Robust noise floor estimation method based on histogram

Abstract

The invention discloses a robust noise floor estimation method based on a histogram, and belongs to the field of electronic reconnaissance, and the method comprises the steps: firstly carrying out the preprocessing of a received radar signal, i.e., A / D sampling and FFT conversion; extracting the FFT data every a plurality of channels, and performing logarithm conversion and amplitude histogram statistics on the extracted channel data; then, carrying out smooth filtering processing to obtain an amplitude distribution envelope of the extracted channel, searching an envelope extreme point, and determining a noise floor estimation value of the extracted channel according to the found local minimum point; and finally, carrying out abnormal value detection, nonlinear conversion and linear interpolation on the noise floor estimation values of the extracted channels, and outputting noise floor estimation results of all channels. The method is insensitive to non-flat noise, missing detection and false alarm in a complex electromagnetic environment are effectively reduced, the calculation complexity is optimized while the detection accuracy is ensured, the data processing pressure of electronic reconnaissance equipment is greatly reduced, and the target signal detection and tracking efficiency is improved.

Description

TECHNICAL FIELD

[0001] The application belongs to the field of electronic reconnaissance, and particularly relates to a histogram-based robust noise floor estimation method. BACKGROUND

[0002] Signal detection is a key technology in non-cooperative communication, and a conventional signal detection method determines the existence of a signal by comparing the signal power and a fixed threshold. Since the electromagnetic environment faced by electronic reconnaissance is increasingly complex, in a non-stationary noise background, fixed threshold detection is prone to cause a large amount of missed detection or false alarm. Therefore, accurate estimation of a non-stationary noise floor is a key to channelization detection.

[0003] An actual channel usually has time-varying and dispersive properties, and is affected by the nonlinear characteristics of devices and other electromagnetic signals in the environment, so that the background noise actually output by a receiver is not ideal Gaussian white noise, and the noise energy difference can reach 5 dB or even more. The document 'Wideband signal detection method based on multi-scale morphological filtering' realizes estimation of the non-stationary background noise power density of a wideband signal based on iterative operation of multi-scale morphological filtering, but this method has a large amount of calculation and poor real-time performance in actual application. The document 'An effective signal detection method for a wideband digital reconnaissance receiver' extracts the minimum value of segmented spectrum data, and performs filtering processing by using a small structure element, and obtains a noise floor estimation value by interpolation. Although this method improves the calculation efficiency, the accuracy is low and affects the subsequent signal detection. Therefore, it is of great significance to study a robust, efficient and easy-to-implement noise floor estimation method for adapting to complex and variable actual reconnaissance environments and improving the reconnaissance performance of electronic countermeasure equipment. SUMMARY

[0004] The application provides a histogram-based robust noise floor estimation method, which realizes accurate and efficient detection and tracking of a target signal in a non-stationary background noise environment.

[0005] The technical solution of the application is as follows: a histogram-based robust noise floor estimation method, steps are as follows:

[0006] Step 1: performing N-point fast Fourier transform on received time domain data frame by frame to obtain FFT data, and proceeding to step 2.

[0007] Step 2: extracting FFT data every several channels to obtain M channel data {X1,...,X m ,...,X M}, M≤N, X mXm(m) represents the FFT result of the mth channel, m∈{1, 2, …, M}, and the dB conversion is performed on the extracted channel data {X1, …, X m ,…,X M} to obtain the channel data in dB, denoted as {X1, …, X m ,…,X M}| dB , and the process goes to step 3.

[0008] Step 3, the amplitude distribution range of the above-mentioned channel data in dB {X1, …, X m ,…,X M}| dB is 0-48 dB, the amplitude distribution range is equally divided into Q intervals as the horizontal coordinate axis of the histogram, the amplitude histogram statistics are performed on the M channel data respectively to obtain the amplitude histogram distribution sequence of the M channel data, and the k-point smoothing filtering is performed on the amplitude histogram distribution sequence of the M channel data to obtain the amplitude distribution envelope of the M channel data after smoothing filtering, and the process goes to step 4.

[0009] Step 4, the extreme value point of the data amplitude distribution envelope obtained in step 3 is found, the noise floor estimation value of the M channels is obtained, and the noise floor estimation values of the channels are output in the extraction order to obtain the noise floor estimation sequence in dB {Thr1, …, Thr m ,…,Thr M}| dB , Thr m represents the noise floor estimation value of the mth channel, and the process goes to step 5.

[0010] Step 5, the abnormal value detection is performed on the noise floor estimation sequence in dB {Thr1, …, Thr m ,…,Thr M}| dB obtained in step 4, and the noise floor estimation abnormal value of the current channel is replaced with the mean value of the noise floor estimation of the adjacent channels to obtain the noise floor estimation sequence after abnormal value correction, and the process goes to step 6.

[0011] Step 6, the non-linear conversion is performed on the noise floor estimation sequence after abnormal value correction obtained in step 5 to obtain the noise floor estimation sequence in amplitude, and the linear interpolation is performed to obtain the noise floor estimation values of other channels, and the noise floor estimation results of the N channels are output

[0012] Compared with the prior art, the present application has the following advantages:

[0013] 1) The present application uses FFT and amplitude histogram statistics to complete the accurate estimation of the noise floor, is not sensitive to non-flat noise, and effectively reduces the missed detection and false alarm in the complex electromagnetic environment.

[0014] 2) The channel data output by the FFT is logarithmically converted and the statistics are extracted, which greatly reduces the amount of calculation data, reduces the data processing pressure of the electronic reconnaissance equipment, and facilitates engineering implementation.

[0015] 3) Based on smoothing filtering and outlier detection, the problem of low resolution and accuracy of the traditional morphological noise floor estimation method is solved, and the detection probability of the target signal is mainly improved. BRIEF DESCRIPTION OF DRAWINGS

[0016] Figure 1 It is a flow chart of a robust noise floor estimation method based on a histogram.

[0017] Figure 2 It is a single channel data amplitude histogram, wherein (a) is a pure noise channel data amplitude histogram, (b) is a continuous wave signal channel data amplitude histogram, and (c) is a pulse signal channel data amplitude histogram.

[0018] Figure 3 It is an amplitude distribution envelope diagram after smoothing filtering, wherein (a) is an amplitude distribution envelope of a pure noise channel data amplitude histogram after smoothing filtering, (b) is an amplitude distribution envelope of a continuous wave signal channel data amplitude histogram after smoothing filtering, and (c) is an amplitude distribution envelope of a pulse signal channel data amplitude histogram after smoothing filtering.

[0019] Figure 4 It is a noise floor estimation sequence, wherein (a) is a noise floor estimation sequence obtained in step 4, (b) is a noise floor estimation sequence after outlier correction obtained in step 5, and (c) is a noise floor estimation result of N channels after linear interpolation obtained in step 6.

[0020] Figure 5 It is a curve diagram of the detection probability changing with the signal-to-noise ratio. DETAILED DESCRIPTION

[0021] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by a person of ordinary skill in the art without creative labor fall within the protection scope of the present application.

[0022] The technical solutions of various embodiments of the present application can be combined with each other, but it must be based on the fact that a person of ordinary skill in the art can realize it, and when the combination of technical solutions appears contradictory or unachievable, it should be considered that the combination of technical solutions does not exist, nor within the protection scope required by the present application.

[0023] The following section will further introduce the specific implementation method, as well as the technical difficulties and inventive points of this invention, using this design example as an example.

[0024] Combination Figure 1 The robust noise floor estimation method based on histograms described in this invention includes the following steps:

[0025] Step 1: Perform an N-point Fast Fourier Transform (FFT) on each frame of the received time-domain data to obtain FFT data, then proceed to Step 2.

[0026] Step 2: Extract FFT data every certain number of channels to obtain M channel data {X1,…,X...} m ,…,X M}, M≤N, X m Let f(x) represent the FFT result of the m-th channel, m∈{1,2,…,M}, and let f(x) represent the FFT result of the extracted channel data {X1,…,X}. m ,…,X M The data is converted to dB to obtain channel data in dB, denoted as {X1,…,X...}. m ,…,X M}| dB .

[0027] Among them, the extracted channel data {X1,…,X m ,…,X M The data is converted to dB to obtain channel data in dB, denoted as {X1,…,X...}. m ,…,X M}| dB Specifically, as follows: {X1,…,X m ,…,X M}| dB ={20log 10 |X1|,…,20log 10 |X m |,…,20log 10 |X M |}, where |·| represents the modulo operation.

[0028] Proceed to step 3.

[0029] Step 3: The main innovation of this invention lies in utilizing the amplitude distribution characteristics of signal and noise to estimate the noise floor by finding the poles of the amplitude histogram distribution. Since the above channel data {X1,…,X...} is in dB... m ,…,X M}| dBThe amplitude distribution range of the data is 0-48dB, the amplitude distribution range is equally divided into Q intervals as the histogram abscissa, the amplitude histogram of the M channel data is counted respectively, the amplitude histogram distribution sequence of the M channel data is obtained, and the amplitude histogram distribution sequence of the M channel data is k-point smoothing filtering, and the amplitude distribution envelope of the M channel data after smoothing filtering is obtained.

[0030] The amplitude histogram distribution sequence of the M channel data is k-point smoothing filtering, and the specific steps are as follows:

[0031] Step 31, input the amplitude histogram distribution sequence of the mth channel data, denoted as {h1, h2, …h q …,h Q}, interval number q=1, 2, …, Q, and initialize the smoothing filter window queue with length k.

[0032] Step 32, the amplitude histogram distribution sequence {h1, h2, …h q …,h Q} is added to the smoothing filter window queue in turn, and when the queue is full, the new sequence data is added to the window while the earliest added data is removed.

[0033] Step 33, calculate the mean value of the k elements in the window And as the smoothing filter output.

[0034] Step 34, repeat steps 32-33 until all sequence data is processed, and the k-point smoothing filtered amplitude distribution envelope

[0035] Go to step 4.

[0036] Step 4, find the extreme value point of the data amplitude distribution envelope obtained in step 3, and obtain the noise floor estimate value of the M channels, and output the noise floor estimate value of each channel according to the extraction order, and obtain the noise floor estimate sequence {Thr1, …, Thr m ,…,Thr M}| dB , Thr m represents the noise floor estimate value of the mth channel.

[0037] Wherein, finding the extreme value point of the data amplitude distribution envelope, obtaining the noise floor estimate value of the M channels, including the following steps:

[0038] Step 41, for the mth channel, the amplitude distribution envelope after k-point smoothing filtering Traverse to find the local maximum value point as the starting point, denoted as Satisfies:

[0039]

[0040] Step 42, from the starting point Start searching for the first local minimum point satisfy:

[0041]

[0042] Local minimum points Let the noise floor estimate of the m-th channel be...

[0043] Proceed to step 5.

[0044] Step 5: The secondary innovation of this invention lies in eliminating the interference of continuous wave signals on the noise floor estimation through outlier detection, thereby ensuring the robustness and accuracy of the noise floor estimation. Specifically, this involves optimizing the noise floor estimation sequence {Thr1,…,Thr} obtained in Step 4 (in dB). m ,…,Thr M}| dB Outlier detection is performed, and the outlier noise floor estimate of the current channel is replaced with the mean noise floor estimate of adjacent channels to obtain the noise floor estimate sequence after outlier correction. The specific steps include:

[0045] Step 51: Traverse the noise floor estimation sequence {Thr1,…,Thr} of M channels in dB. m ,…,Thr M}| dB For Thr m Calculate the mean μ of adjacent channels. m and standard deviation σ m The specific calculation expression is as follows:

[0046]

[0047] Where δ represents the channel decimation interval.

[0048] Step 52: Use the mean μ m and standard deviation σ m The outlier detection threshold Δ = μ was calculated. m +κ·σ m κ is a constant.

[0049] Step 53, compare Thr m and threshold Δ, when Thr m When >Δ, let Thr m =μ m Conversely, retain the original Thr m The value of .

[0050] Go to Step 6.

[0051] Step 6, the present application reduces the data rate by decimation, facilitates hardware implementation, and first proposes to obtain the noise floor estimation results of all channels by linear interpolation, i.e. to perform nonlinear conversion on the noise floor estimation sequence obtained after the abnormal value correction in Step 5, to obtain the noise floor estimation sequence in amplitude units, and to perform linear interpolation to obtain the noise floor estimation values of other channels, and to output the noise floor estimation results of N channels

[0052] Example 1

[0053] In combination Figure 1 , the histogram-based robust noise floor estimation method has the following steps:

[0054] Step 1, A / D sampling is performed on the received radar signal to obtain L=500 frames of time domain data, and N=256 point fast Fourier transform (FFT) is performed to obtain FFT data, as shown in Figure 2 ;

[0055] Step 2, decimation is performed on the FFT data at an interval δ=4 channels, δ is the channel decimation interval, a total of M=64 channels of L frames of data {X1, X 1+δ ,…,X m ,…,X M} are decimated, and dB conversion is performed on the decimated channel data to obtain channel data in dB units {X1, X 1+δ ,…,X m ,…,X M}| dB , which is as follows:

[0056] {X1, X 1+δ ,…,X m ,…,X M}| dB

[0057] = {20log 10 |X1|,20log 10 |X 1+δ |,…,20log 10 |X m |,…,20log 10 |X M |}

[0058] Step 3, the channel data in dB units {X1, X 1+δ ,…,X m ,…,X M}| dBthe amplitude distribution range is 0-48dB, the amplitude distribution range is equally divided into Q intervals as the histogram abscissa, the amplitude histogram of M channel data is counted respectively, and the amplitude histogram distribution sequence of M channel data is obtained, including pure noise channel, continuous wave signal channel and pulse signal channel, as shown in Figure 2 .

[0059] The amplitude histogram distribution sequence of M channel data is respectively filtered by k=5 point smoothing, and the amplitude distribution envelope of M channel data after smoothing is obtained, including the following steps:

[0060] Step 31, input the amplitude histogram distribution sequence of the mth channel data, denoted as {h1, h2,…h q …,h Q}, initialize the smoothing filter window queue with length k;

[0061] Step 32, add the amplitude histogram distribution sequence {h1, h2,…h q …,h Q} to the smoothing filter window queue in turn, and when the queue is full, add the new sequence data to the window while removing the earliest added data.

[0062] Step 33, calculate the mean value of k elements in the window and take it as the smoothing filter output.

[0063] Step 34, repeat steps 32 to 33 until all sequence data is processed, and obtain the amplitude distribution envelope of the amplitude histogram of the mth channel data after k point smoothing

[0064] Step 35, repeat steps 31 to 34 to obtain the amplitude distribution envelope of M channel data after smoothing.

[0065] The amplitude distribution envelope of the data amplitude histogram of the pure noise channel, the continuous wave signal channel and the pulse signal channel after smoothing is shown in Figure 3 .

[0066] Step 4, find the extreme value point of the amplitude distribution envelope of the channel data after smoothing, obtain the noise floor estimation value corresponding to M channels, and output the noise floor estimation value of each channel according to the extraction order to obtain the noise floor estimation sequence in dB, including the following steps:

[0067] Step 41, for the amplitude distribution envelope of the mth channel after k point smoothing , traverse to find the local maximum value point as the starting point, denoted as There are the following three cases:

[0068] The current channel is pure noise, and there is a maximum value point in the amplitude distribution envelope sequence after smoothing filtering;

[0069] The current channel is a continuous wave signal, and there is a maximum value point in the amplitude distribution envelope sequence after smoothing filtering;

[0070] The current channel is a pulse signal, and there are two maximum value points in the amplitude distribution envelope sequence after smoothing filtering;

[0071] For pure noise and continuous wave signals, the unique maximum value point is directly recorded as the starting point. For pulse signals, since the signal amplitude is greater than the noise amplitude, the first maximum value point is noise, and the second maximum value point is signal. The first maximum value point is recorded as the starting point, and the starting point satisfies:

[0072]

[0073] Step 42, starting from the starting point find the first local minimum value point satisfies:

[0074]

[0075] The local minimum value point is taken as the noise floor estimate value of the mth channel, and let

[0076] Step 43, repeat steps 41 to 42 to obtain the noise floor estimate values of M channels, and output the noise floor estimate values of each channel in the extraction order to obtain the noise floor estimate sequence in dB {Thr1,…, Thr m ,…,Thr M}| dB .

[0077] Step 5, perform outlier detection on the noise floor estimate sequence in dB {Thr1,…, Thr m ,…,Thr M}| dB obtained in step 4, and replace the current channel noise floor estimate outlier with the mean value of the noise floor estimates of adjacent channels to obtain the noise floor estimate sequence after outlier correction. The specific steps include the following:

[0078] Step 51, traverse the noise floor estimate sequence in dB {Thr1,…, Thr m ,…,Thr M}| dB of M channels, and for Thr m , calculate the mean value μ of adjacent channelsm and standard deviation σ m The specific calculation expression is:

[0079]

[0080] Step 52, using the mean μ m and standard deviation σ m to calculate the outlier detection threshold Δ = μ m + κ· σ m .

[0081] Step 53, compare Thr m and threshold Δ, when Thr m > Δ, let Thr m = μ m , otherwise keep the original Thr m value, get the outlier corrected noise floor estimation sequence, denoted as denotes the outlier corrected noise floor estimation value of the mth channel.

[0082] Step 6, nonlinear conversion is performed on the outlier corrected noise floor estimation sequence obtained in step 5, and the unit is converted from dB to amplitude, as follows:

[0083]

[0084] obtain the noise floor estimation sequence in amplitude unit and perform linear interpolation, the linear interpolation formula is as follows:

[0085]

[0086] wherein, channel α, β ∈ {1, …, m, …, M}, and are the noise floor estimation values corresponding to the αth and βth channels, denotes the noise floor estimation value of the nth channel obtained by linear interpolation, α < n < β, n ∈ {1, 2, …, N}.

[0087] Insert δ-1 values between adjacent noise floor estimation values to obtain the noise floor estimation values of other channels except the extracted channels, and output the noise floor estimation results of a total of N channels

[0088] Figure 4(a), (b) and (c) are the noise floor estimation sequence obtained in step 4, the noise floor estimation sequence after outlier correction obtained in step 5 and the linear interpolation result of the noise floor estimation of N channels obtained in step 6, respectively. The linear interpolation result of the noise floor estimation of N channels obtained in step 6 is used as a threshold to perform channelization detection on 1000 frames of random data, the signal-to-noise ratio varies from -9dB to 0dB, and the false alarm rate is 10 -2 . The detection probability is shown in Figure 5 . It can be seen that when the signal-to-noise ratio is greater than -7dB, the detection probability can reach more than 90%, providing effective support for subsequent signal processing.

Claims

1. A histogram-based robust noise floor estimation method, characterized in that, The method comprises the following steps: Step 1, performing N-point fast Fourier transform on the received time domain data frame by frame to obtain FFT data, and transferring to step 2; Step 2, decimate the FFT data every several channels to obtain M channel data {X1,..., X m ,…,X M}, M ≤ N, X m represents the FFT result of the mth channel, m ∈ {1, 2,..., M}, and the decimated channel data {X1,..., X m ,…,X M} is converted to dB to obtain channel data in dB, denoted as {X1,..., X m ,…,X M}| dB , and go to Step 3; Step 3: The channel data {X1,…,X} above, in dB. m ,…,X M }| dB The amplitude distribution range is 0 to 48 dB. The amplitude distribution range is divided into Q intervals at equal intervals as the horizontal axis of the histogram. The amplitude histogram statistics are performed on the M channel data respectively to obtain the amplitude histogram distribution sequence of the M channel data. The amplitude histogram distribution sequence of the M channel data is then subjected to k-point smoothing filtering to obtain the amplitude distribution envelope of the M channel data after smoothing filtering. Proceed to step 4. Step 4, find the extreme points of the data amplitude distribution envelope obtained in Step 3, and obtain the noise floor estimates of the M channels, and output the noise floor estimates of the channels in the extraction order to obtain the noise floor estimate sequence {Thr1, Thr2,..., ThrM} in dB. m ,…,Thr M}| dB ,…,Thr m , which represents the noise floor estimate of the mth channel, and go to Step 5. Step 5, perform outlier detection on the noise floor estimate sequence {Thr1, Thr2,..., ThrN} obtained in step 4 in dB, and replace the outlier of the current channel noise floor estimate with the mean of the noise floor estimates of the adjacent channels, to obtain the outlier-corrected noise floor estimate sequence, and go to step 6. m Step 6, perform the following steps on the noise floor estimate sequence obtained in step 5: M Step 7, perform the following steps on the noise floor estimate sequence obtained in step 6: dB Step 8, perform the following steps on the noise floor estimate sequence obtained in step 7: Step 9, perform the following steps on the noise floor estimate sequence obtained in step 8: Step 10, perform the following steps on the noise floor estimate sequence obtained in step 9: Step 11, perform the following steps on the noise floor estimate sequence obtained in step 10: Step Step 6, the noise floor estimation sequence obtained by correcting the outliers in step 5 is subjected to a nonlinear conversion to obtain a noise floor estimation sequence in amplitude units, and linear interpolation is performed to obtain noise floor estimation values of other channels, and noise floor estimation results of N channels are output 2. The histogram-based robust noise floor estimation method of claim 1, wherein, In step 2, the extracted channel data {X1,..., X m ,…,X M} is converted into dB, and the channel data in dB is denoted as {X1,..., X m ,…,X M}| dB , and the conversion is as follows: {X1,...,X m ,...,X M}| dB = {20log 10 |X1|,...,20log 10 |X m |,...,20log 10 |X M |}, Wherein, |·| represents the modulo operation.

3. The histogram-based robust noise floor estimation method of claim 2, wherein, In step 3, the amplitude histogram distribution sequence of the M channel data is k-point smoothing filtering, specifically as follows: Step 31, input the amplitude histogram distribution sequence of the mth channel data, denoted as {h1, h2, … h q …,h Q}, interval sequence number q = 1, 2, …, Q, initialize the smoothing filter window queue with length k; Step 32, the sequence of amplitude histogram distribution {h1, h2,... h q} is added into the smoothing filter window queue, when the queue is full, the newest sequence data is added into the window while the earliest added data is removed from the window. Q ​ Step 33, compute the mean of the k elements in the window and as a smoothing filter output; Step 34, repeat step 32-step 33 until all sequence data is processed to obtain the k-point smoothed amplitude distribution envelope 4. The histogram-based robust noise floor estimation method of claim 3, wherein, In step 4, the extreme points of the data amplitude distribution envelope are found to obtain the noise floor estimation value of the M channels, specifically including the following steps: Step 41, the amplitude distribution envelope of the mth channel after k-point smoothing filtering Traverse Find the local maximum point as the starting point, denoted as Satisfies: Step 42, from the starting point Start looking for the first local minimum point Satisfies: The local minimum point is found Let the noise floor estimate of the mth channel be 5. The histogram-based robust noise floor estimation method of claim 4, wherein, In step 5, the noise floor estimation sequence in dB obtained in step 4 is subjected to outlier detection, and the noise floor estimation mean of the adjacent channels is used to replace the current channel noise floor estimation outlier, specifically including the following steps: Step 51: Traverse the noise floor estimation sequence {Thr1,…,Thr} of M channels in dB. m ,…,Thr M }| dB For Thr m Calculate the mean μ of adjacent channels. m and standard deviation σ m The specific calculation expression is as follows: Wherein, δ represents the channel extraction interval; Step 52, compute the threshold for anomaly detection Δ = μ + κ · σ m , where μ is the mean value and σ is the standard deviation m Step 53, compare the value of the feature with the threshold Δ m Step 54, if the value of the feature is greater than Δ, then the feature is an anomaly m Step 55, if the value of the feature is less than Δ, then the feature is not an anomaly Step 53, compare Thr m to a threshold delta, when Thr m > delta, set Thr m = mu m , otherwise keep the original Thr m value.

Images