CN112929073A

CN112929073A - Method and device for constructing inter-satellite link frequency spectrum cognition machine learning training data set

Info

Publication number: CN112929073A
Application number: CN202110097903.6A
Authority: CN
Inventors: 陈建云; 瞿智; 冯旭哲; 张永刚; 张超; 胡梅; 周超; 王鼎
Original assignee: National University of Defense Technology
Current assignee: National University of Defense Technology
Priority date: 2021-01-25
Filing date: 2021-01-25
Publication date: 2021-06-08
Anticipated expiration: 2041-01-25
Also published as: CN112929073B

Abstract

The application relates to a method and a device for constructing a spectrum cognition machine learning training data set of an inter-satellite link. The method comprises the following steps: establishing requirements according to a preset data set, and setting system parameters; traversing the modulation mode set in the system parameter, generating a baseband signal according to the modulation mode and the system parameter, and generating a modulation signal according to the baseband signal; acquiring a pre-generated inter-satellite channel model, and adding channel information corresponding to the inter-satellite channel model to a modulation signal according to system parameters to obtain a signal set; and packaging the signal sequences in the signal set into a training data set with a preset format. By adopting the method, the inter-satellite link spectrum cognitive machine learning training data set can be effectively generated.

Description

Method and device for constructing inter-satellite link frequency spectrum cognition machine learning training data set

Technical Field

The application relates to the technical field of inter-satellite communication spectrum cognition, in particular to a method and a device for constructing an inter-satellite link spectrum cognition machine learning training data set.

Background

In recent years, with the booming of various commercial space companies and the great adjustment of military space development strategy in the large country around the world, the role of the inter-satellite link technology in various space plans is more and more important. In the commercial space aspect, a lot of satellite communication companies represented by space exploration technology (SpaceX) are creating a lot of business plans for realizing low-orbit constellation networks by means of inter-satellite link technology. The items such as Iridium second generation (Iridium NEXT), "star chain (Starlink)," Telesat, etc. gradually fall on the ground, and a huge number of thousands of constellations are pushed to gradually change from fantasy to reality. In the aspect of military aerospace, the aerospace major country including the United states starts a series of plans for constructing intelligent, comprehensive and low-cost satellite networks, and attempts to acquire the next generation information right advantage by constructing an integrated space network. The use of inter-satellite link technology to construct a heaven-earth integrated network is becoming the development trend of the next generation of spatial communication systems, and the large-scale application of inter-satellite links will become an important factor restricting the development of spatial networks.

According to the carrier frequency division, the inter-satellite link can be divided into a laser link, a microwave link and a millimeter wave link, and the laser link has higher requirements on the aiming, capturing and tracking technology of the inter-satellite signals due to the characteristics of extremely narrow beam and weak interference resistance of an optical system, and is not used much at the present stage. The technology of microwave and millimeter wave in the inter-satellite communication is more mature, and is the mainstream frequency band used by the current inter-satellite link, but the total amount of frequency spectrum resources divided by the frequency band for the inter-satellite communication is insufficient, and the frequency band is the main factor restricting the large-scale application of the microwave and millimeter wave inter-satellite link technology at present.

As is well known, the use and management of electromagnetic spectrum resources in the industry at present mainly depend on the guarantee of various frequency management systems, and the contradiction between supply and demand of the spectrum resources is coordinated through an application-allocation-use flow. The static allocation frequency management mode is simple and effective, an authorized user can enjoy exclusive right to the allocated electromagnetic spectrum resources in the whole time period, the system is ensured not to be interfered to the maximum extent, but the defects of the static allocation frequency management mode are more and more prominent along with the increase of frequency devices. Firstly, the development speed of a new available frequency band is far lower than the increasing speed of the demand of frequency-using equipment on frequency spectrum resources, and the electromagnetic frequency spectrum is gradually crowded. According to the prediction of the International Telecommunications Union (ITU), the gap of only public mobile communication spectrum resources can reach 1100MH in the end of the year. Secondly, in the allocated frequency band, the authorized user can share the frequency spectrum resource alone, so that the frequency spectrum utilization rate is low. A spectrum survey of the berkeley region by the Federal Communications Commission (FCC) in the united states shows that the minimum utilization of the spectrum in the region can only reach 15%, and that less than 3GHz of the spectrum is idle for 70% of the time. New Zealand Shared Spectrum Company (SSC) has performed three and a half days of detection on the indoor Spectrum of a building, and the result shows that the average utilization rate of the 806MHz to 2750MHz frequency band can only reach 6.2%. Similar research in China Mobile and Beijing post and telecommunications university shows that, in the first-line city like Beijing, the average utilization rate of the communication service in the full frequency band can only reach 15.2%, and the utilization rate of the partial frequency band is even less than 5%. In the inter-satellite communication service, the division of the frequency spectrum resources is more crowded, on one hand, because the electromagnetic wave characteristics of different frequency bands are different, the frequency utilization requirements under different scenes are comprehensively considered, and the total amount of the frequency spectrum resources allocated to the inter-satellite link is not large; on the other hand, because the space scale of satellite operation is large, global electromagnetic spectrum compatibility needs to be considered for inter-satellite communication, and the satellite system transmitted in history can generate constraint on the frequency selection problem of the satellite system later. Meanwhile, a static allocation method is adopted for dividing the inter-satellite link frequency spectrum resources, and the utilization rate of the inter-satellite link frequency spectrum resources is not high according to the reasonable inference of the ground frequency utilization efficiency. In conclusion, with the current spectrum allocation and use method, the contradiction between supply and demand of spectrum resources of inter-satellite links becomes more and more prominent, and a scheme for further mining spectrum resource potential by using a proper technical means is necessary.

In recent years, a low-earth-orbit satellite network taking an inter-satellite link technology as a core is rapidly developed, a huge amount of satellite constellations bring new challenges to space electromagnetic spectrum resource management and use, and spectrum resource scarcity and complex signal interference become a new normal state of inter-satellite communication. In the inter-satellite link communication, the spectrum cognitive technology is used, the change of the spectrum environment is sensed in real time, and the frequency band with less interference and longer retention time is selected for communication, so that the system efficiency and the communication quality can be effectively improved. The cognitive radio technology is applied to the inter-satellite communication environment, so that the inter-satellite spectrum resource dynamic allocation is realized, the spectrum resource utilization rate is improved, and the anti-interference capability of inter-satellite communication is enhanced. In the process of electromagnetic spectrum cognition, frequency-using equipment firstly uses a radio frequency receiving device to collect interesting frequency band signals, then analyzes the collected signals through an intelligent algorithm to obtain frequency spectrum characteristic information of the current electromagnetic environment, fully understands the happening matters in the electromagnetic environment and finally comprehensively decides the parameters of signals to be transmitted, so that the transmitted signals can be compatible with the current electromagnetic spectrum environment, and the system efficiency is improved. Aiming at the problems of shortage of frequency resources, insufficient anti-interference capability and the like of a satellite communication system, some research institutions try to apply the cognitive radio idea to a satellite communication network in recent years, and through experiments or simulation verification, part of satellite network cognitive technologies can improve the utilization rate of frequency spectrum resources and the anti-interference performance of the system. Through literature analysis in the past years, although a systematic satellite cognitive network is not established at present, the networking concept has become a common consensus in the industry, and research technical points at the present stage mainly focus on the overall design and key technology implementation of the system. The current research content mainly focuses on the uplink and downlink between the satellite and the ground, and the electromagnetic spectrum cognition of the inter-satellite link is blank.

The method for recognizing the modulation characteristics of the inter-satellite signals by using the modulation recognition method is the first step of realizing dynamic allocation of frequency spectrum resources, the modulation mode recognition algorithm based on machine learning automatically extracts potential characteristics in the signals through a neural network, the signal characteristics acquired by each network layer have no specific physical meaning or statistical significance, and if the output results of each network layer are visualized, the method is not only an expert in signal research but also a very difficult expert to say the specific meaning of each layer. And because of the smoothness of feature extraction, in the method for machine learning, signal feature extraction and classification can be carried out in one network, and the robustness and the identification accuracy of the system are improved.

The method is characterized in that a machine learning method is applied to recognize the modulation characteristics of inter-satellite link signals, and one key point is to train a neural network model by using inter-satellite signal data. Experience in the field of image processing shows that the closer the training data is to the actual situation, the more the model obtained by training can adapt to the actual application environment. The electromagnetic signal modulation mode strictly follows a mathematical model, and under the condition that actual inter-satellite link signals are difficult to obtain, the more truly the actual conditions of the inter-satellite link signals are simulated, the better the model with good identification performance can be obtained. However, the current inter-satellite signals have the problem of insufficient data set in the recognition of modulation characteristics.

Disclosure of Invention

Based on this, it is necessary to provide a method and an apparatus for constructing an inter-satellite link spectrum cognition machine learning training data set, a computer device, and a storage medium for solving the problem that an inter-satellite signal has insufficient data sets in modulation feature cognition.

A method for constructing a training data set for inter-satellite link spectrum cognition machine learning, the method comprises the following steps:

establishing requirements according to a preset data set, and setting system parameters;

traversing the modulation mode set in the system parameter, generating a baseband signal according to the modulation mode and the system parameter, and generating a modulation signal according to the baseband signal;

acquiring a pre-generated inter-satellite channel model, and adding channel information corresponding to the inter-satellite channel model to the modulation signal according to the system parameters to obtain a signal set;

and packaging the signal sequences in the signal set into a training data set with a preset format.

An inter-satellite link spectrum cognition machine learning training data set construction device, the device comprises:

the parameter configuration module is used for constructing requirements according to a preset data set and setting system parameters;

the signal simulation module is used for traversing the modulation mode set in the system parameter, generating a baseband signal according to the modulation mode and the system parameter, and generating a modulation signal according to the baseband signal;

the channel simulation module is used for acquiring a pre-generated inter-satellite channel model and adding channel information corresponding to the inter-satellite channel model to the modulation signal according to the system parameters to obtain a signal set;

and the storage module is used for packaging the signal sequences in the signal set into a training data set in a preset format.

A computer device comprising a memory and a processor, the memory storing a computer program, the processor implementing the following steps when executing the computer program:

A computer-readable storage medium, on which a computer program is stored which, when executed by a processor, carries out the steps of:

According to the method and device for constructing the inter-satellite link frequency spectrum cognition machine learning training data set, the channel information corresponding to the inter-satellite channel model is adjusted through signals generated in different modulation modes, and then the signals are packaged into a preset format and used as driving data of the modulation characteristics of the cognitive signals of a machine learning method in the follow-up process, and any data set can be generated.

Drawings

FIG. 1 is a schematic flow chart illustrating a method for learning a training data set by an inter-satellite link spectrum cognitive machine in one embodiment;

FIG. 2 is a schematic flow chart of system parameter setting in one embodiment;

FIG. 3 is a graph of signal relationships associated with a received signal in one embodiment;

FIG. 4 is a schematic flow chart diagram of signal simulation in one embodiment;

FIG. 5 is a diagram of an inter-satellite channel model architecture in accordance with an embodiment;

FIG. 6 is a block diagram of a communication system model in one embodiment;

FIG. 7 is a schematic diagram of an implementation of a digital mixed orthogonal transform in one embodiment;

FIG. 8 is a block diagram of an inter-satellite link spectrum recognition machine learning training data set in an embodiment;

FIG. 9 is a diagram illustrating an internal structure of a computer device according to an embodiment.

Detailed Description

In order to make the objects, technical solutions and advantages of the present application more apparent, the present application is described in further detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the present application and are not intended to limit the present application.

In one embodiment, as shown in fig. 1, a method for constructing a training data set for spectrum cognition in an inter-satellite link is provided, which includes the following steps:

step one, establishing requirements according to a preset data set, and setting system parameters.

The system parameters refer to the frequency, length, and rate of generation of signals, and can be set according to actual data set requirements.

And step two, traversing the modulation mode set in the system parameter, generating a baseband signal according to the modulation mode and the system parameter, and generating a modulation signal according to the baseband signal.

The total of 11 modulation schemes are 8 digital modulation schemes such as BPSK, QPSK, 8PSK, 16QAM, 64QAM, PAM4, GFSK and CPFSK and 3 analog modulation schemes such as B-FM, DSB-AM and SSB-AM.

And step three, acquiring a pre-generated inter-satellite channel model, and adding channel information corresponding to the inter-satellite channel model to the modulation signal according to the system parameters to obtain a signal set.

The inter-satellite signal model refers to a representation framework of signals, as follows:

r(t)＝s(t)+n(t)

wherein s (t) is the signal component; n (t) is a band-pass Additive White Gaussian Noise (AWGN) process with a bilateral Power Spectral Density (PSD) of N₀/2(W/Hz)。

And step four, packaging the signal sequences in the signal set into a training data set with a preset format.

In the method for constructing the inter-satellite link frequency spectrum cognition machine learning training data set, channel information corresponding to an inter-satellite channel model is adjusted through signals generated in different modulation modes, and then the channel information is packaged into a preset format and used as driving data for subsequently recognizing signal modulation characteristics by using a machine learning method, so that any data set can be generated.

In one embodiment, as shown in fig. 2, a plurality of modulation schemes are selected from the predetermined modulation schemes, and the number of signal sample frames, the baseband symbol rate, the number of sampling points of the baseband symbol, and the number of sampling points of the signal sample frames corresponding to the modulation schemes are set; calculating the number of baseband symbols contained in the signal sample frame according to the number of sampling points of the baseband symbols and the number of sampling points of the signal sample frame; setting the center frequency of the frequency band, the sampling rate, the signal-to-noise ratio sequence, the satellite link distance and the Doppler frequency shift.

Specifically, a general signal model corresponding to the inter-satellite link signal system represents a framework. The bandpass waveforms involved can be written as:

r(t)＝s(t)+n(t)

wherein s (t) is the signal component; n (t) is a band-pass Additive White Gaussian Noise (AWGN) process with a bilateral Power Spectral Density (PSD) of N₀And/2 (W/Hz). The respective components of the formula (1-1) can be written as

Wherein the content of the first and second substances,

is centered at ω_c(rad/s) band pass signal r (t). The AWGN noise process of the complex baseband can be expanded to

Both side power spectral densities are N₀And/2 (W/Hz). Thus, the passband noise process can be rewritten as:

assuming a transmitted signal

The signal is a single-channel amplitude and phase modulation signal with or without residual carrier waves, and the specific form is as follows:

in the formula:

P_dand P_cThe power of the real passband data and the residual carrier signal, respectively.

Is complex modulation of the k-th symbol, where A_kIs a normalized amplitude, satisfies

θ_kIs the phase modulation of the kth symbol;

g (t) denotes subcarrier modulation, and is usually expressed as g (t) sin ω_sct or, where is the subcarrier frequency (in rad/s).

p (t) is the pulse shape, satisfy

T is symbol period and is in seconds

θ_c(t) is the carrier phase

Wherein is shown in the formula

c_k∈{-1,1},A_kA Binary Phase Shift Keying (BPSK) signal when g (T) (1) is 1, for T e {0, T } there is p (T) e { -1,1}, in which case it can be rewritten as:

in the formula, P_t＝P_c+P_dIs the total passband envelopeThe signal power;

is the modulation angle, also called modulation index. For an M-dimensional phase shift keying (M-PSK) signal without a residual carrier, the above equation becomes

In the formula (I), the compound is shown in the specification,

is the phase modulation of the k-th M-PSK symbol. With independent, uniform distribution of q_kE.g. {0, 1., M-1 }. By appropriate definition of d_k(t), Quadrature Amplitude Modulation (QAM) can also be represented by the above formula

And (4) showing.

For the receiver, the clock and carrier frequency phases are initially unknown and noise is present. The residual frequency component is still assumed to be incorrectly estimated at the front-end carrier frequency of the receiver

After conversion to baseband, the signal is still present, resulting in the form of:

where ε is the sign clock of the unknown fractional part. A priori, ε is at [0,1 ]]Upper uniformly distributed, theta_c(t) at [0,2 π]Are uniformly distributed.

As shown in fig. 3, several factors related to the received signal are given, and the signal relation diagram is divided into 3 main parts: a Forward Error Correction (FEC) code encoder, a modulator/amplifier, and a channel. Each of which is affected by several sub-factors.

The FEC code is one of several code patterns. The Consultative Committee for Spatial Data Systems (CCSDS) has customized standard code patterns for deep space or deep space communications, including Reed-solomon (rs) codes, convolutional codes, Turbo codes, BCH codes, cyclic check (CRC) codes. The development of low density parity check codes and improved parity check codes (such as Tornado codes and Raptor codes) has progressed rapidly in theoretical research and various standards enactment, including digital television broadcasting/satellite, Institute of Electrical and Electronics Engineers (IEEE)802.{11n,15,3a,16e } and CCSDS deep space standards.

Associated with each FEC code is its code rate (the portion of the symbol that carries the information) and code length (indicating the number of symbols per codeword). For some patterns, using these two parameters alone is almost sufficient to fully determine its encoding. For example, optimal performance convolutional codes for a given code rate and constraint length are already listed in textbooks, and the convolutional codes used are almost always from these lists. CRC codes of a given length are also typically generated using standard generator polynomials. The RS code is determined by its block length, code rate, domain generator polynomial and code generator polynomial. There are several possibilities for the latter two parameters, but in practical spatial communication systems the ones given in the CCSDS standard are mainly used.

In one embodiment, as shown in fig. 4, a modulation mode set in a system parameter is traversed, and a modulation parameter is set according to the modulation mode and the number of sampling points of a signal sample frame; judging whether the modulation mode is analog modulation or not; if not, generating an ASCII character sequence through the random number, determining a system number of a modulation mode, converting the ASCII character sequence into a bit sequence, converting the bit sequence into a baseband symbol sequence according to the system number, intercepting a baseband signal with the length being the number of sampling points of a signal sample frame from the baseband symbol sequence, and modulating the baseband signal into a modulation signal.

In another embodiment, if yes, intercepting a baseband signal with the length of the number of sampling points of a signal sample frame from a preset audio signal; the baseband signal is modulated into a modulated signal.

Specifically, the purpose of recognizing the inter-satellite link signal modulation characteristics is to understand the spectrum environment, and therefore, objects to be recognized by the signal modulation characteristics include modulation types used in inter-satellite communication and other types of modulation types that can be received by a satellite during the in-orbit period.

During the in-orbit period of the satellite, the satellite can receive the digital modulation signal between the satellites with the appointed frequency and can also receive the digital and analog modulation signals which are partially transmitted by the frequency equipment used on the ground, so the invention constructs the signal data set from the perspective of the digital signal and the analog signal. For analog modulation signals, audio data of a broadcast program is selected for encoding and modulation, which mainly includes the voice of the program host and part of the musical instrument sound. For a digital modulation signal, a random ASCII character string is generated by a random number generator to serve as a signal source, and a binary string of ASCII characters is encoded into a symbol string which can be modulated in different modulation modes and transmitted. In the present invention, the signal modulators each use the two digital or analog signal sources as inputs.

When the baseband signal rates are different, the receiver can obtain different numbers of baseband symbols in fixed-length sampling sequences. For example, in the BPSK modulation scheme, when the baseband rate is 2kHz, the receiver samples at 8kHz, and 128 samples include 32 baseband symbols; when the baseband rate is 1kHz, the receiver samples at 8kHz, and only 16 baseband symbols are contained in 128 samples. The sampling result of the receiver is used as the starting point of the modulation characteristic cognition, and the quantity of baseband symbols contained in the fixed-length sampling sequence directly influences the identification accuracy rate of the modulation mode. Therefore, a normalized sample needs to be selected for each modulation scheme to form a constant normalized symbol rate when constructing the signal modulation scheme data set.

The data set generated by the invention considers that 8 digital modulation modes of BPSK, QPSK, 8PSK, 16QAM, 64QAM, PAM4, GFSK and CPFSK and 3 analog modulation modes of B-FM, DSB-AM and SSB-AM are used, and the base band rates are 2kHz, 20kHz, 200kHz and 2MHz respectively. The scenes corresponding to the respective modulation schemes are shown in table 1:

table 1 common modulation scheme and application scenario

The inter-satellite modulation signal can be divided into analog modulation and digital modulation according to signal property. The analog modulation signal mainly comprises three types of Amplitude Modulation (AM), Phase Modulation (PM) and Frequency Modulation (FM), and the expression thereof is as follows:

y_PM(t)＝Acos[ω_ct+K_ps(t)]

y_FM(t)＝Acos[ω_ct+K_f∫s(τ)dτ]

in the above formula, s (t) is a baseband signal, K_pIs the phase coefficient of the phase modulation, K_fIs the frequency coefficient of the frequency modulation. Wideband frequency modulation (WBFM) is a specific example of an analog signal frequency modulation method, and the maximum value of the instantaneous phase shift amount should satisfy the following condition:

in addition to analog modulation, the satellite receives more of the digitally modulated signal during its orbit, which is expressed as follows:

in the above formula, the first and second carbon atoms are,

and the complex value of N sampling points of the ith digital modulation mode is represented, T represents the symbol width after modulation, and g (-) represents shaping filtering.

Based on the mathematical model of the signal modulation mode, the invention uses a Communication toolkit (Communication Toolbox) to simulate and generate 11 modulation mode signals under an MATLAB platform, and uses a Comm module to simulate free space loss, noise superposition and Doppler frequency shift phenomena of inter-satellite link signals in the transmission process. Experiments show that the quadrature components of the generated modulation signals are zero when the modulation signals do not pass through an analog channel, and after the modulation signals pass through the analog channel, the signals are mixed with channel dynamic characteristics in an inter-satellite link transmission process, and the in-phase components and the quadrature components are changed.

In summary, the set modulation parameters mainly include the modulation mode category, the center frequency, the sampling frequency, and the like; the channel model mainly comprises free space loss, additive white Gaussian noise and Doppler frequency shift, wherein the additive white Gaussian noise is controlled by a signal-to-noise ratio, and the Doppler frequency shift can simulate frequency shift at a certain moment but does not have dynamic variable characteristics; the modulation mode distinguishes between analog modulation and digital modulation, analog modulation (AM-DSB, AM-SSB, WBFM) obtains the source by intercepting the audio signal, digital modulation generates the corresponding sequence by a random number generator.

The inter-satellite link transmitter simulation mainly simulates the modulation process of baseband signals and is divided into four steps. The data whitening is realized by symbol coding, and a specific coding method is adopted to avoid the situations of continuous '0' and continuous '1' of a code stream to be sent; the symbol mapping is the core part of signal modulation, and the baseband symbol is translated into a frequency band signal sampling point by looking up a symbol and radio frequency waveform comparison table; in the step of filtering forming, a raised cosine roll-off filter is adopted to filter digital waveforms, so that the influence of intersymbol interference is reduced, wherein the roll-off factor is 0.35; the signal interpolation is the process of up-sampling the filtered frequency band signal, and transmitting and receiving analog signals.

In one embodiment, as shown in fig. 5, a schematic structural diagram of an inter-satellite channel model is provided, a pre-generated inter-satellite channel model is obtained, and clock offset, doppler shift, free space loss, and gaussian additive white noise corresponding to the inter-satellite channel model are added to the modulation signal according to system parameters, so as to obtain a signal set.

In particular, inter-satellite communications have its own characteristics in electromagnetic environments and transmission channels, compared to traditional air-to-ground satellite communications. The method has the advantages that the atmospheric layer has small signal attenuation effect, influence of factors such as cloud layers and weather is not considered, and free space loss is a main factor of signal attenuation and delay; secondly, the receiving and transmitting parties of the inter-satellite communication are positioned in the sight distance path direction of the opposite end, the communication process basically has no multipath effect, and the Doppler frequency shift generated by the relative motion between the satellites is a main factor of signal frequency shift; and in the inter-satellite communication, the wireless signal path basically has no ionosphere effect, the ionization effect generated by cosmic rays and solar radiation is very little, and the inter-satellite communication is hardly influenced. The invention processes the channel characteristics of the inter-satellite link from three aspects of clock offset, free space loss, Doppler frequency shift and Gaussian additive white noise.

In another embodiment, adding the clock offset corresponding to the inter-satellite channel model to the modulated signal according to the system parameter is:

wherein, Delta_clockRepresenting the clock offset, delta, when a frame signal passes through the analog channel_clockFrom the range [ -max (Δ)_clock)max(Δ_clock)]Randomly generated with uniform distribution, max (Δ)_clock) Is the maximum clock offset;

adding the free space loss corresponding to the inter-satellite channel model to the modulation signal according to the system parameters as follows:

units of the satellite link distance d and the frequency f are km and MHz respectively, and c represents the light speed;

adding the Doppler frequency shift corresponding to the inter-satellite channel model to the modulation signal according to the system parameters as follows:

wherein f is_cRepresenting the carrier frequency, c the speed of light,

the maximum value of the frequency shift of the electromagnetic wave frequency;

adding Gaussian additive white noise corresponding to the inter-satellite channel model to the modulation signal according to the system parameters is as follows:

wherein q (t) represents the quadrature component of the noise, i (t) represents the in-phase component of the noise, x represents the modulation signal, and the covariance matrix epsilon of the gaussian distribution model is:

σ_Iand

respectively representing the standard deviation and variance, σ, of the in-phase component of the noise_QAnd

respectively representing the standard deviation and the variance of the orthogonal component of the noise, p represents the correlation coefficient of two variables, p is 0, and sigma_I＝σ_Q＝σ。

In particular, clock skew is caused by inaccuracies in the internal clock sources of the transmitter and receiver. The clock skew may cause the carrier frequency of the down-conversion of the rf signal to the baseband signal, and the sampling frequency of the dac to deviate from the ideal value. In the inter-satellite link channel simulation, the clock offset factor C is used to represent the offset degree, which can be expressed as:

in the above formula,. DELTA._clockRepresenting the clock offset, delta, when a frame signal passes through the analog channel_clockFrom the range [ -max (Δ)_clock)max(Δ_clock)]Randomly generated with uniform distribution, max (Δ)_clock) Is the maximum clock offset.

In the process of wireless signal propagation, the power of electromagnetic waves is attenuated by the physical action of the propagation path with the increase of the distance from the radiation source, and this phenomenon is called the path loss of electromagnetic waves, and this loss is significant in the conventional satellite communication system, and includes rain attenuation, atmospheric loss, dielectric loss, free space loss, and the like. In the inter-satellite communication, a signal propagation path does not experience the actions of reflection, refraction, absorption, scattering and the like of the earth atmosphere, and the form of signal path loss is mainly represented as free space loss. The electromagnetic wave generated by the satellite antenna is basically not linearly radiated to a fixed point, and is a spot beam antenna with excellent performance, and the radiated electromagnetic wave propagates in a conical shape in space, so that the farther the position is from the antenna, the smaller the received signal energy per unit area is, and the signal power loss is expressed. Like lighting a candle, locations closer to the candle are brighter, and locations further from the candle are dimmer.

In the inter-satellite communication, the satellite transmission signal paths are all in an ideal free space, electromagnetic waves are radiated from the satellite antenna at the transmitting end and then propagate forwards at the speed of light c, and if a satellite receives signals at the axial distance d from the main radiation lobe of the transmitting antenna, the signal power reaching the receiving satellite can be expressed as:

in the above formula, P_rIndicating the power of the signal received by the receiving satellite, P_tRepresenting the power of the signal transmitted by the satellite at the transmitting end, G_tDenotes the antenna gain, G, at the transmitting end_rThe gain of the receiving end antenna is shown, and lambda and d respectively show the wavelength of the electromagnetic wave and the distance between the receiving end satellite and the transmitting end satellite, and are usually taken

Wherein c is the speed of light, and f is the frequency of electromagnetic waves radiated by the satellite antenna at the transmitting end.

When G is_t＝G_rWhen 1, the ratio of the signal power transmitted by the transmitting end satellite to the signal power received by the receiving end satellite is the free space loss, and can be expressed as:

taking the logarithm, the free space loss value is obtained, and is expressed as:

in the above equation, the units of the distance d and the frequency f are km and MHz, respectively, and when the distance d or the frequency f is doubled, the free space loss value is increased by 6 dB.

When there is relative motion between the transmitting end and the receiving end of the electromagnetic wave, there is a deviation between the frequency of the electromagnetic wave received by the receiving end and the frequency of the electromagnetic wave transmitted by the transmitting end, which is called doppler effect, and the deviation between the frequency of the electromagnetic wave of the transmitting end and the frequency of the electromagnetic wave of the receiving end is doppler shift. In the inter-satellite communication, the orbit height, the operating speed, the orbit inclination, the eccentricity and other basic parameters of a transmitting-end satellite and a receiving-end satellite are often different, so that the generated inter-satellite doppler frequency shift often affects the communication quality of an inter-satellite link.

Assuming that a relative movement velocity of the transmitting end satellite and the receiving end satellite is v, an included angle between the relative movement velocity and the propagation direction of the electromagnetic wave is α, and the wavelength of the electromagnetic wave is λ, the doppler shift can be expressed as:

in the above formula, f_cRepresenting the carrier frequency, c the speed of light,

the maximum value of the frequency shift of the electromagnetic wave is called maximum Doppler shift.

In the inter-satellite link, under the condition of normal communication of a receiving satellite and a transmitting satellite, a communication system can be influenced by noise outside the system and noise inside the system. The external noise of the system comprises space ground noise, celestial body radiation interference and the like, and when the external noise of the system is strong, the external noise of the system often causes satellite communication interruption, such as the sun-and-rain interference. The noise inside the system includes receiver and transmitter noise, thermal noise of system elements, etc., which are commonly present in the communication system, and the influence of the noise on the communication quality can be reduced by optimizing the system design, but cannot be completely eliminated. In the system internal noise, the antenna noise of the receiver and the transmitter is an important factor, and is generally expressed by equivalent noise temperature:

in the above formula, T represents the equivalent noise temperature in kelvin, and k is 1.380662 × 10^-23J/K denotes boltzmann's constant, B denotes the receive antenna bandwidth, and p denotes the probability of the receiver receiving noise.

When inter-satellite link signal simulation is carried out, under the condition of normal communication, the cosmic space bottom noise is Gaussian white noise, the external noise burstiness of other systems is strong, and an explicit expression is not easy to construct; the noise inside the system is mainly thermal noise, so that the system channel noise can be simplified into an additive white gaussian noise model.

An additive white gaussian noise model is a common model for signal processing in communication systems, and is characterized by a fixed spectral density and zero-mean gaussian amplitude distribution. Assuming that the time domain expression of additive white gaussian noise is n (t) ═ i (t) + jq (t), the probability density function can be expressed as:

According to the probability theory correlation theory, the in-phase component I (t) and the quadrature component Q (t) of the noise n (t) after orthogonal decomposition can be used as two independent random variables to participate in calculation, and the two random variables are independent and distributed in the same way and meet the Gaussian distribution model. At this time, the covariance matrix ε can be expressed as:

in the above formula, σ_IAnd

respectively representing the standard deviation and the variance of the orthogonal components of the noise, and p represents the correlation coefficient of the two variables. Here, the in-phase component i (t) and the quadrature component q (t) satisfy an independent equal distribution, that is, ρ is 0, σ_I＝σ_Qσ, its probability density function can be simply expressed as:

after the modulation signal is processed, the inter-satellite link signal needs to be preprocessed, which is specifically as follows:

1. signal normalization

In the inter-satellite link spectrum cognitive system, the relative position between the sensing node and the transmitting node is in dynamic change, and the transmitting power of signals in various modulation modes is different, so that the difference of the power of electromagnetic signals received by the sensing node is large. The core idea of adopting the neural network to recognize the signal modulation characteristics is data driving, unreasonable data characteristics can make the training model deviate from a set target, the signal power is obviously not the obvious characteristic of the modulation mode, the power difference of the training signal is eliminated by using signal normalization processing, and the effective characteristic of the modulation mode can be favorably kept in network parameter learning.

As shown in fig. 6, a model block diagram of a communication system is provided, and from the viewpoint of signal processing flow, the modulation scheme characteristic cognitive model studied in the present invention operates between signal reception and demodulation. Before the received signal enters the recognition model, the power of the input signal is equivalent to that of the training signal through normalization processing, and the recognition effect can be optimized.

The normalization can be implemented by the following means:

one is the overall power normalization of the received signal segment, as shown in the following equation:

in the above equation, r (i) represents a sampling point of the received signal, N represents a sampling length of the received signal, and r_p(i) And the values of the sampling points subjected to normalization processing are shown.

Another normalization method is to process the amplitude of the signal, and the calculation is simple, and the normalized signal of this method can be expressed as:

in the above equation, | r (i) | represents the instantaneous amplitude value of the signal r (i), and max represents the maximum sampling value in the received signal segment.

Because the relative size of the signal sampling value contains the potential characteristics of the modulation mode, the absolute size has almost no influence on the neural network training, and the amplitude value normalization method is used when the data set is generated and the signal modulation mode inference is carried out.

2. Signal orthogonal transformation

The signal orthogonal transformation refers to a process of decomposing a real signal into an in-phase component and an orthogonal component in an analog or digital mode, and in the research content of the invention, the orthogonal transformation is mainly used in a signal preprocessing scene of data set generation and modulation mode identification.

The characteristics of the modulation mode on the signal are mainly reflected on the instantaneous frequency, the instantaneous amplitude and the instantaneous phase, which are important factors for distinguishing the modulation classes. However, these transient characteristics of the actual signal are not particularly obvious in time series, and the transient characteristics of the modulated signal can be enhanced by decomposing the actual signal into an in-phase component and a quadrature component using a method of orthogonal transformation.

If the amplitude of a narrowband signal is a (t), the phase is

The carrier being f₀Then its time domain expression is:

the analytical expression of the signal after orthogonal transformation can be written as:

by means of the in-phase component sequence i (t) and the quadrature component sequence q (t), the instantaneous parameter expression of the signal can be easily obtained:

therefore, the fact that the received signal time sequence is transformed into the in-phase component and the quadrature component in an analytic form is an implicit extraction of instantaneous characteristics of a modulation mode, deep characteristics of a corresponding modulation mode can be obtained by training a neural network through the transformed sequence, and the theory basis of the identification of the modulation mode by using an I/Q signal is also provided.

As shown in fig. 7, a schematic diagram for implementing digital frequency mixing orthogonal transformation is provided, in which a radio frequency front end receives an analog signal x (t), and then forms a digital sequence x (n) through analog-to-digital conversion, and then forms a local oscillation digital sequence cos ω orthogonal to each other₀n、sinω₀n, and finally filtering the product by a digital low-pass filter to obtain the corresponding in-phase component I (n) and quadrature component Q (n).

In one embodiment, the power of the simulation signal in the signal set is normalized to obtain a normalized signal; the normalized signal is stored as an n-dimensional vector using numpy and cPickle and saved as a training data set.

Specifically, after the data set signal is modulated and transmitted, due to the existence of free space loss and noise, the signals received by the satellite at the receiving end of the inter-satellite link are uneven in power or amplitude, and if the signals are taken as training samples at this time, the amplitude or power characteristics may be taken as effective characteristic processing of modulation signal classification in the training process by the neural network, so that the identification of the signal modulation characteristics is affected, and therefore normalization processing needs to be performed on the power of the simulation signal before the data set is stored. Method of signal power normalization as described above, in the received 128 point-sized samples, each signal sample is scaled to unity energy when stored.

The purpose of signal data set storage is to facilitate its ease of use in machine learning environments outside of the software-radio software ecosystem. A common approach is to store datasets as N-dimensional vectors using numpy, a popular format for storing reasonably sized datasets in the machine learning community, and cpikle, which provides and processes the N-dimensional Array object. The support of the cPickle package in the MATLAB environment is not ideal, and the invention selects the MAT and Pickle files to store the generated signal data sets.

The data set used by the invention can be characterized from three dimensions of modulation mode type, baseband rate and signal-to-noise ratio, the modulation mode comprises 8 digital signal modulations and 3 analog signal modulations, the signal-to-noise ratio is simulated once every 2dB interval from-20 dB to 18dB, the baseband rate selects 2kHz, 20kHz, 200kHz and 2MHz to be simulated respectively, each modulation mode, baseband rate and signal-to-noise ratio combination comprises 20000I/Q signal samples, and the whole data set has 880000 signal samples.

After the baseband signal is modulated and propagated, the system samples the baseband signal at a rate of 8MHz to form a received signal stream, and windowing, normalizing and converting operations form the received signal stream into 128-point complex-valued vector samples, each sample representing a data length of 16 μ s and containing 4 to 4000 symbols. Considering that the tensor form used by the current deep neural network common toolkit (Keras, Theano, TensorFlow, etc.) is a 4-dimensional real floating-point vector, the data set packs the obtained sample values into N_examples×N_channels×Dim₁×Dim₂Form (b) wherein N_examplesRepresenting the number of samples of the data set, N_channelsIndicating the number of channels, corresponding to the RGB channels in the image processing, Dim₁Indicating that each sample has two paths of values, Dim, of in-phase component I (t) and quadrature component Q (t)₂Representing the number of sample points per sample. Thus, the dimensions of the inventive dataset are 880000 × 1 × 2 × 128.

It should be understood that, although the steps in the flowchart of fig. 1 are shown in order as indicated by the arrows, the steps are not necessarily performed in order as indicated by the arrows. The steps are not performed in the exact order shown and described, and may be performed in other orders, unless explicitly stated otherwise. Moreover, at least a portion of the steps in fig. 1 may include multiple sub-steps or multiple stages that are not necessarily performed at the same time, but may be performed at different times, and the order of performance of the sub-steps or stages is not necessarily sequential, but may be performed in turn or alternately with other steps or at least a portion of the sub-steps or stages of other steps.

In one embodiment, as shown in fig. 8, there is provided an inter-satellite link spectrum cognition machine learning training data set construction apparatus, including: a parameter configuration module 802, a signal simulation module 804, a channel simulation module 806, and a storage module 808, wherein:

a parameter configuration module 802, configured to establish requirements according to a preset data set, and set system parameters;

a signal simulation module 804, configured to traverse the modulation mode set in the system parameter, generate a baseband signal according to the modulation mode and the system parameter, and generate a modulation signal according to the baseband signal;

a channel simulation module 806, configured to obtain a pre-generated inter-satellite channel model, and add channel information corresponding to the inter-satellite channel model to the modulation signal according to the system parameter to obtain a signal set;

a storage module 808, configured to package the signal sequences in the signal set into a training data set in a preset format.

In one embodiment, the parameter configuration module 802 is further configured to select multiple modulation modes from the given modulation modes, and set the number of signal sample frames, the baseband symbol rate, the number of sampling points of the baseband symbol, and the number of sampling points of the signal sample frames corresponding to the modulation modes; calculating the number of baseband symbols contained in the signal sample frame according to the number of sampling points of the baseband symbols and the number of sampling points of the signal sample frame; setting the center frequency of the frequency band, the sampling rate, the signal-to-noise ratio sequence, the satellite link distance and the Doppler frequency shift.

In one embodiment, the signal simulation module 804 is further configured to traverse the modulation mode set in the system parameter, and set a modulation parameter according to the modulation mode and the number of sampling points of the signal sample frame; judging whether the modulation mode is analog modulation or not; if not, generating an ASCII character sequence through a random number, determining a system number of the modulation mode, converting the ASCII character sequence into a bit sequence, converting the bit sequence into a baseband symbol sequence according to the system number, and intercepting a baseband signal with the length being the number of sampling points of the signal sample frame from the baseband symbol sequence; and modulating the baseband signal into a modulation signal.

In one embodiment, the signal simulation module 804 is further configured to, if yes, intercept a baseband signal with a length equal to the number of sampling points of the signal sample frame from a preset audio signal; and modulating the baseband signal into a modulation signal.

In one embodiment, the channel simulation module 806 is further configured to obtain a pre-generated inter-satellite channel model, and add clock offset, doppler shift, free space loss, and white gaussian additive noise corresponding to the inter-satellite channel model to the modulation signal according to the system parameter to obtain a signal set.

In one embodiment, the channel simulation module 806 is further configured to add a clock offset corresponding to the inter-satellite channel model to the modulation signal according to the system parameter as follows:

wherein f is_cRepresenting the carrier frequency, c the speed of light,

the maximum value of the frequency shift of the electromagnetic wave frequency;

σ_Iand

In one embodiment, the storage module 808 is further configured to perform normalization processing on the power of the simulation signal in the signal set to obtain a normalized signal; the normalized signal is stored as an n-dimensional vector using numpy and cPickle and saved as a training data set.

For specific limitations of the inter-satellite link spectrum cognition machine learning training data set construction device, reference may be made to the above limitations on the inter-satellite link spectrum cognition machine learning training data set construction method, which are not described herein again. All modules in the inter-satellite link spectrum cognition machine learning training data set construction device can be completely or partially realized through software, hardware and a combination thereof. The modules can be embedded in a hardware form or independent from a processor in the computer device, and can also be stored in a memory in the computer device in a software form, so that the processor can call and execute operations corresponding to the modules.

In one embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as shown in fig. 9. The computer device includes a processor, a memory, a network interface, and a database connected by a system bus. Wherein the processor of the computer device is configured to provide computing and control capabilities. The memory of the computer device comprises a nonvolatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operation of an operating system and computer programs in the non-volatile storage medium. The database of the computer device is used for storing a training data set. The network interface of the computer device is used for communicating with an external terminal through a network connection. The computer program is executed by a processor to realize a method for constructing a training data set for inter-satellite link spectrum cognition machine learning.

Those skilled in the art will appreciate that the architecture shown in fig. 9 is merely a block diagram of some of the structures associated with the disclosed aspects and is not intended to limit the computing devices to which the disclosed aspects apply, as particular computing devices may include more or less components than those shown, or may combine certain components, or have a different arrangement of components.

In an embodiment, a computer device is provided, comprising a memory storing a computer program and a processor implementing the steps of the method in the above embodiments when the processor executes the computer program.

In an embodiment, a computer-readable storage medium is provided, on which a computer program is stored, which computer program, when being executed by a processor, carries out the steps of the method in the above-mentioned embodiments.

It will be understood by those skilled in the art that all or part of the processes of the methods of the embodiments described above can be implemented by hardware instructions of a computer program, which can be stored in a non-volatile computer-readable storage medium, and when executed, can include the processes of the embodiments of the methods described above. Any reference to memory, storage, database, or other medium used in the embodiments provided herein may include non-volatile and/or volatile memory, among others. Non-volatile memory can include read-only memory (ROM), Programmable ROM (PROM), Electrically Programmable ROM (EPROM), Electrically Erasable Programmable ROM (EEPROM), or flash memory. Volatile memory can include Random Access Memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms such as Static RAM (SRAM), Dynamic RAM (DRAM), Synchronous DRAM (SDRAM), Double Data Rate SDRAM (DDRSDRAM), Enhanced SDRAM (ESDRAM), Synchronous Link DRAM (SLDRAM), Rambus Direct RAM (RDRAM), direct bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

The technical features of the above embodiments can be arbitrarily combined, and for the sake of brevity, all possible combinations of the technical features in the above embodiments are not described, but should be considered as the scope of the present specification as long as there is no contradiction between the combinations of the technical features.

The above-mentioned embodiments only express several embodiments of the present application, and the description thereof is more specific and detailed, but not construed as limiting the scope of the invention. It should be noted that, for a person skilled in the art, several variations and modifications can be made without departing from the concept of the present application, which falls within the scope of protection of the present application. Therefore, the protection scope of the present patent shall be subject to the appended claims.

Claims

1. A method for constructing a training data set for inter-satellite link spectrum cognition machine learning, the method comprises the following steps:

2. The method of claim 1, wherein the building requirements from a pre-set data set, setting system parameters, comprises:

selecting a plurality of modulation modes from the established modulation modes, and setting the number of signal sample frames, the rate of baseband symbols, the number of sampling points of the baseband symbols and the number of sampling points of the signal sample frames corresponding to the modulation modes;

calculating the number of baseband symbols contained in the signal sample frame according to the number of sampling points of the baseband symbols and the number of sampling points of the signal sample frame;

setting the center frequency of the frequency band, the sampling rate, the signal-to-noise ratio sequence, the satellite link distance and the Doppler frequency shift.

3. The method of claim 2, wherein traversing the modulation scheme set in the system parameter, generating a baseband signal according to the modulation scheme and the system parameter, and generating a modulation signal according to the baseband signal comprises:

traversing the modulation mode set in the system parameter, and setting a modulation parameter according to the modulation mode and the number of sampling points of the signal sample frame;

judging whether the modulation mode is analog modulation or not;

if not, generating an ASCII character sequence through a random number, determining a system number of the modulation mode, converting the ASCII character sequence into a bit sequence, converting the bit sequence into a baseband symbol sequence according to the system number, and intercepting a baseband signal with the length being the number of sampling points of the signal sample frame from the baseband symbol sequence;

and modulating the baseband signal into a modulation signal.

4. The method of claim 3, further comprising:

if so, intercepting a baseband signal with the length of the sampling point number of the signal sample frame from a preset audio signal;

and modulating the baseband signal into a modulation signal.

5. The method according to claim 2, wherein the obtaining a pre-generated inter-satellite channel model, and adding channel information corresponding to the inter-satellite channel model to the modulation signal according to the system parameter to obtain a signal set comprises:

and acquiring a pre-generated inter-satellite channel model, and adding clock offset, Doppler frequency shift, free space loss and Gaussian additive white noise corresponding to the inter-satellite channel model to the modulation signal according to the system parameters to obtain a signal set.

6. The method of claim 5, wherein adding clock offset, Doppler shift, free space loss, and white Gaussian additive noise corresponding to an inter-satellite channel model to the modulated signal according to the system parameters comprises:

adding the clock skew corresponding to the inter-satellite channel model to the modulation signal according to the system parameters as follows:

wherein f is_cRepresenting the carrier frequency, c the speed of light,

the maximum value of the frequency shift of the electromagnetic wave frequency;

σ_Iand

7. The method according to any one of claims 1 to 6, wherein said packaging the signal sequences in the signal set into a training data set of a preset format comprises:

normalizing the power of the simulation signals in the signal set to obtain normalized signals;

the normalized signal is stored as an n-dimensional vector using numpy and cPickle and saved as a training data set.

8. An inter-satellite link spectrum cognition machine learning training data set construction device is characterized by comprising the following steps:

9. A computer device comprising a memory and a processor, the memory storing a computer program, wherein the processor implements the steps of the method of any one of claims 1 to 7 when executing the computer program.

10. A computer-readable storage medium, on which a computer program is stored, which, when being executed by a processor, carries out the steps of the method of any one of claims 1 to 7.