WO2021098501A1 - 基于生成对抗网络的无线信道建模实现方法 - Google Patents

基于生成对抗网络的无线信道建模实现方法 Download PDF

Info

Publication number
WO2021098501A1
WO2021098501A1 PCT/CN2020/126042 CN2020126042W WO2021098501A1 WO 2021098501 A1 WO2021098501 A1 WO 2021098501A1 CN 2020126042 W CN2020126042 W CN 2020126042W WO 2021098501 A1 WO2021098501 A1 WO 2021098501A1
Authority
WO
WIPO (PCT)
Prior art keywords
generator
discriminator
channel data
input
channel
Prior art date
Application number
PCT/CN2020/126042
Other languages
English (en)
French (fr)
Inventor
孙彦赞
朱文星
张舜卿
徐树公
吴雅婷
王涛
Original Assignee
上海大学
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 上海大学 filed Critical 上海大学
Publication of WO2021098501A1 publication Critical patent/WO2021098501A1/zh

Links

Images

Classifications

    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04B—TRANSMISSION
    • H04B17/00—Monitoring; Testing
    • H04B17/30—Monitoring; Testing of propagation channels
    • H04B17/391—Modelling the propagation channel
    • H04B17/3912—Simulation models, e.g. distribution of spectral power density or received signal strength indicator [RSSI] for a given geographic region
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks

Definitions

  • the present invention relates to a technology in the field of wireless communication, in particular to a method for implementing wireless channel modeling based on a generative confrontation network.
  • wireless channel modeling has always been a basic task for the theoretical analysis and practical application of wireless communication systems.
  • An accurate channel model can help understand the physical effects of different wireless channels on transmitted signals.
  • the existing channel generation methods mainly rely on certain parameters to characterize the quality of the complex wireless channel environment. These "parameterized" channels are obviously not suitable for the evaluation of network performance. Take the vehicle channel as an example. During the course of the vehicle itself, a certain dispersion effect and Doppler effect will be caused. There are many wireless channel environment parameters that may be affected.
  • the design and generation of the channel model must be complex enough to accurately reflect the characteristics of the channel. .
  • the present invention addresses the defects and deficiencies of the prior art for high-density and high-mobile communication channels, which are difficult to obtain accurate channel model parameters through traditional theoretical closed-form derivation due to the complexity of the channel model, and proposes a generation-based confrontation
  • the channel model implementation method of the network uses a generative confrontation network for wireless channel modeling, learns the statistical characteristics of the real channel, and can generate channel data with the same statistical characteristics and preprocess the input data of the discriminator. The statistical parameters of the channel data are extracted, and the input dimension of the discriminator is reduced.
  • the present invention uses real channel data and channel data generated by the generator to alternately train the discriminator and generator of the confrontation network until the discriminator cannot distinguish between the real channel data and the generated data, and the generator learns the channel data, thereby being used for Generate channel data with the same statistical characteristics to achieve the goal of establishing a stable channel and a generalized non-stationary channel model for the channel.
  • the real channel data refers to: After selecting the scene of the channel model and the corresponding channel parameters, the real channel data is obtained through the simulation platform or using a dedicated channel data collection tool.
  • the channel data generated by the generator refers to: using batches of random noise as the input of the generator to obtain the channel data samples generated by the generator.
  • the generator preferably uses uniformly distributed random noise as input.
  • the confrontation network includes a generator and a discriminator that both use a fully connected neural network, where: the discriminator includes two hidden layers, each hidden layer is 10 neurons, the activation function is sigmod, and the input of the discriminator is The real channel data and the mean, variance, kurtosis and skewness of the channel data generated by the generator; the generator includes two hidden layers, each of which is 5 neurons, the activation function is tanh, and the input of the generator is random noise.
  • the alternate training refers to: training the discriminator while fixing the parameters of the generator or training the generator while fixing the parameters of the discriminator, and the two are alternately iteratively trained until the discriminator cannot distinguish between false samples and real samples.
  • Figure 1 is a schematic diagram of an urban channel scene
  • Figure 2 is a schematic diagram of the Jakes channel model
  • Figure 3 is a schematic diagram of a framework for generating a confrontation network channel modeling
  • Figure 4 is a schematic diagram of the experimental results.
  • FIG. 1 is a schematic diagram of an urban channel scenario, and the actual implementation of the present invention is not limited to this channel scenario.
  • the specific steps of channel modeling for this scenario include:
  • Step 1 Select the scene of the channel model and the corresponding channel parameters:
  • the base station acts as a transmitter and is located in a fixed position; the car acts as a transmitter and drives at a constant speed in a fixed direction.
  • the movement of the car and the reflection of the transmitted signal cause the Doppler effect and the multipath effect, and because of the occlusion of the building, there is no line-of-sight link.
  • the channel can be regarded as a jake channel model.
  • the Doppler frequency is set to 926 Hz
  • the sampling time is set to 10 -6 s
  • 5 ⁇ 10 4 samples are collected each time.
  • Step 2 In order to better verify the accuracy of the result, simulate the channel model through the Matlab simulation platform according to the set parameters. By running the Matlab program, the real channel data can be obtained.
  • the real channel data obeys specific statistical characteristics. As can be seen from Figure 2, the amplitude of the channel value obeys the Ruili distribution, and the phase obeys the uniform distribution.
  • Step 3 the experimental platform used is Ubuntu16.04, Python3.6, and PyTorch0.4 GPU framework. Both the generation network and the discrimination network adopt the fully connected neural network.
  • the generator network structure and the discriminator network structure are as follows Show:
  • N is 5000.
  • Step 4 The statistical value obtained in step 3 is used as the input of the discriminator, and a random noise is used as the input of the generator.
  • a simple and lightweight generative confrontation network is constructed and trained.
  • the training method is that the discriminator network and the generator network are alternately iteratively trained 5000 times and merged. After the training, the network model parameters of the generator are retained, and each round includes :
  • b) Use the generated channel data as the input of the discriminator to calculate the loss: initialize the generator parameters randomly. Use noise as the input of the generator to get the generated channel data samples.
  • the generator adopts batch input, that is, input 5000 batches, and 5000 batches of generated samples can be obtained. Input the generated samples into the discriminator for training, set the label to false, and calculate the loss loss fake through the cross-entropy loss function.
  • Step 5 In the test phase, by loading the network model parameters of the trained generator, and inputting random noise to the generator network, the channel data that obeys the distribution characteristics of the real channel data can be obtained.
  • FIG. 4 which is the experimental result of this embodiment, it can be seen that after training the generation countermeasure network by sampling real channel data, the channel data generated by the generator approximately obeys the distribution characteristics of real channel data.
  • Fig. 4(a) shows that the amplitude value of the generated channel data approximately obeys the Ruili distribution, and Fig. 4(b) the phase approximately obeys the uniform distribution.
  • Figure 4(c) uses real channel data that obeys a normal distribution to train the generative countermeasure network, and the generated channel data is also approximately normal distribution.
  • the JS divergence is less than 0.08, indicating that the generated channel data can Accurately obey the real channel data distribution.

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • General Health & Medical Sciences (AREA)
  • Computing Systems (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Molecular Biology (AREA)
  • Artificial Intelligence (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Health & Medical Sciences (AREA)
  • Spectroscopy & Molecular Physics (AREA)
  • Electromagnetism (AREA)
  • Computer Networks & Wireless Communication (AREA)
  • Signal Processing (AREA)
  • Mobile Radio Communication Systems (AREA)

Abstract

一种基于生成对抗网络的信道模型实现方法，通过使用真实信道数据和生成器生成的信道数据对对抗网络的鉴别器和生成器交替训练，直至鉴别器无法分辨真实信道数据和生成数据，达到生成器对信道数据的学习，从而用于产生具有相同的统计特性的信道数据，实现对信道的建立平稳信道及广义非平稳信道模型的目标。本发明生成的信道数据JS散度均小于0.08，能够准确的服从真实信道数据分布。

Description

基于生成对抗网络的无线信道建模实现方法 技术领域
本发明涉及的是一种无线通信领域的技术，具体是一种基于生成对抗网络的无线信道建模实现方法。
背景技术
对于无线通信系统，无线信道建模对于无线通信系统的理论分析和实际应用一直是一个基础任务，准确的信道模型能够帮助理解不同无线信道对于传输信号的物理影响。现有的信道生成方式主要依赖于某些参数去表征复杂的无线信道环境的质量。这些“参数化”方式生成的信道在网络性能的评估过程中，显然是不太适用的。以车车信道为例，车辆本身的行进过程中，会造成一定的色散效应和多普勒效应，可能影响的无线信道环境参数非常多，设计生成信道的模型必须足够复杂才能精确反应信道的特征。
发明内容
本发明针对现有技术对于高密集、高移动的通信信道，由于其信道模型的复杂性，难以通过传统的理论闭式推导获得其精准的信道模型参数的缺陷和不足，提出一种基于生成对抗网络的信道模型实现方法，使用生成对抗网络用于无线信道建模，学习出真实信道的统计特性，并能够生成具有相同统计特性的信道数据且对于鉴别器的输入数据进行了预处理的工作，提取出信道数据的统计参数，减少鉴别器的输入维度。
本发明是通过以下技术方案实现的：
本发明通过使用真实信道数据和生成器生成的信道数据对对抗网络的鉴别器和生成器交替训练，直至鉴别器无法分辨真实信道数据和生成数据，达到生成器对信道数据的学习，从而用于产生具有相同的统计特性的信道数据，实现对信道的建立平稳信道及广义非平稳信道模型的目标。
所述的真实信道数据是指：选定信道模型的场景和相应的信道参数后通过仿真平台或者使用专用的信道数据采集工具，得到真实的信道数据。
所述的生成器生成的信道数据是指：使用批次的随机噪声作为生成器的输入，得到其生成的信道数据样本。
所述的生成器优选使用均匀分布的随机噪声作为输入。
所述的对抗网络，包括均采用全连接神经网络的生成器和鉴别器，其中：鉴别器包括两个隐藏层，每个隐藏层为10个神经元，激活函数为sigmod，鉴别器的输入为真实信道数据和 生成器生成的信道数据的均值、方差、峰度和偏度；生成器包括两个隐藏层，每个隐藏层为5个神经元，激活函数为tanh，生成器的输入为随机噪声。
所述的交替训练是指：固定生成器的参数的同时训练鉴别器或固定鉴别器的参数的同时训练生成器，两者交替迭代训练直至鉴别器无法鉴别虚假样本和真实样本。
附图说明
图1为城市信道场景示意图；
图2为jakes信道模型示意图；
图3为生成对抗网络信道建模的框架示意图；
图4为实验结果示意图。
具体实施方式
本实施例场景如图1所示，为城市信道场景的示意图，本发明实际实现不仅限于该信道场景。针对该场景信道建模的具体步骤包括：
步骤一、选定信道模型的场景和相应的信道参数：在该场景中基站作为发射机，位于固定位置；汽车作为发射机，沿着固定方向匀速行驶。在该场景中分别因为汽车的移动和发射信号的反射导致存在多普勒效应和多径效应，而且因为楼房的遮挡，使得不存在视距链路。使用传统的信道建模方法，该信道可视为jakes信道模型。在仿真中将多普勒频率设定为926Hz，采样时间设定为10 -6s，每次采集5×10 4个样本。
步骤二、为了更好的验证结果的准确性，根据设定的参数，通过Matlab仿真平台仿真该信道模型。通过运行Matlab程序，可以得到真实的信道数据。真实信道数据是服从特定的统计特性的，从图2可以看出，信道值的幅度服从瑞丽分布，相位服从均匀分布。
步骤三、在本实施例中，采用的实验平台为Ubuntu16.04、Python3.6、PyTorch0.4GPU框架，生成网络和鉴别网络均采用全连接神经网络，生成器网络结构和鉴别器网络结构如下所示：
  生成器 鉴别器
学习率 0.001 0.001
激活函数 Tanh Sigmod
批次大小 5000 1
输入大小 1 4
隐藏层层数 2 2
神经元数量 5,5 10,10
输出大小 1 1
训练回合 5000 5000
计算5000批次大小的生成信道数据或者采样的5000批次真实信道数据x i的均值
Figure PCTCN2020126042-appb-000001
方差
Figure PCTCN2020126042-appb-000002
峰度
Figure PCTCN2020126042-appb-000003
和偏度
Figure PCTCN2020126042-appb-000004
其中：N为5000。
步骤四、将步骤三得到的统计值作为鉴别器的输入，将一个随机噪声作为生成器的输入。通过上述参数设置构建得到简单轻量的生成对抗网络并进行训练，训练方式为鉴别器网络和生成器网络交替迭代训练5000回合并在训练结束后保留生成器的网络模型参数，其中每一回合包括：
a)使用真实信道数据作为鉴别器输入，计算损失：将真实样本进行数据处理得到四个统计值输入到鉴别器中，将真实信道数据的输出标签设为真，通过交叉熵损失函数计算损失loss real。
b)使用生成的信道数据作为鉴别器输入，计算损失：对生成器参数进行随机初始化。使用噪声作为生成器的输入，得到生成的信道数据样本。对于生成器采用批次的输入，即输入5000个批次，可得到5000个批次的生成样本。将生成样本输入到鉴别器中训练，标签设为假，通过交叉熵损失函数计算损失loss fake。
c)将步骤a和步骤b中的损失相加得到鉴别器的损失，即loss adv＝loss real+loss real；固定生成器网络模型的参数，通过对鉴别器损失函数进行反向传播，仅对鉴别器进行训练并重复该训练步骤20次。
i)使用批次的随机噪声作为生成器的输入，得到生成的信道数据样本，将批次的生成信道数据样本进行数据处理后，输入到鉴别器中，将生成信道数据的输出结果标签设为真；
ii)根据交叉熵损失函数计算生成器的损失loss gen，固定鉴别器网络模型的参数，通过对鉴别器损失函数进行反向传播，仅对生成器进行训练并重复该训练步骤20次。
步骤五、在测试阶段，通过加载训练后的生成器的网络模型参数，对生成器网络输入随机噪声便可得到服从真实信道数据分布特性的信道数据。
如图4所示，为本实施例实验结果，可以看出通过采样真实信道数据对生成对抗网络进行训练后，生成器生成的信道数据近似服从真实信道数据分布特性。图4(a)表示生成的信道数据幅度值近似服从瑞丽分布，图4(b)相位近似服从均匀分布。图4(c)使用服从正态分布真实信道数据对生成对抗网络进行训练，生成的信道数据同样近似正态分布。
通过常用的衡量数据概率分布的数值指标JS散度度量生成的信道数据分布与真实的信道数据分布，对于图4中的三对信道数据分布的JS散度均小于0.08，表明生成的信道数据能够准确的服从真实信道数据分布。
上述具体实施可由本领域技术人员在不背离本发明原理和宗旨的前提下以不同的方式对其进行局部调整，本发明的保护范围以权利要求书为准且不由上述具体实施所限，在其范围内的各个实现方案均受本发明之约束。

Claims (5)

  1. 一种基于生成对抗网络的信道模型实现方法，其特征在于，通过使用真实信道数据和生成器生成的信道数据对对抗网络的鉴别器和生成器交替训练，直至鉴别器无法分辨真实信道数据和生成数据，达到生成器对信道数据的学习，从而用于产生具有相同的统计特性的信道数据，实现对信道的建立平稳信道及广义非平稳信道模型的目标；
    所述的生成器生成的信道数据是指：使用批次的随机噪声作为生成器的输入，得到其生成的信道数据样本。
  2. 根据权利要求1所述的方法，其特征是，所述的生成器使用均匀分布的随机噪声作为输入。
  3. 根据权利要求1所述的方法，其特征是，所述的对抗网络，包括均采用全连接神经网络的生成器和鉴别器，其中：鉴别器包括两个隐藏层，每个隐藏层为10个神经元，激活函数为sigmod，鉴别器的输入为真实信道数据和生成器生成的信道数据的均值、方差、峰度和偏度；生成器包括两个隐藏层，每个隐藏层为5个神经元，激活函数为tanh，生成器的输入为随机噪声。
  4. 根据权利要求1所述的方法，其特征是，所述的交替训练是指：固定生成器的参数的同时训练鉴别器或固定鉴别器的参数的同时训练生成器，两者交替迭代训练直至鉴别器无法鉴别虚假样本和真实样本。
  5. 根据权利要求1或4所述的方法，其特征是，所述的交替训练包括以下步骤：
    1)计算所有样本的均值、方差、峰度和偏度；
    2)分别对鉴别器和生成器进行与样本个数相同次数的训练并在训练结束后保留生成器的网络模型参数，其中每一回合包括：
    a)使用真实信道数据作为鉴别器输入，计算损失：将真实样本的均值、方差、峰度和偏度输入到鉴别器中，将真实信道数据的输出标签设为真，通过交叉熵损失函数计算损失loss real；
    b)使用生成的信道数据作为鉴别器输入，计算损失：对生成器参数进行随机初始化；使用噪声作为生成器的输入，得到生成的信道数据样本；对于生成器采用批次的输入得到生成样本；将生成样本输入到鉴别器中训练，标签设为假，通过交叉熵损失函数计算损失loss fake；
    c)将步骤a和步骤b中的损失相加得到鉴别器的损失，即loss adv＝loss real+loss real；固定生成器网络模型的参数，通过对鉴别器损失函数进行反向传播，仅对鉴别器进行训练并重复该训练步骤20次，具体为：
    i)使用批次的随机噪声作为生成器的输入，得到生成的信道数据样本，将批次的生成信道数据样本进行数据处理后，输入到鉴别器中，将生成信道数据的输出结果标签设为真；
    ii)根据交叉熵损失函数计算生成器的损失loss gen，固定鉴别器网络模型的参数，通过对鉴别器损失函数进行反向传播，仅对生成器进行训练并重复该训练步骤20次。
PCT/CN2020/126042 2019-11-19 2020-11-03 基于生成对抗网络的无线信道建模实现方法 WO2021098501A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201911131740.8 2019-11-19
CN201911131740.8A CN110875790A (zh) 2019-11-19 2019-11-19 基于生成对抗网络的无线信道建模实现方法

Publications (1)

Publication Number Publication Date
WO2021098501A1 true WO2021098501A1 (zh) 2021-05-27

Family

ID=69718191

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2020/126042 WO2021098501A1 (zh) 2019-11-19 2020-11-03 基于生成对抗网络的无线信道建模实现方法

Country Status (2)

Country Link
CN (1) CN110875790A (zh)
WO (1) WO2021098501A1 (zh)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113449468A (zh) * 2021-06-22 2021-09-28 北京交通大学 一种锂离子电池组的多参数分布特性建模和参数生成方法
CN116389287A (zh) * 2023-05-29 2023-07-04 北京理工大学 一种模分复用通信系统的信道构建方法
CN116996148A (zh) * 2023-07-17 2023-11-03 哈尔滨工程大学 基于生成对抗网络的极地环境水下声信道建模方法及装置

Families Citing this family (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110875790A (zh) * 2019-11-19 2020-03-10 上海大学 基于生成对抗网络的无线信道建模实现方法
CN111355675B (zh) * 2020-03-11 2021-05-18 南京航空航天大学 一种基于生成对抗网络的信道估计增强方法、装置和系统
CN112446002B (zh) * 2020-11-13 2022-11-15 天津大学 一种面向时序kpi数据的异常检测方法
WO2023070675A1 (zh) * 2021-11-01 2023-05-04 Oppo广东移动通信有限公司 数据处理的方法及装置
CN117678172A (zh) * 2021-11-02 2024-03-08 Oppo广东移动通信有限公司 信道数据生成方法、装置、设备及存储介质
WO2023097636A1 (zh) * 2021-12-03 2023-06-08 Oppo广东移动通信有限公司 数据处理的方法和装置
WO2023097645A1 (zh) * 2021-12-03 2023-06-08 Oppo广东移动通信有限公司 数据获取方法、装置、设备、介质、芯片、产品及程序
CN114584230B (zh) * 2022-03-07 2023-06-20 东南大学 一种基于对抗网络与长短期记忆网络的预测信道建模方法
CN115208494B (zh) * 2022-07-25 2023-09-22 中国人民解放军国防科技大学 一种基于条件生成对抗网络的信道切换策略学习方法

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107293289A (zh) * 2017-06-13 2017-10-24 南京医科大学 一种基于深度卷积生成对抗网络的语音生成方法
CN108696331A (zh) * 2018-03-19 2018-10-23 西安电子科技大学 一种基于生成对抗网络的信号重构方法
WO2019161342A1 (en) * 2018-02-17 2019-08-22 Regeneron Pharmaceuticals, Inc. Gan-cnn for mhc peptide binding prediction
CN110289927A (zh) * 2019-07-01 2019-09-27 上海大学 基于条件生成对抗网络的信道模拟实现方法
CN110875790A (zh) * 2019-11-19 2020-03-10 上海大学 基于生成对抗网络的无线信道建模实现方法

Family Cites Families (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2973292B2 (ja) * 1996-12-10 1999-11-08 日本電気株式会社 Ｐｃｍ試験信号発生器の試験信号パターン設定装置

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107293289A (zh) * 2017-06-13 2017-10-24 南京医科大学 一种基于深度卷积生成对抗网络的语音生成方法
WO2019161342A1 (en) * 2018-02-17 2019-08-22 Regeneron Pharmaceuticals, Inc. Gan-cnn for mhc peptide binding prediction
CN108696331A (zh) * 2018-03-19 2018-10-23 西安电子科技大学 一种基于生成对抗网络的信号重构方法
CN110289927A (zh) * 2019-07-01 2019-09-27 上海大学 基于条件生成对抗网络的信道模拟实现方法
CN110875790A (zh) * 2019-11-19 2020-03-10 上海大学 基于生成对抗网络的无线信道建模实现方法

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113449468A (zh) * 2021-06-22 2021-09-28 北京交通大学 一种锂离子电池组的多参数分布特性建模和参数生成方法
CN113449468B (zh) * 2021-06-22 2024-03-12 北京交通大学 一种锂离子电池组的多参数分布特性建模和参数生成方法
CN116389287A (zh) * 2023-05-29 2023-07-04 北京理工大学 一种模分复用通信系统的信道构建方法
CN116389287B (zh) * 2023-05-29 2023-08-18 北京理工大学 一种模分复用通信系统的信道构建方法
CN116996148A (zh) * 2023-07-17 2023-11-03 哈尔滨工程大学 基于生成对抗网络的极地环境水下声信道建模方法及装置

Also Published As

Publication number Publication date
CN110875790A (zh) 2020-03-10

Similar Documents

Publication Publication Date Title
WO2021098501A1 (zh) 基于生成对抗网络的无线信道建模实现方法
CN108805269B (zh) 一种基于lstm循环神经网络拾取震相到时的方法
US20210197335A1 (en) Data Augmentation Method Based On Generative Adversarial Networks In Tool Condition Monitoring
CN110189769B (zh) 基于多个卷积神经网络模型结合的异常声音检测方法
CN107247259B (zh) 基于神经网络的k分布海杂波形状参数估计方法
CN108919353B (zh) 一种微震波形初至到时的自动分级拾取与优选方法
CN105160678A (zh) 基于卷积神经网络的无参考立体图像质量评估方法
CN106353651A (zh) 基于bp网络的gis局放声电联合检测故障定位方法
CN108761287B (zh) 一种变压器局部放电类型识别方法
CN106408088A (zh) 一种基于深度学习理论的旋转机械设备故障诊断方法
US9465121B2 (en) System and method for on-site instant seismic analysis
CN105334495B (zh) 一种无线网络中基于信号到达时间的非视距稳健定位方法
CN112949387B (zh) 基于迁移学习的智能抗干扰目标检测方法
CN104683051B (zh) 一种无线多径衰落信道仿真方法及信道仿真器
CN112990082B (zh) 一种水声脉冲信号的检测识别方法
CN104634872A (zh) 高铁钢轨伤损在线监测方法
CN111580151A (zh) 一种基于SSNet模型的地震事件到时识别方法
CN110988263B (zh) 一种基于改进的Stacking模型的车辆尾气浓度估计方法
CN110398775B (zh) 隧道突涌水灾害微震事件信号波动初至拾取方法及系统
CN113640712B (zh) 一种舰船垂向感应磁场垂向分量的预测方法
CN110057918A (zh) 强噪声背景下的复合材料损伤定量识别方法及系统
CN112146882B (zh) 基于迁移学习振动信号图像识别的轴承故障诊断方法
EP3321674B1 (en) Structure evaluation system, structure evaluation device, and structure evaluation method
Qian et al. Application of new training method combined with feedforward artificial neural network for rolling bearing fault diagnosis
CN112364296A (zh) 一种基于深度学习的p波到时自动拾取方法

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 20890817

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 20890817

Country of ref document: EP

Kind code of ref document: A1

32PN Ep: public notification in the ep bulletin as address of the adressee cannot be established

Free format text: NOTING OF LOSS OF RIGHTS PURSUANT TO RULE 112(1) EPC (EPO FORM 1205A DATED 04/11/2022)

122 Ep: pct application non-entry in european phase

Ref document number: 20890817

Country of ref document: EP

Kind code of ref document: A1