FBMC waveform design method of design mode intelligent parameter accurate value

Abstract

The invention provides an FBMC (Filter Bank Management Controller) waveform design method for determining intelligent parameter values of a design mode, which is used for a DFT-preread FBMC communication system. According to the method, mapping from system performance indexes to prototype filter design parameters is achieved through the neural network, and the problem of parameter exhaustion search in a traditional iterative least square method is avoided. A neural network is trained through an offline data set, performance requirements are converted into design parameters, a prototype filter is obtained through an iterative least square (ILS) method, and the efficiency and flexibility of FBMC waveform design are improved. Waveform design can be realized according to system performance requirements, and the design cost is reduced.

Description

Technical Field

[0001] The invention relates to the technical field of mobile communications, and in particular to an FBMC waveform design method for intelligent parameter value confirmation of a design mode. Background Art

[0002] With the surge in the number of network users and the emergence of new network services, the global demand for network traffic is growing at an alarming rate. Although traditional Orthogonal Frequency Division Multiplexing (OFDM) technology plays an important role in modern communication systems, it has some limitations, including high out-of-band leakage, dependence on cyclic prefixes, high peak-to-average power ratio, and strict requirements for system synchronization. These shortcomings limit the performance of OFDM technology in certain application scenarios.

[0003] FBMC technology has attracted the attention of researchers due to its low sidelobe power and high spectrum efficiency, and has been applied in many fields such as the Internet of Things and underwater acoustic communications. In the field of the Internet of Things, FBMC technology can efficiently utilize scattered spectrum resources, which is crucial for modern communication systems with increasingly tight spectrum resources. In the field of underwater acoustic communications, FBMC technology effectively suppresses out-of-band power leakage, reduces the need for virtual carrier technology, reduces crosstalk between spectrums, thereby improving spectrum efficiency and reducing the requirements for band spacing.

[0004] The core advantage of FBMC technology lies in the selection of its prototype filter, which makes it have obvious advantages over OFDM technology in terms of side lobes and spectrum efficiency. Different prototype filter design methods can optimize the FBMC system in various performance indicators, such as Signal to Interference plus Noise Ratio (SINR), Peak-to-Average Power Ratio (PAPR) and Out-of-Band Emission (OoBE), to meet the needs of different application scenarios. However, traditional FBMC prototype filter design methods have the problems of insufficient flexibility or too many adjustment parameters, which makes it difficult to accurately match the system performance requirements. Therefore, developing new prototype filter design methods to improve the flexibility and performance of FBMC systems has become an important direction of current research. Summary of the invention

[0005] In order to make the discrete Fourier transform (DFT) spread filter bank multi-carrier (DFT-spread FBMC) system meet the established performance index requirements (that is, the numerical requirements of SINR, OoBE, and PAPR performance) through prototype filter design, the present invention provides an FBMC waveform design method with intelligent parameter value confirmation of the design mode, which realizes the mapping of system performance index requirements to prototype filter design parameters through neural networks, and can avoid the problem of parameter exhaustive search. The design parameters obtained by the neural network can be directly used in the prototype filter design to obtain the required prototype filter.

[0006] In order to improve the flexibility of DFT-spread FBMC waveform to meet the performance requirements in different scenarios, the present invention adopts an iterative least squares method (ILS) with a perfect reconstruction error and filter out-of-band energy weighted as a cost function to realize FBMC prototype filter design; further, under specific system performance requirements, in order to reduce the cost of design implementation (searching for design parameters so that the resulting system performance meets the requirements), a database constructed offline is used to complete a neural network that can realize the mapping of system performance indicators to design parameters, and then the obtained design parameters are used in waveform design, so that the required FBMC prototype filter design can be realized through a single iterative least squares method.

[0007] The beneficial effects of the present invention are mainly manifested in: using the iterative least squares method to implement the FBMC prototype filter design, compared with the traditional method, it can effectively achieve flexible control of system performance by changing the design parameter value. The constructed neural network can convert specific system performance index requirements into design parameters required by the design mode, reducing the process of exhaustive search of parameters in the previous design mode to make the system performance meet the requirements, and can effectively improve the efficiency of FBMC waveform design. BRIEF DESCRIPTION OF THE DRAWINGS

[0008] Figure 1 This is the block diagram of the DFT-spread FBMC system;

[0009] Figure 2 Design patterns for iterative least squares unconstrained optimization;

[0010] Figure 3 This is the implementation process of the FBMC waveform design method according to the present invention;

[0011] Figure 4 To display the design results. DETAILED DESCRIPTION

[0012] In order to make the purpose, technical solutions and advantages of the present invention more clear, the present invention is described in detail below with reference to the accompanying drawings and specific embodiments.

[0013] It should be noted that, in order to avoid blurring the present invention due to some unimportant details, only the structures and processing steps closely related to the scheme of the present invention are shown in the drawings, and the remaining details that are not closely related to the present invention are omitted. It should be understood that the specific embodiments described herein are only used to explain the embodiments of the present invention, and are not used to limit the embodiments of the present invention.

[0014] The present application provides an FBMC waveform design method with intelligent parameter confirmation of the design mode, which is aimed at the iterative least squares unconstrained optimization design mode. This design mode is adopted because only two design parameters, weight value and filter length, need to be adjusted, but the adjustment of the two can achieve flexible compromise exchange of system performance indicators. It includes setting design parameters - solving design problems to obtain prototype filter coefficients - testing system performance indicators. This application will use a neural network to replace this mode to achieve the mapping of design parameters and system performance indicators, including the following steps:

[0015] First, for a specific DFT-spread FBMC system (such as Figure 1 As shown), based on Figure 2 The FBMC prototype filter design mode shown in the figure is used to prepare for the collection of offline data sets. That is, the design parameters, including the weight value α and the filter length value N, are transformed, and an iterative least squares solution is used to solve unconstrained optimization with the weighted cost function of the perfect reconstruction error and the filter out-of-band energy.

[0016] Further, in Figure 2 In the iterative least squares solution unconstrained optimization process shown in FIG, the square root raised cosine impulse function is first used to obtain the initial prototype filter coefficient g 0 The matrix parameters B and W can be calculated by the prototype filter. From the operation Q = B H W H Q g WB (which represents the entire DFT-spread FBMC process) is rewritten because

[0017]

[0018] in:

[0019]

[0020] Thus, Q can be rewritten as

[0021]

[0022] On the other hand, the iterative least squares solution unconstrained optimization process is an unconstrained optimization of the following cost function, where the first two terms represent the perfect reconstruction error and the last term is the out-of-band energy of the filter.

[0023]

[0024] in,

[0025] In order to achieve fast unconstrained optimization, the original problem (3) is transformed into the following form:

[0026]

[0027] Where Λ=diag(1 1×(KM)2 / 4 ,α 1×N ), α 1×N =[α,…,α],

[0028] q=[1,0 1×(KM / 2) ,1,0 1×(KM / 2) ,1,…,1,0 1×N ] T , C s It's Q s Cholesky decomposition of Because of the iterative calculation, it is calculated from the current g Used for LS solution of problem (4)

[0029] g=(Ψ T ΛΨ) -1 Ψ T Λq (5)

[0030] to get updates on g.

[0031] Get the prototype filter coefficient vector g = [g(0), g(1), ..., g(N-1)] T After that, the system performance index test is carried out to obtain the SINR, PAPR, and OoBE result values.

[0032] The data set consists of {α, N, SINR, PAPR, OoBE} values, see the table below

[0033] Weighted value Filter length SINR PAPR OoBE 32 28 11.3574 5.6302 0.0088 34 28 11.3962 4.2016 0.0088 36 28 11.5025 5.7798 0.0084 38 28 11.5298 3.7888 0.0088 40 28 11.3742 4.9742 0.0081 42 28 11.3922 4.8832 0.0088 44 28 11.5332 3.8120 0.0085 46 28 11.3913 4.5737 0.0088 48 28 11.5207 5.4471 0.0097 50 28 11.3797 5.6975 0.0093

[0034] The dataset is a set of discrete sample points in space. Figure 3 The neural network training and testing shown in .

[0035] Further, the calculation formula of the SINR is:

[0036]

[0037] Among them, Y S (k), Y I (k), Y n (k) are the signal part, interference part and noise part of the demodulated signal respectively.

[0038] Furthermore, the calculation formula of the PAPR is:

[0039]

[0040] Among them, s(n) is the transmitted signal and L is the signal length.

[0041] Furthermore, the calculation formula of OoBE is:

[0042]

[0043] Among them, P s (f) is the power spectral density function of the transmitted signal, and In-band width and out-band width

[0044] Secondly, construct Figure 3 The neural network shown in Figure 1 is the input of the network and the required values ​​of the system performance indicators {SINR, PAPR, OoBE}, and the output is the design parameters (weighting value α and filter length N) required to implement the FBMC prototype filter design. After splitting the data set, a neural network capable of achieving high-precision mapping is obtained through the training and testing of the neural network.

[0045] Furthermore, the neural network includes an input layer, a hidden layer and an output layer. The aforementioned data set is used to train and test the neural network. This is accomplished by splitting the data set, tuning the network structure and parameters; wherein the tuning objects include the number of hidden layers, the number of neurons, the activation function, the connection mode, the loss function, the optimizer, the initialization parameters, the learning rate, the batch size, etc., and the tuning is performed to ensure a high-accuracy mapping from the system performance indicators to the design parameters.

[0046] Then, for the specific application scenario, given the requirements for system performance {SINR, PAPR, OoBE}, the required values ​​are increased by about 10%-20% for the sake of mapping accuracy; and the improved values ​​are input into the trained neural network to obtain the design parameters required for the design mode (that is, iterative least squares solution for unconstrained optimization).

[0047] Finally, the design parameters are used in the design method to perform iterative least squares operation to obtain the prototype filter required by the FBMC system.

[0048] like Figure 3 As shown in the figure, the {SINR, PAPR, OoBE} values ​​are input into the neural network to obtain the parameters {α, N} required for the design mode, and then the ILS method is used to solve the prototype filter coefficients. In the test, the {SINR, PAPR, OoBE} requirement is set to {12dB, 2, 0.001}, and the obtained parameters {α, N} are {9, 33}. The prototype filter designed by ILS is applied to DFT-spreadFBMC, and the power spectral density of the obtained waveform is shown in Figure 4 After calculation, the results of the system {SINR, PAPR, OoBE} are {13.0272dB, 1.9597, 0.00097} respectively, which shows that the results meet the set requirements.

[0049] The embodiment of the present application is applicable to a DFT-spread FBMC communication system. For a waveform design mode that uses iterative least squares, a neural network is used to replace the entire process from the original design parameter input to the system performance test result output to achieve the mapping between the design parameters and the system performance indicators. Flexible FBMC waveform design under multiple performance requirements can be completed, avoiding the problem of exhaustive parameter search under the original design mode. The embodiment of the present application also implements the training and testing of the neural network through an offline design data set. In the face of specific performance requirements of different application scenarios, the specific performance requirements are input into the neural network to obtain the parameters required for the design mode, and then the prototype filter required for the FBMC waveform is obtained through the iterative least squares method to achieve an FBMC waveform that meets the performance requirements.

[0050] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, a person skilled in the art should understand that without departing from the concept of the present invention, several simple deductions or substitutions can be made, which should all be deemed to fall within the scope of protection determined by the claims submitted for the present invention.

Claims

1. A FBMC waveform design method with intelligent parameter determination of design mode, suitable for filter bank multi-carrier communication system using discrete Fourier transform extension, characterized in that: The method comprises the following steps: a) Collect offline data sets based on the FBMC prototype filter design mode of iterative least squares method, transform design parameters, and perform unconstrained optimization by iterative least squares solution with the weighted cost function of perfect reconstruction error and filter out-of-band energy; After obtaining the coefficient vector of the FBMC prototype filter, the system performance index test is carried out to obtain the signal-to-interference-noise ratio, peak-to-average power ratio, and out-of-band leakage result values, and finally a data set consisting of different design parameter values ​​is obtained; b) constructing a neural network, whose input is the required value of the system performance index and whose output is the design parameters required for the design of the FBMC prototype filter; After splitting the data set obtained in step a), a neural network capable of achieving high-precision mapping is obtained through a neural network training and testing process; c) For specific application scenarios, given the requirements for system performance, the required values ​​are increased by 10%-20%; and the increased values ​​are input into the trained neural network to obtain the design parameters required for the design model; d) using the design parameters obtained in step c) to perform an iterative least squares method to perform an unconstrained optimization operation to obtain a FBMC prototype filter.

2. The FBMC waveform design method according to claim 1, characterized in that: The iterative least squares solution unconstrained optimization process in step a) uses a square root raised cosine impulse function as the initial prototype filter coefficients.

3. The FBMC waveform design method according to claim 1 or 2, characterized in that: The calculation parameters include weighting values ​​and filter length values.

4. The FBMC waveform design method according to claim 1, characterized in that: The neural network structure in step b) includes an input layer, a hidden layer and an output layer; by splitting the data set, tuning the network structure and parameters, high-precision mapping from system performance indicators to design parameters is ensured.

5. The FBMC waveform design method according to claim 4, characterized in that: The data set used for the neural network training is a data set consisting of different design parameters and corresponding system performance index result values ​​obtained by transforming the design parameters and performing iterative least squares solution unconstrained optimization.

6. The FBMC waveform design method according to claim 5, characterized in that: The iterative least squares solution unconstrained optimization process includes transforming the original problem and performing iterative calculation using Cholesky decomposition to obtain updated prototype filter coefficients.

Images