AI network model matching method and device, storage medium, and user equipment

By adjusting the size of the initial channel estimation matrix to be consistent with the input size of the AI ​​network model, the problem of channel estimation of different PDSCH resource sizes is solved, the number of AI models is reduced, and the accuracy and performance of channel estimation is improved.

CN114070675BActive Publication Date: 2024-12-06SPREADTRUM SEMICON (NANJING) CO LTD
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202010779215.3
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2020-08-05
Publication Date
2024-12-06
Estimated Expiration
2040-08-05

AI Technical Summary

Technical Problem

The existing AI network model can only be used for channel estimation of fixed PDSCH resource sizes, and channel estimation of different PDSCH resource sizes cannot be implemented, resulting in an increase in the number of required AI models and limited resource configuration and accuracy.

Method used

By obtaining the initial channel estimation matrix and adjusting it according to the input size of the current AI network model, including expansion or intercepting of boundary data in the time and frequency domains, ensuring that the matrix size is consistent with the input size of the AI ​​network model, thereby achieving different PDSCHs Channel estimation of resource size.

Benefits of technology

Different PDSCH resource sizes are implemented to use the same AI network model for channel estimation, reducing the number of AI models required, saving resources, and improving the accuracy and performance of channel estimation.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN114070675B_ABST
    Figure CN114070675B_ABST
Patent Text Reader

Abstract

An AI network model matching method and device, storage medium, and user equipment, the AI ​​network model matching method comprising: obtaining an initial channel estimation matrix to be input into the current AI network model, the size of the initial channel estimation matrix being the transmission resource size of the PDSCH; if the size of the initial channel estimation matrix is ​​smaller than the input size of the current AI network model, adjusting the size of the initial channel estimation matrix according to the input size of the current AI network model so that the size of the adjusted initial channel estimation matrix is ​​consistent with the input size of the current AI network model. The technical solution of the present invention can realize that different PDSCH resource sizes can use the same AI network model for channel estimation to reduce the number of required AI models.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of communication technologies, and in particular to an AI network model matching method and device, a storage medium, and a user equipment. Background Art

[0002] In channel estimation, the purpose of applying artificial intelligence (AI) algorithms is to compare the channel estimation process, that is, the process of estimating all channel values ​​from pilot signals, to the traditional image restoration / denoising process, and use deep learning algorithms for image restoration / denoising to complete channel estimation.

[0003] At present, AI network models have strict requirements on the size of data input and output, that is, the data size input to the AI ​​network model (such as dimension size, the value of each dimension) needs to be consistent with the input size corresponding to the AI ​​network model.

[0004] However, in the communication system, the size of the physical downlink shared channel (PDSCH) (such as bandwidth / number of time domain symbols) is dynamically indicated by the downlink control information (DCI). Therefore, for a specific AI model, only channel estimation with a fixed PDSCH transmission resource size can be achieved. How to achieve channel estimation for different PDSCH sizes using the same AI network model to reduce the number of AI models required is a problem that needs to be overcome in the current application of AI to channel estimation. Summary of the invention

[0005] The technical problem solved by the present invention is how to achieve channel estimation using the same AI network model for different PDSCH resource sizes to reduce the number of required AI models.

[0006] To solve the above technical problems, an embodiment of the present invention provides an AI network model matching method, which includes: obtaining an initial channel estimation matrix to be input into the current AI network model, the size of the initial channel estimation matrix being the transmission resource size of the PDSCH; if the size of the initial channel estimation matrix is ​​smaller than the input size of the current AI network model, adjusting the size of the initial channel estimation matrix according to the input size of the current AI network model so that the size of the adjusted initial channel estimation matrix is ​​consistent with the input size of the current AI network model.

[0007] Optionally, adjusting the size of the initial channel estimation matrix according to the input size of the current AI network model includes: at least expanding the initial channel matrix using boundary data in the time domain and / or frequency domain so that the size of the expanded initial channel estimation matrix is ​​consistent with the input size of the current AI network model.

[0008] Optionally, the expansion using at least the boundary data of the initial channel matrix in the time domain and / or frequency domain includes: selecting the boundary data of the initial channel matrix in the time domain and / or frequency domain, the boundary data of the initial channel matrix in the time domain occupies a single symbol, and the boundary data of the initial channel matrix in the frequency domain occupies a single subcarrier; and using the boundary data to expand the initial channel matrix in the time domain and / or frequency domain.

[0009] Optionally, the use of the boundary data to perform time domain and / or frequency domain expansion on the initial channel matrix includes: copying the boundary data located in the time domain to the boundary position of the initial channel matrix in the time domain, and / or copying the boundary data located in the frequency domain to the boundary position of the initial channel matrix in the frequency domain, until the size of the expanded initial channel estimation matrix is ​​consistent with the input size of the current AI network model.

[0010] Optionally, using the boundary data to perform time domain and / or frequency domain expansion on the initial channel matrix includes: if the size of the initial channel estimation matrix is ​​different from the input size of the current AI network model in both the time domain and the frequency domain, first copying the boundary data located in the time domain to the boundary position of the initial channel matrix in the time domain, and then copying the boundary data located in the frequency domain to the boundary position of the initial channel matrix in the frequency domain, until the size of the expanded initial channel estimation matrix is ​​consistent with the input size of the current AI network model in both the time domain and the frequency domain; or, if the size of the initial channel estimation matrix is ​​different from the input size of the current AI network model in both the time domain and the frequency domain, first copying the boundary data located in the frequency domain to the boundary position of the initial channel matrix in the time domain, and then copying the boundary data located in the time domain to the boundary position of the initial channel matrix in the frequency domain, until the size of the expanded initial channel estimation matrix is ​​consistent with the input size of the current AI network model in both the time domain and the frequency domain.

[0011] Optionally, the expansion using at least the boundary data of the initial channel matrix in the time domain and / or frequency domain includes: intercepting the data of the initial channel matrix in the time domain and / or frequency domain, the difference between the transmission resource size of the PDSCH in the time domain and / or frequency domain and the input size of the current AI network model in the time domain and / or frequency domain is an integer multiple of the intercepted data size; and using the intercepted time domain and / or frequency domain data to expand the initial channel matrix in the time domain and / or frequency domain.

[0012] Optionally, the use of the intercepted time domain and / or frequency domain data to expand the initial channel matrix in the time domain and / or frequency domain includes: if the size of the initial channel estimation matrix is ​​different from the input size of the current AI network model in both the time domain and the frequency domain, the intercepted data located in the time domain is first copied to the boundary position of the initial channel matrix in the time domain, and then the intercepted data located in the frequency domain is copied to the boundary position of the initial channel matrix in the frequency domain, until the size of the expanded initial channel estimation matrix is ​​consistent with the input size of the current AI network model in both the time domain and the frequency domain; or, if the size of the initial channel estimation matrix is ​​different from the input size of the current AI network model in both the time domain and the frequency domain, the intercepted data located in the frequency domain is first copied to the boundary position of the initial channel matrix in the time domain, and then the intercepted data located in the time domain is copied to the boundary position of the initial channel matrix in the frequency domain, until the size of the expanded initial channel estimation matrix is ​​consistent with the input size of the current AI network model in both the time domain and the frequency domain.

[0013] Optionally, the expansion of at least the initial channel matrix using boundary data in the time domain and / or frequency domain includes: if the difference between the transmission resource size of the PDSCH in the time domain and / or frequency domain and the input size of the current AI network model in the time domain and / or frequency domain is an integer multiple of the size of the initial channel matrix, then the initial channel matrix itself is used to expand the initial channel matrix.

[0014] Optionally, before adjusting the size of the initial channel estimation matrix according to the input size of the current AI network model, the method also includes: selecting one or more AI network models whose number of symbols occupied by the input size is closest to the number of symbols occupied by the resources of the PDSCH as a first AI network model set; selecting an AI network model in the first AI network model set whose number of subcarriers occupied by the input size is closest to the number of subcarriers occupied by the resources of the PDSCH as the current AI network model.

[0015] Optionally, before adjusting the size of the initial channel estimation matrix according to the input size of the current AI network model, the method also includes: selecting one or more AI network models whose number of subcarriers occupied by the input size is closest to the number of subcarriers occupied by the resources of the PDSCH as a second AI network model set; selecting an AI network model in the second AI network model set whose number of symbols occupied by the input size is closest to the number of symbols occupied by the resources of the PDSCH as the current AI network model.

[0016] Optionally, before adjusting the size of the initial channel estimation matrix according to the input size of the current AI network model, the method also includes: if the difference in the number of symbols between the input size of the current AI network model and the transmission resource size of the PDSCH in the time domain is greater than a first preset threshold, and / or the difference in the number of subcarriers between the input size of the current AI network model and the transmission resource size of the PDSCH in the frequency domain is greater than a second preset threshold, then using a traditional channel estimation method to perform channel estimation.

[0017] Optionally, before performing channel estimation using the traditional channel estimation method, the method also includes: receiving the first preset threshold and / or the second preset threshold from the base station, or determining the first preset threshold and / or the second preset threshold based on the performance of the current AI network model.

[0018] To solve the above technical problems, an embodiment of the present invention further discloses an AI network model matching device, the AI ​​network model matching device comprising: an acquisition module, used to acquire an initial channel estimation matrix to be input into the current AI network model, the size of the initial channel estimation matrix being the transmission resource size of the PDSCH; an adjustment module, used to adjust the size of the initial channel estimation matrix according to the input size of the current AI network model if the size of the initial channel estimation matrix is ​​smaller than the input size of the current AI network model, so that the size of the adjusted initial channel estimation matrix is ​​consistent with the input size of the current AI network model.

[0019] An embodiment of the present invention further discloses a storage medium on which a computer program is stored. When the computer program is executed by a processor, the steps of the AI ​​network model matching method are executed.

[0020] An embodiment of the present invention also discloses a user device, including a memory and a processor, wherein the memory stores a computer program that can be run on the processor, and the processor executes the steps of the AI ​​network model matching method when running the computer program.

[0021] Compared with the prior art, the technical solution of the embodiment of the present invention has the following beneficial effects:

[0022] In the technical solution of the present invention, an initial channel estimation matrix to be input into the current AI network model is obtained, and the size of the initial channel estimation matrix is ​​the transmission resource size of the PDSCH; if the size of the initial channel estimation matrix is ​​smaller than the input size of the current AI network model, the size of the initial channel estimation matrix is ​​adjusted according to the input size of the current AI network model, so that the size of the adjusted initial channel estimation matrix is ​​consistent with the input size of the current AI network model. When the size of the initial channel estimation matrix does not match the input size of the current AI network model, the technical solution of the present invention adjusts the size of the initial channel estimation matrix so that the size of the adjusted initial channel estimation matrix is ​​consistent with the input size of the current AI network model, thereby achieving different PDSCH resource sizes using the same AI network model for channel estimation, thereby reducing the number of required AI models, saving configuration resources on the user equipment side, and improving the accuracy of channel estimation.

[0023] Furthermore, the technical solution of the present invention can also configure multiple AI network models on the user equipment side. When the input sizes of the multiple AI network models are different, the difference between the transmission resource size of PDSCH and the input size of the current AI network model in the time domain and / or frequency domain is used to select the optimal AI network model for channel estimation, thereby improving the performance of channel estimation.

[0024] Furthermore, in some cases, the UE does not perform well in channel estimation using the AI ​​network model, and the UE needs to fall back to the traditional channel estimation method. The present invention also designs a channel estimation fallback mechanism to address this problem, that is, when certain conditions are met, the UE falls back to the traditional channel estimation algorithm. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 is a flow chart of an AI network model matching method according to an embodiment of the present invention;

[0026] Figure 2 is a schematic diagram of a specific application scenario of an embodiment of the present invention;

[0027] Figure 3 This is a partial flow chart of an AI network model matching method according to an embodiment of the present invention;

[0028] Figure 4 It is a partial flow chart of another AI network model matching method according to an embodiment of the present invention;

[0029] Figure 5 is a schematic diagram of another specific application scenario of an embodiment of the present invention;

[0030] Figure 6 This is a partial flow chart of another AI network model matching method according to an embodiment of the present invention;

[0031] Figure 7 It is a structural diagram of an AI network model matching device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0032] As described in the background technology, in the communication system, the size of the physical downlink shared channel (PDSCH) (such as bandwidth / number of time domain symbols) is dynamically indicated by DCI. Then for a specific AI model, only channel estimation of PDSCH transmission resources of a fixed size can be achieved. How to achieve channel estimation using the same AI network model for different PDSCH sizes to reduce the number of required AI models is a problem that needs to be overcome in the current application of AI to channel estimation.

[0033] When the size of the initial channel estimation matrix does not match the input size of the current AI network model, the technical solution of the present invention adjusts the size of the initial channel estimation matrix so that the size of the adjusted initial channel estimation matrix is ​​consistent with the input size of the current AI network model, thereby achieving different PDSCH resource sizes using the same AI network model for channel estimation, thereby reducing the number of required AI models, saving configuration resources on the user equipment side, and improving the accuracy of channel estimation.

[0034] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0035] Figure 1 It is a flowchart of an AI network model matching method according to an embodiment of the present invention.

[0036] Figure 1 The AI ​​network model matching method shown can be used on the user equipment (UE) side, that is, the steps of the method can be performed by the user equipment. The AI ​​network model matching method can specifically include the following steps:

[0037] Step S101: obtaining an initial channel estimation matrix to be input into the current AI network model, wherein the size of the initial channel estimation matrix is ​​the transmission resource size of the PDSCH;

[0038] Step S102: If the size of the initial channel estimation matrix is ​​smaller than the input size of the current AI network model, the size of the initial channel estimation matrix is ​​adjusted according to the input size of the current AI network model so that the size of the adjusted initial channel estimation matrix is ​​consistent with the input size of the current AI network model.

[0039] It should be pointed out that the serial numbers of the steps in this embodiment do not limit the execution order of the steps.

[0040] The current AI network model is used to perform channel estimation according to the input channel estimation matrix, which can be implemented using any appropriate AI network model, for example, it can be a model obtained by training based on historical data. In this embodiment, the user equipment side can only set a single AI network model, that is, the current AI network model, to reduce the overhead of the AI ​​network model and the storage overhead of the user equipment. Alternatively, the user equipment sets multiple AI network models, and different AI network models have different input sizes. When channel estimation is required, the user equipment selects an AI network model from multiple AI network models as the current AI network model.

[0041] The specific selection method of the current AI network model can be random selection or selecting an optimal AI network model. The specific implementation method of the user equipment selecting the optimal AI network model from multiple AI network models can refer to the subsequent embodiments and will not be repeated here.

[0042] It should be noted that the output size and input size of the AI ​​network model are consistent.

[0043] In the specific implementation of step S101, the input to the AI ​​network model is the initial channel estimation matrix, which can be obtained by linear interpolation based on the least squares (LS) pilot estimation value. The size of the initial channel estimation matrix is ​​the transmission resource size of the PDSCH, that is, the size of the initial channel estimation matrix is ​​N×T, where N is the number of subcarriers of the scheduled PDSCH and T is the number of time domain symbols of the PDSCH. The output of the AI ​​network model is the final channel estimation matrix, and the size of the final channel estimation matrix is ​​consistent with the size of the initial channel estimation matrix, that is, N×T.

[0044] Compared with traditional channel estimation methods, the initial channel estimation is filtered / denoised by the AI ​​network model, making the output channel estimation value more accurate.

[0045] The size of the initial channel estimation matrix and the input size of the current AI network model both include two dimensions, time domain and frequency domain. If the size of the initial channel estimation matrix is ​​inconsistent with the input size of the current AI network model, such as the number of time domain symbols is inconsistent, or the number of frequency domain subcarriers is inconsistent, or both the number of time domain symbols and the number of frequency domain subcarriers are inconsistent, it will lead to inaccurate channel estimation. Therefore, in the specific implementation of step S102, it is necessary to adjust the size of the initial channel estimation matrix according to the input size of the current AI network model so that the size of the adjusted initial channel estimation matrix is ​​consistent with the input size of the current AI network model.

[0046] Specifically, when the size of the initial channel estimation matrix is ​​smaller than the input size of the current AI network model, the size of the initial channel estimation matrix may be expanded.

[0047] In a non-limiting embodiment of the present invention, Figure 1 The step S102 shown may include the following steps a: at least using the boundary data of the initial channel matrix in the time domain and / or frequency domain to expand so that the size of the expanded initial channel estimation matrix is ​​consistent with the input size of the current AI network model.

[0048] In this embodiment, the boundary data refers to the estimated value corresponding to the boundary resource element (RE) in the initial channel matrix. The boundary resource element may be an RE at the boundary position of the initial channel, for example, an RE at the boundary position of the time domain or an RE at the boundary position of the frequency domain.

[0049] If the number of symbols of the initial channel matrix in the time domain is less than the number of symbols occupied by the input size of the current AI network model in the time domain, the initial channel matrix can be expanded using the boundary data of the initial channel matrix in the time domain. Correspondingly, if the number of subcarriers of the initial channel matrix in the frequency domain is less than the number of subcarriers occupied by the input size of the current AI network model in the frequency domain, the initial channel matrix can be expanded using the boundary data of the initial channel matrix in the frequency domain.

[0050] The boundary data of the initial channel matrix in the time domain may refer to the data at the boundary position of the initial channel matrix in the time domain dimension, and the boundary data may occupy one or more symbols; accordingly, the boundary data of the initial channel matrix in the frequency domain may refer to the data at the boundary position of the initial channel matrix in the frequency domain dimension, and the boundary data may occupy one or more subcarriers.

[0051] Combination Figure 2 , the left and right directions in the figure represent the time domain dimension, and the up and down directions represent the frequency domain dimension. The shaded part shown in the figure is the boundary data of the initial channel matrix in the frequency domain.

[0052] For further information, please refer to Figure 3 The above step a may specifically include the following steps:

[0053] Step S301: selecting boundary data of the initial channel matrix in the time domain and / or frequency domain, wherein the boundary data of the initial channel matrix in the time domain occupies a single symbol, and the boundary data of the initial channel matrix in the frequency domain occupies a single subcarrier;

[0054] Step S302: using the boundary data to perform time domain and / or time domain expansion on the initial channel matrix.

[0055] In this embodiment, the boundary data occupies a single symbol or a single subcarrier. When the boundary data is used to expand the initial channel matrix, the number of required boundary data can be determined based on the difference between the size of the initial channel matrix and the input size of the current AI network model. For example, if the number of symbols of the initial channel matrix in the time domain is 4, and the number of symbols of the current AI network model in the time domain is 8, then the initial channel matrix needs to be expanded in the time domain by 4 time domain boundary data; similarly, if the number of subcarriers of the initial channel matrix in the frequency domain is 8, and the number of subcarriers of the current AI network model in the frequency domain is 12, then the initial channel matrix needs to be expanded in the frequency domain by 4 frequency domain boundary data.

[0056] Finally, the size of the expanded initial channel matrix is ​​consistent with the input size of the current AI network model in both time and frequency domains.

[0057] Furthermore, step S302 may include the following steps: copying the boundary data in the time domain to the boundary position of the initial channel matrix in the time domain, and / or copying the boundary data in the frequency domain to the boundary position of the initial channel matrix in the frequency domain, until the size of the expanded initial channel estimation matrix is ​​consistent with the input size of the current AI network model.

[0058] This embodiment describes in detail the specific method of expanding the initial channel matrix using the boundary data, that is, the expansion of the initial channel matrix in the time domain or frequency domain is achieved by copying the boundary data.

[0059] Combination Figure 2 It can be seen that the boundary data occupies a single subcarrier. The difference between the number of subcarriers in the time domain of the initial channel matrix and the current AI network model in the time domain is 4. Then, at the time domain boundary position of the initial channel matrix, that is, Figure 2 The upper boundary of the initial channel matrix shown is extended by 4 boundary data.

[0060] For further information, please refer to Figure 4 The above step a may specifically include the following steps:

[0061] Step S401: intercepting data of the initial channel matrix in the time domain and / or frequency domain, and the difference between the transmission resource size of the PDSCH in the time domain and / or frequency domain and the input size of the current AI network model in the time domain and / or frequency domain is an integer multiple of the intercepted data size;

[0062] Step S402: using the intercepted time domain and / or frequency domain data to perform time domain and / or frequency domain expansion on the initial channel matrix.

[0063] Unlike the aforementioned embodiment where the boundary data only occupies a single symbol / subcarrier, the intercepted data in the embodiment of the present invention may occupy multiple symbols or subcarriers. Expanding the initial channel matrix using boundary data requires repeating the boundary data multiple times, while expanding the initial channel matrix using intercepted data can reduce the number of repeated copies.

[0064] In a specific implementation, the data intercepted in the time domain may be located at the time domain boundary position of the initial channel matrix, and the data intercepted in the frequency domain may be located at the frequency domain boundary position of the initial channel matrix.

[0065] Refer to Figure 5 , the difference between the number of subcarriers of the initial channel matrix in the time domain and the current AI network model in the time domain is 4, then the data with the number of symbols 4 is intercepted directly at the time domain boundary of the initial channel matrix, and at the time domain boundary of the initial channel matrix, that is, Figure 5 The upper bound of the original channel matrix is ​​shown to extend the truncated data.

[0066] Furthermore, if the difference between the transmission resource size of the PDSCH in the time domain and / or frequency domain and the input size of the current AI network model in the time domain and / or frequency domain is an integer multiple of the size of the initial channel matrix, the initial channel matrix itself is used to expand the initial channel matrix.

[0067] In this embodiment, the data used to expand the initial channel matrix is ​​the initial channel matrix itself, which is performed on the premise that the difference between the transmission resource size of the PDSCH and the input size of the current AI network model is an integer multiple of the size of the initial channel matrix.

[0068] In a specific implementation, the number of expansions may be an integer multiple of the difference and the size of the initial channel matrix. For example, if the difference is 1 times the size of the initial channel matrix, the initial channel matrix is ​​expanded once using the initial channel matrix itself; if the difference is 2 times the size of the initial channel matrix, the initial channel matrix is ​​expanded twice using the initial channel matrix itself.

[0069] Further, if the size of the initial channel estimation matrix is ​​different from the input size of the current AI network model in both the time domain and the frequency domain, the initial channel matrix itself is first copied to the boundary position of the initial channel matrix in the time domain, and then the initial channel matrix itself is copied to the boundary position of the initial channel matrix in the frequency domain, until the size of the expanded initial channel estimation matrix is ​​consistent with the input size of the current AI network model in both the time domain and the frequency domain; or, if the size of the initial channel estimation matrix is ​​different from the input size of the current AI network model in both the time domain and the frequency domain, the initial channel matrix itself is first copied to the boundary position of the initial channel matrix in the time domain, and then the initial channel matrix itself is copied to the boundary position of the initial channel matrix in the frequency domain, until the size of the expanded initial channel estimation matrix is ​​consistent with the input size of the current AI network model in both the time domain and the frequency domain.

[0070] Continue to refer to Figure 5 , the difference between the number of subcarriers of the initial channel matrix in the time domain and the current AI network model in the time domain is 4, and the number of symbols occupied by the initial channel matrix is ​​also 4, then the initial channel matrix is ​​directly expanded at the time domain boundary of the initial channel matrix. Of course, if the difference between the number of subcarriers of the initial channel matrix in the time domain and the current AI network model in the time domain is 8, then the initial channel matrix can be copied twice at the time domain boundary of the initial channel matrix.

[0071] In a non-limiting embodiment of the present invention, please refer to Figure 6 , Figure 1 The method may further include the following steps:

[0072] Step S601: selecting one or more AI network models whose number of symbols occupied by the input size is closest to the number of symbols occupied by the PDSCH resources as a first AI network model set;

[0073] Step S602: Selecting an AI network model in the first AI network model set whose number of subcarriers occupied by the input size is closest to the number of subcarriers occupied by the PDSCH resources as the current AI network model.

[0074] In order to improve the performance of AI channel estimation, multiple sets of AI network models will be saved on the user equipment side. The larger the difference between the PDSCH Size scheduled by DCI and the input Size corresponding to the AI ​​network model, the worse the AI ​​channel estimation performance; the difference between the PDSCH Size and the input Size corresponding to the AI ​​network model refers to the difference between the number of PDSCH subcarriers / number of time domain symbols scheduled by DCI and the input Size corresponding to the AI ​​network model. Therefore, in this embodiment, when the user equipment sets multiple AI network models, an optimal AI network model can be selected as the current AI network model.

[0075] In this embodiment, the UE first selects the AI ​​network model subset whose input size occupies the closest number of time domain symbols according to the number of time domain symbols of the DCI-scheduled PDSCH, that is, the first AI network model set, and the first AI network model set may include one or more AI network models. Then, in the case where the first AI network model set includes multiple AI network models, the UE selects the AI ​​network model whose input size occupies the closest number of frequency domain subcarriers according to the DCI-scheduled PDSCH bandwidth (that is, the number of frequency domain subcarriers) in the first AI network model set as the current AI network model. Of course, if the first AI network model set includes only one AI network model, step S602 is no longer executed.

[0076] In a specific implementation, when selecting the closest AI network model, it can be determined based on the difference between the number of symbols occupied by the input size and the number of symbols occupied by the PDSCH resources, or the difference between the number of subcarriers occupied by the input size and the number of subcarriers occupied by the PDSCH resources and a preset threshold. For example, if the above difference is less than the preset threshold, the AI ​​network model is considered to be the closest AI network model.

[0077] It is understandable that the size of the preset threshold can be set according to the adaptability of the actual application environment, and the embodiment of the present invention does not limit this.

[0078] In an alternative embodiment, Figure 6 The steps S601 and S602 shown may also be replaced by the following steps respectively:

[0079] Selecting one or more AI network models whose number of subcarriers occupied by the input size is closest to the number of subcarriers occupied by the PDSCH resources as the second AI network model set;

[0080] An AI network model whose number of symbols occupied by the input size is closest to the number of symbols occupied by the resources of the PDSCH is selected from the second AI network model set as the current AI network model.

[0081] Different from the aforementioned embodiment, the UE first selects the closest AI network model subset, that is, the second AI network model set, according to the DCI-scheduled PDSCH bandwidth (i.e., the number of subcarriers in the frequency domain). The second AI network model set may include one or more AI network models. Then, in the case where the second AI network model set includes multiple AI network models, within the first AI network model set, the AI ​​network model whose number of symbols occupied by the input size is closest to the number of symbols occupied by the PDSCH is selected as the current AI network model. Of course, if the second AI network model set includes only one AI network model, the second step is no longer performed.

[0082] For more information about the working principle and working method of the embodiments of the present invention, please refer to Figure 6 The relevant description in will not be repeated here.

[0083] In a non-limiting embodiment, if the difference in the number of symbols between the input size of the current AI network model and the transmission resource size of the PDSCH in the time domain is greater than a first preset threshold, and / or the difference in the number of subcarriers between the input size of the current AI network model and the transmission resource size of the PDSCH in the frequency domain is greater than a second preset threshold, a traditional channel estimation method is used for channel estimation.

[0084] If the difference between the PDSCH Size scheduled by DCI and the input size of the AI ​​network model is larger, then even if the size of the adjusted initial channel estimation matrix is ​​made consistent with the input size of the current AI network model through the aforementioned expansion method, the performance of the AI-based channel estimation method will be worse. In this case, a fallback mechanism can be set, that is, when the difference between the PDSCH Size scheduled by DCI and the input size of the AI ​​network model meets certain conditions, the UE should fall back to the traditional algorithm for channel estimation.

[0085] In this embodiment, if the difference between the number of time domain symbols of the PDSCH scheduled by the network and the number of time domain symbols corresponding to the configured AI network model input Size is greater than the first preset threshold, the UE uses the traditional channel estimation algorithm, or if the difference between the number of frequency domain subcarriers of the PDSCH scheduled by the network and the number of frequency domain subcarriers corresponding to the configured AI network model input Size is greater than the second preset threshold, the UE uses the traditional channel estimation algorithm. The UE determines whether to use the AI ​​network model for channel estimation based on the first preset threshold and the second preset threshold.

[0086] Further, the first preset threshold and / or the second preset threshold are received from the base station, or the first preset threshold and / or the second preset threshold are determined according to the performance of the current AI network model.

[0087] In this embodiment, the first preset threshold and / or the second preset threshold may be determined by the UE itself according to the performance of the AI ​​network model channel estimation. Alternatively, the first preset threshold and / or the second preset threshold may also be configured by the base station and sent to the UE, and the threshold configured by the base station is applied to all AI network models on the UE side.

[0088] Please refer to Figure 7 The embodiment of the present invention further discloses an AI network model matching device 70. The AI ​​network model matching device 70 may include an acquisition module 701 and an adjustment module 702.

[0089] Among them, the acquisition module 701 is used to obtain an initial channel estimation matrix to be input into the current AI network model, and the size of the initial channel estimation matrix is ​​the transmission resource size of the PDSCH; the adjustment module 702 is used to adjust the size of the initial channel estimation matrix according to the input size of the current AI network model if the size of the initial channel estimation matrix is ​​smaller than the input size of the current AI network model, so that the size of the adjusted initial channel estimation matrix is ​​consistent with the input size of the current AI network model.

[0090] In an embodiment of the present invention, when the size of the initial channel estimation matrix does not match the input size of the current AI network model, the size of the initial channel estimation matrix is ​​adjusted so that the size of the adjusted initial channel estimation matrix is ​​consistent with the input size of the current AI network model, thereby achieving different PDSCH resource sizes using the same AI network model for channel estimation, thereby reducing the number of required AI models and saving configuration resources on the user equipment side.

[0091] For more information about the working principle and working method of the AI ​​network model matching device 70, please refer to Figures 1 to 6 The relevant description in will not be repeated here.

[0092] The embodiment of the present invention further discloses a storage medium, which is a computer-readable storage medium and stores a computer program. When the computer program is run, it can execute Figure 1 , 3 , 4, and 6. The storage medium may include ROM, RAM, a magnetic disk, or an optical disk, etc. The storage medium may also include a non-volatile memory or a non-transitory memory, etc.

[0093] The embodiment of the present invention further discloses a user equipment, which may include a memory and a processor, wherein the memory stores a computer program that can be run on the processor. When the processor runs the computer program, it can execute Figure 1 , 3 , 4, and 6. The user equipment includes but is not limited to mobile phones, computers, tablet computers and other terminal devices.

[0094] Although the present invention is disclosed as above, the present invention is not limited thereto. Any person skilled in the art can make various changes and modifications without departing from the spirit and scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the scope defined by the claims.

Claims

1. An AI network model matching method, characterized in that: include: Obtain an initial channel estimation matrix to be input into the current AI network model, where the size of the initial channel estimation matrix is ​​the transmission resource size of the PDSCH; If the size of the initial channel estimation matrix is ​​smaller than the input size of the current AI network model, adjusting the size of the initial channel estimation matrix according to the input size of the current AI network model so that the size of the adjusted initial channel estimation matrix is ​​consistent with the input size of the current AI network model; Before adjusting the size of the initial channel estimation matrix according to the input size of the current AI network model, the method further includes: Selecting one or more AI network models whose number of symbols occupied by the input size is closest to the number of symbols occupied by the resources of the PDSCH as the first AI network model set; Selecting an AI network model whose number of subcarriers occupied by the input size is closest to the number of subcarriers occupied by the PDSCH resources from the first AI network model set as the current AI network model; or, Selecting one or more AI network models whose number of subcarriers occupied by the input size is closest to the number of subcarriers occupied by the PDSCH resources as the second AI network model set; An AI network model whose number of symbols occupied by the input size is closest to the number of symbols occupied by the resources of the PDSCH is selected from the second AI network model set as the current AI network model.

2. The AI ​​network model matching method according to claim 1, characterized in that: The adjusting the size of the initial channel estimation matrix according to the input size of the current AI network model includes: At least the initial channel estimation matrix is ​​expanded using boundary data in the time domain and / or frequency domain so that the size of the expanded initial channel estimation matrix is ​​consistent with the input size of the current AI network model, and the boundary data refers to the estimated value corresponding to the boundary resource unit in the initial channel estimation matrix.

3. The AI ​​network model matching method according to claim 2, characterized in that: The step of at least expanding the initial channel estimation matrix using boundary data in the time domain and / or frequency domain comprises: Selecting boundary data of the initial channel estimation matrix in the time domain and / or the frequency domain, wherein the boundary data of the initial channel estimation matrix in the time domain occupies a single symbol, and the boundary data of the initial channel estimation matrix in the frequency domain occupies a single subcarrier; The initial channel estimation matrix is ​​expanded in the time domain and / or in the time domain using the boundary data.

4. The AI ​​network model matching method according to claim 3, characterized in that: The using the boundary data to perform time domain and / or time domain expansion on the initial channel estimation matrix comprises: The boundary data in the time domain is copied to the boundary position of the initial channel estimation matrix in the time domain, and / or the boundary data in the frequency domain is copied to the boundary position of the initial channel estimation matrix in the frequency domain, until the size of the expanded initial channel estimation matrix is ​​consistent with the input size of the current AI network model.

5. The AI ​​network model matching method according to claim 3, characterized in that: The using the boundary data to perform time domain and / or time domain expansion on the initial channel estimation matrix comprises: If the size of the initial channel estimation matrix is ​​different from the input size of the current AI network model in both the time domain and the frequency domain, first copy the boundary data in the time domain to the boundary position of the initial channel estimation matrix in the time domain, and then copy the boundary data in the frequency domain to the boundary position of the initial channel estimation matrix in the frequency domain, until the size of the expanded initial channel estimation matrix is ​​consistent with the input size of the current AI network model in both the time domain and the frequency domain; Alternatively, if the size of the initial channel estimation matrix is ​​different from the input size of the current AI network model in both the time domain and the frequency domain, the boundary data located in the frequency domain is first copied to the boundary position of the initial channel estimation matrix in the time domain, and then the boundary data located in the time domain is copied to the boundary position of the initial channel estimation matrix in the frequency domain, until the size of the expanded initial channel estimation matrix is ​​consistent with the input size of the current AI network model in both the time domain and the frequency domain.

6. The AI ​​network model matching method according to claim 2, characterized in that: The step of at least expanding the initial channel estimation matrix using boundary data in the time domain and / or frequency domain comprises: The data of the initial channel estimation matrix in the time domain and / or frequency domain is intercepted, and the difference between the transmission resource size of the PDSCH in the time domain and / or frequency domain and the input size of the current AI network model in the time domain and / or frequency domain is an integer multiple of the intercepted data size; The initial channel estimation matrix is ​​expanded in the time domain and / or frequency domain using the intercepted time domain and / or frequency domain data.

7. The AI ​​network model matching method according to claim 6, characterized in that: The step of performing time domain and / or frequency domain expansion on the initial channel estimation matrix using the intercepted time domain and / or frequency domain data comprises: If the size of the initial channel estimation matrix is ​​different from the input size of the current AI network model in both the time domain and the frequency domain, the intercepted data in the time domain is first copied to the boundary position of the initial channel estimation matrix in the time domain, and then the intercepted data in the frequency domain is copied to the boundary position of the initial channel estimation matrix in the frequency domain, until the size of the expanded initial channel estimation matrix is ​​consistent with the input size of the current AI network model in both the time domain and the frequency domain; Alternatively, if the size of the initial channel estimation matrix is ​​different from the input size of the current AI network model in both the time domain and the frequency domain, the intercepted data located in the frequency domain is first copied to the boundary position of the initial channel estimation matrix in the time domain, and then the intercepted data located in the time domain is copied to the boundary position of the initial channel estimation matrix in the frequency domain, until the size of the expanded initial channel estimation matrix is ​​consistent with the input size of the current AI network model in both the time domain and the frequency domain.

8. The AI ​​network model matching method according to claim 2, characterized in that: The step of at least expanding the initial channel estimation matrix using boundary data in the time domain and / or frequency domain comprises: If the difference between the transmission resource size of the PDSCH in the time domain and / or frequency domain and the input size of the current AI network model in the time domain and / or frequency domain is an integer multiple of the size of the initial channel estimation matrix, the initial channel estimation matrix itself is used to expand the initial channel estimation matrix.

9. The AI ​​network model matching method according to claim 1, characterized in that: Before adjusting the size of the initial channel estimation matrix according to the input size of the current AI network model, the method further includes: If the difference in the number of symbols between the input size of the current AI network model and the transmission resource size of the PDSCH in the time domain is greater than a first preset threshold, and / or the difference in the number of subcarriers between the input size of the current AI network model and the transmission resource size of the PDSCH in the frequency domain is greater than a second preset threshold, a traditional channel estimation method is used for channel estimation.

10. The AI ​​network model matching method according to claim 9, characterized in that: Before the channel estimation is performed using a traditional channel estimation method, the following step is also included: Receive the first preset threshold and / or the second preset threshold from a base station, or determine the first preset threshold and / or the second preset threshold according to the performance of the current AI network model.

11. The AI ​​network model matching method according to claim 10, characterized in that: The receiving the first preset threshold and / or the second preset threshold from the base station comprises: The first preset threshold and / or the second preset threshold is determined by receiving a system message block or an RRC message broadcast by the base station.

12. An AI network model matching device, characterized in that: include: An acquisition module, used to acquire an initial channel estimation matrix to be input into the current AI network model, where the size of the initial channel estimation matrix is ​​the transmission resource size of the PDSCH; An adjustment module, configured to adjust the size of the initial channel estimation matrix according to the input size of the current AI network model if the size of the initial channel estimation matrix is ​​smaller than the input size of the current AI network model, so that the size of the adjusted initial channel estimation matrix is ​​consistent with the input size of the current AI network model; A module for selecting one or more AI network models whose number of symbols occupied by the input size is closest to the number of symbols occupied by the resources of the PDSCH, as a first AI network model set; A module for selecting, from the first AI network model set, an AI network model whose number of subcarriers occupied by the input size is closest to the number of subcarriers occupied by the PDSCH resources, as the current AI network model; or, Used to select one or more AI network models whose number of subcarriers occupied by the input size is closest to the number of subcarriers occupied by the resources of the PDSCH, as a module of the second AI network model set; used to select an AI network model whose number of symbols occupied by the input size is closest to the number of symbols occupied by the resources of the PDSCH in the second AI network model set, as a module of the current AI network model.

13. A storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the AI ​​network model matching method according to any one of claims 1 to 11 are executed.

14. A user equipment, comprising a memory and a processor, wherein the memory stores a computer program executable on the processor, characterized in that: When the processor runs the computer program, the steps of the AI ​​network model matching method described in any one of claims 1 to 11 are performed.

Citation Information

Patent Citations

  • Accelerated operation method and server for convolutional neural network, and storage medium

    CN107451654A

  • Machine Learning for Channel Estimation

    US20190356516A1