Radiation source signal sample training data enhancement method and system
By simulating the random multipath effect, random truncation and random signal-to-noise ratio, diverse radiation source signal-to-noise samples are generated, which solves the problem of insufficient generalization of deep learning networks caused by insufficient sample size in the prior art, and improves the adaptability and generalization ability of deep learning networks to radiation source signals.
Patent Information
- Application Number
- CN202510188609.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-02-20
- Publication Date
- 2025-06-10
AI Technical Summary
In the prior art, the insufficient number of samples of the radiation source signal leads to insufficient generalization during training and the inability to effectively process the radiation source signal.
By simulating the random multipath effect, the start time of the random cutoff signal, and the random signal-to-noise ratio, a large number of radiation source signal-to-noise samples with different characteristics are generated. Specific steps include multipath effect processing, random truncation, noise superposition and multiple random sampling.
The generated diversified samples can cover various complex situations that may be encountered in practical applications of radiation source signals, allowing deep learning networks to be exposed to a wider range of signal characteristics during training, thereby improving their adaptability and generalization capabilities to unknown radiation source signal data.
Smart Images
Figure CN120123768A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of signal processing, and in particular to a method and system for enhancing training data of radiation source signal samples. Background Art
[0002] With the development of artificial intelligence deep learning technology, its research in the field of radiation source signal processing is becoming more and more extensive. However, in actual receiving systems, it is very difficult to obtain radiation source signals. The reception of radiation source signals is easily affected by the environment. In different electromagnetic environments, there will be different signal transmission paths. Due to the differences in transmission paths, the same transmitted signal will result in a huge difference in the received signal, and it is almost impossible to cover the signal reception in all scenarios. Therefore, the number of samples of radiation source signals is difficult to meet the requirements of deep learning training, resulting in insufficient generalization of the trained deep learning network. Therefore, there is an urgent need to construct a data enhancement algorithm for radiation source signals to improve the performance of the deep network for intelligent processing of radiation source signals. Summary of the Invention
[0003] To overcome the deficiencies of the prior art, the present invention provides a method and system for enhancing training data of radiation source signal samples, and solves the problems existing in the prior art such as insufficient generalization of deep learning network training due to insufficient sample size.
[0004] The technical solutions adopted by the present invention to solve the above problems are as follows:
[0005] A method for enhancing training data of radiation source signal samples includes the following steps:
[0006] S1, multipath effect processing: performing multipath effect processing on the radiation source signal received by the receiving system.
[0007] As a preferred technical solution, in step S1, the expression of the signal after multipath effect processing is:
[0008]
[0009] where s 1 (m) represents the signal after multipath effect processing, m represents the serial number of the signal sampling point, m = 1, 2,..., M, M represents the length of the signal, s(m) represents the radiation source signal received by the receiving system, k represents the transmission path number, K represents the number of transmission paths, s k (m) represents the signal of the kth transmission path, A k represents the amplitude attenuation of each transmission path, Delay k represents the delay of each transmission path, D k represents the delay sampling points corresponding to Delay k corresponding to s(m - Dk ) represents the delayed signal, represents rounding down, dt represents the sampling interval, exp represents the exponential operation, j represents the imaginary unit, π represents the pi, and RF represents the carrier frequency of the radiation source signal received by the receiving system.
[0010] As a preferred technical solution, in step S1, the Rayleigh distribution multipath effect model or the Rice distribution multipath effect model is used for multipath effect processing.
[0011] As a preferred technical solution, it includes the following steps:
[0012] S2, random truncation: randomly truncate the start time of the signal.
[0013] As a preferred technical solution, in step S2, the expression for randomly truncating the start time of the signal is:
[0014] s 2 (m) = s 1 (m - k + 1)
[0015] where, s 2 (m) represents the signal after random truncation, m = 1, 2,..., M - k + 1, and k represents the number of sampled points for truncation.
[0016] As a preferred technical solution, it includes the following steps:
[0017] S3, noise addition: add noise to the signal.
[0018] As a preferred technical solution, in step S3, add random Gaussian white noise to the signal.
[0019] As a preferred technical solution, in step S3, the expression for the signal after adding random Gaussian white noise is: s 3 (m) = s 2 (m) + a · [rand(1, M - k + 1) + j · rand(1, M - k + 1)]
[0020] where, s 3 (m) represents the signal after adding random Gaussian white noise, a represents the amplitude of the added random Gaussian white noise, a ∈ [0, A max , A max represents the maximum amplitude of the Gaussian white noise, and rand(1, *) represents the Gaussian random function.
[0021] As a preferred technical solution, it includes the following steps:
[0022] S4. Multiple random samplings: Steps S1 to S3 are looped N times; where N ≥ 2 and N is an integer.
[0023] A radiation source signal sample training data enhancement system for implementing the described radiation source signal sample training data enhancement method, including the following modules:
[0024] Multipath effect processing module: For performing multipath effect processing on the radiation source signal received by the receiving system.
[0025] Compared with the prior art, the present invention has the following beneficial effects:
[0026] By simulating conditions such as random multipath effects, random sampling start times, and random signal-to-noise ratios, the present invention generates a large number of radiation source signal samples with different characteristics. These diverse samples can cover various complex situations that the radiation source signal may encounter in actual applications, enabling the deep learning network to come into contact with a wider range of signal characteristics during the training process, thereby improving the adaptability and generalization ability of the deep learning network model to unknown radiation source signal data. Description of the Drawings
[0027] Figure 1 It is a step schematic diagram of a radiation source signal sample training data enhancement method described in the present invention. Detailed Embodiments
[0028] The following combines embodiments and the accompanying drawings to further elaborate on the present invention in detail, but the implementation manners of the present invention are not limited thereto.
[0029] Embodiment 1
[0030] As Figure 1 shown, a small sample training data enhancement method for radiation source signals based on multipath effects. First, then for different electromagnetic environments of the receiving system, different multipath models such as Rayleigh distribution and Rice distribution are used for multipath enhancement, then the start time of the signal is randomly truncated to enhance different signal start times, and finally random Gaussian white noise is superimposed to enhance different received powers of the signal. This method can solve the problem of insufficient generalization in the training of deep learning networks due to insufficient sample size in the actual application process of radiation source signals.
[0031] Assume that the radiation source signal received by the receiving system is represented as s(m) (m = 1, 2,..., M). The sampling interval is dt, and the carrier frequency of the signal is RF.
[0032] Step 1: Perform multipath effect processing on the signal. Assume that there are a total of K transmission paths for the radiation source signal, and the amplitude attenuation and delay of each transmission path are A k and Delayk Let the signal on the k-th transmission path be denoted as s k (m). Then the delay Delay k The corresponding number of delay sampling points is where represents rounding down. The delay Delay k The corresponding residual phase shift is: Phi k = 2π·RF·(Delay k - D·dt). Then
[0033] s k (m) = A k ·s(m - D k )·exp(-j·Phi k )
[0034] = A k ·s(m - D k )·exp[-j·2π·RF·(Delay k - D·dt)]
[0035] For a total of K multipaths, the data enhanced signal after the multipath effect is
[0036]
[0037] The multipath effect is processed using the Rayleigh distribution multipath effect model or the Rice distribution multipath effect model.
[0038] Step 2: Randomly truncate the start time of the signal for data enhancement with different sampling start times. Assume the maximum length of the random truncation is P, and the start time after random truncation is p, where p ∈ [1, P] and p is uniformly distributed. Then let the signal after randomly truncating the start time be s 2 (m). Then s 2 (m) = s 1 (m - k + 1), (m = 1, 2, …, M - P + 1). Here, k represents the number of sampling points for truncation.
[0039] Step 3: Add random Gaussian white noise to the signal for data enhancement with different received signal-to-noise ratios. Assume the maximum amplitude of the added Gaussian white noise is A max , and the amplitude of the added Gaussian white noise is a, where a ∈ [0, A max and a is uniformly distributed. Then let the signal after randomly adding white noise be s 3 (m). Then s 3 (m) = s 2(m) + a·[rand(1, M - k + 1) + j·rand(1, M - k + 1)], where rand(1, *) is a Gaussian random function.
[0040] Step 4: Perform multiple random samplings on all received radiation source sample signals according to Steps 1 to 3. Each signal sample is randomly sampled 100 times, which is equivalent to expanding the sample size by 100 times, obtaining a large number of diverse samples with random multipath, random sampling start times, and random signal-to-noise ratios, thereby solving the problem of insufficient generalization of deep learning network training due to insufficient sample size in the actual application process of radiation source signals.
[0041] The present invention generates a large number of radiation source signal samples with different characteristics by simulating conditions such as random multipath effects, random sampling start times, and random signal-to-noise ratios. These diverse samples can cover various complex situations that radiation source signals may encounter in the actual application process, enabling the deep learning network to expose to a wider range of signal characteristics during training, thereby improving the adaptability and generalization ability of the deep learning network model to unknown radiation source signal data.
[0042] Embodiment 2
[0043] As Figure 1 shown, based on Embodiment 1, this embodiment provides a more refined implementation method.
[0044] Assume that the radiation source signal received by the receiving system is represented as s(m) (m = 1, 2,..., M). The sampling interval is dt, and the carrier frequency of the signal is RF = 3 GHz.
[0045] Step 1: Perform multipath effect processing on the signal. Assume that there are a total of K transmission paths for the radiation source signal, and the relative amplitude and delay of each path are A k and Delay k . Let the signal of the k-th multipath be represented as s k (m), then the number of delayed sampling points corresponding to the delay Delay k is where represents rounding down. The residual phase shift amount corresponding to the delay Delay k is: Phi k = 2π·RF·(Delay k - D·dt). Then
[0046] s k (m) = A k ·s(m - D k )·exp(-j·Phi k )
[0047] = Ak ·s(m - D k )·exp[-j·2π·RF·(Delay k - D·dt)]
[0048] For a total of K multipaths, the data - enhanced signal after the multipath effect is
[0049]
[0050] The multipath effect is processed using a Rayleigh - distributed multipath effect model or a Rice - distributed multipath effect model.
[0051] Step 2: Randomly truncate the start time of the signal for data - enhanced different sampling start times. Assume that the maximum length of the random truncation is P = 50, and the start time after random truncation is p, where p ∈ [1, P] and p is uniformly distributed. Then let the signal after random truncation of the start time be s 2 (m), then s 2 (m)=s 1 (m - k + 1), (m = 1, 2, …, M - P + 1).
[0052] Step 3: Add random Gaussian white noise to the signal for data - enhanced different received signal - to - noise ratios. Assume that the maximum amplitude of the added Gaussian white noise is A max = 0.03, corresponding to the minimum signal - to - noise ratio of about 30 dB. The amplitude of the added Gaussian white noise is a, where a ∈ [0, A max and a is uniformly distributed. Then let the signal with randomly added white noise be s 3 (m), then s 3 (m)=s 2 (m)+a·[rand(1, M - k + 1)+j·rand(1, M - k + 1)], where rand(1, *) is a Gaussian random function.
[0053] Step 4: Randomly sample all received radiation source sample signals multiple times according to Steps 1 - 3 to obtain a large number of diverse samples with random multipaths, random sampling start times, and random signal - to - noise ratios, thereby solving the problem of insufficient generalization in the training of deep - learning networks due to insufficient sample size in the actual application process of radiation source signals.
[0054] As described above, the present invention can be preferably implemented.
[0055] All features disclosed in all embodiments in this specification, or all steps in all methods or processes implicitly disclosed, except for mutually exclusive features and / or steps, can be combined and / or extended, replaced in any manner.
[0056] The above are only the preferred embodiments of the present invention, and do not impose any form of limitation on the present invention. Based on the technical essence of the present invention, any simple modifications, equivalent replacements, and improvements made to the above embodiments within the spirit and principles of the present invention still fall within the protection scope of the technical solution of the present invention.
Claims
1. A method for enhancing radiation source signal sample training data, characterized in that: The following steps are involved: S1, multipath effect processing: multipath effect processing is performed on the radiation source signal received by the receiving system.
2. The method for enhancing the radiation source signal sample training data according to claim 1, characterized in that: In step S1, the expression of the signal after multipath effect processing is: Where s1(m) represents the signal after multipath effect processing, m represents the serial number of the signal sampling point, m = 1, 2, ..., M, M represents the length of the signal, s(m) represents the radiation source signal received by the receiving system, k represents the transmission path number, K represents the number of transmission paths, s k (m) represents the kth transmission path signal, A k Indicates the amplitude attenuation of each transmission path, Delay k Denotes the delay of each transmission path, D k Delay k The corresponding delayed sampling points, s(mD k ) represents the delayed signal, represents rounding down, dt represents the sampling interval, exp represents the exponential operation, j represents the imaginary unit, π represents the ratio of pi, and RF represents the carrier frequency of the radiation source signal received by the receiving system.
3. The method for enhancing the radiation source signal sample training data according to claim 2, characterized in that: In step S1, a Rayleigh distribution multipath effect model or a Rice distribution multipath effect model is used to perform multipath effect processing.
4. The method for enhancing the radiation source signal sample training data according to claim 2, characterized in that: The following steps are involved: S2, random truncation: Randomly truncate the starting time of the signal.
5. The method for enhancing the radiation source signal sample training data according to claim 4, characterized in that: In step S2, the expression for randomly truncating the signal at the start time of the signal is: s2(m)=s1(m-k+1) Wherein, s2(m) represents the signal after random truncation, m=1,2,…,M-k+1, and k represents the number of truncated sampling points.
6. The method for enhancing the radiation source signal sample training data according to claim 5, characterized in that: The following steps are involved: S3, noise superposition: adding noise to the signal.
7. The method for enhancing the radiation source signal sample training data according to claim 6, characterized in that: In step S3, random Gaussian white noise is added to the signal.
8. The method for enhancing the radiation source signal sample training data according to claim 7, characterized in that: In step S3, the signal expression after adding random Gaussian white noise is: s3(m)=s2(m)+a·[rand(1,M-k+1)+j·rand(1,M-k+1)] Among them, s3(m) represents the signal after adding random Gaussian white noise, a represents the amplitude of the added random Gaussian white noise, a∈[0,A max ], A max represents the maximum amplitude of Gaussian white noise, and rand(1,*) represents the Gaussian random function.
9. A radiation source signal sample training data enhancement method according to any one of claims 6 to 8, characterized in that: The following steps are involved: S4, multiple random sampling: Steps S1 to S3 are executed N times in a loop; wherein N≥2 and N is an integer.
10. A radiation source signal sample training data enhancement system, characterized in that: A method for enhancing radiation source signal sample training data according to any one of claims 1 to 9, comprising the following modules: Multipath effect processing module: used to perform multipath effect processing on the radiation source signal received by the receiving system.