CN113848532A - FMCW radar signal noise reduction system and method based on noise reduction model - Google Patents
FMCW radar signal noise reduction system and method based on noise reduction model Download PDFInfo
- Publication number
- CN113848532A CN113848532A CN202111127976.1A CN202111127976A CN113848532A CN 113848532 A CN113848532 A CN 113848532A CN 202111127976 A CN202111127976 A CN 202111127976A CN 113848532 A CN113848532 A CN 113848532A
- Authority
- CN
- China
- Prior art keywords
- noise reduction
- signal
- fmcw radar
- time domain
- module
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01S—RADIO DIRECTION-FINDING; RADIO NAVIGATION; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION OR RERADIATION OF RADIO WAVES; ANALOGOUS ARRANGEMENTS USING OTHER WAVES
- G01S7/00—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00
- G01S7/02—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00 of systems according to group G01S13/00
- G01S7/36—Means for anti-jamming, e.g. ECCM, i.e. electronic counter-counter measures
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- General Engineering & Computer Science (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Biomedical Technology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Computer Networks & Wireless Communication (AREA)
- Radar, Positioning & Navigation (AREA)
- Remote Sensing (AREA)
- Radar Systems Or Details Thereof (AREA)
Abstract
The invention relates to a radar signal noise reduction technology, and discloses an FMCW radar signal noise reduction system and method based on a noise reduction model, which directly reduce noise of an original time domain signal so as to improve radar detection distance and detection precision. The system comprises: the time domain signal preprocessing module is used for preprocessing the FMCW radar signal time domain data; the encoder module is used for encoding the preprocessed FMCW radar signal time domain data to obtain a characteristic vector on a characteristic space; a separator module for calculating a mask of the feature vectors on the feature space obtained by the encoding; and the decoder module is used for generating the time domain signal after noise reduction according to the feature vector and the mask on the feature space.
Description
Technical Field
The invention relates to a radar signal noise reduction technology, in particular to an FMCW (Frequency-Modulated Continuous-Wave) radar signal noise reduction system and method based on a noise reduction model.
Background
FMCW radar refers to radar that transmits a continuous wave signal, the frequency of which is modulated by a particular signal. The FMCW radar mainly comprises a radio frequency front end consisting of a transmitting antenna, a receiving antenna, a power divider mixer and a frequency mixer, a triangular wave generator and a back end processing part for AD sampling and signal processing. In the use process of the radar, thermal noise and interference are phenomena which exist all the time; noise and interference can cause problems of raising detection threshold, generating false target and the like, and can seriously affect the use of radar. On one hand, the minimum detectable signal-to-noise ratio of a receiver can be reduced, and the radar detection range is increased; on the other hand, the signal-to-noise ratio is improved, so that the signal and the noise can be distinguished more obviously, and the discovery probability is improved under the condition of a certain false alarm probability, so that a weak target can be detected more easily.
The traditional denoising method is based on signal processing and statistical means, and mainly comprises FIR filtering, IIR filtering, median filtering, wavelet transformation and the like. The general steps can be summarized as follows: 1. preprocessing the radar signal according to a selected method; 2. and performing noise reduction calculation on the radar signal by using a filtering or wavelet transformation method.
The noise reduction method using conventional signal processing also has significant drawbacks: the filtering method is suitable for the condition that the overlapping of signal and noise frequency bands is very small, and can only inhibit the noise of a fixed frequency band; however, in practical situations, the frequency bands of the difference frequency signal and the noise of the FMCW radar are mixed together, and meanwhile, the frequency band of the difference frequency signal also changes along with different target positions; the filtering method cannot automatically modify the filter parameters according to the characteristics of the frequency domain distribution of the difference frequency signal, and the filtering effect is not ideal. The wavelet transform method can effectively inhibit noise and improve the signal-to-noise ratio to a certain extent, but the principle is complex, the calculation amount is large, hardware is not easy to realize, and the method is not widely applied to an FMCW radar system.
In recent years, researchers have attempted to use neural network methods to denoise FMCW radar signals. At present, two ideas exist in the field of deep learning for noise reduction of FMCW data, and the main direction is to process an RD (Range-Doppler) graph and an mD (micro-Doppler) graph obtained after radar signals are processed in an image noise reduction mode, such as a convolutional neural network or a generation countermeasure network, but the method needs to process the signals into an image format, and the use scene is limited.
Disclosure of Invention
The technical problem to be solved by the invention is as follows: the FMCW radar signal noise reduction system and method based on the noise reduction model are provided, and the original time domain signals are directly subjected to noise reduction so as to improve the radar detection distance and the detection precision.
The technical scheme adopted by the invention for solving the technical problems is as follows:
an FMCW radar signal noise reduction system based on a noise reduction model, comprising:
the time domain signal preprocessing module is used for preprocessing the FMCW radar signal time domain data;
the encoder module is used for encoding the preprocessed FMCW radar signal time domain data to obtain a characteristic vector on a characteristic space;
a separator module for calculating a mask of the feature vectors on the feature space obtained by the encoding;
and the decoder module is used for generating the time domain signal after noise reduction according to the feature vector and the mask on the feature space.
As a further optimization, the encoder module consists of a one-dimensional time-sequential convolutional neural network.
As a further optimization, the separator module is formed by stacking one-dimensional expansion convolution neural network modules, and each expansion convolution neural network module is formed by combining a depth separable convolution and skip-connection (jump transfer) structure.
As a further optimization, the decoder module consists of a one-dimensional time-sequential convolutional neural network.
In addition, based on the system, the invention also provides an FMCW radar signal noise reduction method based on the noise reduction model, which comprises the following steps:
A. training an FMCW radar time domain signal noise reduction model, the model comprising an encoder module, a separator module and a decoder module:
a1, preprocessing the selected FMCW radar signal time domain data;
a2, encoding the preprocessed FMCW radar signal time domain data by using an encoder module to obtain a characteristic vector on a characteristic space;
a3, learning a mask of an interference signal in a feature vector on a feature space by using a separator module;
a4, generating a noise-reduced time domain signal by a decoder module according to the feature vector and the mask on the feature space;
a5, adopting a signal-to-noise ratio with unchanged scale as a loss function of the noise reduction model, and maximizing the loss function to carry out optimization training until a trained FMCW radar time domain signal noise reduction model is obtained;
B. and taking an FMCW radar time domain signal to be denoised as input, and obtaining the denoised time domain signal by using the trained FMCW radar time domain signal denoising model.
As a further optimization, in step a1, the preprocessing includes a normalization method or a zero-averaging method.
As a further optimization, in step a2, the encoding, by the encoder module, of the preprocessed FMCW radar signal time domain data specifically includes: for an input time domain signal x, the signal is first cut into T segments x of length L that can be overlappedkK 1, 2.. times.t, then for each slice data xkObtaining an N-dimensional feature expression vector w of the input signal in a feature space by using one-dimensional convolutionk。
As a further optimization, in step A3, the splitter module learns the masks m corresponding to the clean signal and the noise signal in the feature coding through the stacked dilation convolution and skip-connection structures1And m2。
As a further optimization, in step a4, the generating, by the decoder module, a noise-reduced time-domain signal according to the feature vector and the mask in the feature space specifically includes: the decoder module mixes the feature vector and the mask corresponding to the clean signal: m is1⊙wkThen decoded by a decoder to obtainAnd (4) reducing the noise of the clean signal.
As a further optimization, in step a5, in the optimization training process, when the loss value is not in the set threshold range, the model parameters are adjusted and training is continued until the loss value falls to the set threshold range, and the model at this time is taken as the FMCW radar time domain signal noise reduction model.
The invention has the beneficial effects that:
the FMCW radar signal time domain data are subjected to direct noise reduction through the training noise reduction model, signals do not need to be converted into image formats in advance, and therefore more application scenes are possessed; in practical application, only the time domain data of the original FMCW radar signal needs to be input, and then a clean time domain signal can be obtained, so that the radar detection range and accuracy are improved, the noise reduction process is automatically processed, and the efficiency is high.
Drawings
FIG. 1 is a block diagram of a noise reduction system of an FMCW radar signal based on a noise reduction model according to the present invention;
FIG. 2 is a flow chart of noise reduction model training in the present invention.
Detailed Description
The invention aims to provide an FMCW radar signal noise reduction system and method based on a noise reduction model, which directly reduce noise of an original time domain signal so as to improve radar detection distance and detection precision.
In particular, as shown in fig. 1, the FMCW radar signal noise reduction system based on the noise reduction model in the present invention includes: the device comprises a time domain signal preprocessing module, an encoder module, a separator module and a decoder module;
the time domain signal preprocessing module is used for preprocessing the FMCW radar signal time domain data;
the encoder module is used for encoding the preprocessed FMCW radar signal time domain data to obtain a characteristic vector on a characteristic space;
a separator module for calculating a mask of the feature vectors on the feature space obtained by the encoding;
and the decoder module is used for generating the time domain signal after noise reduction according to the feature vector and the mask on the feature space.
The encoder module consists of a one-dimensional convolutional neural network and is used for encoding the preprocessed data to obtain a feature vector on a feature space. And the separator module is formed by stacking one-dimensional expansion convolution neural network modules, and each convolution module is formed by combining a depth separable convolution and skip-connection structure. The splitter learns the mask of the clean signal in feature encoding by means of stacked dilation convolution and skip-connection structures. And the decoder module consists of a one-dimensional convolutional neural network. The decoder module reconstructs a noise-free signal by adding the encoder features to the mask obtained by the separator.
Based on the system, the FMCW radar signal noise reduction method based on the noise reduction model comprises two parts of model training and noise reduction processing by utilizing the model;
firstly, training an FMCW radar time domain signal noise reduction model, wherein the model comprises an encoder module, a separator module and a decoder module, before training, setting a model loss function, the number of one-dimensional convolution blocks, the size of a receptive field, setting a method for iteratively updating model parameters, initializing parameters of each layer in the model, connecting and aligning network layers, selecting model training parameters and the like.
The training process is shown in fig. 2, and includes the following steps:
firstly, selecting FMCW radar signal time domain data with a proper length, and preprocessing the FMCW radar signal time domain data;
in this step, the selected signal length is generally in units of chirp of the FMCW radar, and the preprocessing method may select a normalization method of subtracting the minimum value from the time domain signal and then dividing the difference between the maximum value and the minimum value, or a zero-averaging method of subtracting the mean value from the square difference.
Secondly, carrying out feature space coding on the data obtained by preprocessing by using a coder;
in this step, the flow of processing the input data by the encoder is as follows: for an input time domain signal x, the signal is first cut into T segments x of length L that can be overlappedk，k＝1, 2, T, then data x for each slicekObtaining an N-dimensional feature expression vector w of the input signal in a feature space by using one-dimensional convolutionk。
Thirdly, learning the mask of the clean signal in the feature vector on the feature space by using a separator module;
in this step, the separator learns the mask m corresponding to the clean signal and the noise signal in the feature coding through the stacked expansion convolution and skip-connection structure1And m2. The encoding vector dimension is N-dimensional, then the mask vector dimension is also N-dimensional.
Fourthly, generating a time domain signal after noise reduction by using a decoder module according to the feature vector and the mask on the feature space;
in this step, the step of noise reduction by the decoder is to mix the feature vector and the mask corresponding to the clean signal: m is1⊙wkThen, a decoder is used for decoding to obtain a clean signal after noise reduction.
Fifthly, judging whether the loss value is in a threshold range;
in the step, the signal-to-noise ratio with unchanged scale is used as a loss function of the noise reduction model, the loss function is maximized to carry out optimization training, and in the training process, when the loss value is not within the threshold range, model parameters are adjusted and training is continued until the loss value is within the threshold range, and the model is used as the signal noise reduction model.
Based on the steps, a trained FMCW radar signal noise reduction model can be obtained, in practical application, a section of FMCW radar time domain signal is given as input, and the model can be used for directly obtaining the noise-reduced time domain signal.
Claims (10)
1. FMCW radar signal noise reduction system based on a noise reduction model, comprising:
the time domain signal preprocessing module is used for preprocessing the FMCW radar signal time domain data;
the encoder module is used for encoding the preprocessed FMCW radar signal time domain data to obtain a characteristic vector on a characteristic space;
a separator module for calculating a mask of the feature vectors on the feature space obtained by the encoding;
and the decoder module is used for generating the time domain signal after noise reduction according to the feature vector and the mask on the feature space.
2. The noise reduction model-based FMCW radar signal noise reduction system of claim 1,
the encoder module is composed of a one-dimensional time sequence convolution neural network.
3. The noise reduction model-based FMCW radar signal noise reduction system of claim 1,
the separator module is formed by stacking one-dimensional expansion convolution neural network modules, and each expansion convolution neural network module is formed by combining a depth separable convolution and skip-connection structure.
4. The noise reduction model-based FMCW radar signal noise reduction system of claim 1,
the decoder module is composed of a one-dimensional time sequence convolution neural network.
5. A FMCW radar signal noise reduction method based on a noise reduction model, applied to the system of any one of claims 1-4, comprising the steps of:
A. training an FMCW radar time domain signal noise reduction model, the model comprising an encoder module, a separator module and a decoder module:
a1, preprocessing the selected FMCW radar signal time domain data;
a2, encoding the preprocessed FMCW radar signal time domain data by using an encoder module to obtain a characteristic vector on a characteristic space;
a3, learning a mask of an interference signal in a feature vector on a feature space by using a separator module;
a4, generating a noise-reduced time domain signal by a decoder module according to the feature vector and the mask on the feature space;
a5, adopting a signal-to-noise ratio with unchanged scale as a loss function of the noise reduction model, and maximizing the loss function to carry out optimization training until a trained FMCW radar time domain signal noise reduction model is obtained;
B. and taking an FMCW radar time domain signal to be denoised as input, and obtaining the denoised time domain signal by using the trained FMCW radar time domain signal denoising model.
6. A FMCW radar signal noise reduction method based on a noise reduction model as set forth in claim 5,
in step a1, the preprocessing includes a normalization method or a zero-averaging method.
7. A FMCW radar signal noise reduction method based on a noise reduction model as set forth in claim 5,
in step a2, the encoding, by the encoder module, of the preprocessed FMCW radar signal time domain data specifically includes: for an input time domain signal x, the signal is first cut into T segments x of length L that can be overlappedkK 1, 2, …, T, and then for each slice data xkObtaining an N-dimensional feature expression vector w of the input signal in a feature space by using one-dimensional convolutionk。
8. A FMCW radar signal noise reduction method based on a noise reduction model as set forth in claim 5,
in step A3, the splitter module learns the masks m corresponding to the clean signal and the noise signal in the feature coding through the stacked dilation convolution and skip-connection structure1And m2。
9. The method of claim 5, wherein the step A4 of generating the noise-reduced time-domain signal according to the eigenvector and the mask in the eigenspace by the decoder module specifically comprises: decoder moduleThe block first mixes the feature vector and the mask corresponding to the clean signal: m is1⊙wkThen, a decoder is used for decoding to obtain a clean signal after noise reduction.
10. A FMCW radar signal noise reduction method based on a noise reduction model as set forth in claim 5,
in step a5, in the optimization training process, when the loss value is not in the set threshold range, adjusting the model parameters and continuing training until the loss value is reduced to the set threshold range, and taking the model at this time as the FMCW radar time-domain signal noise reduction model.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111127976.1A CN113848532A (en) | 2021-09-26 | 2021-09-26 | FMCW radar signal noise reduction system and method based on noise reduction model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111127976.1A CN113848532A (en) | 2021-09-26 | 2021-09-26 | FMCW radar signal noise reduction system and method based on noise reduction model |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN113848532A true CN113848532A (en) | 2021-12-28 |
Family
ID=78979496
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111127976.1A Pending CN113848532A (en) | 2021-09-26 | 2021-09-26 | FMCW radar signal noise reduction system and method based on noise reduction model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113848532A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115991171A (en) * | 2023-03-20 | 2023-04-21 | 凯晟动力技术（嘉兴）有限公司 | Vehicle body controller and method based on remote control key start |
-
2021
- 2021-09-26 CN CN202111127976.1A patent/CN113848532A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115991171A (en) * | 2023-03-20 | 2023-04-21 | 凯晟动力技术（嘉兴）有限公司 | Vehicle body controller and method based on remote control key start |
| CN115991171B (en) * | 2023-03-20 | 2023-08-15 | 凯晟动力技术（嘉兴）有限公司 | Vehicle body controller and method based on remote control key start |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111142105B (en) | ISAR imaging method for complex moving target | |
| Fuchs et al. | Automotive radar interference mitigation using a convolutional autoencoder | |
| CN108051812B (en) | Satellite-borne SAR moving target detection method based on two-dimensional speed search | |
| CN112087774B (en) | Communication radiation source individual identification method based on residual error neural network | |
| CN109581352A (en) | A kind of super-resolution angle measuring system based on millimetre-wave radar | |
| CN111580107B (en) | Radio frequency interference suppression method and device for SAR echo signal and imaging method | |
| CN107831473B (en) | Distance-instantaneous Doppler image sequence noise reduction method based on Gaussian process regression | |
| CN102393423A (en) | Lamb wave denoising method based on adaptive threshold value orthogonal wavelet transform | |
| CN115359771B (en) | Underwater sound signal noise reduction method, system, equipment and storage medium | |
| CN113534151B (en) | Dual-band ISAR imaging method based on off-grid sparse Bayesian learning | |
| CN110320535A (en) | A kind of more disturbance restraining methods of satellite navigation receiver based on wavelet package transforms and spatial processing | |
| Chen et al. | A DNN autoencoder for automotive radar interference mitigation | |
| CN113848532A (en) | FMCW radar signal noise reduction system and method based on noise reduction model | |
| CN107102305B (en) | Robust cognitive radar transmitting and receiving joint design method in clutter environment | |
| CN113205564B (en) | SAR intelligent target edge reconstruction method | |
| CN114814840A (en) | Method and system for image screening and interference position detection of interference-containing synthetic aperture radar | |
| CN108196238B (en) | Clutter map detection method based on adaptive matched filtering under Gaussian background | |
| CN103426145A (en) | Synthetic aperture sonar speckle noise suppression method based on multiresolution analysis | |
| CN111044996A (en) | LFMCW radar target detection method based on dimension reduction approximate message transfer | |
| CN102509268B (en) | Immune-clonal-selection-based nonsubsampled contourlet domain image denoising method | |
| CN114696788B (en) | Wave form and filter combined cognition design method for resisting multi-main lobe interference | |
| CN114881215A (en) | Millimeter wave radar data screening neural network processing method | |
| CN115097391A (en) | Synthetic aperture radar interference suppression method based on automatic encoder | |
| CN115356729A (en) | Direct sparse imaging method for near-field non-uniform sampling | |
| CN115345216A (en) | FMCW radar interference elimination method fusing prior information |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |