Research On Transformer Model In UWB Channel Estimation

Ultra wideband(UWB)technology has many advantages in wireless communication systems,both in terms of transmission rate,large capacity,and strong anti multipath capability.In the modern era of rapid development of communication,the research of UWB technology has become an indispensable and important component in the direction of modern wireless communication,and is also a key link in the future development of wireless communication.At the beginning of the development of ultra wideband(UWB),UWB was limited to carrier based multi band pulse radio ultra wideband(IR-UWB)schemes.However,with the development of communication systems,multi band orthogonal frequency division multiplexing(MB-OFDM UWB)schemes have also been added to ultra wideband communications.In wireless communication systems,signals enter the channel through modulation,coding,and digital to analog conversion.After exiting the channel,they undergo digital to analog conversion.Before being demodulated,channel estimation is required to reduce the impact of the channel on the signal.Therefore,the study of channel estimation is crucial in communication systems.In the development of channel estimation research,blind channel estimation was first proposed,but the estimation effect was poor.Then,someone studied pilot based non blind channel estimation,and then combined the two estimates to propose a semi blind channel estimation that did not fully utilize pilots.The research on ultra wideband channel estimation in thesis is based on the research of MB-OFDM UWB systems.The estimation algorithm is a non blind channel research using pilot insertion.Based on the algorithms previously studied,the popular in-depth learning method research this year is introduced.Through the introduction of traditional algorithms and the understanding of compressed sensing channel estimation,the foundation and ideas for subsequent algorithm research and design are provided.The establishment of basic models for ultra wideband channels is introduced in time,from the basic channel model Poisson model to the model that can describe multipath modelsΔ-K model,then S-V model that describes indoor channels more accurately.Finally,use Matlab to simulate the channel impulse response of IEEE802.15.4a model,and then calculate and simulate its characteristic parameters to understand the channel characteristics of CM1 to CM4 channels proposed by IEEE802.15.4a model.For fully connected networks in deep learning,cyclic neural networks are introduced,and convolutional neural networks are introduced to study the characteristics of ultra wideband channels.Their convolutional operations are used to more accurately classify the four special channels in the IEEE 802.15.4a model.A channel estimation structure based on CNN and Transformer module is proposed to replace the pilot insertion channel estimation method for channel estimation.By comparing its mean square error and error rate performance with traditional channel estimation methods through simulation,the availability of this scheme for channel estimation is verified.