Detection method for orthogonal frequency division multiplexing index modulation

Abstract

The invention relates to a detection method for orthogonal frequency division multiplexing index modulation. Comprising the following steps: receiving a receiving signal and channel state information generated by an orthogonal frequency division multiplexing index modulation system; performing equalization processing according to the received signal and the channel state information to generate a data set; processing the data set through a neural network to perform network feature extraction; taking the output of the neural network as the input of a Transform module to construct global context information; and classifying the output of the Transform module to generate a detection result. The CNN-Transform network is used for signal detection of a signal OFDM-IM system, good performance can still be achieved under the low signal-to-noise ratio, compared with a traditional ZF detector, the performance is greatly improved, cost is reduced, robustness is achieved, related and irrelevant information can be filtered, and information processing efficiency is improved. Recursive memory and optimal polynomial projection are combined, and the performance of processing the long sequence and long distance dependency relationship is improved.

Description

Technical Field

[0001] The embodiments of the present invention relate to the field of communication technology, and in particular to a detection method for orthogonal frequency division multiplexing index modulation. Background Art

[0002] The deep learning (DL) method is used to solve the signal detection problem of the orthogonal frequency division multiplexing (OFDM) system. Traditional signal detection usually estimates the channel response and then compensates the received signal to detect and restore the original signal. It is usually based on a priori assumptions about the signal pattern and uses a neural network structure that imitates the human brain to process and analyze a large amount of data. It may perform poorly in the case of noise, interference and signal deformation through a large amount of reverse. OFDM is a digital multi-carrier modulation method that is widely used in the communication field, especially in wireless systems such as Wi-Fi, 4G LTE and 5G. Traditional signal detection methods are inefficient. Unlike traditional signal detection, DL-based signal detection does not estimate the channel response, but treats the neural network as a black box, which uses propagation to update the weight value in the network, and finally adapts to the input data and finds the best result. It has great advantages even under low signal-to-noise ratio. Deep learning can handle complex nonlinear problems and conduct self-organized learning, which can effectively solve huge data sets.

[0003] Regarding the above technical solution, the inventors found that there are at least the following technical problems:

[0004] The performance is very poor at low signal-to-noise ratios, the extraction of data features is not sufficient, and it is not robust. The performance varies greatly when the value of M is changed. Alternatively, instead of converting the received signal into the final estimated information bit, the neural network provides soft probabilities of different symbols as intermediate results, which requires subsequent processing steps, and has high complexity and cost. Memory limitations and efficiency issues may be encountered when processing long sequences.

[0005] Therefore, it is necessary to improve one or more problems existing in the above-mentioned related technical solutions.

[0006] It should be noted that this section is intended to provide background or context for the technical solutions of the present invention stated in the claims. The description herein is not admitted to be prior art by inclusion in this section. Summary of the invention

[0007] The object of the present invention is to provide a detection method for OFDM index modulation, thereby solving one or more problems caused by the limitations and defects of the above-mentioned related technologies at least to a certain extent.

[0008] In a first aspect, the present invention provides a detection method for orthogonal frequency division multiplexing index modulation, comprising:

[0009] receiving a received signal and channel state information generated by an orthogonal frequency division multiplexing index modulation system;

[0010] Performing equalization processing according to the received signal and channel state information to generate a data set;

[0011] Processing the data set through a neural network to extract network features;

[0012] Using the output of the neural network as the input of the Transformer module to construct global context information;

[0013] The output of the Transformer module is classified to generate a detection result.

[0014] Optionally, the step of performing equalization processing according to the received signal and the channel state information to generate a data set includes:

[0015] The received signal and channel state information are input into a preprocessing module, and zero forced equalization is used to perform equalization processing to generate a data set.

[0016] Optionally, the step of processing the data set through a neural network to extract network features includes:

[0017] The neural network adopts multiple convolutional layers, and the input and output size of each convolutional layer changes according to the change of input data.

[0018] Optionally, the step of processing the data set through a neural network to extract network features includes:

[0019] The neural network uses a convolutional layer to extract features and performs a pooling operation before a fully connected layer.

[0020] Optionally, the step of using the output of the neural network as the input of the Transformer module to construct global context information includes:

[0021] The Transformer module uses a multi-head attention mechanism to process input data in parallel.

[0022] Optionally, the step of using the output of the neural network as the input of the Transformer module to construct global context information includes:

[0023] The output of the neural network is layer-normalized, decomposed through multiple parallel linear layers, and then input into the multi-head attention for processing.

[0024] Optionally, the step of using the output of the neural network as the input of the Transformer module to construct global context information includes:

[0025] The data processed by multi-head attention are uniformly normalized.

[0026] Optionally, the step of using the output of the neural network as the input of the Transformer module to construct global context information includes:

[0027] The data processed by multi-head attention is processed by a feedforward network for sequence representation and used as the output of the Transformer module.

[0028] Optionally, the step of classifying the output of the Transformer module to generate a detection result includes:

[0029] The output of the Transformer module is trained using a Sigmoid activation function, and the loss function is minimized.

[0030] The technical solution provided by the present invention may include the following beneficial effects:

[0031] In the present invention, by using CNN and Transformer network for signal detection of OFDM-IM system, good performance can still be achieved under low signal-to-noise ratio, which has greatly improved performance compared with the traditional ZF detector, reduced cost, and has robustness, can filter relevant and irrelevant information, and improve the efficiency of information processing. Combining recursive memory and optimal polynomial projection improves the performance of processing long sequences and long-distance dependencies. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] The accompanying drawings herein are incorporated into and constitute a part of the specification, illustrate embodiments consistent with the present invention, and together with the specification are used to explain the principles of the present invention. Obviously, the accompanying drawings described below are only some embodiments of the present invention, and for those of ordinary skill in the art, other accompanying drawings can be obtained based on these accompanying drawings without creative work.

[0033] Figure 1 A schematic flow chart showing a detection method for orthogonal frequency division multiplexing index modulation in an exemplary embodiment of the present invention;

[0034] Figure 2 A schematic diagram showing an OFDM-IM transmitter in an exemplary embodiment of the present invention;

[0035] Figure 3 A schematic diagram showing the judgment logic of a detection method for OFDM index modulation in an exemplary embodiment of the present invention is shown;

[0036] Figure 4 A schematic diagram showing a neural network and a Transformer module in an exemplary embodiment of the present invention. DETAILED DESCRIPTION

[0037] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, these embodiments are provided so that the present invention will be more comprehensive and complete and fully convey the concepts of the example embodiments to those skilled in the art. The described features, structures, or characteristics may be combined in any suitable manner in one or more embodiments.

[0038] In addition, the accompanying drawings are only schematic illustrations of embodiments of the present invention and are not necessarily drawn to scale. The same reference numerals in the drawings represent the same or similar parts, and their repeated descriptions will be omitted. Some of the block diagrams shown in the accompanying drawings are functional entities and do not necessarily correspond to physically or logically independent entities.

[0039] The present invention provides a detection method for orthogonal frequency division multiplexing index modulation, referring to Figure 1 As shown in , the following steps are included:

[0040] Step S100: receiving a reception signal and channel state information generated by an OFDM system.

[0041] Step S200: performing equalization processing according to the received signal and channel state information to generate a data set.

[0042] Step S300: Processing the data set through a neural network to extract network features.

[0043] Step S400: Using the output of the neural network as the input of the Transformer module to construct global context information.

[0044] Step S500: classify the output of the Transformer module to generate a detection result.

[0045] It is important to understand that Orthogonal Frequency Division Multiplexing (OFDM) is a multi-carrier modulation technology that divides a high-speed data stream into multiple low-speed data streams and modulates them onto different subcarriers. The subcarriers are orthogonal, which means they will not interfere with each other, thereby improving spectrum utilization.

[0046] Index Modulation (IM) is a technology that uses the activation state of a subcarrier (i.e., whether it is transmitting a signal) to transmit information. In IM, not all subcarriers transmit signals, but some subcarriers are selectively activated according to certain rules.

[0047] Orthogonal Frequency Division Multiplexing with Index Modulation (OFDM-IM) is a promising multi-carrier system that can replace conventional OFDM. In OFDM-IM, subsets of subcarriers are active and their indices are used to carry information bits. This makes OFDM-IM more reliable and energy efficient than conventional OFDM because no bandwidth or additional power is required when using the indices of active subcarriers to carry data bits. OFDM-IM combines the high spectral efficiency of OFDM with the high energy efficiency of IM. In OFDM-IM, data is transmitted not only by the modulation of the subcarriers but also by the activation state (index) of the subcarriers. OFDM-IM can modulate the activated subcarriers using different modulation schemes such as QPSK, 16-QAM, etc.

[0048] It is also necessary to understand that the present invention is applicable to the OFDM-IM system model. By introducing index modulation in OFDM symbols, OFDM-IM can improve spectrum utilization while maintaining the performance of traditional OFDM systems. The signal detection part is replaced by a hybrid neural network, which greatly improves the detection accuracy and efficiency.

[0049] refer to Figure 2 As shown, a general OFDM-IM system model is considered. On a correlated double selective fading channel, the system model adopts an OFDM-IM system with Q subcarriers. All subcarriers are equally divided into R subblocks, and each subblock contains N = Q / R subcarriers. Each subblock can be represented by Y r =[Y r,1 ,…,Y r,N ], r = 1, ...R represents, then the transmitted signal is: Each sub-block Y r The N subcarriers in the subblock are adjacent to each other. If Q is large, it can be assumed that the channel coefficients of the subcarriers in each subblock are basically correlated. However, if the ratio of the coherent bandwidth to the total number of subblocks R is less than the subcarrier spacing, it can be assumed that the subblock Y r The channel coefficients of the subcarriers are independent. In the OFDM-IM system, the subblock is the basic unit of information modulation and can be written as: r =[0,…0,A r,1 ,0,…0,A r,2 ,0,…0,…,A r,k ,0,…0] T Among them A r,k(k=1,…,K) represents the M-ary quadrature amplitude modulation (QAM) signal, Y r ∈∧, is the set of all possible signal vectors. r Indicates that each sub-block has bits to activate K subcarriers, represents the binomial coefficient, and p2=K[log2 M] bits are mapped to the M-ary signal constellation to select the data symbols that modulate the active subcarriers. Therefore, each subblock modulates p=p1+p2 bits of information, and all symbols of the OFDM-IM system modulate R(p1+p2) bits of information.

[0050] It is also important to understand that Convolutional Neural Networks (ConvNets) are a type of neural network (CNN), which is one of the best learning algorithms for understanding image content and performs well in image segmentation, classification, detection, and retrieval related tasks. CNN is a feedforward neural network with a convolutional structure. The convolutional structure can reduce the amount of memory occupied by deep networks. The three key operations are local receptive field, weight sharing, and pooling layer, which effectively reduce the number of network parameters and alleviate the overfitting problem of the model.

[0051] It is also important to understand that the Transformer model is an advanced deep learning architecture that is specifically designed to process sequence data, and performs particularly well in natural language processing (NLP) tasks. The model consists of an encoder and a decoder, each of which consists of multiple layers, each of which includes a self-attention mechanism and a feedforward network. Since its introduction, the Transformer model has achieved breakthrough progress in multiple NLP fields, including but not limited to machine translation, text summarization, question-answering systems, text generation, and speech recognition. In addition, its variants, such as BERT, GPT, etc., have achieved remarkable results in various tasks, promoting the development of natural language processing technology. The introduction of the Transformer model not only provides a new perspective on processing sequence data in theory, but also proves its powerful capabilities in practice, especially in processing long-distance dependencies and parallel computing. With the deepening of research, the Transformer model and its variants will continue to play an important role in the field of artificial intelligence.

[0052] It is also important to understand that the encoder converts the input sequence into a series of continuous vector representations, while the decoder generates the target sequence based on the encoder's output and the previously generated output sequence. This architecture enables the Transformer to effectively handle sequence-to-sequence tasks.

[0053] By using the above detection method of OFDM index modulation, the CNN-Transformer network is used for signal detection of OFDM-IM system, which can still achieve good performance under low signal-to-noise ratio, greatly improving the performance of traditional ZF detector, reducing costs, and being robust, able to filter relevant and irrelevant information, and improving the efficiency of information processing. Combining recursive memory and optimal polynomial projection improves the performance of processing long sequences and long-distance dependencies.

[0054] Next, we will refer to Figures 1 to 4 Each step of the above-mentioned OFDM index modulation detection method in this exemplary embodiment is described in more detail.

[0055] In some embodiments, reference Figure 3 As shown in , step S200 includes:

[0056] The received signal and channel state information are input into a preprocessing module, and zero forced equalization is used to perform equalization processing to generate a data set.

[0057] It should be understood that the received signal y and the channel state information H generated by the OFDM-IM system are first sent to the preprocessing module, and the preprocessing module uses zero-forcing (ZF) equalization to obtain the signal

[0058] In some embodiments, reference Figure 3 As shown in , step S300 includes:

[0059] The neural network adopts multiple convolutional layers, and the input and output size of each convolutional layer changes according to the change of input data.

[0060] It is important to understand that convolutional neural networks usually use two-dimensional or three-dimensional images, and the input matrix of the CNN module is where Re(·) and Im(·) represent the imaginary and real parts of the complex number, respectively. yes The input and output size of each layer of the CNN module changes according to the input data. Each convolution kernel can learn specific features from the input data. By extracting and analyzing these features, the detector can better understand the structure and meaning of the data and make more accurate decisions.

[0061] In some embodiments, reference Figure 4 As shown in , step S300 includes:

[0062] The neural network uses a convolutional layer to extract features and performs a pooling operation before the fully connected layer.

[0063] It is important to understand that the convolution layer extracts features and then performs pooling operations. Since each neuron in the output feature map of the convolution layer is locally connected to its input, and the corresponding connection weights are weighted and summed with the local input plus the bias value to obtain the neuron input value, this process is equivalent to the convolution process.

[0064] In some embodiments, reference Figure 4 As shown in , step S400 includes:

[0065] The Transformer module uses a multi-head attention mechanism to process input data in parallel.

[0066] It is important to understand that the self-attention mechanism is at the heart of the Transformer, which allows the model to consider information about all other elements in the sequence while processing each element of the sequence. This mechanism works by computing the attention weights of each element on other elements, thereby capturing the complex relationships between elements. A key advantage of the self-attention mechanism is that it can process all elements in the sequence in parallel, in stark contrast to traditional sequence models such as recurrent neural networks, which need to process each element step by step in sequence. The Transformer further introduces a multi-head attention mechanism, which allows the model to learn information in multiple different representation subspaces in parallel. This mechanism enhances the model's ability to capture different types of information, for example, one head may focus on capturing syntactic information while another head may focus on capturing semantic information.

[0067] Since convolutional neural networks often only focus on local features and cannot capture global dependencies well, Transformer can do this well. Combining the two can well identify effective features in data. Since the present invention is a signal detector, signal detection is described as a classification task, so only the encoder part of the Transformer is used. The output matrix of the convolutional neural network is Then use the Transformer module to extract the matrix The global dependencies in the matrix First, it is fed into a linear fully connected layer with a ReLu activation function to play the role of embedding, and the embedding matrix E is sent as input to the Transformer encoder, which consists of a self-attention mechanism, a feedforward block, layer normalization, and a residual connection.

[0068] In some embodiments, reference Figure 4 As shown in , step S400 includes:

[0069] The output of the neural network is layer-normalized, decomposed through multiple parallel linear layers, and then input into the multi-head attention for processing.

[0070] It is important to understand that since the self-attention mechanism itself does not contain information about the order of elements in the sequence, the Transformer uses position encoding to provide this information. Position encoding is usually implemented as a linear combination of sine and cosine functions, ensuring that the model can understand the relative position of elements in the sequence.

[0071] In some embodiments, reference Figure 4 As shown in , step S400 includes:

[0072] The data processed by multi-head attention are uniformly normalized.

[0073] It is important to understand that the output of each sub-layer in the Transformer model is added to the input of the sub-layer to form a residual connection. This design allows gradients to flow more efficiently in the network, which helps train deep networks. In addition, the output of each sub-layer is processed by layer normalization to stabilize the training process and accelerate convergence.

[0074] In some embodiments, reference Figure 4 As shown in , step S400 includes:

[0075] The data processed by multi-head attention is processed by a feedforward network for sequence representation and used as the output of the Transformer module.

[0076] It is important to understand that in each encoder, after the self-attention layer, there is a feed-forward network for further processing of the sequence representation. These networks usually consist of several fully connected layers that perform nonlinear transformations on the output of the self-attention layer.

[0077] In some embodiments, reference Figure 3 As shown in , step S500 includes:

[0078] The output of the Transformer module is trained using a Sigmoid activation function, and the loss function is minimized.

[0079] It is important to understand that the final extracted features The data is transferred to the output layer for fine classification. The output layer is a fully connected layer with a Sigmoid activation function. The final result is a vector of size p. The loss function uses the cross entropy loss function Among them I Subis the target label, θ is a learnable parameter in the network, and the loss function can be effectively minimized through training, so that the predicted label The highest accuracy is achieved.

[0080] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example" or "some examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine the different embodiments or examples described in this specification.

[0081] Those skilled in the art will readily appreciate other embodiments of the present invention after considering the specification and practicing the invention disclosed herein. The present invention is intended to cover any variations, uses or adaptations of the present invention that follow the general principles of the present invention and include common knowledge or customary techniques in the art that are not disclosed by the present invention. The specification and examples are to be considered exemplary only, and the true scope and spirit of the present invention are indicated by the appended claims.

Claims

1. A detection method for orthogonal frequency division multiplexing index modulation, characterized in that: include: receiving a received signal and channel state information generated by an orthogonal frequency division multiplexing index modulation system; Performing equalization processing according to the received signal and channel state information to generate a data set; Processing the data set through a neural network to extract network features; Using the output of the neural network as the input of the Transformer module to construct global context information; The output of the Transformer module is classified to generate a detection result.

2. The detection method of OFDM index modulation according to claim 1, characterized in that: The step of performing equalization processing according to the received signal and the channel state information to generate a data set comprises: The received signal and channel state information are input into a preprocessing module, and zero forced equalization is used to perform equalization processing to generate a data set.

3. The detection method of OFDM index modulation according to claim 2, characterized in that: The step of processing the data set by a neural network to extract network features comprises: The neural network adopts multiple convolutional layers, and the input and output size of each convolutional layer changes according to the change of input data.

4. The detection method of OFDM index modulation according to claim 3, characterized in that: The step of processing the data set by a neural network to extract network features comprises: The neural network uses a convolutional layer to extract features and performs a pooling operation before a fully connected layer.

5. The detection method of OFDM index modulation according to claim 1, characterized in that: The step of using the output of the neural network as the input of the Transformer module to construct global context information includes: The Transformer module uses a multi-head attention mechanism to process input data in parallel.

6. The detection method of OFDM index modulation according to claim 5, characterized in that: The step of using the output of the neural network as the input of the Transformer module to construct global context information includes: The output of the neural network is layer-normalized, decomposed through multiple parallel linear layers, and then input into the multi-head attention for processing.

7. The detection method of OFDM index modulation according to claim 6, characterized in that: The step of using the output of the neural network as the input of the Transformer module to construct global context information includes: The data processed by multi-head attention are uniformly normalized.

8. The detection method of OFDM index modulation according to claim 7, characterized in that: The step of using the output of the neural network as the input of the Transformer module to construct global context information includes: The data processed by multi-head attention is processed by a feedforward network for sequence representation and used as the output of the Transformer module.

9. The detection method of OFDM index modulation according to any one of claims 1 to 8, characterized in that: The step of classifying the output of the Transformer module to generate a detection result comprises: The output of the Transformer module is trained using a Sigmoid activation function, and the loss function is minimized.

Images