An OFDM-IM Signal Detection Method Based on FFDNet

Abstract

The present invention belongs to the field of signal detection in radio communication, and specifically relates to an OFDM-IM signal detection method based on FFDNet, including: denoising the signal combined with the noise level through the FFDNet denoising network to obtain the denoised signal and the energy distribution of the denoised signal; recombining the denoised signal and the energy distribution of the denoised signal to obtain the recombined signal to be detected and the recombined energy distribution; using the detection network to detect the recombined signal to be detected, and judging the position of the activated carrier through the recombined energy distribution to obtain the signal estimate value, and making a decision on the estimated value of the signal to obtain the estimated bit stream. The present invention flexibly detects OFDM-IM signals in various signal-to-noise ratio ranges through the noise distribution coefficient, the energy distribution coefficient, and the signal recombination coefficient, and flexibly and accurately detects the signal to be detected under the condition of using a single model.

Description

Technical Field

[0001] The present invention belongs to the field of signal detection in wireless communication, and particularly relates to a method for detecting OFDM-IM signals based on FFDNet. Background Art

[0002] Orthogonal frequency division multiplexing technology has been widely used in 5G and 4G due to its advantages such as being able to transmit a large amount of data using less frequency band and resisting inter-symbol interference caused by multipath effects. Index Modulation (IM) technology uses the media in the communication system to transmit additional information and is a new type of multi-dimensional modulation technology. These media are entities such as antennas, subcarriers, time slots, spreading codes, etc. Since index modulation can transmit additional information without consuming energy or consuming only a little energy, and can be flexibly balanced with spectral efficiency and has a high energy efficiency (EE), IM technology has attracted wide attention from scholars and may become a candidate technology for future wireless communication. Index modulation technology can be divided into frequency-domain index modulation technology and spatial-domain index modulation technology from the perspective of dimension. Spatial-domain index modulation is also called Spatial Modulation (SM) technology, which uses the index of antennas to control the activation of antennas, and the activated part performs normal constellation symbol mapping. Recently, an OFDM-IM technology developed from spatial modulation technology has been proposed. The emergence of Orthogonal Frequency Division Multiplexing Index Modulation (OFDM-IM) has become a modulation method to replace or supplement existing OFDM systems. In an OFDM-IM system, the modulation dimension of OFDM is extended from one to two. OFDM-IM divides all subcarriers into two parts, one part is silent subcarriers, and the other part is active subcarriers. The modulated constellation point information is placed on the active subcarriers, and the silent and non-silent carriers can also carry additional information. Compared with traditional OFDM, OFDM-IM has many advantages. First, OFDM-IM modulation can make a trade-off between spectral efficiency and transmission performance. Second, the peak-to-average power ratio of OFDM-IM signals is lower than that of OFDM signals when transmitting. Finally, OFDM-IM signals greatly reduce inter-carrier interference and improve the robustness of the system. In general, OFDM-IM technology has good application prospects and is worthy of our in-depth exploration.

[0003] In an OFDM-IM system, for the research on signal detection, scholars use machine learning to detect OFDM-IM signals, and the development of noise reduction technology has greatly promoted the development of communication.

[0004] Soltani et al. (M. Soltani, V. Pourahmadi, A. Mirzaei and H. Sheikhzadeh, "Deep Learning-Based Channel Estimation," in IEEE Communications Letters, vol. 23, no. 4, pp. 652-655, April 2019, doi: 10.1109 / LCOMM.2019.2898944.) proposed a two-stage cascaded network ChannelNet, which equivalent the pilot to a low-resolution image, improves the resolution through a super-resolution algorithm and denoises through the DNCNN network (K. Zhang, W. Zuo, Y. Chen, D. Meng and L. Zhang, "Beyond a Gaussian Denoiser: Residual Learning of Deep CNN for Image Denoising," in IEEE Transactions on Image Processing, vol. 26, no. 7, pp. 3142-3155, July 2017, doi: 10.1109 / TIP.2017.2662206.), and finally realizes channel estimation. This research shows that denoising technology has important research value for improving communication performance, especially the development of deep learning denoising technology plays an increasingly important role in the research of the communication field.

[0005] Due to the high complexity of DNCNN, the detection of OFDM-IM signals based on DNCNN will bring certain network delays in communication and is not flexible enough. To overcome the shortcomings of the CNN-based denoising method that is not flexible enough and has too high time complexity, Zhang et al. (K. Zhang, W. Zuo and L. Zhang, "FFDNet: Toward a Fast and Flexible Solution for CNN-Based Image Denoising," in IEEE Transactions on Image Processing, vol. 27, no. 9, pp. 4608-4622, Sept. 2018, doi: 10.1109 / TIP.2018.2839891.) proposed a fast and flexible denoising convolutional neural network (FFDNet).

[0006] T.V. Luong et al. (T.V. Luong, Y. Ko, N.A. Vien, D.H.N. Nguyen, and M. Matthaiou, “Deep learning-based detector for OFDM-IM,” IEEE Wireless Commun. Lett., vol. 8, no. 4, pp. 1159–1162, Aug. 2019.) proposed a machine learning-based OFDM-IM detector, DeepIM. This paper uses deep learning methods for signal detection in OFDM-IM systems, and can achieve near-optimal bit error rate performance with very low model training time. The proposed DeepIM structure only requires two fully-connected (FC) non-linear layers to efficiently perform detection tasks under Rayleigh fading channels. In addition, the number of nodes in the hidden layer can be adaptively adjusted to make a compromise between performance and complexity. By training DeepIM offline, minimizing the bit error rate using simulated data, and then the trained model can be used as an online detector with very low running time. Although this method can effectively recover information using a low-complexity network, there is no obvious improvement in reducing the bit error rate performance.

[0007] Based on the current research situation of OFDM-IM signal detection, it is found that there are still some challenges in the process of signal detection using denoising networks and detection networks:

[0008] 1. The OFDM-IM signal detection method based on DNCNN obtains better bit error rate performance compared to directly performing OFDM-IM detection under high signal-to-noise ratio conditions. However, an obvious defect of its network is that the time complexity of the network is relatively high, which will bring network latency in communication and is not conducive to its popularization.

[0009] 2. Due to the characteristics of DNCNN itself, if the best performance is to be achieved in different signal-to-noise ratio environments, the OFDM-IM signal detection based on DNCNN needs to train models for each signal-to-noise ratio range, so it is impossible to use a single model to complete signal detection in all signal-to-noise ratio environments. Summary of the Invention

[0010] To solve the above technical problems, the present invention proposes an OFDM-IM signal detection method based on FFDNet, including the following steps:

[0011] S1: In the OFDM-IM system, the transmitting end sends the signal X through the channel H to the receiving end, obtains the received signal Y, and performs equalization processing on the received signal Y to obtain an equalized signal in two-dimensional format.

[0012] S2: Perform data enhancement on the signals on the silent subcarriers of the equalized signal in two-dimensional format to obtain the enhanced equalized signal

[0013] S3: Establish an FFDNet noise reduction network; the FFDNet noise reduction network includes: a downsampling layer, a graph combination layer, a CNN noise reduction layer, and a recombination layer;

[0014] S4: Downsample the enhanced equalized signal through the downsampling layer to obtain 4 subgraphs of the same size. If the noise level is δ, the noise is noise = δ / 255×randn(size(label)). Meanwhile, obtain the noise level map M of the signal-to-noise ratio corresponding to the 4 subgraphs, and the size of the noise level map M is the same as that of the subgraphs;

[0015] S5: Combine the subgraphs of the enhanced equalized signal and the noise level map M corresponding to the subgraphs to obtain the training data samples

[0016] S6: Use the signal X as the label and input it together with the training data samples into the CNN noise reduction layer for training of the CNN noise reduction layer;

[0017] S7: Input the equalized signal in two-dimensional format into the trained CNN noise reduction layer for data noise reduction to obtain the noise-reduced signal Y de and the energy distribution E of the noise-reduced signal de ;

[0018] S8: Use the noise estimation level δ to calculate the quantity distribution coefficient γ and the signal combination coefficient β respectively. Recombine the noise-reduced signal Y de and the energy distribution E of the noise-reduced signal de through the recombination layer to obtain the recombined signal to be detected and the recombined energy distribution;

[0019] S9: Construct a detection network and train the detection network;

[0020] S10: Use the trained detection network to detect the recombined signal to be detected, and judge the position of the active carrier through the recombined energy distribution to obtain the estimated value of the signal of the signal estimation and make a decision on the estimated value of the signal

[0021] Advantages of the present invention:

[0022] The present invention elastically detects OFDM-IM signals in various signal-to-noise ratio ranges through a noise distribution coefficient, an energy distribution coefficient, and a signal recombination coefficient. Secondly, under high signal-to-noise ratio conditions, by reducing the depth of the noise reduction network, the detection network can still achieve good performance and the network latency is further reduced. For problems such as reducing the network complexity affecting the detection performance of FFD-Denoising-IM under low signal-to-noise ratio conditions, considering that the performance of directly detecting using the OFDM-IM signal and detecting using the OFDM-IM signal after noise reduction is relatively close, through the organic combination of the received signal and the noise-reduced signal, the noise-reduced signal can be more effectively used for signal detection. Brief Description of the Drawings

[0023] Figure 1 It is a schematic diagram of the OFDM-IM signal detection method based on the FFDNet noise reduction network of the present invention;

[0024] Figure 2 It is a schematic diagram of the OFDM-IM system of the present invention. Detailed Embodiments

[0025] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0026] An OFDM-IM signal detection method based on FFDNet, as Figure 1 shown, includes:

[0027] S1: In the OFDM-IM system, the transmitting end sends the signal X to the receiving end through the channel H, obtains the received signal Y, and performs equalization processing on the received signal Y to obtain an equalized signal in a two-dimensional format

[0028] S2: Perform data enhancement on the signals on the silent subcarriers of the equalized signal in two-dimensional format to obtain the enhanced equalized signal

[0029] S3: Establish an FFDNet noise reduction network; the FFDNet noise reduction network includes: a downsampling layer, a graph combination layer, a CNN noise reduction layer, and a recombination layer;

[0030] S4: Through the downsampling layer, the enhanced equalized signal Downsample to obtain 4 subgraphs of the same size. If the noise level is δ, the noise is noise = δ / 255 × randn(size(label)). Meanwhile, obtain the noise level map M of the signal-to-noise ratio corresponding to the 4 subgraphs, and the size of the noise level map M is the same as that of the subgraphs;

[0031] S5: Combine the enhanced equalized signal subgraphs and the subgraph corresponding noise level map M to obtain the training data samples

[0032] S6: Use the signal X as the label and input the training data samples into the CNN noise reduction layer for training of the CNN noise reduction layer;

[0033] S7: Input the two-dimensional format equalized signal into the trained CNN noise reduction layer for data noise reduction to obtain the denoised signal Y de and the energy distribution E of the denoised signal de ;

[0034] S8: Use the noise estimation level δ to calculate the quantity distribution coefficient γ and the signal combination coefficient β respectively. Recombine the denoised signal Y de and the energy distribution E of the denoised signal de through the recombination layer to obtain the recombined signal to be detected and the recombined energy distribution;

[0035] S9: Construct a detection network and train the detection network;

[0036] S10: Use the trained detection network to detect the recombined signal to be detected, and judge the position of the active carrier through the recombined energy distribution to obtain the estimated value of the signal of the signal estimation to make a decision to obtain the estimated bitstream

[0037] As Figure 2 shown, assume that the channel information H is known at the receiving end, the received OFDM-IM signal is Y, and the transmitted signal is X; Focusing on the single-input single-output (SISO) communication link, the OFDM-IM system consists of N C subcarriers, where N C subcarriers are divided into L groups, and the number of subcarriers in each group is N L , then it can be known that N L = N C / L. At the transmitting end, the processing method for each group is the same. Only discuss a certain group: In each transmission, NL Only m sub - carriers out of the sub - carriers are activated, and the total number of bits transmitted is P bits, where P1 is the index bit and P2 is the data bit. Then P = P1+P2, where: P1 = log2C(N L , m). In the present invention, (N L , m, S) is defined. (N L , m, S) means that when transmitting information each time, m sub - carriers are activated on N L carriers for data transmission, and every 2 bits are mapped through the S - order. For a set of received signals Assume that only 1 sub - carrier is activated, then Y m is the received signal at the position of the activated carrier.

[0038] To simplify the training process, a constant θ(δ)={θ de (δ), θ im} is defined, where θ de (δ) is the constant of the noise reduction network, θ im is the constant of the detection network, f de and f im are the noise reduction function and the detection function respectively. Before training the network, a data bit stream b needs to be randomly generated, and the transmitted signal X can be obtained through OFDM - IM modulation.

[0039] Data enhancement is performed on the signals on the silent sub - carriers of the two - dimensional format equalized signal, including:

[0040]

[0041] Among them, represents the enhanced equalized signal, n1, n NL represent the positions of the silent carriers and the noise distribution of N L respectively, is the real part of the equalized signal, is the imaginary part of the equalized signal.

[0042] On the silent sub - carriers, we continue to add complex noise to make the symbols of the silent carriers larger or make the symbols of the activated carriers smaller. In the training stage, in this way, the model can learn a more complex noise environment, and the model will be more robust and show better performance.

[0043] The FFDNet denoising network includes: a downsampling layer, a graph combination layer, a CNN denoising layer, and a recombination layer; the CNN denoising layer consists of a CNN structure composed of 12 convolutional layers. The network is composed of three types of operations: convolutional layer (Conv), activation function (ReLU), and batch normalization (BN). The first layer uses 64 filters of size 3×3×1, followed by a Relu activation function. The middle 10 layers use convolutional layers, ReLU activation functions, and batch normalization, and each layer has 64 filters of size 3×3×64. The last layer is a convolutional layer that uses a 3×3×64 filter to upsample the output of the network to obtain the denoised signal.

[0044] Use the transmitted signal X as the label along with the training data samples Input it into the CNN denoising layer to train the CNN denoising layer, including:

[0045]

[0046] Among them, Loss1 represents the loss function of the denoising network, M1 represents the size of the training set of the denoising network, f de represents the denoising function, represents the training data samples, θ de (δ) represents the constant of the denoising network, δ represents the noise estimation level, and X represents the transmitted signal.

[0047] Input the equalized signal in two-dimensional format into the trained CNN denoising layer for data denoising to obtain the denoised signal Y de and the energy distribution E of the denoised signal de , including:

[0048] The denoised signal Y de :

[0049]

[0050] Among them, Y de represents the denoised signal, f de represents the denoising function, represents the equalized signal in two-dimensional format, θ de (δ) represents the constant of the denoising network, and δ represents the noise estimation level;

[0051] The energy distribution E of the denoised signal de :

[0052]

[0053] Among them, E de represents the energy distribution of the denoised signal, and Y deRepresents the signal after noise reduction, and respectively represent the noise-reduced signals at the corresponding carrier positions.

[0054] Through the CNN noise reduction layer, the signal after noise reduction can be obtained. The signal after noise reduction has a clearer energy distribution, but there is also a possibility of misjudgment, which will misjudge the position of the activated carrier. In order to obtain a more effective energy distribution, the combination E of the energy distribution of the noise-reduced equalized signal and the energy distribution of the received signal is used Sum to determine the activated position.

[0055] Using the noise estimation level δ, calculate the energy distribution coefficient γ and the signal recombination coefficient β respectively, including:

[0056] Energy distribution coefficient γ:

[0057]

[0058] Signal recombination coefficient β:

[0059]

[0060] Among them, δ represents the noise estimation level, and SNR represents the signal-to-noise ratio in the actual channel environment.

[0061] For the signal Y after noise reduction de and the energy distribution E of the signal after noise reduction de perform recombination to obtain the recombined signal to be detected and the recombined energy distribution, including:

[0062] Recombined signal to be detected:

[0063]

[0064] Recombined energy distribution:

[0065]

[0066] Among them, Y d represents the recombined signal to be detected, Y de represents the signal after noise reduction, represents the equalized signal in two-dimensional format, β represents the detection signal recombination coefficient, E sum represents the energy distribution of the reconstructed signal, Y represents the received signal, and γ represents the energy distribution coefficient.

[0067] The detection network includes a DNN structure composed of two layers of fully connected networks. The first layer has 128 neurons with the activation function being the Tanh function, and the second layer has S neurons with the activation function being the Sigmoid function, where S represents the number of output bits.

[0068] Training the detection network includes:

[0069]

[0070] Among them, Loss2 represents the loss function of the detection network, M2 represents the size of the training set of the detection network, L represents the output of the detection network, b represents the randomly generated data bit stream, and |||| represents the norm operation.

[0071] Using the trained detection network to detect the reorganized signal to be detected, and judging the position of the active carrier through the reorganized energy distribution to obtain the estimated value of the signal Including:

[0072]

[0073] Among them, represents the estimated value of the output signal of the detection network, f im respectively represent the detection function, θ im represents the constant of the detection function, Y d represents the reorganized signal to be detected, E sum represents the energy distribution of the reorganized signal.

[0074] For the estimated value of the signal Make a decision to obtain the estimated bit stream Including: calculating the estimated signal The Euclidean distances from the transmitted bits 0 and 1, and taking the bit value corresponding to the shortest Euclidean distance as the final bit stream

[0075] Although the embodiments of the present invention have been shown and described, for those of ordinary skill in the art, it can be understood that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for detecting OFDM-IM signals based on FFDNet, characterized in that, Including: S1: In the OFDM-IM system, the transmitting end sends the signal X to the receiving end through the channel H, obtains the received signal Y, and performs equalization processing on the received signal Y to obtain the equalized signal in two-dimensional format. S2: Perform data enhancement on the signals on the silent subcarriers of the equalized signal in two-dimensional format to obtain the enhanced equalized signal S3: Establish an FFDNet denoising network; The FFDNet denoising network includes: a downsampling layer, a graph combination layer, a CNN denoising layer, and a recombination layer; S4: Downsample the enhanced balanced signal through a downsampling layer to obtain 4 subgraphs of the same size. If the noise level is δ, the noise is noise = δ / 255 × randn(size(label)). Meanwhile, obtain the noise level map M of the signal-to-noise ratio corresponding to the 4 subgraphs, and the size of the noise level map M is the same as that of the subgraphs; S5: Combine the subgraph of the enhanced equalization signal and the subgraph corresponding noise level graph M to obtain a training data sample S6: Use the signal X as a label together with the training data samples Input them into the CNN noise reduction layer and train the CNN noise reduction layer; S7: Input the equalization signal in two-dimensional format into the trained CNN noise reduction layer for data noise reduction, and obtain the denoised signal Y de and the energy distribution E of the denoised signal de ; S8: Calculate the quantity distribution coefficient γ and the signal combination coefficient β respectively using the noise estimation level δ, and pass the denoised signal Y de and the energy distribution E of the denoised signal de through the recombination layer for recombination to obtain the recombined signal to be detected and the recombined energy distribution; S9: Construct a detection network and train the detection network; S10: Detect the recombined signal to be detected using the trained detection network, and determine the position of the activated carrier by the recombined energy distribution to obtain the value of the signal estimation Make a decision to obtain the estimated bitstream 2. The OFDM-IM signal detection method based on FFDNet according to claim 1, wherein, Perform data augmentation on the signals on the silent subcarriers of the equalized signal in two-dimensional format, including: Among them, represents the enhanced equalization signal, n1, respectively represent the silent carrier position, N L 's noise distribution, is the real part of the equalization signal, is the imaginary part of the equalization signal.

3. A method for detecting OFDM-IM signals based on FFDNet according to claim 1, characterized in that, Use the transmitted signal X as a label together with the training data samples Input into the CNN noise reduction layer and perform the training of the CNN noise reduction layer, including: Among them, Loss1 represents the loss function of the noise reduction network, M1 represents the size of the training set of the noise reduction network, and f de represents the noise reduction function, represents the training data sample, and θ de (δ) represents the constant of the noise reduction network, δ represents the noise estimation level, and X represents the transmitted signal.

4. A method for detecting OFDM-IM signals based on FFDNet according to claim 1, characterized in that, Input the equalization signal in two-dimensional format into the trained CNN noise reduction layer for data noise reduction to obtain the noise-reduced signal Y de and the energy distribution E of the noise-reduced signal de , including: The denoised signal Y de : Among them, Y de represents the signal after noise reduction, f de represents the noise reduction function, represents the equalized signal in two-dimensional format, θ de (δ) represents the constant of the noise reduction network, and δ represents the noise estimation level; Energy distribution E of the signal after noise reduction de : Among them, E de represents the energy distribution of the signal after noise reduction, Y de represents the signal after noise reduction, and respectively represent the noise-reduced signals at the corresponding carrier positions.

5. A method for detecting OFDM-IM signals based on FFDNet according to claim 1, characterized in that Calculate the energy distribution coefficient γ and the signal recombination coefficient β respectively using the noise estimation level δ, including: Energy distribution coefficient γ: Signal recombination coefficient β: Where δ represents the noise estimation level and SNR represents the signal-to-noise ratio in the actual channel environment.

6. A method for detecting OFDM-IM signals based on FFDNet according to claim 1, characterized in that, For the denoised signal Y de and the energy distribution E of the denoised signal de perform recombination to obtain the recombined detected signal and the recombined energy distribution, including: The signal to be detected after recombination: The energy distribution after recombination: Among them, Y d represents the signal to be detected after recombination, Y de represents the signal after noise reduction, represents the equalized signal in two-dimensional format, β represents the recombination coefficient of the detection signal, E sum represents the energy distribution of the reconstructed signal, Y represents the received signal, and γ represents the energy distribution coefficient.

7. A method for detecting OFDM-IM signals based on FFDNet according to claim 1, characterized in that The detection network includes: a DNN structure composed of two layers of fully connected networks. The first layer has 128 neurons and the activation function is the Tanh function. The second layer has S neurons and the activation function is the Sigmoid function, where S represents the number of output bits.

8. A method for detecting OFDM-IM signals based on FFDNet according to claim 1, wherein Training the detection network includes: Where Loss2 represents the loss function of the detection network, M2 represents the size of the training set of the detection network, L represents the output of the detection network, and b represents the randomly generated data bit stream.

9. A method for detecting OFDM-IM signals based on FFDNet according to claim 1, characterized in that, Use the trained detection network to detect the recombined signal to be detected, and determine the position of the activated carrier by the recombined energy distribution to obtain the estimated value of the signal including: Among them, represents the estimated value of the output signal of the detection network, and f im respectively represent the detection function, and θ im represents the constant of the detection function, and Y d represents the signal to be detected after recombination, and E sum represents the energy distribution of the signal after reconstruction.

10. A method for detecting OFDM-IM signals based on FFDNet according to claim 1, characterized in that, The estimated value of the signal Perform a decision to obtain the estimated bitstream including: calculating the estimated signal The Euclidean distances from the transmitted bits 0 and 1, and taking the bit value corresponding to the shortest Euclidean distance as the final bitstream

Images