FIR (Finite Impulse Response) filter order estimation method and system by virtue of neural network
Through the data set trained by deep learning neural networks, the accurate estimation of the order of FIR filters is achieved, solving the problems of inaccurate estimation and complex design in the existing methods, and reducing design costs.
Patent Information
- Application Number
- CN202411876590.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-19
- Publication Date
- 2025-04-11
AI Technical Summary
The existing FIR filter order estimation method has the problem of inaccurate estimation, which requires manual adjustment and repeated testing, which increases design cost and complexity.
Using deep learning neural networks, deep neural networks are trained to achieve estimation mapping of frequency and performance to order by constructing data sets containing filter order, passband frequency, stopband frequency, passband attenuation, stopband attenuation and fluctuation weighting values.
It improves the accuracy of FIR filter order estimation, reduces the repeated processes of subsequent adjustment and testing, and reduces design cost and complexity.
Smart Images
Figure CN119813993A_ABST
Abstract
Description
Technical Field
[0001] The invention relates to the technical field of digital signal processing and communication, and relates to a method and system for estimating the order of an FIR filter by means of a neural network. Background Art
[0002] As a type of digital filter, FIR filters are widely used in modulation communication systems, multi-carrier communication systems, image denoising and other fields due to their precise linear phase and unique stability. When designing FIR filters, engineers usually hope to reduce the order of the filter as much as possible while meeting performance requirements to reduce implementation costs and complexity.
[0003] Common FIR filter optimization design methods include min-max and least squares. The design of both methods requires order estimation under given frequency parameters and performance requirements. Existing estimation methods are formulaic estimations based on design experience and have strict determinism. Under strict performance requirements, existing methods have inaccurate estimates and require manual order adjustment and repeated testing based on design results, which increases design costs. Summary of the invention
[0004] In view of the deficiencies in the prior art, the present invention provides a method and system for estimating the order of an FIR filter using a deep learning neural network. The present invention can achieve an estimated mapping of frequency and performance to order through a neural network, and its high accuracy can avoid the repeated process of subsequent adjustment and testing.
[0005] In a first aspect, the present invention provides a method for estimating the order of an FIR filter using a neural network, the method comprising the following steps:
[0006] Constructing a data set for a deep learning neural network, the data set comprising a filter order, a passband frequency, a stopband frequency, a passband attenuation, a stopband attenuation, and a fluctuation weight value;
[0007] Constructing a deep neural network, taking the passband frequency, stopband frequency, passband attenuation, and stopband attenuation as inputs, taking the filter order and fluctuation weight value as outputs, and using the data set to train and test the neural network;
[0008] Taking the FIR filter design requirements as input, using the trained deep neural network, we can get an estimate of the filter order.
[0009] The estimated filter order and fluctuation weight value are combined with the given frequency for minimum-maximum design or least-squares design to obtain specific FIR filter coefficient results.
[0010] In a second aspect, the present invention provides a FIR filter order estimation system using a neural network, comprising:
[0011] A data set construction unit, used to construct a data set for a deep learning neural network, wherein the data set includes a filter order, a passband frequency, a stopband frequency, a passband attenuation, a stopband attenuation, and a fluctuation weight value;
[0012] A network model building unit, used for building a deep neural network, taking the passband frequency, stopband frequency, passband attenuation, and stopband attenuation as inputs, taking the filter order and fluctuation weight value as outputs, and using the data set to train and test the neural network;
[0013] An order estimation unit, which takes the FIR filter design requirements as input and uses the trained deep neural network to obtain an estimate of the filter order;
[0014] The filter coefficient acquisition unit is used to combine the estimated filter order and fluctuation weighted value with the given frequency for minimum maximum design or least squares design to obtain specific FIR filter coefficient results.
[0015] In a third aspect, the present invention provides an intelligent terminal, comprising a memory, a processor, and a FIR filter order estimation program with the aid of a neural network, which is stored in the memory and can be run on the processor, and the FIR filter order estimation program with the aid of a neural network, when executed by the processor, implements the steps of the FIR filter order estimation method with the aid of a neural network.
[0016] In a fourth aspect, the present invention provides a computer-readable storage medium, on which is stored a FIR filter order estimation program using a neural network, which, when executed by a processor, implements the steps of the FIR filter order estimation method using a neural network.
[0017] The beneficial effects of the present invention are mainly manifested in: it can overcome the problem of narrow applicability of existing formula estimation methods, and can be applied to different FIR filter design methods through training of different deep neural networks, thereby achieving more accurate filter order estimation and avoiding excessive manual adjustments and repeated tests. BRIEF DESCRIPTION OF THE DRAWINGS
[0018] Figure 1 A schematic diagram for realizing order estimation with the help of neural network;
[0019] Figure 2 The filter design results after order estimation are shown. DETAILED DESCRIPTION
[0020] In the following description, specific details such as specific system structures and technologies are provided for illustration rather than limitation, so as to provide a thorough understanding of the embodiments of the present application. However, it should be clear to those skilled in the art that the present application can also be implemented in other embodiments without these specific details.
[0021] According to existing theories, there is an implicit numerical relationship between the order of FIR filters and their frequency and performance. The traditional method is to infer the filter order from the frequency and performance requirements through a certain mathematical formula relationship. The present invention aims to improve traditional filter design methods, such as the maximum-minimum method and the least squares method, and achieve accurate estimation of the filter order by introducing deep learning neural network technology. This method can ensure that the predetermined performance standards are met with a lower order, thereby avoiding the need for manual adjustment and search of the design order, and reducing the design cost and complexity. It has practical application value for quickly deploying digital filters as prototype filters to filtering multi-carrier systems.
[0022] An FIR filter order estimation method using a deep learning neural network provided in an embodiment of the present application is applicable to the design of a digital finite impulse response (FIR) filter, comprising the following steps:
[0023] First, construct a data set for deep learning neural networks, that is, obtain filters through minimum and maximum design or least squares design under given parameters, and perform filter performance tests to obtain a data set consisting of order, passband frequency, stopband frequency, passband attenuation, stopband attenuation, and fluctuation weighted values. Given parameters are discrete sample points in space, and the obtained data set is not complete, which will be used for training and testing deep neural networks.
[0024] Secondly, a deep neural network is constructed with passband frequency, stopband frequency, passband attenuation, and stopband attenuation as input, and filter order and fluctuation weighted value as output. The data set is used to train and test the neural network to achieve high-precision filter order estimation.
[0025] Then, the FIR low-pass filter design is performed according to specific requirements. The FIR filter design requirements, namely frequency and attenuation, are used as input, and the trained deep neural network is used to obtain an estimate of the filter order.
[0026] Finally, the estimated filter order and fluctuation weighted value are combined with the given frequency for minimum maximum design or least squares design to obtain a specific FIR filter coefficient result. The minimum maximum design and least squares design respectively use the estimated filter order and fluctuation weighted value to design the filter coefficient.
[0027] This application utilizes the highly nonlinear ability and generalized mapping ability of deep neural networks to realize the estimation of filter order through neural networks, which to a certain extent breaks through the limitations of existing formula estimation methods.
[0028] Furthermore, the data set in this embodiment is constructed by using the minimum maximum design and the least squares design, and the corresponding MATLAB functions used in the two are firpm() and firls(), and performance tests are performed.
[0029] During the design, the function filter order, passband and stopband frequencies, and fluctuation weight values need to be given. The function parameters are set based on the consideration that the FIR filter is applied to the filtering multi-carrier system as a prototype filter. After the filter coefficient values are designed, the passband and stopband attenuation are calculated as well as the actual passband and stopband cutoff frequencies. Finally, a data set consisting of filter order, passband frequency, stopband frequency, passband attenuation, stopband attenuation, and fluctuation weight is obtained, and some of the data are shown in Tables 1 and 2.
[0030] Table 1 Dataset constructed for min-max design
[0031] 43 0.0312 0.0906 3.0109 43.0369 21 43 0.0312 0.0906 3.0527 43.6737 23 43 0.0312 0.0906 3.0936 43.3556 25 43 0.0312 0.0906 3.1302 43.9792 27 43 0.0312 0.0906 3.1637 43.5181 29
[0032] Table 2 Dataset constructed for least squares design
[0033] 32 0.0531 0.1057 5.4484 21.2392 7 32 0.0531 0.1049 5.5366 21.7440 9 32 0.0531 0.1043 5.5944 22.0881 11 32 0.0531 0.1040 5.6352 22.3379 13 32 0.0531 0.1037 5.6655 22.5275 15
[0034] Among them, the first column is the filter order, the second column is the passband cutoff frequency, the third column is the stopband cutoff frequency, the fourth column is the passband attenuation, the fifth column is the stopband attenuation, and the sixth column is the fluctuation weight value.
[0035] Furthermore, based on the data set, with the help of the matlab deep neural network toolbox, a neural network is constructed for mapping from passband frequency, stopband frequency, passband attenuation, stopband attenuation to filter order and fluctuation weight value. Considering the requirements of mapping accuracy and network implementation complexity, the test in this embodiment uses a 4-input 2-output, 2 hidden layers, and the output layer uses a deep neural network containing intlayer. The process of realizing filter order estimation is as follows: Figure 1 shown.
[0036] Furthermore, based on the constructed neural network, the filter frequency and attenuation requirements are input into the network to obtain the filter order and fluctuation weighted value. In the test, the typical IS-95 filter design requirements were used, the normalized passband frequency was set to 0.2373, the normalized stopband frequency was set to 0.3008, and the passband and stopband attenuation were required to be 1dB and 40dB respectively. Through the neural network suitable for minimum and maximum design, the filter order and fluctuation weighted value were estimated to be 46 and 6 respectively; using the obtained values, they were used in the firpm() function, and the filter design results were obtained as shown below. Figure 2 As shown (labeled as pm). Through the neural network suitable for least squares design, the filter order and fluctuation weight value are estimated to be 61 and 73 respectively; using the obtained values, they are used in the firls() function, and the filter design results are as follows Figure 2 As shown in (marked as ls). From the results in the figure, both designs after order estimation meet the attenuation requirement of {1dB, 40dB}, which shows that the order estimation method of this embodiment is effective.
[0037] Corresponding to the above-mentioned FIR filter order estimation method using a neural network, an embodiment of the present application further provides a FIR filter order estimation system using a neural network, including:
[0038] A data set construction unit, used to construct a data set for a deep learning neural network, wherein the data set includes a filter order, a passband frequency, a stopband frequency, a passband attenuation, a stopband attenuation, and a fluctuation weight value;
[0039] A network model building unit, used for building a deep neural network, taking the passband frequency, stopband frequency, passband attenuation, and stopband attenuation as inputs, taking the filter order and fluctuation weight value as outputs, and using the data set to train and test the neural network;
[0040] An order estimation unit, which takes the FIR filter design requirements as input and uses the trained deep neural network to obtain an estimate of the filter order;
[0041] The filter coefficient acquisition unit is used to combine the estimated filter order and fluctuation weighted value with the given frequency for minimum maximum design or least squares design to obtain specific FIR filter coefficient results.
[0042] It should be noted that the specific structure and implementation of the above-mentioned FIR filter order estimation system using a neural network and its various modules or units can refer to the corresponding description in the above-mentioned method embodiment, and will not be repeated here.
[0043] It should be noted that the division method of each module of the above-mentioned FIR filter order estimation system using a neural network is not unique and is not specifically limited here.
[0044] Based on the above embodiments, the present application also provides an intelligent terminal, which includes a processor, a memory, a network interface and a display screen connected through a system bus. Among them, the processor of the intelligent terminal is used to provide computing and control capabilities. The memory of the intelligent terminal includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a FIR filter order estimation program using a neural network. The internal memory provides an environment for the operation of the operating system and the FIR filter order estimation program using a neural network in the non-volatile storage medium. The network interface of the intelligent terminal is used to communicate with an external terminal through a network connection. When the FIR filter order estimation program using a neural network is executed by the processor, the steps of any of the above-mentioned FIR filter order estimation methods using a neural network are implemented. The display screen of the intelligent terminal can be a liquid crystal display screen or an electronic ink display screen.
[0045] An embodiment of the present application also provides a computer-readable storage medium, on which is stored a FIR filter order estimation program using a neural network. When the FIR filter order estimation program using a neural network is executed by a processor, the steps of any one of the FIR filter order estimation methods using a neural network provided in the embodiments of the present application are implemented.
[0046] The embodiments described above are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. However, these modifications or replacements do not deviate from the spirit and scope of the technical solutions of the embodiments of the present application, and should be included in the protection scope of the present application.
Claims
1. A method for estimating the order of an FIR filter using a neural network, characterized in that: The following steps are involved: Constructing a data set for a deep learning neural network, the data set comprising a filter order, a passband frequency, a stopband frequency, a passband attenuation, a stopband attenuation, and a fluctuation weight value; Constructing a deep neural network, taking the passband frequency, stopband frequency, passband attenuation, and stopband attenuation as inputs, taking the filter order and fluctuation weight value as outputs, and using the data set to train and test the neural network; Taking the FIR filter design requirements as input, using the trained deep neural network, we can get an estimate of the filter order. The estimated filter order and fluctuation weight value are combined with the given frequency for minimum-maximum design or least-squares design to obtain specific FIR filter coefficient results.
2. The method according to claim 1, characterized in that The data set is constructed using min-max design and least squares design, and the filter design is performed using matlab functions firpm() and firls(), respectively, and performance tests are performed.
3. The method according to claim 1, characterized in that: It also includes obtaining a filter through minimum-maximum design or least-squares design under given parameters, and performing filter performance testing to obtain a data set consisting of filter order, passband frequency, stopband frequency, passband attenuation, stopband attenuation, and fluctuation weighted value.
4. The method according to claim 1, 2 or 3, characterized in that: The deep neural network is a neural network with four inputs, two outputs and two hidden layers.
5. The method according to claim 4, characterized in that The output layer of the neural network comprises an intlayer.
6. The method according to claim 1, characterized in that The FIR filter design requirements include normalized passband frequency, normalized stopband frequency, passband attenuation, and stopband attenuation.
7. The method according to claim 1, characterized in that The minimum maximum design and the least square design respectively use the estimated filter order and the fluctuation weight value to design the filter coefficients.
8. A FIR filter order estimation system using a neural network, characterized in that: include: A data set construction unit, used to construct a data set for a deep learning neural network, wherein the data set includes a filter order, a passband frequency, a stopband frequency, a passband attenuation, a stopband attenuation, and a fluctuation weight value; A network model building unit, used for building a deep neural network, taking the passband frequency, stopband frequency, passband attenuation, and stopband attenuation as inputs, taking the filter order and fluctuation weight value as outputs, and using the data set to train and test the neural network; An order estimation unit, which takes the FIR filter design requirements as input and uses the trained deep neural network to obtain an estimate of the filter order; The filter coefficient acquisition unit is used to combine the estimated filter order and fluctuation weighted value with the given frequency for minimum maximum design or least squares design to obtain specific FIR filter coefficient results.
9. An intelligent terminal, characterized in that: The intelligent terminal includes a memory, a processor, and a FIR filter order estimation program with the help of a neural network stored in the memory and executable on the processor. When the FIR filter order estimation program with the help of a neural network is executed by the processor, the steps of the FIR filter order estimation method with the help of a neural network as described in any one of claims 1 to 7 are implemented.
10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a FIR filter order estimation program using a neural network, and when the FIR filter order estimation program using a neural network is executed by a processor, the steps of the FIR filter order estimation method using a neural network as described in any one of claims 1 to 7 are implemented.