Wireless signal identification system training sample simulation method

Abstract

The invention provides a training sample simulation method for a wireless signal identification system. The method comprises the following steps: carrying out baseband signal modeling based on frame elements; generating an intermediate frequency signal based on a random interference variable; sample generation based on superposition of a plurality of intermediate frequency signals. According to the invention, a baseband signal is constructed in a modular mode based on frame elements, samples are endowed with sufficient diversity by using uniformly distributed interference variables, and superposition of intermediate frequency signals of the same type and different types can be realized at the same time. The method is based on a standard sample simulation process, simulation signals of any number can be generated, the requirements of a training set of an identification system for sample diversity and multiple signals contained in a single sample are met, and direct data support can be provided for intelligent wireless signal processing.

Description

Technical Field

[0001] The invention belongs to the field of data engineering, and specifically designs a training sample simulation method for a wireless signal recognition system. Background Art

[0002] The signal recognition system based on deep learning requires a large number of high-quality training samples, but due to the limitations of actual acquisition, it is impossible to provide samples of sufficient quantity and quality. On the other hand, wireless signals are naturally generated by computers and transmitted through intermediate frequency, radio frequency devices and antennas, and propagated through channels. Therefore, the actual collected signals are homologous to the computer simulated signals. Therefore, it is necessary and feasible to use computer simulation methods to generate wireless signal samples.

[0003] China Telecom Co., Ltd. disclosed a wireless signal simulation method, device, electronic device and medium, which generates iso-strength lines and iso-strength surfaces based on the signal strength of each position point in the test data of the antenna to be tested, and simulates the wireless signal; the Bengbu Tank Academy of the Chinese People's Liberation Army disclosed a radio station simulation communication electromagnetic environment simulation device, which has a noise and radiation simulation signal generation module, a signal synthesis module, a user input and display module, and a control module, which can realize that the voice signal output by the actual radio station is processed by the simulated electromagnetic environment simulation and output to the audio terminal; the National University of Defense Technology of the Chinese People's Liberation Army disclosed a radio signal test simulation system, including: an arbitrary waveform module, which is used to configure the signal generator virtual instrument to generate different types of type of simulation signal, and display the time domain waveform of the simulation signal; a closed-field path module, used to configure various types of RF connectors, RF connecting lines and attenuators to simulate the closed-field path; a spectrum analysis module, used to configure a spectrum analyzer virtual instrument to perform spectrum analysis on the simulation signal, and display the results of the spectrum analysis; the Chinese People's Liberation Army has disclosed a general information processing platform suitable for wireless simulation equipment in various frequency bands, which uses hardware modules to build a public platform and replaces hardware with software, focusing on simulating the functions and operations of various types of wireless communication equipment in actual applications with software instead of hardware, and using Ethernet technology to achieve system upgrades of wireless simulation equipment connected to the network on a terminal device.

[0004] However, the simulated signal generated by the above-mentioned wireless signal simulation system cannot directly support the training of the signal recognition system based on deep learning. On the one hand, the lack of randomness of the variables in the simulated signal leads to the diversity of samples that cannot meet the training requirements. On the other hand, the simulated signal is a sampling signal at a single intermediate frequency, which does not match the mode of the signal received by the recognition system (wideband sampling signal). Summary of the invention

[0005] In view of the problems existing in the prior art, the present invention provides a wireless signal recognition system training sample simulation method, which specifically includes the following steps:

[0006] Step 1: Baseband signal modeling based on frame elements;

[0007] Frame elements include start flag, address information, data field, time slot structure, and end flag;

[0008] The baseband signal of the frame element is assembled based on the frame element;

[0009] Step 1: Determine the baseband signal frame type. The baseband signal frame type is divided into continuous signal, periodic frame signal and non-periodic frame signal. Determine the frame elements contained, including frame header, synchronization header, special code, time slot interval, data segment and frame tail;

[0010] Step 2: Define frame elements and determine parameters such as modulation type, symbol rate, coding type, and information sequence construction;

[0011] Step 3: Determine the timing of frame elements, assemble frame elements, and construct the basic frame structure;

[0012] Step 4: If the baseband signal is a continuous signal, use one frame structure to construct the baseband signal; if the baseband signal is a periodic frame signal, use three or more frame structures to construct the baseband signal, and the frame interval is a fixed time slot; if the baseband signal is a non-periodic frame signal, use three or more frame structures to construct the baseband signal, and the frame interval is a random time slot;

[0013] Step 2: Generation of intermediate frequency signal based on random interference variables;

[0014] The parameters involved in the IF signal simulation include signal appearance time, signal duration, signal average power, noise standard deviation, signal center frequency, and signal carrier phase. Since the IF signal parameters are independent of the signal type, the above parameters are interference parameters in the identification system. For the observation time T d For the intermediate frequency signal, let the following interference variables obey uniform distribution:

[0015] (1) Signal appearance time T start :T start ~U[-T d ,T d ]

[0016] (2) Signal duration T c :T c ~U[0,2T d ]

[0017] (3) Average signal power P: P~U[P min ,Pmax ]

[0018] (4) Noise standard deviation σ N :σ N ~U[0,σ Nmax ]

[0019] (5) Signal center frequency f c :f c ~U[0,f s / 2)

[0020] (6) Signal carrier phase Among them, ~U[·] represents that the random variable follows a uniform distribution, P min and P max represents the possible maximum and minimum values ​​of the average power of the simulated signal, σ Nmax represents the maximum possible value of the noise standard deviation, f s is the signal sampling frequency; adding channels and interference factors are not considered, and only the noise type is Gaussian white noise;

[0021] The generation of the intermediate frequency signal is a random process. Let the baseband signal generated in step 1 be s(t). The specific process is as follows:

[0022] Step 1: Randomization of signal occurrence time and duration;

[0023] In order to ensure the randomness of the appearance time and duration of the intermediate frequency signal, two interceptions are performed:

[0024] First time: Randomly intercept the length of T c The signal segment s(t|T c ), fill in the front and back zeros to construct the signal sequence s1(t) of the intermediate process:

[0025]

[0026] Second: Randomly intercept s1(t) to form an observation time of T d The randomized signal sequence s2(t) of the signal occurrence time and duration;

[0027] Step 2: Randomization of signal average power;

[0028] Normalize s2(t) by dividing it by the standard deviation and multiplying it by the amplitude gain to get the random power signal s3(t):

[0029]

[0030] Among them, std() represents the standard deviation of the signal;

[0031] Step 3: Randomization of noise variance;

[0032] The random power signal s3(t) is superimposed with Gaussian white noise of random standard deviation to obtain the noisy intermediate frequency signal s4(t):

[0033] s4(t)=s3(t)+σ N *randn

[0034] In the formula, randn represents a random variable with a value range of [0,1], a mean of 0.5, and a variance of 1 that follows a normal distribution;

[0035] Step 4: Randomization of signal center frequency and carrier phase;

[0036] The noisy intermediate frequency signal s4(t) is converted into an intermediate frequency signal s5(t) by setting a random center frequency and carrier phase:

[0037]

[0038] The above process is described using the joint probability density function:

[0039]

[0040] Since the above random variables are unrelated, according to the uniform distribution of the above variables, we have:

[0041]

[0042] In the formula, f(T start )、f(T c )、f(P)、f(σ N )、f(f c ), Represents the variable T start , T c , P, σ N 、f c , The probability density function of

[0043] Step 3: Sample generation based on the superposition of multiple intermediate frequency signals;

[0044] Define the intermediate frequency signal generated by one round of operation in step 2 as Where I represents the number of intermediate frequency signals within the sampling bandwidth; multiple intermediate frequency signals are superimposed to obtain the simulated sample S(t):

[0045]

[0046] Determine whether the index i of the signal reaches the set signal quantity I. If so, execute sample output and output S(t); if not, determine whether the next signal is of the same type as the current signal; if so, jump to step 2, if not, jump to step 1.

[0047] In one embodiment of the present invention, in step 1, for the Link-4A baseband signal in the CAINS mode, the following steps are specifically included:

[0048] Step 1: Determine that the Link-4A baseband signal in CAINS mode is a periodic frame signal, which contains five frame elements: synchronization pulse string, protection interval, start bit, data bit, and non-keying signal;

[0049] Step 2: In CAINS mode, the Link-4A frame elements are defined as follows: the modulation type is 2FSK modulation; the frequency interval is 40kHz, the symbol 0 is a negative frequency deviation of -20kHz, and the symbol 1 is a positive frequency deviation of 20kHz; the symbol rate is 10kBd, and the pulse train is synchronized; 5kBd guard interval, start bit, data bit, unkeyed signal, the coding type is even parity, and the information sequence structure is shown in Table 1;

[0050] Table 1 Link-4A (CAINS mode) frame element information sequence structure

[0051]

[0052]

[0053] Step 3: Determine that the basic frame structure of Link-4A in CAINS mode is arranged in the order of synchronization pulse train, protection interval, start bit, data bit and non-keying signal;

[0054] Step 4: In CAINS mode, the Link-4A baseband signal is a periodic frame signal. The baseband signal is constructed using more than three frame structures, and the frame interval is a fixed time slot of 2ms.

[0055] The present invention uses computer simulation to realize data set generation in the field of deep learning, which can make up for the problems of small number and low quality of actually collected signal samples.

[0056] The present invention has the following advantages:

[0057] (1) The baseband signal generation based on frame elements complies with the standard computer simulation process, giving signal generation more freedom and scalability;

[0058] (2) From the perspective of statistical distribution, the randomness of interference variables in the sample is introduced, which greatly expands the diversity of the sample;

[0059] (3) The samples are generated by superimposing multiple intermediate frequency signals, which conforms to the signal receiving mode of the actual recognition system and supports the training of multi-signal recognition modalities of the signal recognition system based on deep learning. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] Figure 1 The baseband signal modeling process based on frame elements is shown;

[0061] Figure 2 Shows the construction of the basic frame structure of Link-4A (CAINS mode);

[0062] Figure 3 Shows the construction of Link-4A (CAINS mode) baseband signal;

[0063] Figure 4 The realization process of the intermediate frequency signal considering the statistical distribution of the interference variables is shown;

[0064] Figure 5 A sample example showing the superposition of multiple intermediate frequency signals. DETAILED DESCRIPTION

[0065] The present invention will be described in detail below with reference to the accompanying drawings.

[0066] The present invention proposes a wireless signal recognition system training sample simulation method, which specifically includes the following steps:

[0067] Step 1: Baseband signal modeling based on frame elements;

[0068] A frame element refers to the smallest constituent unit of a baseband signal that has a single information sequence structure, coding type, modulation type, symbol rate and other parameters, and has relatively independent functions and structures, including a start flag, address information, data field, time slot structure, end flag, etc.

[0069] The baseband signal modeling method of frame elements adopts a modular design, which can flexibly set the combination mode and order of frame elements, and present the combination of frame elements in a form that is easy to interact and customize, so as to construct any required baseband signal.

[0070] The baseband signal is assembled based on frame elements, the process is as follows Figure 1 The following takes Link-4A (CAINS mode) baseband signal generation as an example to illustrate the specific process:

[0071] Step 1: Determine the baseband signal frame type. The baseband signal frame type can usually be divided into continuous signal, periodic frame signal and non-periodic frame signal. At the same time, determine the frame elements contained, including frame header, synchronization header, special code, time slot interval, data segment, frame tail, etc.

[0072] For example, the Link-4A (CAINS mode) baseband signal is a periodic frame signal, which includes five frame elements: synchronization pulse string, protection interval, start bit, data bit, and non-keying signal.

[0073] Step 2: Define frame elements and determine parameters such as modulation type, symbol rate, coding type, and information sequence construction;

[0074] For example, the Link-4A (CAINS mode) frame elements are defined as follows: the modulation type is 2FSK modulation, the frequency interval is 40kHz (symbol 0 is a negative frequency deviation of -20kHz, and symbol 1 is a positive frequency deviation of 20kHz), the symbol rate is 10kBd (synchronous pulse train) and 5kBd (guard interval, start bit, data bit, unkeyed signal), the coding type is even parity, and the information sequence structure is shown in Table 1.

[0075] Table 1 Link-4A (CAINS mode) frame element information sequence structure

[0076] Information Type Number of bits (time slot) Code Type Overall length 78bit(14ms) Synchronous pulse train 16bit(1.6ms) 0101010101010101 Guard Interval 4bit(0.8ms) 0000 Start bit 1 bit (0.2 ms) 1 Data bits 56bit(11.2ms) Random pattern No keying signal 1 bit (0.2 ms) 0

[0077] Step 3: Determine the timing of frame elements, assemble frame elements, and construct the basic frame structure;

[0078] For example, the basic frame structure of Link-4A (CAINS mode) is arranged in the order of synchronization pulse string, protection interval, start bit, data bit and non-keying signal, such as Figure 2 shown.

[0079] Step 4: If the baseband signal is a continuous signal, use 1 frame structure to construct the baseband signal; if the baseband signal is a periodic frame signal, use 3 or more frame structures to construct the baseband signal, and the frame interval is a fixed time slot; if the baseband signal is a non-periodic frame signal, use 3 or more frame structures to construct the baseband signal, and the frame interval is a random time slot.

[0080] For example, the Link-4A (CAINS mode) baseband signal is a periodic frame signal, which uses more than three frame structures to construct the baseband signal, and the frame interval is a fixed time slot of 2ms, such as Figure 3 shown.

[0081] Step 2: Generation of intermediate frequency signal based on random interference variables;

[0082] Modulation type, symbol rate, frame interval and other baseband signal parameters are sensitive parameters of the signal recognition system (parameters related to the signal type). The parameters involved in the intermediate frequency signal simulation include signal appearance time, signal duration, signal average power, noise standard deviation, signal center frequency and signal carrier phase. Since the intermediate frequency signal parameters are independent of the signal type, the above parameters are interference parameters in the recognition system. For this reason, the parameters in the baseband signal modeling process are set to fixed values. In the intermediate frequency signal implementation, the interference factors are considered to be random variables that conform to a certain statistical distribution. During training, the recognition system learns this distribution from the samples and is insensitive to samples that meet this distribution. For the observation time T d For the intermediate frequency signal, the following interference variables can be made to obey uniform distribution:

[0083] (1) Signal appearance time T start :T start ~U[-T d ,T d ]

[0084] (2) Signal duration T c :T c ~U[0,2T d ]

[0085] (3) Average signal power P: P~U[P min ,P max ]

[0086] (4) Noise standard deviation σ N :σ N ~U[0,σ Nmax ]

[0087] (5) Signal center frequency f c :f c ~U[0,f s / 2)

[0088] (6) Signal carrier phase Among them, ~U[·] represents that the random variable follows a uniform distribution, P min and P max represents the possible maximum and minimum values ​​of the average power of the simulated signal, σ Nmax represents the maximum possible value of the noise standard deviation, f s It should be noted that, since the channel and interference are relatively complicated, in order to simplify the description, the present invention does not consider adding channel and interference factors, and only considers the noise type as Gaussian white noise.

[0089] Therefore, the generation of the intermediate frequency signal is defined as a random process, which is achieved by adding uniformly distributed interference variables on the basis of the baseband signal. Let the baseband signal generated in step 1 be s(t). The specific process is as follows:

[0090] Step 1: Randomization of signal occurrence time and duration;

[0091] In order to ensure the randomness of the appearance time and duration of the intermediate frequency signal, two interceptions are required:

[0092] First time: Randomly intercept the length of T c The signal segment s(t|T c ), fill in the front and back zeros to construct the signal sequence s1(t) of the intermediate process:

[0093]

[0094] Second: Randomly intercept s1(t) to form an observation time of T d A randomized signal sequence s2(t) with signal occurrence time and duration.

[0095] Step 2: Randomization of signal average power;

[0096] The signal samples in the training set should have a random average power, that is, s2(t) is normalized by dividing it by the standard deviation and then multiplied by the amplitude gain to obtain the random power signal s3(t):

[0097]

[0098] Among them, std() represents the standard deviation of the signal.

[0099] Step 3: Randomization of noise variance;

[0100] The random power signal s3(t) is superimposed with Gaussian white noise of random standard deviation to obtain the noisy intermediate frequency signal s4(t):

[0101] s4(t)=s3(t)+σ N *randn

[0102] In the formula, randn represents a random variable with a value range of [0,1], a mean of 0.5, and a variance of 1 that follows a normal distribution.

[0103] Step 4: Randomization of signal center frequency and carrier phase;

[0104] The noisy intermediate frequency signal s4(t) still belongs to the category of baseband signal. It is converted into the intermediate frequency signal s5(t) by setting a random center frequency and carrier phase:

[0105]

[0106] The above process can be described using the joint probability density function:

[0107]

[0108] Since the above random variables are unrelated, according to the uniform distribution of the above variables, we have:

[0109]

[0110] In the formula, f(T start )、f(T c )、f(P)、f(σ N )、f(f c ), Represents the variable T start , T c , P, σ N 、f c , The probability density function of .

[0111] Step 3: Sample generation based on the superposition of multiple intermediate frequency signals;

[0112] It is worth noting that the intermediate frequency signal s5(t) is a sampled signal at a single frequency. However, the signal receiving bandwidth in the recognition system is usually (much) larger than the signal bandwidth, and there is a high probability that there is more than one intermediate frequency signal in the received signal, such as Figure 4 In particular, cluster communication signals usually have multiple intermediate frequency signals on the sampling spectrum, so sample generation needs to consider the superposition of multiple intermediate frequency signals.

[0113] Define the intermediate frequency signal generated by one round of operation in step 2 as Where I represents the number of intermediate frequency signals within the sampling bandwidth. Multiple intermediate frequency signals are superimposed to obtain the simulated sample S(t):

[0114]

[0115] Next, determine whether the signal index i reaches the set signal number I. If so, execute sample output and output S(t); if not, determine whether the next signal is of the same type as the current signal; if so, jump to step 2, if not, jump to step 1, the flowchart is as follows: Figure 5 shown.

Claims

1. A wireless signal recognition system training sample simulation method, characterized in that: The specific steps include: Step 1: Baseband signal modeling based on frame elements; Frame elements include start flag, address information, data field, time slot structure, and end flag; The baseband signal of the frame element is assembled based on the frame element; Step 1: Determine the baseband signal frame type. The baseband signal frame type is divided into continuous signal, periodic frame signal and non-periodic frame signal. Determine the frame elements contained, including frame header, synchronization header, special code, time slot interval, data segment and frame tail; Step 2: Define frame elements and determine parameters such as modulation type, symbol rate, coding type, and information sequence construction; Step 3: Determine the timing of frame elements, assemble frame elements, and construct the basic frame structure; Step 4: If the baseband signal is a continuous signal, use one frame structure to construct the baseband signal; if the baseband signal is a periodic frame signal, use three or more frame structures to construct the baseband signal, and the frame interval is a fixed time slot; if the baseband signal is a non-periodic frame signal, use three or more frame structures to construct the baseband signal, and the frame interval is a random time slot; Step 2: Generation of intermediate frequency signal based on random interference variables; The parameters involved in the IF signal simulation include signal appearance time, signal duration, signal average power, noise standard deviation, signal center frequency, and signal carrier phase. Since the IF signal parameters are independent of the signal type, the above parameters are interference parameters in the identification system. For the observation time T d For the intermediate frequency signal, let the following interference variables obey uniform distribution: (1) Signal appearance time T start :T start ~U[-T d ,T d ] (2) Signal duration T c :T c ~U[0,2T d ] (3) Average signal power P: P~U[P min ,P max ] (4) Noise standard deviation σ N :σ N ~U[0,σ Nmax ] (5) Signal center frequency f c :f c ~U[0,f s / 2) (6) Signal carrier phase Among them, ~U[·] represents that the random variable follows a uniform distribution, P min and P max represents the possible maximum and minimum values ​​of the average power of the simulated signal, σ Nmax represents the maximum possible value of the noise standard deviation, f s is the signal sampling frequency; adding channels and interference factors are not considered, and only the noise type is Gaussian white noise; The generation of the intermediate frequency signal is a random process. Let the baseband signal generated in step 1 be s(t). The specific process is as follows: Step 1: Randomization of signal occurrence time and duration; In order to ensure the randomness of the appearance time and duration of the intermediate frequency signal, two interceptions are performed: First time: Randomly intercept the length of T c The signal segment s(t|T c ), fill in the front and back zeros to construct the signal sequence s1(t) of the intermediate process: Second: Randomly intercept s1(t) to form an observation time of T d The randomized signal sequence s2(t) of the signal occurrence time and duration; Step 2: Randomization of signal average power; Normalize s2(t) by dividing it by the standard deviation and multiplying it by the amplitude gain to get the random power signal s3(t): Among them, std() represents the standard deviation of the signal; Step 3: Randomization of noise variance; The random power signal s3(t) is superimposed with Gaussian white noise of random standard deviation to obtain the noisy intermediate frequency signal s4(t): s4(t)=s3(t)+σ N *randn In the formula, randn represents a random variable with a value range of [0,1], a mean of 0.5, and a variance of 1 that follows a normal distribution; Step 4: Randomization of signal center frequency and carrier phase; The noisy intermediate frequency signal s4(t) is converted into an intermediate frequency signal s5(t) by setting a random center frequency and carrier phase: The above process is described using the joint probability density function: Since the above random variables are unrelated, according to the uniform distribution of the above variables, we have: In the formula, f(T start )、f(T c )、f(P)、f(σ N )、f(f c ), Represents the variable T start , T c , P, σ N 、f c , The probability density function of Step 3: Sample generation based on the superposition of multiple intermediate frequency signals; Define the intermediate frequency signal generated by one round of operation in step 2 as i=1,2,...I, where I represents the number of intermediate frequency signals within the sampling bandwidth; multiple intermediate frequency signals are superimposed to obtain the simulated sample S(t): Determine whether the index i of the signal reaches the set signal quantity I. If so, execute sample output and output S(t); if not, determine whether the next signal is of the same type as the current signal; if so, jump to step 2, if not, jump to step 1.

2. The wireless signal recognition system training sample simulation method according to claim 1, characterized in that: In step 1, for the Link-4A baseband signal in CAINS mode, the following steps are specifically included: Step 1: Determine that the Link-4A baseband signal in CAINS mode is a periodic frame signal, which contains five frame elements: synchronization pulse string, protection interval, start bit, data bit, and non-keying signal; Step 2: In CAINS mode, the Link-4A frame elements are defined as follows: the modulation type is 2FSK modulation; the frequency interval is 40kHz, the symbol 0 is a negative frequency deviation of -20kHz, and the symbol 1 is a positive frequency deviation of 20kHz; the symbol rate is 10kBd, and the pulse train is synchronized; 5kBd guard interval, start bit, data bit, unkeyed signal, the coding type is even parity, and the information sequence structure is shown in Table 1; Table 1 Link-4A (CAINS mode) frame element information sequence structure Step 3: Determine that the basic frame structure of Link-4A in CAINS mode is arranged in the order of synchronization pulse train, protection interval, start bit, data bit and non-keying signal; Step 4: In CAINS mode, the Link-4A baseband signal is a periodic frame signal. The baseband signal is constructed using more than three frame structures, and the frame interval is a fixed time slot of 2ms.

Images