CN-116388905-B - Millimeter wave and terahertz channel modeling method based on migration generation countermeasure network

Abstract

A millimeter wave and terahertz channel modeling method based on migration generation countermeasure network includes constructing a generated countermeasure network model in an off-line stage, pre-training by adopting a simulation data set, and performing migration learning and fine tuning on the pre-trained generated countermeasure network model to obtain millimeter wave and terahertz channels of the migration countermeasure network model. Compared with a standard channel model in the traditional third-generation mobile communication cooperation plan, the invention obtains good performance in the aspect of channel modeling, the root mean square error is improved by 9dB, the structural similarity index measurement is higher, the problem of rare measured data is solved, and the invention is verified in the actually measured channel.

Inventors

• HAN CHONG
• HU ZHENGDONG
• LI YUANBO

Assignees

• 上海交通大学

Dates

Publication Date

20260505

Application Date

20230404

Claims (1)

1. 1. A millimeter wave and terahertz channel modeling method based on migration generation of an countermeasure network is characterized by comprising the steps of constructing a generated countermeasure network model in an off-line stage, pretraining by adopting a simulation data set, and performing migration learning and fine tuning on the pretrained countermeasure network model to obtain millimeter wave and terahertz channels of the migration countermeasure network model, wherein a large amount of channel data matched with measured data are generated in an on-line stage, and the method specifically comprises the following steps: step 1, generating a simulation data set according to a standard channel model of a third generation mobile communication cooperation plan, wherein parameters of the simulation data set come from measurement data, and the parameters comprise delay spread, an angle domain and path loss parameters; the analog data set comprises 10000 channel data, each channel data is 401-dimensional power delay distribution, and the power in 400ns receiving time corresponds to the power; Step 2, training the simulation data set generated in the step 1 to generate an countermeasure network; The generating countermeasure network comprises a generator G and a discriminator D, wherein the generator G maps random noise into analog power delay distribution according to 100-dimensional random noise; The generator G and the discriminator D are respectively composed of five fully-connected layers, and the number of neurons is 128,128,128,128,401 and 512,256,128,64,1 respectively; the training is that the generator and the discriminator alternately train 10000 rounds in total, and the loss function is that Where E is the data expectation, x is the true data entered, G (z) is the false data generated, For randomly linear sampling between x and G (z), i.e. = x+ , Is an adjustable parameter for controlling gradient loss; And step 3, retraining the trained generated countermeasure network on the measured data to obtain the migration generated countermeasure network T-GAN, so as to generate channel data matched with the measured data in an online stage.

Description

Millimeter wave and terahertz channel modeling method based on migration generation countermeasure network Technical Field The invention relates to a technology in the field of wireless communication, in particular to a millimeter wave and terahertz channel modeling method based on migration generation of an countermeasure network. Background The existing random channel modeling method has the problem of low precision under certain assumed distribution and experience parameters. For example, a geometric location-based random channel model assumes that the scattered locations follow some statistical distribution, e.g., the transmitters and receivers are uniformly distributed within a circle, however, the scattered locations are difficult to characterize by some statistical distribution, making the geometric location-based random channel model inaccurate for use in the millimeter wave and terahertz bands. Furthermore, obtaining extensive channel measurements for millimeter wave and terahertz channel modeling is time consuming and expensive, and therefore lacks a large amount of measurement data. Existing high-frequency channel modeling techniques based on generating an countermeasure network generally improve the original channel modeling by generating the countermeasure network, and the techniques need a large amount of data for supervised learning and cannot be implemented in a small sample scene. Disclosure of Invention Aiming at the defects in the prior art, the invention provides a millimeter wave and terahertz channel modeling method based on migration generation of an countermeasure network, migration learning is applied to channel modeling based on generation of the countermeasure network, a large amount of millimeter wave and terahertz channel data generated by migration can be obtained by using a small amount of millimeter wave and terahertz channel measurement data sets, and the method can be used for assisting in the design of a communication system. The power delay distribution generated by the migration generation countermeasure network is well matched with the measurement result, compared with a standard channel model in the traditional third-generation mobile communication cooperation plan, the migration generation countermeasure network has good performance in the aspect of channel modeling, the root mean square error is improved by 9dB, the structural similarity index measurement is higher, the problem of rare measurement data is solved, and the migration generation countermeasure network is verified in an actually measured channel. The invention is realized by the following technical scheme: The invention relates to a millimeter wave and terahertz channel modeling method based on migration generation countermeasure network, which is characterized in that a countermeasure network model is built and generated in an off-line stage, an analog data set is adopted for pretraining, and then migration learning and fine tuning are carried out on the pretrained countermeasure network model to obtain millimeter wave and terahertz channels of the migration countermeasure network model, and a large amount of channel data which are consistent with measurement data can be generated in an on-line stage, so that the problem that the millimeter wave and terahertz lack of the measurement data is solved, and the design of millimeter wave and terahertz communication systems is facilitated. The simulated data set is generated by a standard channel model of a third generation mobile communication cooperation program. The pre-training refers to training the generation of the countermeasure network using the simulated data set. The transfer learning refers to transferring the knowledge learned from the simulation data set to the measurement data set. By fine tuning is meant retraining a generated challenge model that has been trained on a simulated dataset using the measured dataset. Technical effects According to the invention, the knowledge learned from the analog data set is migrated to the measurement data set by using migration learning, and the countermeasure network is generated based on the migration obtained by the migration learning, so that channel data which is highly matched with the measurement data can be generated, and the measurement data set is greatly expanded. Compared with the prior art, the invention greatly expands the measurement data set, thereby solving the problem of rare millimeter wave and terahertz data volume and being beneficial to the design of millimeter wave and terahertz communication systems. Drawings FIG. 1 is a schematic diagram of a generating countermeasure network in accordance with the present invention; FIG. 2 is a schematic diagram of transfer learning; fig. 3 is a measurement scenario of channel data; FIG. 4 is an average power delay profile; Fig. 5 is a schematic diagram of a similarity measure (SSIM) of the generated power delay profile and the real measurement data.