Underwater Wireless Optical Communication System and Method Based on Generalized Arbitrary Entropy Value TCM and Focal Loss Neural Network

Abstract

The present invention discloses a general underwater wireless optical communication system and method based on generalized arbitrary entropy value TCM and focal loss neural network, which includes a transmitting end and a receiving end. The transmitting end adopts a generalized arbitrary entropy value TCM scheme to generate orthogonal amplitude modulation symbols of arbitrary entropy values ​​that conform to approximate Gaussian distribution, so as to approximate the AWGN channel capacity under specific signal-to-noise ratio conditions and improve the transmission rate of the communication system. The receiving end adopts a focal loss-based neural network to achieve equalization and demodulation of the received signal, further improving the convergence rate of the neural network during the training process. The present invention can have the anti-noise and anti-interference capabilities of traditional channel coding, and can also approach the communication capacity of the underwater wireless optical communication system.

Description

Technical Field

[0001] The present invention belongs to the technical field of underwater wireless optical communication, and specifically relates to an underwater wireless optical communication system and method based on generalized arbitrary entropy value TCM and focal loss neural network. Background Art

[0002] With the booming development of the marine economy, the enthusiasm of humans for exploring and developing the ocean is getting higher and higher, and the pace is getting faster and faster, resulting in an increasing number of applications of various underwater wireless communication technologies. Compared with radio frequency communication with extremely short transmission distances and underwater acoustic communication with extremely low rates underwater, the transmission distance of underwater wireless optical communication can reach hundreds of meters, and the data rate can reach the level of hundreds of Mbps. At the same time, it also has the characteristics of high confidentiality and low cost, and is very suitable for the rapid interaction and transmission of large-scale underwater data. However, the maturity and inherent characteristics of the optical transceiver devices in long-distance underwater wireless optical communication systems often lead to limited bandwidth, and the multipath effect caused by underwater scattering and turbulence will further limit the effective bandwidth of the system. Although traditional channel coding schemes can effectively alleviate low signal-to-noise ratio and various random interferences, the redundancy introduced by them will further occupy bandwidth, thereby reducing the effective transmission rate of the communication system. In contrast, trellis-coded modulation (TCM), as a scheme that does not waste bandwidth and can provide coding gain at the same time, can well make up for the above disadvantages. However, the probabilities of the appearance of each symbol in traditional TCM technology are the same, which does not meet the characteristics of using discrete modulation formats to approximate the continuous Gaussian distribution to achieve the Gaussian white noise (AWGN) channel capacity. At the same time, the average entropy value of a single symbol is an integer and does not have arbitrariness, and cannot meet the requirements of general underwater wireless communication systems. In view of the above existing problems, there is an urgent need for a TCM technology that can achieve arbitrary entropy values to cope with this situation. The underwater wireless optical communication joint signal processing technology based on generalized arbitrary entropy value TCM and focal loss neural network proposed by the present invention can better solve the above difficulties. Summary of the Invention

[0003] The technical problem to be solved by the present invention is as follows: At the sending end, a generalized arbitrary entropy value TCM scheme is jointly composed of a distribution matcher using probability shaping technology (PS) and a convolutional encoder to generate quadrature amplitude modulation (QAM) symbols with arbitrary entropy values that conform to an approximate Gaussian distribution, meeting the requirements of general underwater wireless communication systems, approaching the AWGN channel capacity under specific signal-to-noise ratio (SNR) conditions, and improving the transmission rate of the communication system. At the receiving end, a neural network based on focal loss is used to achieve the equalization and demodulation of the received signal. Compared with traditional neural networks using mean square error (MSE) loss, the introduction of the focal loss function can correspond to the non-uniform symbols generated by the distribution matcher at the sending end, further improving the convergence rate of the neural network during the training process and meeting the requirements of strong real-time performance of the actual communication system.

[0004] To solve the above technical problems, the present invention is achieved through the following technical solutions:

[0005] The present invention discloses a general underwater wireless optical communication system based on generalized arbitrary entropy - valued TCM and focal - loss neural network, which includes a transmitting end and a receiving end. The transmitting end includes the input of a random binary sequence, a generalized arbitrary entropy TCM scheme jointly composed of a distribution matcher (mapping uniform bits to non - uniform symbols), a convolutional encoder, and QAM mapping, a complex - to - real device (C2RT), a digital - to - analog converter (DAC), and a laser light source for realizing long - distance underwater transmission. The receiving end includes a detector for receiving optical signals, an analog - to - digital converter (ADC), a neural network based on the focal - loss function for restoring the received signal into a non - uniform bit distribution, an inverse distribution matcher for inverse - mapping the non - uniform bit distribution into a uniform bit distribution, the output of a binary sequence, and the calculation of the final bit - error rate (BER). The transmitting end inputs the random binary sequence into the generalized arbitrary entropy - valued TCM module jointly composed of a distribution matcher, a convolutional encoder, and QAM mapping. The module inputs the sequence into the C2RT device to realize the transformation from complex numbers to real numbers. The sequence is transmitted through the underwater channel after being emitted by the ADC and the laser, and then the signal is received by the detector and the DAC. The received signal is restored into a non - uniform bit distribution through the neural network based on the focal - loss function, and then the non - uniform bit distribution is inverse - mapped into a uniform bit distribution through the inverse distribution matcher. Finally, the input and output signals are used for BER calculation.

[0006] A general underwater wireless optical communication method based on generalized arbitrary entropy - valued TCM and focal - loss neural network specifically includes the following steps:

[0007] Step 1: Build the transmitting end of the underwater wireless optical communication system based on the generalized arbitrary entropy - valued TCM scheme.

[0008] Specifically, it includes:

[0009] Step 1.1: For the QAM modulation system adopting the generalized arbitrary entropy - valued TCM scheme, the input random binary sequence first takes k bits. Among them, 2 bits are generated into 2 non - uniform - distributed bits (non - coded bits of non - uniform distribution) through a distribution matcher based on probability shaping technology, and the remaining k - 2 bits are generated into k - 1 bits (coded bits of uniform distribution) through a convolutional encoder with a code rate of R=(k - 2) / (k - 1).

[0010] Step 1.2, the merged k + 1 - bit mapping is mapped to a single QAM symbol, and the QAM symbol at this time is of order M×2. It can be found that although the convolutional coding process is experienced, due to the synchronous increase in the modulation order of the QAM symbol, the bit information carried by a single QAM symbol remains unchanged, so bandwidth is not wasted. At the same time, due to the introduction of the non - uniform probability bit output of the distribution matcher, the distribution of the final QAM symbol is no longer uniform, so QAM symbols with arbitrary entropy values can be generated through this scheme.

[0011] Step 2, build the receiving end of an underwater wireless optical communication system based on a focal - loss neural network. Specifically, it includes:

[0012] Step 2.1, the QAM - mapped symbols based on the generalized arbitrary - entropy - value TCM scheme are non - uniform, and the corresponding bits after inverse mapping are also non - uniform. Traditional neural networks usually adopt the MSE loss function, which assumes that the output result is uniformly distributed;

[0013] Step 2.2, for the above - mentioned result of non - uniform bit output, construct the focal - loss function for two - bit classification as:

[0014]

[0015] where y = {0, 1} is the true bit label, p is the predicted probability, and γ is the attention parameter. For example, in the case where 1 appears more in the non - uniform - distribution bits, taking γ greater than 1 will make the loss weight of the predicted value of 1 smaller, so the neural network will pay more attention to the samples with the predicted value of 0, improving its accuracy on few - samples, and vice versa. For uniformly - distributed coding bits, taking γ = 1, the loss function at this time is no different from the general cross - entropy loss function.

[0016] To verify the effectiveness of this generalized arbitrary - entropy - value TCM scheme, simulation experiments under the AWGN channel were carried out, verifying that the coding gain of PS - TCM - 16QAM based on this scheme is feasible, and it is significantly better than general 8QAM and uniformly - distributed TCM - 16QAM.

[0017] The beneficial effects of the present invention are:

[0018] 1. The present invention is compatible with all current underwater wireless optical communication systems. The receiving end only needs to replace the traditional digital signal processing process with a focal - loss neural network.

[0019] 2. The transmitting end of the present invention adopts the generalized arbitrary - entropy - value TCM scheme, which does not waste bandwidth compared with traditional channel coding schemes, and can generate QAM symbols with arbitrary entropy values at the same time.

[0020] 3. The receiving end of the present invention adopts a focal loss neural network, replacing the original multiple digital signal processing modules in the traditional system. At the same time, it has the excellent non-linear fitting ability of DNN, can achieve better bit error performance, and the introduction of the focal loss function enables the neural network to converge more quickly when processing non-uniform bits, meeting the actual requirements. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 is a schematic structural diagram of the present invention;

[0022] Figure 2 is a statistical probability diagram and constellation diagram of non-uniformly distributed 16QAM symbols in the present invention;

[0023] Figure 3 is an AWGN simulation result diagram in the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0024] The following detailed description of the specific embodiments of the present invention is made with reference to the accompanying drawings. The following embodiments are used to illustrate the present invention but are not intended to limit the scope of the present invention.

[0025] Embodiment 1, as Figure 1 shown, a general underwater wireless optical communication system based on generalized arbitrary entropy value TCM and focal loss neural network, which includes a transmitting end and a receiving end. The transmitting end includes the input of a random binary sequence, a generalized arbitrary entropy TCM scheme jointly composed of a distribution matcher (mapping uniform bits to non-uniform symbols), a convolutional encoder, and a QAM mapper, a complex-to-real converter (C2RT), a digital-to-analog converter (DAC), and a laser light source for realizing long-distance underwater transmission. The receiving end includes a detector for receiving optical signals, an analog-to-digital converter (ADC), a neural network based on the focal loss function for restoring the received signal into a non-uniform bit distribution, an inverse distribution matcher for inverse mapping the non-uniform bit distribution into a uniform bit distribution, an output of a binary sequence, and a final bit error rate (BER) calculation. The transmitting end inputs the random binary sequence into the generalized arbitrary entropy value TCM module jointly composed of the distribution matcher, the convolutional encoder, and the QAM mapper. The module inputs the sequence into the C2RT device to realize the transformation from complex numbers to real numbers. The sequence is transmitted through the underwater channel after being transmitted by the ADC and the laser, and the signal is received using the detector and the DAC. The received signal is restored into a non-uniform bit distribution through the neural network based on the focal loss function, and the non-uniform bit distribution is inverse mapped into a uniform bit distribution through the inverse distribution matcher. Finally, the input and output signals are used for BER calculation.

[0026] A general underwater wireless optical communication method based on generalized arbitrary entropy value TCM and focal loss neural network specifically includes the following steps:

[0027] Step 1: Build the transmitter of an underwater wireless optical communication system based on the generalized arbitrary entropy value TCM scheme.

[0028] Specifically, it includes:

[0029] 1.1 For the QAM modulation system adopting the generalized arbitrary entropy value TCM scheme, first take k bits from the input random binary sequence. Among them, 2 bits are generated into 2 non-uniformly distributed bits (non-coded bits with non-uniform distribution) through a distribution matcher based on probability shaping technology, and the remaining k - 2 bits are generated into k - 1 bits (coded bits with uniform distribution) through a convolutional encoder with a code rate of R=(k - 2) / (k - 1).

[0030] 1.2 The combined k + 1 bits are mapped to a single QAM symbol, and at this time the QAM symbol is of M×2 order. It can be found that although the convolutional coding process is experienced, due to the synchronous increase in the modulation order of the QAM symbol, the bit information carried by a single QAM symbol remains unchanged, so the bandwidth is not wasted. At the same time, due to the non-uniform probability bit output of the distribution matcher being introduced, the distribution of the final QAM symbol is no longer uniform, so QAM symbols with arbitrary entropy values can be generated through this scheme.

[0031] Step 2: Build the receiver of an underwater wireless optical communication system based on the focal loss neural network. Specifically, it includes:

[0032] 2.1 The QAM mapping symbols based on the generalized arbitrary entropy value TCM scheme are non-uniform, and the corresponding bit positions after inverse mapping are also non-uniform. Traditional neural networks usually adopt the MSE loss function, which assumes that the output result is uniformly distributed;

[0033] 2.2 For the above results of non-uniform bit output, construct the focal loss function for two-bit classification as:

[0034]

[0035] where y = {0, 1} is the true bit label, p is the predicted probability, and γ is the attention parameter. For example, for the case where 1 appears more in the non-uniformly distributed bits, taking γ greater than 1 will make the loss weight of the predicted value of 1 smaller, so the neural network will pay more attention to the samples with the predicted value of 0 and improve its accuracy on few samples, and vice versa. For the uniformly distributed coded bits, take γ = 1, and at this time the loss function is no different from the general cross-entropy loss function.

[0036] For the traditional TCM decoding terminal, it is necessary to go through channel equalization, real-to-complex transformation (R2CT) and Viterbi decoding in sequence, and the process is relatively complex. While using a neural network can incorporate channel equalization and decoding together to achieve joint optimization. Without loss of generality, except for using the focal loss function as the loss function, other aspects such as network structure design, backpropagation process, and descent algorithm selection of this neural network can utilize traditional methods.

[0037] As Figure 2 shown, it shows the probability statistical chart and constellation diagram of non-uniformly distributed 16QAM symbols generated after adopting the generalized arbitrary entropy value TCM scheme. Since the distributions of its symbols are not equal, the entropy value is not an integer and has arbitrariness, meeting the requirements of the general underwater wireless communication system.

[0038] As Figure 3 shown, it shows the BER comparison chart of PS-TCM-16QAM, general 8QAM, and uniformly distributed TCM-16QAM based on this scheme under the AWGN channel. It can be seen that the performance of PS-TCM-16QAM is the best, verifying the feasibility of this scheme.

[0039] Although the specific embodiments of the present invention have been described in detail above, the present invention is not limited to the above embodiments. Within the scope of knowledge possessed by those of ordinary skill in the art, various changes can be made without departing from the purpose of the present invention, and modifications or deformations without creative labor are still within the protection scope of the present invention.

Claims

1. A general underwater wireless optical communication system based on generalized arbitrary entropy value TCM and focal loss neural network, which includes a transmitter and a receiver. Characterized in that, The transmitter includes a random binary sequence input module, a complex-to-real converter, a digital-to-analog converter, a laser light source for realizing long-distance underwater transmission, and a generalized arbitrary entropy value TCM module composed of a distribution matcher, a convolutional encoder, and a QAM mapping module. The receiver includes a detector for receiving optical signals, an analog-to-digital converter, a neural network based on the focal loss function for restoring the received signal into a non-uniform bit distribution, an inverse distribution matcher for inverse mapping the non-uniform bit distribution into a uniform bit distribution, a binary sequence output module, and a final bit error rate calculation module; the transmitter inputs the random binary sequence into the generalized arbitrary entropy value TCM module composed of a distribution matcher, a convolutional encoder, and a QAM mapping module. The generalized arbitrary entropy value TCM module inputs the sequence into the complex-to-real converter to realize the transformation from complex to real. The sequence is transmitted through the underwater channel after being converted by the digital-to-analog converter and emitted by the laser. The signal is received by the detector and the digital-to-analog converter, and the received signal is restored into a non-uniform bit distribution through the neural network based on the focal loss function, and the non-uniform bit distribution is inverse mapped into a uniform bit distribution through the inverse distribution matcher, and finally the input and output signals are used for BER calculation.

2. A method for a general underwater wireless optical communication system based on generalized arbitrary entropy value TCM and focal loss neural network as described in claim 1, specifically including the following steps: Step 1, build the transmitter of the underwater wireless optical communication system based on the generalized arbitrary entropy value TCM scheme; specifically including: Step 1.1, for the QAM modulation system using the generalized arbitrary entropy value TCM scheme, the input random binary sequence first takes k bits. Among them, 2 bits are generated into 2 non-uniformly distributed bits through the distribution matcher based on probability shaping technology, and the remaining k - 2 bits are generated into k - 1 bits through a convolutional encoder with a code rate of R=(k - 2) / (k - 1). Step 1.2, the combined k + 1 bits are mapped to a single QAM symbol. At this time, the QAM symbol is of M×2 order. Since the modulation order of the QAM symbol synchronously increases, the bit information carried by a single QAM symbol remains unchanged and does not waste bandwidth; at the same time, the non-uniform probability bit output of the distribution matcher is introduced, and the distribution of the final QAM symbol is no longer uniform, and QAM symbols with arbitrary entropy values can be generated. Step 2, build the receiver of the underwater wireless optical communication system based on the focal loss neural network; specifically including: Step 2.1, the QAM mapping symbols based on the generalized arbitrary entropy value TCM scheme are non-uniform, and the corresponding inverse-mapped bit positions are also non-uniform. Traditional neural networks usually use the MSE loss function, which assumes that the output result is uniformly distributed. Step 2.2, for the above results of non-uniform bit output, construct the focal loss function for two-bit classification as: where y = {0, 1} is the true bit label, p is the predicted probability, and γ is the attention parameter.

Images