipment is able to pre-compensate for Doppler frequency shift and track the Doppler rate changes caused by the mobility of the considered NTN node, with only a residual frequency synchronization error affecting the received cyclic prefix orthogonal frequency division multiplexing waveform.

3. A channel prediction method applicable to non-terrestrial networks according to claim 1, characterized in that: In step S2, for a user equipment, N sc The subcarriers of dimension 1 are used for uplink transmission, and the number of subcarriers is N. sym , the dimension of the entire transmission end resource grid X is (N sc ×N sym ).

4. A channel prediction method applicable to non-terrestrial networks according to claim 3, characterized in that: In step S2, in one time slot, M π The QAM pilot symbol is inserted into each subcarrier, and the position index is recorded in the OFDM resource grid as The number of OFDM pilot symbols per time slot is 7; And every other time slot, all pilot symbols are removed to reduce the pilot overhead.

5. The channel prediction method applicable to non-terrestrial networks according to claim 1, characterized in that: In step S3, the CFR matrix is ​​represented by H=FFT[h(t,τ)], where FFT represents fast Fourier transform, and the OFDM resource grid at the receiving end is represented as: Y=H⊙X+W; Where W represents the impact of AWGN on the OFDM grid, N0 represents the noise power spectrum density, and ⊙ is the Hadamard multiplication operator of the matrix.

6. A channel prediction method applicable to non-terrestrial networks according to claim 1, characterized in that: In step S4, for the selected system model, interpolation covers not only the data symbol positions in the first time slot, but also all symbol positions in the second time slot, and there is no pilot in the second time slot, and channel prediction is adopted; The execution logic of step S4 is as follows: first, the data symbols on the first time slot are equalized and demapped using the obtained channel estimation value; then the obtained data bits are remapped to M-QAM symbols, which are used to perform least squares on the entire first time slot, thereby forming an overall data and pilot estimation scheme.

7. A channel prediction method applicable to non-terrestrial networks according to claim 6, characterized in that: The demapping error may lead to inaccurate quantization of the estimated channel matrix, and the channel predictor is used to identify and equalize such errors and predict the channel; After the entire resource grid is equalized by the estimated or predicted channel coefficients, the data symbols are demapped and the resulting data is decoded.

8. A channel prediction method applicable to non-terrestrial networks according to claim 7, characterized in that: The main structure of the channel predictor includes three parts: encoding layer, prediction layer, and decoding layer; Flatten is a custom layer that flattens the tensor on the frequency axis, with each frequency as a dimension of a tensor; TimeMirror is a custom layer that mirrors the generated tensor on the time axis and arranges it by time; the Relu layer is an activation function with the formula f(x)=max(0,x); the BatchNorm layer is a normalization operation, and the units of LSTM are configured as 32.

9. A channel prediction device applicable to a non-terrestrial network, applied to the channel prediction method applicable to a non-terrestrial network as claimed in any one of claims 1 to 8, characterized in that: The device includes: a memory, a processor, and computer program instructions stored in the memory and executable on the processor, and when the processor executes the computer program instructions, the channel prediction method applicable to non-terrestrial networks as described in any one of claims 1-8 above is implemented.

10. A computer-readable storage medium, characterized in that: A computer program is stored which can be loaded by a processor and execute the channel prediction method applicable to a non-terrestrial network as claimed in any one of claims 1 to 8.

Images