THz Waveform Design and Deep Learning-Based Receiver Design

Sensing Integrated DFT-Spread OFDM Waveform and Deep Learning-Powered Receiver Design for Terahertz Integrated Sensing and Communication Systems. Yongzhi Wu et.al. IEEE Transactions on Communications, January 2023 (pdf) (Citations 7)

  • 主要是设计了一种 ==flexible guard interval (FGI)== 帧结构来替换CP，取得了一定的效果。
  • 设计SensingNet and ComNet (both are dense network) 来对感知（距离和速度）和通信组件进行替换。

Quick Overview

  • proposed a SI-DFT-s-OFDM waveform for THz ISAC system.
  • SensingNet: Multiple-input average-output neural network for sensing: block-wise processing and multiple-wise processing.
    • block-wise for range estimation
    • subcarrier-wise for velocity estimation
  • ComNet: Two-level communication neural network, the first level is designed to extract channel information. The second level is designed to recover the transmitted data symbols.

  • joint sensing and communication neural network by incorporating the proposed SensingNet and ComNet.

  • lower PARA.

  • 插入参考块（reference blocks）以满足感知参数估计和通信信道估计的要求

  • Active Sensing and passive sensing:

    • Active sensing, Tx-Rx or Tx-target-Tx(用于感知)

    • Passive sensing, transmits a signal that is jointly designed and used for communication and sensing. This is followed by the received communication signal serving as the sensing signal that carries the information of the

      transmitter, such as its distance.

System model

Communication channel model

image-20240117130122857

Sensing channel model

image-20240117130454912

Power attenuation

image-20240117130519723

Waveform design

image-20240117132106602

  • data matrix 横着看有个block，每个block长。每个格子是QAM调制符号。
  • the reference block is used for sensing parameter estimation and signal recovery.
  • 增加循环前缀image-20240117133102792

​ 即将一个block的最后采样点再移到前面。

​ CP长度一般要比最大时延扩展长，以消除IBI （inter-block interference）

==由于THz时延扩展可能不确定==，使得CP可能长度没有固定的需求。这就是本文的Motivation之一。然后帧结构变为：

image-20240117135042344

Reference block is generated by fixed reference symbols and inserted into the data blocks. 并且插入数据序列reference data 是，并且使得后面几位都基本相同（==和我们单载波加保护间隔是同一个目的==）

image-20240117135337563

SensingNet

Regard the received reference signals as input features (Complex, imaginary part and real part)

  • block-wise input processing for range estimation. Inputs.
  • subcarrier-wise input processing for velocity estimation. Inputs.

全是 Dense layer，并且block-wise的所有输入共享一个network，多个输入结果再进行平均作为最终输出。同理对于subcarrier-wise也是这样。

image-20240117141932651

ComNet

image-20240117142259818

two-level：

  • received pilot（reference block）处理信道信息

  • received data block 级联 上一级的信道信息一起输入，进行处理。

    image-20240117142516384

然后一起训练

image-20240117142749458