End-to-end information transmission system and method based on artificial intelligence

Abstract

The present invention belongs to the technical field of wireless communication networks, and discloses an end-to-end information transmission system and method based on artificial intelligence. The method comprises the steps of: constructing a model library; determining an algorithms library of a model library; and developing the algorithms library to a deep neural network. The system comprises anAI emitter configured to determine an algorithms library constraint model library range, reduce the unknown parameters of a cost function library at a signal source coding module, channel coding module and a digital modulation module and eclectically select signal source coding, channel coding and digital modulation key parameters; and an AI receiver responding to the emitter and being configuredto determine mathematical mapping corresponding to the signal source coding, the channel coding and the digital modulation. The end-to-end information transmission system and method mainly solve theproblems that the modularization performances are severely lost, the ideal AWGN channel modeling is not accurate and description of other models is not accurate in the current mobile communication system, reduce the modularization information gain loss, effectively reduce the error code rate, improve the algorithm universality, reduce the algorithm complexity and improve the network performance.

Description

technical field [0001] The invention belongs to the technical field of wireless communication networks, and in particular relates to an end-to-end information transmission system and method based on artificial intelligence. Background technique [0002] At present, the existing technology commonly used in the industry is as follows: under the condition that the traditional end-to-end information transmission system makes full use of transmission resources (ie, bandwidth, power, complexity), select the sending and receiving scheme to approach the limit given by Shannon . For the linear Gaussian channel model, a method close to Shannon channel capacity has been found (using Turbo code, LDPC code under this model). In the wireless channel, non-linearity, Doppler frequency shift, fading, shadow effect and other user interference make the wireless channel cannot be modeled by a simple AWGN channel, and the characteristics of the wireless channel make it more challenging to find ...

AI Technical Summary

Problems solved by technology

2. Traditional channel coding improves communication quality by increasing information redundancy bits, but for mobile channels with strictly limited bandwidth and extremely poor transmission characteristics, it often has to sacrifice transmission rate and bandwidth

[0004] (1) Prior art 1 utilizes the method of establishing a function model to determine the main influencing factors of the strategy, but only considers the combination of the source coding module and the channel coding module, and does not consider the combination of coding and modulation and the pull-through of all modules situation; at present, there is no technology to realize the end-to-end information transmission system, and the learning ability of the neural network is not used, which cannot reflect the intelligence and versatility

[0005] (2) The second prior art only uses the neural network to propose a self-encoding method for the signal whose source is an image, and does not consider using a neural network learning method for other types of sources. At present, deep learning is mainly used in image processing and speech Partially optimized performance has been achieved in recognition, and only some modules of the end-to-end information transmission system have been partially optimized, without considering the joint optimization of each module, falling into the narrow direction of local optimality, which cannot reflect the versatility and low complexity of the algorithm sex

[0006] (3) The three existing technologies only use the neural network to learn new modulation signals and modulation methods of digital signals, but only a small module in the end-to-end information transmission system has achieved universality, that is, it only considers local optimality, and It has not risen to the global optimum of the end-to-end information transmission system, which brings technical problems in the design of the end-to-end information transmission system

Moreover, most signal processing algorithms in communication can only roughly capture the model, while deep learning does not require an accurate model, which provides the possibility for end-to-end information transmission. Therefore, deep learning can be used to design an end-to-end information transmission system based on artificial intelligence and method, which can effectively solve the problems of severe modular performance loss, inaccurate ideal AWGN channel modeling and inaccurate description of other models in existing mobile communication systems

Images