Autoencoder-Based Transceiver Design for OWC Systems in Log-Normal Fading Channel

Abstract

In this paper, we construct the autoencoder (AE) for optical wireless communication (OWC) systems with non-negativity and peak power constraints, which provides effective transceiver design in log-normal channel. We consider the cases where perfect channel state information (CSI) or noisy CSI can be obtained under three kinds of communication rate, which is defined as the ratio of channel use number to bit number. Meanwhile, we present the block error rate (BLER) performance to further demonstrate our transceivers' superior performance than common model-based methods. The learned constellation points distribution is provided to understand the transmitter's performance. Numerical simulations are conducted to ensure the best convergence. The results indicate that AE-based transceivers can achieve model-based optimal BLER performance or provide significantly better BLER performance.

Model-Aware End-to-End Learning for SISO Optical Wireless Communication Over Poisson Channel

“…Content may change prior to final publication. Citation information: DOI 10.1109/JPHOT.2020.3038534, IEEE Photonics Journal IEEE Photonics Journal Model-aware End-to-End Learning Such phenomenon also appears in our previous work [29] and results in about 1.5 dB performance loss compared with the (7,4) Hamming code with MLD. This dilemma should be blamed to the vanishing gradient problem of the sigmoid activation, which is located in the last layer of the transmitter component and leads to the update cessation of the transmitter network when it learns a kind of binary permutation.…”

“…3(a), we feed the one-hot version of the message m into the transmitter network. Determined by the system constraints of the normalized peak-power and the number of channel use, the last layer is an N -unit dense layer with sigmoid activation, which is the most common activation function for embedding the peak-limited constraint into the network [22], [29]. The last layer of the receiver network is also fixed as an M -unit dense layer with softmax activation, enabling the receiver network to deal with M -category classification (i.e., signal detection).…”

Low complexity deep learning algorithms for compensating atmospheric turbulence in the free space optical communication system

“…In the context of FSO communication, many algorithms are used to detect the received signal after propagating through atmospheric turbulence. Depending on the deployed algorithm, these works can be divided into three categories; classical machine learning-based methods [16][17][18][19], convolutional neural network (CNN)-based methods [20][21][22][23][24][25][26][27][28][29][30][31][32], and deep neural network (DNN)-based methods [33][34][35][36][37][38][39][40]. Table 1 shows the main differences between these works.…”

Trainable Communication Systems: Concepts and Prototype

“…EXIT chart-based degree profile matching requires knowledge of the mutual information between the output LLRs of the demapper and the transmitted bits 3 which can be simulated via Monte-Carlo methods as shown later. In the case of IDD, the demapper observes y and additionally lE as side information (so-called a priori knowledge denoted by the mutual information ĨA ).…”

“…This approach enables joint optimization of the transmitter and receiver for a specific channel model without extensive mathematical analysis. Autoencoder-based communication systems have first been proposed in the context of wireless communications [1], and have subsequently been extended towards other settings, such as optical fiber [2], optical wireless [3], and molecular communications [4]. Most of these approaches are optimized on the symbol-wise categorical cross entropy (CE), which is equivalent to maximizing the mutual information between the channel input and output [5].…”