5G FR1 shaping filter system and design method

Abstract

The invention provides a 5G FR1 shaping filter system and a design method, and relates to the technical field of shaping filters, and the system comprises a BBU end intelligent prediction module, an RRU end configuration management and dynamic coefficient switching module, and an RRU end unified structure channel filter module. The BBU intelligent prediction module predicts an optimal FIR filter coefficient group in real time through a lightweight DNN model and issues the optimal FIR filter coefficient group to the RRU end; the RRU end configuration management and dynamic coefficient switching module realizes flexible switching of coefficients by combining a dynamic configuration instruction based on a pre-stored FR1 full-band initial filter coefficient group; and the RRU end unified structure channel filter module calls the target coefficient group into the FIR structure and then performs spectrum shaping filtering processing on the downlink CP-OFDM signal. Through an AI-driven dynamic optimization mechanism and a unified hardware architecture, the problems that in the prior art, a filter structure is scattered, and static configuration flexibility is insufficient are solved.

Description

TECHNICAL FIELD

[0001] The present document relates to the technical field of shaped filter, in particular to a universal 5G FR1 shaped filter system and design method. BACKGROUND

[0002] 5G NR system realizes 3MHz-100MHz single carrier bandwidth adjustment in FR1 frequency band by virtue of multiple sub-carrier spacing and flexible bandwidth configuration, so as to adapt to complex spectrum deployment requirements. In the base station transmit link, the digital channel filter located between the IFFT and the digital up-conversion module undertakes the important task of downlink signal spectrum shaping, and meets the spectrum specification requirements by limiting the bandwidth and suppressing the out-of-band power.

[0003] Current mainstream filter design adopts fixed parameters and static configuration, which is only suitable for a single frequency band or bandwidth. This customized solution has significant defects: new bandwidth configuration requires redesign of coefficients and hardware logic, resulting in rising system development and verification costs, and difficulty in upgrading and maintenance. Although some FPGA-based filters have a certain flexibility, they cannot be compatible with the full bandwidth configuration of the FR1 frequency band, and it is difficult to meet the high-flexible multiplexing requirements of 5G public network / special network and industry customized deployment.

[0004] In digital communication systems, the design of digital filters for multi-bandwidth signal processing has been widely studied. In particular, in the context of high-speed link processing and frequency resource diversification configuration, some schemes improve processing capacity and configuration flexibility through interpolation structure or reconfigurable coefficient switching mechanism. Although these designs have certain practicality in engineering, they mainly focus on computational efficiency or static structure multiplexing, and have not yet focused on the key requirements such as spectral shaping accuracy, real-time adaptive ability, and structural uniformity required by the channel filter in the 5G NR base station transmit link.

[0005] For example, patent CN118041302A proposes a digital filter structure combining 4 times interpolation and FIFO scheduling mechanism, aiming to improve the parallel processing capacity and throughput efficiency of the system. This scheme has high processing performance under fixed bandwidth conditions, but the interpolation multiple is fixedly configured, lacking adaptability to multi-bandwidth scenarios. The filter structure needs to be designed and verified independently for each bandwidth, resulting in high repeated development cost of the system, non-reusable architecture, and difficulty in supporting the dynamic requirements of frequent bandwidth switching in 5G NR systems.

[0006] The patent CN116545412A proposes a reconfigurable FIR filter scheme, which supports multiple carrier bandwidth configurations by loading different coefficient tables, and has certain parameter flexibility. However, this scheme usually needs to configure multiple filter modules for parallel switching, lacks structural uniformity, and the resource scheduling and control logic is complex, which can easily cause FPGA resource redundancy, and its coefficient configuration is still a static design, lacking the ability to adaptively optimize according to the runtime channel conditions.

[0007] In summary, the existing related schemes focus on throughput performance or parameter multiplexing, and cannot effectively unify the structural generality, multi-bandwidth support capability and intelligent optimization performance, making it difficult to meet the comprehensive requirements of efficient spectrum shaping and dynamic operation adaptation under the 5G FR1 frequency band. Therefore, there is an urgent need for a channel filter design scheme that integrates an AI adaptive mechanism under a unified architecture, supports real-time channel sensing and has dynamic parameter optimization capability to meet the flexible deployment requirements of 5G base stations in multiple scenarios and multiple configuration environments. SUMMARY

[0008] The embodiments of the present application provide a 5G FR1 shaping filter system and design method, which realizes a unified architecture to cover the FR1 full frequency band and bandwidth configuration, quickly adapts to different bandwidths by establishing a flexible coefficient configuration mechanism, and intelligently adjusts the filter parameters based on the channel state by introducing AI dynamic optimization to improve the spectrum shaping precision and resource utilization.

[0009] According to the embodiments of the present application, a 5G FR1 shaping filter system is provided, comprising: a BBU-end intelligent prediction module, an RRU-end configuration management and dynamic coefficient switching module, and an RRU-end unified structure channel filter module; The BBU-end intelligent prediction module is configured to collect channel context information in the wireless link operation process, input the channel context information into a pre-trained lightweight DNN model, output an optimal FIR filter coefficient set, and send the optimal FIR filter coefficient set to the RRU-end configuration management and dynamic coefficient switching module. The RRU-end configuration management and dynamic coefficient switching module is configured to receive the optimal FIR filter coefficient set, store the optimal FIR filter coefficient set after format conversion based on the initial filter coefficient set of the preset FR1 full carrier bandwidth, and obtain the corresponding target coefficient set in the optimal FIR filter coefficient set based on the obtained configuration instruction and the preset mapping table. The RRU-end unified structure channel filter module adjusts the target coefficient set into the FIR structure and performs spectrum shaping filter processing on the downlink CP-OFDM signal.

[0010] According to the embodiment of the present application, a 5G FR1 forming filter design method is provided, comprising: S1, collecting channel context information in the wireless link operation process, inputting the channel context information into the pre-trained lightweight DNN model, outputting the optimal FIR filter coefficient set, and sending the optimal FIR filter coefficient set to the RRU end configuration management and dynamic coefficient switching module; S2, receiving the optimal FIR filter coefficient set, storing the optimal FIR filter coefficient set after format conversion based on the initial filter coefficient set of the preset FR1 full carrier bandwidth, and based on the obtained configuration instruction and the preset mapping table, acquiring the corresponding target coefficient set in the optimal FIR filter coefficient set; S3, adjusting the target coefficient set into the FIR structure to perform spectral shaping filter processing on the downlink CP-OFDM signal.

[0011] By using the embodiment of the present application, a channel filter hardware platform covering FR1 full frequency band and all bandwidth configurations is realized through unified order design and module instantiation, the system consistency and logical multiplexing efficiency are improved, a plurality of groups of static filter coefficients are preloaded to the FPGA module, and combined with the configuration control logic, the use is switched as needed, the repeated development is reduced, the bandwidth adaptability is enhanced, the lightweight deep neural network model is deployed on the baseband processing unit (BBU) side, the channel context (CQI, load, PRB distribution, etc.) is perceived in real time, the optimal filter parameter coefficient set is predicted, and the spectral shaping precision and resource utilization rate are improved. BRIEF DESCRIPTION OF DRAWINGS

[0012] In order to more clearly illustrate the technical solutions in the one or more embodiments of the present application or the prior art, the drawings needed to be used in the embodiment or prior art description will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments described in the present application, and other drawings can also be obtained by those skilled in the art without creative labor.

[0013] Figure 1 A 5G FR1 forming filter system block diagram of the embodiment of the present application; Figure 2 A 5G FR1 forming filter design method flow chart of the embodiment of the present application; Figure 3 A 5G FR1 forming filter working principle flow chart of the embodiment of the present application. DETAILED DESCRIPTION

[0014] In order to make the person in the art better understand the technical scheme in one or more embodiments of the present specification, the technical scheme in one or more embodiments of the present specification will be clearly and completely described below in combination with the drawings in one or more embodiments of the present specification. Obviously, the described embodiments are only a part of the embodiments of the present specification, rather than all the embodiments. Based on one or more embodiments of the present specification, all other embodiments obtained by those skilled in the art without creative labor should belong to the protection scope of the present document.

[0015] The existing filter design lacks a unified architecture, which is difficult to reuse and maintain: most current filter designs adopt a customized solution for a specific bandwidth, and the filter structure and coefficients need to be developed and verified for each carrier bandwidth, resulting in scattered modules and poor resource reuse. The system structure is complex, and the cost of on-site deployment, upgrading and expansion, and long-term operation and maintenance is high, which makes it difficult to support the rapid deployment and flexible expansion requirements of 5G networks.

[0016] Lack of dynamic adaptation capability: traditional filters use static configuration methods and cannot dynamically adjust filtering performance according to different business loads, modulation methods, spectrum utilization conditions, and other operating parameters, resulting in unstable spectrum shaping effects, low spectrum utilization efficiency, and inability to adapt to changing wireless environments.

[0017] Lack of intelligent optimization mechanism: existing solutions do not introduce artificial intelligence or context-aware technology, relying on offline static modeling and manual experience configuration, and cannot dynamically optimize passband ripple, stopband attenuation, or transition band characteristics based on real-time channel quality, limiting further improvement of spectrum shaping precision and system performance.

[0018] The present application proposes a 5G FR1 digital channel filter design and implementation method combining a general architecture and an artificial intelligence optimization mechanism, and constructs a solution that can solve the above problems at the same time.

[0019] System embodiment According to an embodiment of the present application, a 5G FR1 shaping filter system is provided, Figure 1 is a 5G FR1 shaping filter system framework of an embodiment of the present application, according to Figure 1 shown, the 5G FR1 shaping filter system of the embodiment of the present application specifically includes: a BBU-end intelligent prediction module, a RRU-end configuration management and dynamic coefficient switching module, and a RRU-end unified structure channel filter module, the baseband unit is the BBU-end, and the radio frequency unit is the RRU-end.

[0020] The BBU-end intelligent prediction module is configured to collect channel context information in a wireless link operation process, input the channel context information into a pre-trained lightweight DNN model, output an optimal FIR filter coefficient set, and send the optimal FIR filter coefficient set to a RRU-end configuration management and dynamic coefficient switching module. The BBU-end intelligent prediction module is deployed on a processor resource at the BBU end. Multi-dimensional context parameters such as link quality, spectrum usage, and transmission configuration are obtained from the physical layer and the MAC layer, normalized, and input into a pre-trained lightweight DNN model. The model adopts a three-layer network structure of an input layer, a hidden layer, and an output layer, outputs a coefficient set of a fixed-order FIR filter, and finally sends the coefficient set to the RRU through a front-end interface.

[0021] The BBU-end intelligent prediction module is specifically configured to: obtain context parameters and perform standardization processing on the context parameters, wherein the context parameters include link quality parameters, spectrum usage parameters, and transmission configuration parameters; input the preprocessed context parameters into a preset lightweight DNN model to output an optimal FIR filter coefficient set, and the construction process of the preset lightweight DNN model includes: constructing a network structure: adopting an MLP structure including an input layer, a hidden layer, and an output layer, the dimension of the input layer is determined by the context parameters, the hidden layer includes a plurality of nodes and the activation function is ReLU, the number of nodes of the output layer is consistent with the order of the FIR filter, and the structure of the lightweight DNN model is as follows: ; wherein, is an input feature vector, d is an input dimension, and includes channel context information; is an output FIR filter coefficient set, the length N is consistent with the filter order, is a feedforward neural network (MLP) function with a parameter set θ; training the model: based on the target passband ripple, stopband attenuation, and transition band indicators, generating optimal filter coefficients as a supervision target, adjusting the model parameters through the calculation of the mean square error between the predicted coefficients and the target coefficients as a loss function, and using the back propagation algorithm, the loss function used is as follows: ; represents the i-th target FIR filter coefficient, and in the training stage, the back propagation algorithm is used to continuously adjust, so that the mean square error between and the target coefficient is minimized.

[0022] The RRU end configuration management and dynamic coefficient switching module is configured to receive an optimal FIR filter coefficient set, perform format conversion on the optimal FIR filter coefficient set based on a preset initial filter coefficient set of an FR1 full carrier bandwidth, and store the format-converted coefficient set, obtain a corresponding target coefficient set from the optimal FIR filter coefficient set based on a preset mapping table and a configuration instruction obtained, and control the FIR coefficient set in the internal memory to be called into an IP core filter module according to the instruction content and the preset mapping table. The RRU end configuration management and dynamic coefficient switching module is deployed in the internal FPGA logic resource of the RRU, and is responsible for monitoring configuration instructions such as "bandwidth index" or "AI prediction ID" transmitted through a front-end interface (for example, an eCPRI link). According to the instruction content and a preset mapping table, the FIR coefficient set in the internal memory is controlled to be called into an IP core filter module.

[0023] The preset initial filter coefficient set of the FR1 full carrier bandwidth includes the following steps: The transmission bandwidth, sampling rate and normalized transition band ratio of each carrier bandwidth and subcarrier spacing combination in the FR1 frequency band are calculated to generate a normalized transition band ratio table; The carrier bandwidth and subcarrier spacing combination corresponding to the minimum normalized transition band ratio are selected from the normalized transition band ratio table as a design constraint condition; According to the design constraint condition, the unified filter order meeting the FR1 full frequency band requirement is calculated by a filter design tool in combination with preset passband ripple and stopband attenuation indicators; For each carrier bandwidth, the combination with the minimum normalized transition band ratio in all subcarrier spacings thereof is selected as a design condition, and the FIR filter coefficient of the corresponding bandwidth is generated using a digital filter design tool, and all coefficients adopt the unified order; The generated coefficient sets of each bandwidth are stored in a preset format to form an initial filter coefficient set covering the FR1 full frequency band.

[0024] The preset mapping table includes the following steps: The carrier bandwidth and subcarrier spacing combinations in the FR1 frequency band are grouped, and a unique coefficient serial number is assigned; In the FPGA, a storage space block is allocated for each group of coefficient sets, and the starting address and length are recorded; A mapping relationship between the associated carrier bandwidth, subcarrier spacing, coefficient serial number and storage block number is established.

[0025] The RRU end unified structure channel filter module calls the target coefficient set into the FIR structure to perform spectral shaping filter processing on the downlink CP-OFDM signal.

[0026] The RRU-side unified channel filter module is implemented primarily on an FPGA, using an FIR IP core for filter calculation logic. The FPGA design initializes and loads multiple FIR filter coefficient sets supporting all 5G FR1 bandwidths, dynamically switching the active coefficients based on parameter configuration or AI prediction input. The filter structure remains unified, and the processing clock settings are compatible with all FR1 sampling rates. Both input and output data use two's complement fixed-point format.

[0027] The RRU-end unified structure channel filter module is specifically used for: Loading the target coefficient group into the FIR IP core of the FPGA; According to the bandwidth index or AI prediction ID in the configuration instruction, the storage address of the target coefficient group is determined through the mapping table, and the activated filter coefficients are switched in real time; Using FPGA as the hardware carrier and the two's complement fixed-point data format, the downlink CP-OFDM signal is subjected to spectrum shaping and filtering processing. The processing clock frequency is set to an integer multiple of the highest sampling rate of FR1.

[0028] The embodiments of the present invention have the following beneficial effects: Through unified order design and module instantiation, a channel filter hardware platform covering the entire FR1 frequency band and all bandwidth configurations is implemented, improving system consistency and logic multiplexing efficiency. It supports preloading multiple sets of static filter coefficients into the FPGA module, and combined with configuration control logic, it can be switched on demand, reducing repetitive development and enhancing bandwidth adaptability. By deploying a lightweight deep neural network model on the baseband processing unit (BBU) side, channel context such as CQI, load, PRB distribution, etc. is perceived in real time, and the optimal filter parameter coefficient group is predicted, thereby improving spectrum shaping accuracy and resource utilization.

[0029] Method Example According to an embodiment of the present invention, a 5G FR1 shaping filter design method is provided. Figure 2 Flowchart of the 5G FR1 shaping filter design method according to an embodiment of the present invention. Figure 2 As shown, the 5G FR1 shaping filter design method of the embodiment of the present invention specifically includes: S1. Collect channel context information during the operation of the wireless link, input the channel context information into a pre-trained lightweight DNN model, output an optimal FIR filter coefficient set, and send the optimal FIR filter coefficient set to the RRU configuration management and dynamic coefficient switching module; Said S1 specifically includes: Acquiring context parameters and performing normalization on the context parameters, wherein the context parameters include: link quality parameters, spectrum usage parameters, and transmission configuration parameters; The preprocessed context parameters are input into a preset lightweight DNN model, and an optimal FIR filter coefficient set is output, and the construction process of the preset lightweight DNN model comprises: Constructing a network structure: an MLP structure comprising an input layer, a hidden layer and an output layer is adopted, the dimension of the input layer is determined by the context parameters, the hidden layer comprises a plurality of nodes and the activation function is ReLU, the number of nodes of the output layer is consistent with the order of the FIR filter, and the structure of the lightweight DNN model is as follows: ; wherein, is an input feature vector, d is an input dimension, and contains channel context information; is an output FIR filter coefficient set, the length N is consistent with the filter order, is a feedforward neural network (MLP) function with a parameter set θ; Model training: based on the target passband ripple, stopband attenuation and transition band indicators, the optimal filter coefficients are generated as the supervision target, the mean square error between the predicted coefficients and the target coefficients is calculated as the loss function, the model parameters are adjusted by using the back propagation algorithm, and the loss function adopted is as follows: ; represents the i-th target FIR filter coefficient, and in the training stage, the back propagation algorithm is used to continuously adjust, so that the FIR coefficient vector output by the model minimizes the MSE error with the target coefficient.

[0030] S2, receiving an optimal FIR filter coefficient set, performing format conversion on the optimal FIR filter coefficient set based on a preset initial filter coefficient set of the FR1 full carrier bandwidth, and storing the format-converted optimal FIR filter coefficient set, obtaining a corresponding target coefficient set in the optimal FIR filter coefficient set based on a preset mapping table and an obtained configuration instruction; The preset initial filter coefficient set of the FR1 full carrier bandwidth comprises: Calculate the transmission bandwidth, sampling rate and normalized transition band ratio of each carrier bandwidth and subcarrier spacing combination in the FR1 frequency band, and generate a normalized transition band ratio table; Select the carrier bandwidth and subcarrier spacing combination corresponding to the minimum normalized transition band ratio from the normalized transition band ratio table as the design constraint condition; According to the design constraint condition, the preset passband ripple and stopband attenuation indicators are combined, and the uniform filter order meeting the FR1 full-band requirement is calculated by using a filter design tool; For each carrier bandwidth, select the combination with the smallest normalized transition band ratio among all subcarrier spacing combinations as the design condition, use a digital filter design tool to generate the FIR filter coefficient of the corresponding bandwidth, and all coefficients use the uniform order number; Store the generated bandwidth coefficient set in a preset format to form an initial filter coefficient set covering the entire FR1 frequency band.

[0031] The target coefficient set corresponding to the optimal FIR filter coefficient set is obtained based on the obtained configuration instruction and the preset mapping table, and specifically includes: Group the carrier bandwidth and subcarrier spacing combinations in the FR1 frequency band, and assign a unique coefficient serial number; In the FPGA, a storage space block is allocated for each coefficient set, and the starting address and length are recorded; Establish a mapping relationship between the associated carrier bandwidth, subcarrier spacing, coefficient serial number and storage block number.

[0032] S3, after loading the target coefficient set into the FIR structure, performing spectral shaping filter processing on the downlink CP-OFDM signal.

[0033] The S3 specifically includes: Load the target coefficient set into the FIR IP core of the FPGA; According to the bandwidth index or AI prediction ID in the configuration instruction, determine the storage address of the target coefficient set through the mapping table, and switch the activated filter coefficient in real time; Using the FPGA as a hardware carrier, using the complement fixed-point data format, performing spectral shaping filter processing on the downlink CP-OFDM signal, and setting the processing clock frequency to be an integer multiple of the highest sampling rate covering FR1.

[0034] As Figure 3 As shown in the flow chart of the working principle of the 5G FR1 forming filter of the embodiment of the application, when the 5G FR1 forming filter designed by the embodiment of the application works, the specific process is as follows: Step 1: Design of initial filter coefficient with uniform order number Step 1.1, analyze the filter requirements under each carrier bandwidth and subcarrier spacing combination of 5G FR1 frequency band, and use the following method: Based on 3GPP TS 38.104 Table 5.3.2-1 and 3GPP TS 38.104 Table B.5.2-1 / -2 / -3 / -4, the transmission bandwidth, sampling rate and normalized transition band ratio under all bandwidth and subcarrier spacing conditions of 5G FR1 are obtained according to the following calculation steps, and the results are summarized to generate Table 1.

[0035] Table 1 Normalized transition ratio table of each bandwidth and subcarrier combination of FR1

[0036] In Table 1, each parameter has the following meanings: TB is the abbreviation of Transmission Bandwidth, and the Chinese name is transmission bandwidth, ; N RB is the abbreviation of Number of Resource Block, and the Chinese name is resource block number; RB is the abbreviation of Resource Block, and the Chinese name is resource block; SCS is the abbreviation of Sub Carrier Space, and the Chinese name is subcarrier spacing; CB is the abbreviation of Channel Bandwidth, and the Chinese name is channel bandwidth; N of each CB and SCS of FR1 RB Refer to 3GPP TS 38.104 Table 5.3.2-1; SR = FFT Size*SCS / 1000; SR is the abbreviation of Sample Rate, and the Chinese name is sampling rate. In the formula, the unit of SCS is KHz, so the unit of SR is MSPS, and the calculation method is: ; The FFT Size of each CB and SCS of FR1 refers to 3GPP TS 38.104 Table B.5.2-1 / -2 / -3 / -4; TR is the abbreviation of normalized Transition Ratio, and the Chinese name is normalized transition ratio, that is, the normalized ratio of the transition band to the sampling rate, TR = (CB / 2-TB / 2) / (SR / 2).

[0037] Step 1.2, analyze and determine the coefficient uniform order value, as follows: Analyze Table 1 to find the bandwidth and subcarrier combination corresponding to the minimum ratio. This combination requires the steepest filter transition band, and it is also the combination with the highest filter coefficient order requirement; Take the channel bandwidth, transmission bandwidth and sampling rate of this combination as the filter coefficient design input parameters, combine the filter passband ripple index and stopband minimum attenuation index requirements, and based on the firpmord function of MATLAB, the order of the filter coefficient that meets the requirements can be obtained. This order can meet the uniform order of all bandwidths and subcarriers of FR1.

[0038] Step 1.3, generate the initial filter coefficient set corresponding to the bandwidth based on MATLAB tools, design a unified filter coefficient that meets all subcarrier combinations of FR1 same bandwidth as follows: Further analyze the transition band relative sampling rate normalization ratio of each group of bandwidth and subcarrier in step 1, compare and find the minimum TR value among all subcarrier intervals (15kHz / 30kHz / 60kHz) under each bandwidth, for example, when the transition band of 20M bandwidth 15kHz subcarrier is the steepest, design the filter coefficient according to this combination condition to design a unified filter coefficient for 20M bandwidth carrier, that is, the coefficient is generated only according to the difference of each bandwidth; The initial filter coefficient is designed according to all RB resources of each bandwidth being scheduled for use.

[0039] Step 2: intelligent prediction model design, training and deployment Lightweight DNN (MLP structure) is used to model the downlink channel state of 5G base station and predict the optimal FIR channel filter coefficient. The specific design is as follows: Step 2.1, construct a feature vector: The input of the model is a group of channel context parameters collected by the BBU in real time, which is divided into three categories: Link quality: CQI, BLER, path loss, UE distance; Spectrum usage: PRB occupancy, system load, center frequency; Transmission configuration: transmit power, modulation method (MCS), carrier bandwidth, subcarrier spacing.

[0040] The above context parameters are standardized for continuous values, and the discrete categories are encoded using One-hot encoding. The final feature vector dimension is 15~30.

[0041] Step 2.2, network structure design Lightweight DNN (MLP structure) is used, including: Input layer dimension: d∈[15,30], selected according to specific features.

[0042] Hidden layer 1: h1 (such as 64) nodes, activation function ReLU; Hidden layer 2: h2 (such as 32) nodes, activation function ReLU; Output layer: N nodes, corresponding to N order FIR coefficient, activation function Linear or Tanh.

[0043] The model function form is as follows: ; Where: ∈R^d: is the input feature vector, d is the input dimension, and contains channel context information such as CQI, BLER, path loss, PRB occupancy, transmit power, etc., in encoded and normalized form; ∈R^N is the output FIR filter coefficient group, and the length N is consistent with the filter order; is a feedforward neural network (MLP) function with parameter set θ.

[0044] The parameter set θ includes the following: first hidden layer parameters (weight matrix and bias term): W(1) ∈ R^(h1×d), b(1) ∈ R^h1; second hidden layer parameters (weight matrix and bias term): W(2) ∈ R^(h2×h1), b(2) ∈ R^h2; output layer parameters (weight matrix and bias term): W(3) ∈ R^(N×h2), b(3) ∈ R^N; that is: θ = {W(1), b(1), W(2), b(2), W(3), b(3)}; Step 2.3: Perform model training as follows: During the training phase, offline design tools (such as MATLAB) are used to generate the optimal filter coefficients for each set of input channel context parameters based on the target passband ripple, stopband attenuation, and transition band indicators using the firpm or remez algorithm. These coefficients serve as the supervision target for model training.

[0045] Loss function design: mean squared error (MSE) between the predicted and target coefficients; ; The above model is continuously adjusted during the training phase through the back propagation algorithm (such as Adam or SGD) so that the FIR coefficient vector output by the model Minimize the MSE error with the target coefficients to optimize the parameter set θ.

[0046] Step 2.4: Online reasoning and interface output After model training is completed, it is deployed on the BBU side and called by the scheduling module; Read current channel context parameters in real time 实时 ; Perform forward reasoning and output filter coefficients ; The coefficient group is compressed and quantized and then sent to the RRU through the fronthaul interface (such as the eCPRI interface).

[0047] Step 3: Design and implement the FPGA channel FIR filter module, as follows: Step 3.1. Generate a coefficient file containing full bandwidth coefficients Based on the initial filter coefficient set generated in step 1 for the FR1 full carrier bandwidth, the data format of the coefficient set is converted and saved in a coefficient file (such as a.coe format) supported by the FPGA FIR IP core. According to the order of the coefficient set in the coefficient file, assuming that there are N sets of coefficients, the corresponding coefficient sequence numbers are 0~(N-1). Subsequently, a certain set of coefficients can be selected by the coefficient sequence number, and the coefficient sequence number corresponds to the carrier bandwidth one by one. A coefficient mapping table is established to record the mapping relationship between the coefficient sequence number, the coefficient set, and the carrier bandwidth.

[0048] Step 3.2, parameter instantiation of the FPGA FIR IP core The channel filter is a FIR filter processing module implemented using the FIR IP core. The FIR IP core is an integrated module of the FPGA device, which is not the content of the present application. It implements FIR filtering processing based on externally input coefficients, and needs to be parameterized when an instance of the IP core is generated, specifically including 2 points: Import the coefficient file (such as a.coe format) containing all bandwidth coefficients, and store them in the IP core internal coefficient storage space in the order of the file content, each set of coefficients occupying the same size of storage space block. Number these storage space blocks in order from small to large address, and determine the corresponding storage space block number according to the coefficient sequence number. Add the mapping relationship between the coefficient sequence number and the coefficient storage space block number in the coefficient mapping table; The input clock frequency is designed to be an integer power of 2 times the highest sampling rate in the FR1 full carrier bandwidth (for example, 4 times the highest sampling rate), to support all carrier sampling rates.

[0049] Step 4: FPGA configuration module design and implementation, as follows: The configuration module supports two operations: update and switch. The switching process does not reconfigure the filter structure, and meets the real-time adaptation under multiple bandwidths. The configuration module implements: Through the configuration interface, receive the control instructions or AI prediction results from the BBU, and parse the configuration content, including: input carrier bandwidth, operation type, and new coefficient data; According to the coefficient mapping table, design a lookup table, through which, input the bandwidth, determine the selected coefficient sequence number, and the selected coefficient storage space block; If it is a coefficient update instruction, the configuration module writes the new coefficient data into the selected coefficient storage space block through the update interface; if it is a coefficient switching instruction, the configuration module controls the coefficient storage module to output the coefficient corresponding to the selected coefficient sequence number through the switching interface.

[0050] By adopting the embodiment of the present application, the following beneficial effects are achieved: The application constructs a filter structure with uniform order, covers all typical bandwidths in FR1 frequency band, and is uniformly loaded in FPGA to avoid structural redundancy and resource waste, and improve the consistency and later maintainability of system design. The application innovatively introduces a lightweight DNN (such as MLP) model, realizes real-time perception of link running state, combines a pre-trained model to output optimal FIR coefficients, and enables the filter to have the ability to optimize with environmental changes. This dynamic optimization mechanism greatly improves the spectral shaping accuracy and interference control ability of the system, which is much better than the traditional fixed coefficient filter or static coefficient table loading scheme. The same structure supports online switching and updating of filter coefficients through a configuration module. This mechanism not only simplifies the design of hardware resources, but also enhances the flexible deployment capability of the system in public networks / special networks and industry-specific scenarios. The application deeply couples the design among the filter algorithm, AI model and FPGA hardware platform, including uniform order, coefficient format quantization, model output encoding and other detailed optimizations, to ensure the end-to-end real-time performance from model prediction to hardware update. At the same time, the soft and hard decoupling architecture improves the expansion capability and engineering implementability of the system, facilitating actual commercial deployment. Through the linkage of AI model prediction at the BBU end + configuration module at the RRU end + dynamic coefficient injection mechanism, the application realizes a closed-loop control process of intelligent spectral shaping, has good algorithm expansion and feedback iteration capability, and can continuously optimize the model accuracy with the actual network operation effect.

[0051] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the application, and not to limit them; although the application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement for part or all of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the application.

Claims

1. A 5G FR1 shaping filter system, characterized in that: include: BBU-side intelligent prediction module, RRU-side configuration management and dynamic coefficient switching module, and RRU-side unified structure channel filter module; The BBU-side intelligent prediction module is used to collect channel context information during the operation of the wireless link, input the channel context information into a pre-trained lightweight DNN model, output an optimal FIR filter coefficient group, and send the optimal FIR filter coefficient group to the RRU-side configuration management and dynamic coefficient switching module; The RRU-side configuration management and dynamic coefficient switching module is configured to receive an optimal FIR filter coefficient group, convert the format of the optimal FIR filter coefficient group based on an initial filter coefficient group of a preset FR1 full carrier bandwidth and store the converted data, and obtain a corresponding target coefficient group from the optimal FIR filter coefficient group based on the obtained configuration instruction and a preset mapping table; The RRU-end unified structure channel filter module adjusts the target coefficient group into the FIR structure and performs spectrum shaping and filtering processing on the downlink CP-OFDM signal.

2. The system according to claim 1, wherein: The BBU-side intelligent prediction module is specifically used to: Acquiring context parameters and performing normalization on the context parameters, wherein the context parameters include: link quality parameters, spectrum usage parameters, and transmission configuration parameters; The preprocessed context parameters are input into a preset lightweight DNN model to output an optimal FIR filter coefficient group. The preset lightweight DNN model construction process includes: Constructing the network structure: An MLP structure consisting of an input layer, a hidden layer, and an output layer is adopted. The dimension of the input layer is determined by the context parameter. The hidden layer contains several nodes and the activation function is ReLU. The number of nodes in the output layer is consistent with the order of the FIR filter. The structure of the lightweight DNN model is as follows: ； in, ∈R^d is the input feature vector, d is the input dimension, and contains channel context information; ∈R^N is the output FIR filter coefficient group, and the length N is consistent with the filter order. is a feedforward neural network (MLP) function with parameter set θ; Model training: Based on the target passband ripple, stopband attenuation, and transition band indicators, the optimal filter coefficients are generated as the supervision target. The mean square error between the output FIR filter coefficient group and the target coefficients is calculated as the loss function, and the back propagation algorithm is used to adjust the model parameters. The loss function used is as follows: ； in, Represents the i-th target FIR filter coefficient, which is continuously adjusted by the back propagation algorithm during the training phase. Minimize the mean square error between the coefficients and the target coefficients.

3. The system according to claim 1, wherein: The process of constructing the initial filter coefficient group of the preset FR1 full carrier bandwidth includes: Calculate the transmission bandwidth, sampling rate, and normalized transition band ratio for each carrier bandwidth and subcarrier spacing combination in the FR1 frequency band, and generate a normalized transition band ratio table; Selecting a carrier bandwidth and subcarrier spacing combination corresponding to a minimum value of the normalized transition band ratio from the normalized transition band ratio table as a design constraint; Based on the design constraints, combined with the preset passband ripple and stopband attenuation indicators, the filter design tool is used to calculate the unified filter order that meets the requirements of the full FR1 frequency band; For each carrier bandwidth, select the combination with the smallest normalized transition band ratio among all its subcarrier spacings as the design condition, use a digital filter design tool to generate FIR filter coefficients for the corresponding bandwidth, and all coefficients use the unified filter order; The generated bandwidth coefficient groups are stored in a preset format to form an initial filter coefficient group covering the entire FR1 frequency band.

4. The system according to claim 1, wherein: The preset mapping table construction process includes: Group the carrier bandwidth and subcarrier spacing within the FR1 frequency band and assign unique coefficient numbers; Allocate a block of storage space for each coefficient group in the FPGA and record the starting address and length; A mapping relationship between associated carrier bandwidth, subcarrier spacing, coefficient sequence number and storage block number is established.

5. The system according to claim 1, wherein: The RRU-end unified structure channel filter module is specifically used for: Loading the target coefficient group into the FIR IP core of the FPGA; According to the bandwidth index or AI prediction ID in the configuration instruction, the storage address of the target coefficient group is determined through the mapping table, and the activated filter coefficients are switched in real time; Using FPGA as the hardware carrier and the two's complement fixed-point data format, the downlink CP-OFDM signal is subjected to spectrum shaping and filtering processing. The processing clock frequency is set to an integer multiple of the highest sampling rate of FR1.

6. A 5G FR1 shaping filter design method based on the 5G FR1 shaping filter system according to any one of claims 1 to 5, characterized in that: include: S1. Collect channel context information during the operation of the wireless link, input the channel context information into a pre-trained lightweight DNN model, output an optimal FIR filter coefficient set, and send the optimal FIR filter coefficient set to the RRU configuration management and dynamic coefficient switching module; S2. Receive an optimal FIR filter coefficient group, convert the format of the optimal FIR filter coefficient group based on an initial filter coefficient group of a preset FR1 full carrier bandwidth and store the converted data, and obtain a corresponding target coefficient group from the optimal FIR filter coefficient group based on the obtained configuration instruction and a preset mapping table; S3. After the target coefficient group is transferred into the FIR structure, spectrum shaping and filtering processing is performed on the downlink CP-OFDM signal.

7. The method according to claim 6, characterized in that Said S1 specifically includes: Acquiring context parameters and performing normalization on the context parameters, wherein the context parameters include: link quality parameters, spectrum usage parameters, and transmission configuration parameters; The preprocessed context parameters are input into a preset lightweight DNN model to output an optimal FIR filter coefficient group. The preset lightweight DNN model construction process includes: Constructing the network structure: An MLP structure consisting of an input layer, a hidden layer, and an output layer is adopted. The dimension of the input layer is determined by the context parameter. The hidden layer contains several nodes and the activation function is ReLU. The number of nodes in the output layer is consistent with the order of the FIR filter. The structure of the lightweight DNN model is as follows: ； in, ∈R^d is the input feature vector, d is the input dimension, and contains channel context information; ∈R^N is the output FIR filter coefficient group, and the length N is consistent with the filter order. is a feedforward neural network (MLP) function with parameter set θ; Model training: Based on the target passband ripple, stopband attenuation, and transition band indicators, the optimal filter coefficients are generated as the supervision target. The mean square error between the output FIR filter coefficient group and the target coefficients is calculated as the loss function, and the back propagation algorithm is used to adjust the model parameters. The loss function used is as follows: ； In the training phase, the back propagation algorithm is used to continuously adjust Minimize the mean square error between the coefficients and the target coefficients.

8. The method according to claim 6, characterized in that The method for constructing the initial filter coefficient group of the preset FR1 full carrier bandwidth includes: Calculate the transmission bandwidth, sampling rate, and normalized transition band ratio for each carrier bandwidth and subcarrier spacing combination in the FR1 frequency band, and generate a normalized transition band ratio table; Selecting a carrier bandwidth and subcarrier spacing combination corresponding to a minimum value of the normalized transition band ratio from the normalized transition band ratio table as a design constraint; Based on the design constraints, combined with the preset passband ripple and stopband attenuation indicators, the filter design tool is used to calculate the unified filter order that meets the requirements of the full FR1 frequency band; For each carrier bandwidth, select the combination with the smallest normalized transition band ratio among all its subcarrier spacings as the design condition, use a digital filter design tool to generate FIR filter coefficients for the corresponding bandwidth, and all coefficients use the unified filter order; The generated bandwidth coefficient groups are stored in a preset format to form an initial filter coefficient group covering the entire FR1 frequency band.

9. The method according to claim 6, characterized in that The step of obtaining the corresponding target coefficient group from the optimal FIR filter coefficient group based on the obtained configuration instruction and the preset mapping table specifically includes: Group the carrier bandwidth and subcarrier spacing within the FR1 frequency band and assign unique coefficient numbers; Allocate a block of storage space for each coefficient group in the FPGA and record the starting address and length; A mapping relationship between associated carrier bandwidth, subcarrier spacing, coefficient sequence number and storage block number is established.

10. The method according to claim 6, characterized in that The S3 specifically includes: Loading the target coefficient group into the FIR IP core of the FPGA; According to the bandwidth index or AI prediction ID in the configuration instruction, the storage address of the target coefficient group is determined through the mapping table, and the activated filter coefficients are switched in real time; Using FPGA as the hardware carrier and the two's complement fixed-point data format, the downlink CP-OFDM signal is subjected to spectrum shaping and filtering processing. The processing clock frequency is set to an integer multiple of the highest sampling rate of FR1.

Images