超声波穿金属无线通信噪声智能抑制方法研究

来源：计算机技术与发展杂志2023年第09期北京时间：

作者：朱晓雨;曹自平;崔红涛

单位：南京邮电大学 通信与信息工程学院,江苏 南京 210003 Author(s): ZHU Xiao-yu;CAO Zi-ping;CUI Hong-tao School of Communication and Information Engineering,Nanjing University of Posts and Telecommunications,Nanjing 210003,China 关键词: 神经网络;超声波通信;回波消除;深度学习;盲源分离 Keywords: neural network;ultrasonic communication;echo cancellation;deep learning;blind source separation 分类号: TP399 DOI: 10. 3969 / j. issn. 1673-629X. 2023. 09. 028 摘要: 超声波是面向密闭金属腔体内外间进行无线通信时一种较为理想的信息传输媒介,回波是超声波穿金属无线通信系统的主要噪声来源。 针对现有超声波穿金属无线通信系统中回波消除技术的局限,该文引入深度学习方法来实现一种新的回波消除技术方案进行通信接收端信号信噪比的改善。 搭建了由摄像头、发射端 FPGA、发射端超声换能器、金属铝块、接收端超声换能器和接收端 FPGA 组成的试验型超声波穿金属无线通信系统,作为信道的铝块其厚度为 50 mm,基带频率为 10 MHz。 使用全卷积时域音频分离网络对超声波穿金属通信系统接收端的信号进行盲源分离,模型经过 45 次训练后较好实现了去除混合信号中的无用噪声而增强超声波信号的信噪比,在通信速率 2 Mbps 条件下实现了 13. 57 dB的 SDR 提升量和 39. 70 dB 的 SI-SDR 提升量。 Abstract: Ultrasonic wave is a kind of ideal information transmission medium for wireless communication between the inside and outsideof closed metal cavity. Echo is the?main noise source of ultrasonic wireless communication system through metal. Aiming at thelimitation of echo cancellation technology in the existing ultrasonic?metal-penetrating wireless communication system,we introduce a newecho cancellation technology scheme to improve the signal to noise ratio of communication receiver. An experimental ultrasonic metal-penetrating wireless communication system composed of camera, transmitter FPGA, transmitter ultrasonic transducer, metal - aluminumblock,receiver ultrasonic transducer and receiver FPGA was built. The thickness of the aluminum block as the channel was 50 mm and　the base　band frequency was 10 MHz. The blind source separation of the signals at the receiving end of the ultrasonic metalcommunication system is carried out by using the full convolution time domain audio separation network. After 45 training,the model caneffectively remove the useless noise in the mixed signal and enhance the signal - to - noise ratio of the ultrasonic signal. Under thecondition of communication rate of 2 Mbps,13. 57 dB SDR boost and 39. 70 dB SI-SDR boost are achieved.

摘要：超声波是面向密闭金属腔体内外间进行无线通信时一种较为理想的信息传输媒介,回波是超声波穿金属无线通信系统的主要噪声来源。 针对现有超声波穿金属无线通信系统中回波消除技术的局限,该文引入深度学习方法来实现一种新的回波消除技术方案进行通信接收端信号信噪比的改善。 搭建了由摄像头、发射端 FPGA、发射端超声换能器、金属铝块、接收端超声换能器和接收端 FPGA 组成的试验型超声波穿金属无线通信系统,作为信道的铝块其厚度为 50 mm,基带频率为 10 MHz。 使用全卷积时域音频分离网络对超声波穿金属通信系统接收端的信号进行盲源分离,模型经过 45 次训练后较好实现了去除混合信号中的无用噪声而增强超声波信号的信噪比,在通信速率 2 Mbps 条件下实现了 13. 57 dB的 SDR 提升量和 39. 70 dB 的 SI-SDR 提升量。

关键词：神经网络;超声波通信;回波消除;深度学习;盲源分离