CN-122027053-A - Signal-to-noise ratio estimation method based on IQ sequence and eigenvalue characterization learning

CN122027053ACN 122027053 ACN122027053 ACN 122027053ACN-122027053-A

Abstract

The invention discloses a signal-to-noise ratio estimation method based on IQ sequence and eigenvalue characterization learning, which solves the problem that the estimation accuracy and the calculation complexity of the existing signal-to-noise ratio estimation mode are difficult to be compatible. The method comprises the steps of constructing a feature sample set according to the corresponding relation between an IQ input matrix of each historical complex baseband signal and covariance eigenvalue vectors to train feature extraction sub-models, obtaining predicted values of covariance eigenvalue vectors corresponding to each historical complex baseband signal by utilizing the trained feature extraction sub-models, constructing a signal-to-noise ratio sample set according to the corresponding relation between the predicted values of covariance eigenvalue vectors and signal-to-noise ratios to train signal-to-noise ratio regression sub-models, carrying out cascade connection on the trained feature extraction sub-models and the signal-to-noise ratio regression sub-models to obtain signal-to-noise ratio estimation models, carrying out fine adjustment on network parameters of the signal-to-noise ratio estimation models, and carrying out signal-to-noise ratio estimation on real-time complex baseband signals by utilizing the fine-adjusted signal-to-noise ratio estimation models.

Inventors

CHEN TAO
CHEN LU
ZHANG LU
ZHOU XIAOLONG
WANG HAILUN

Assignees

衢州学院

Dates

Publication Date: 20260512
Application Date: 20260129

Claims (10)

1. An signal-to-noise ratio estimation method based on IQ sequence and eigenvalue characterization learning, which is characterized by comprising the following steps: Constructing a characteristic sample set according to the corresponding relation between the IQ input matrix of each historical complex baseband signal and the covariance characteristic value vector; Obtaining a predicted value of a covariance eigenvalue vector corresponding to each historical complex baseband signal by utilizing the feature extraction submodel which is trained, constructing a signal-to-noise ratio sample set according to the corresponding relation between the predicted value of the covariance eigenvalue vector of each historical complex baseband signal and the signal-to-noise ratio; Cascading the feature extraction sub-model and the signal to noise ratio regression sub-model which are trained to obtain a signal to noise ratio estimation model; And estimating the signal-to-noise ratio of the real-time complex baseband signal by using the signal-to-noise ratio estimation model after fine adjustment.
2. The signal-to-noise ratio estimation method based on IQ sequence and eigenvalue characterization learning according to claim 1, wherein the constructing of the characteristic sample set is performed by: IQ input matrix and covariance eigenvalue vector of each historical complex baseband signal are obtained respectively. And constructing a characteristic sample set by taking the IQ input matrix of each historical complex baseband signal as a characteristic sample and taking the corresponding covariance characteristic value vector as a label value of the characteristic sample.
3. The signal-to-noise ratio estimation method based on IQ sequence and eigenvalue characterization learning according to claim 2, wherein the training of the feature extraction sub-model with the feature sample set is performed by: And training the feature extraction submodel by taking an IQ input matrix of each historical complex baseband signal as input and a corresponding covariance eigenvalue vector as output to obtain a trained feature extraction submodel.
4. The signal-to-noise ratio estimation method based on IQ sequence and eigenvalue characterization learning according to claim 3 wherein the loss function of the feature extraction submodel Expressed as: (1) Wherein, the 、 Respectively the first Predicted values and label values of covariance eigenvalue vectors corresponding to the individual eigenvalues, Representing the training batch size.
5. The signal-to-noise ratio estimation method based on IQ sequence and eigenvalue characterization learning according to claim 4 wherein the feature extraction sub-model employs a residual neural network; The residual neural network sequentially comprises a two-dimensional convolution layer, a plurality of residual blocks, a self-adaptive average pooling layer and a full-connection layer.
6. The signal-to-noise ratio estimation method based on IQ sequence and eigenvalue characterization learning according to claim 5, wherein the constructing a signal-to-noise ratio sample set is performed by: and summarizing all the signal-to-noise ratio samples and the labels thereof by taking the predicted value of the covariance eigenvalue vector of each historical complex baseband signal as the signal-to-noise ratio sample and the corresponding signal-to-noise ratio as the label of the signal-to-noise ratio sample, so as to construct a signal-to-noise ratio sample set.
7. The signal-to-noise ratio estimation method based on IQ sequence and eigenvalue characterization learning of claim 6, wherein regression loss function of signal-to-noise ratio regression sub-model Expressed as: (2) Wherein, the Represent the first Signal-to-noise ratio labeling values corresponding to the signal-to-noise ratio samples, Represent the first And predicting the signal-to-noise ratio estimated value output by the signal-to-noise ratio regression submodel corresponding to each signal-to-noise ratio sample.
8. The signal-to-noise ratio estimation method based on IQ sequence and eigenvalue characterization learning according to claim 7, wherein the fine tuning of network parameters of the signal-to-noise ratio estimation model is performed by: And selecting an IQ input matrix of part of the historical complex baseband signals as input and a corresponding signal-to-noise ratio as output, and performing fine adjustment on network parameters of the signal-to-noise ratio estimation model.
9. The method for estimating signal-to-noise ratio based on IQ sequence and eigenvalue characterization learning according to claim 8, wherein a loss function of a signal-to-noise ratio estimation model The definition is as follows: (3) Wherein, the Is a weight coefficient.
10. The signal-to-noise ratio estimation method based on IQ sequence and eigenvalue characterization learning according to claim 9, wherein the performing signal-to-noise ratio estimation on the real-time complex baseband signal using the fine-tuned signal-to-noise ratio estimation model is performed by: inputting the IQ input matrix of the real-time complex baseband signal into a signal-to-noise ratio estimation model after fine adjustment, and predicting and outputting a corresponding signal-to-noise ratio estimation result by the signal-to-noise ratio estimation model.

Description

Signal-to-noise ratio estimation method based on IQ sequence and eigenvalue characterization learning Technical Field The invention relates to the technical field of communication signal identification, in particular to a signal-to-noise ratio estimation method based on IQ sequence and eigenvalue characterization learning. Background The Signal-to-Noise Ratio (SNR) is an important parameter for measuring the relative relation between the useful Signal power and the Noise power in the received Signal, and is a key index for characterizing the channel quality and the receiving performance in the communication system. The accurate signal-to-noise ratio estimation has important roles in the functions of signal detection, adaptive modulation and coding, power control, spectrum sensing and the like, so that the stable and accurate signal-to-noise ratio estimation is realized in a complex wireless environment, and has important engineering significance. The existing signal-to-noise ratio estimation method mainly comprises a traditional method based on statistical analysis and a method based on deep learning. The conventional statistical method generally uses moment characteristics, power characteristics or related characteristics of the received signal to perform modeling, and has the advantages of simple implementation structure and clear theoretical model, but often depends on noise distribution assumptions or priori conditions. When the signal-to-noise ratio is low, the noise presents non-Gaussian distribution or the channel condition changes, the estimation accuracy and the robustness of the signal-to-noise ratio are easy to be reduced, and the signal-to-noise ratio is difficult to adapt to the complex and changeable actual wireless environment. In recent years, a signal-to-noise ratio estimation method based on deep learning is paid attention to, and the method generally takes in-phase components and quadrature components (IQ sequences) of a received signal as input, automatically learns a nonlinear relation between the signal and noise through a neural network, and shows better modeling capability under a low signal-to-noise ratio condition. However, the direct IQ sequence-based method generally has the problems of high input dimension and large feature redundancy, which results in high model parameter scale and computation complexity, and is unfavorable for deployment in a real-time system or resource-constrained equipment. To reduce feature dimensions and enhance statistical stability, the prior art attempts to introduce covariance matrices and their eigenvalue vectors as intermediate features for signal-to-noise ratio estimation. The eigenvalue vector can effectively describe signal energy distribution and noise structure under lower dimensionality, and has better physical interpretability and discrimination capability. However, the existing scheme generally needs to explicitly construct a covariance matrix and perform eigenvalue decomposition operation in an inference stage, and an additional matrix operation process increases system calculation burden and processing time delay, so that the application of the covariance matrix in an actual communication system is limited. Therefore, how to avoid explicit eigenvalue calculation and realize efficient estimation from IQ sequence to signal-to-noise ratio while maintaining the advantages of eigenvalue statistical properties is still a technical problem to be solved in the art. Disclosure of Invention In view of the above analysis, the embodiment of the invention aims to provide a signal-to-noise ratio estimation method based on IQ sequence and eigenvalue characterization learning, which is used for solving the problem that the estimation accuracy and the calculation complexity of the existing signal-to-noise ratio estimation mode are difficult to be compatible. The invention provides a signal-to-noise ratio estimation method based on IQ sequence and eigenvalue characterization learning, which is characterized by comprising the following steps: Constructing a characteristic sample set according to the corresponding relation between the IQ input matrix of each historical complex baseband signal and the covariance characteristic value vector; Obtaining a predicted value of a covariance eigenvalue vector corresponding to each historical complex baseband signal by utilizing the feature extraction submodel which is trained, constructing a signal-to-noise ratio sample set according to the corresponding relation between the predicted value of the covariance eigenvalue vector of each historical complex baseband signal and the signal-to-noise ratio; Cascading the feature extraction sub-model and the signal to noise ratio regression sub-model which are trained to obtain a signal to noise ratio estimation model; And estimating the signal-to-noise ratio of the real-time complex baseband signal by using the signal-to-noise ratio estimation model after fine adjustment. Base