Description:DESCRIPTION

Technical Field of the Invention

The present invention is associated with field of signal interception, which is related to detecting week unknown signals. In particular the invention is about detecting the signals which are with very low Signal to Noise and Interference Ratio, by exploiting the instantaneous characteristics of signal in time, frequency, phase and angle of arrival.

Background of the Invention

In COMINT and ELINT applications, detecting signals from distant and low-power sources with low Signal-to-Noise Ratio (SNR) presents a significant challenge. Existing signal detection methods primarily focus on one or two dimensions of the received signal characteristics, limiting their effectiveness in accurately detecting signals. There is an opportunity to exploit the behavior of signals across multiple dimensions and develop a better method for detecting the presence or absence of signals using suitable weighing functions. Furthermore, existing detection algorithms that perform well at low SNRs are often computationally complex and have poor real-time response, making them impractical for many applications.

Prior arts:
[1] Richard W. Middlestead, “Digital Communications with Emphasis on Data Modems: Theory, Analysis, Design, Simulation, Testing, and Applications,” First Edition, 2017, PP 755-766.

[2]USH2222H1 titled “Normalized matched filter—a low rank approach”, and assigned to AIR FORCE, GOVERNMENT OF THE UNITED STATES OF AMERICA.

[3] US6087977A titled “False alarm rate and detection probability in a receiver”, and assigned to CARDION NEWCO, INC.

[4]Xiaolei Fan, Yurong Wan, Tao Li, and Zengping Chen, “ An Adaptive Method of Pulse Detection Based on Frequency-domain CFAR, ” 2016 Progress In Electromagnetic Research Symposium (PIERS), Shanghai, China, 8–11 August.

[5] LANCON Fabienne, HILLION Alain, SAOUDI Samir, “RADAR Signal Extraction Using Correlation”.

[6] “Probability of Intercept in Electronic Countermeasures Receivers”, NAVAL POSTGRADUATE SCHOOL, Monterey, California.

Prior art includes various detection techniques such as envelop detection, matched filtering, and CFAR-based approaches. However, envelop detection, the simplest and most commonly used method, requires high SNR for optimal signal interception. Other algorithms show effectiveness on certain signal types but suffer from computational complexity and limited real-time performance.

The present invention addresses these limitations and provides an improved solution for pulse detection. By utilizing a multi-domain machine learning-based approach, the invention leverages the properties of phase, time, and frequency to enhance detection accuracy and adaptability. The system can effectively detect signals even at very low SNRs, reducing false alarms and achieving high probability of detection.

The prior art references [1] to [5] evaluate different detection methods, including envelop detection, low rank matched filter, CFAR-based approaches, and correlation-based methods. While these techniques have demonstrated certain capabilities, they still exhibit limitations in terms of sensitivity to SNR, computational complexity, and real-time response.

The need for the present invention arises from the demand for a pulse detection system that overcomes the drawbacks of existing methods. The proposed solution combines multi-domain analysis with machine learning techniques, enabling improved accuracy, real-time responsiveness, and adaptability to different signal types. By exploiting the behavior of signals across multiple dimensions, the present invention offers a practical and efficient approach to pulse detection in electronic warfare systems, radar applications, and communication systems.

Brief Summary of the Invention

According to an aspect of the present invention, a multi-domain machine learning-based pulse detection method and system that utilizes the Phase, Time, and Frequency properties of the signal for accurate classification and detection of pulses. The system is designed to operate in challenging environments such as electronic warfare operations, where reliable detection and classification of pulse signals amidst high levels of background noise are crucial.

In accordance with the aspect of the present invention, the multi-domain machine learning-based pulse detection system, comprises:
a two-channel receiver hardware configured to receive signals in the frequency range of 400 to 6000 MHz;
a detection algorithm implemented in three levels, including a first, second, and third level of computing steps;
in the first level, summing the Fast Fourier Transform (FFT) data of all channels bin-wise to obtain a single vector;
in the second level, computing the phased difference among two phase spectrums for all combinations of channels and providing the variance output to a variance block to exploit the signal's spread across multiple bins in a given intercept;
in the third level, processing the phase difference using the phase spectrum of the FFT between consecutive frames of the same channel and providing the output to a statistical block for the final detection;
detecting the final signal by applying a weighted sum to the three levels of processed data.

In accordance with the aspect of the present invention, the system involving a multi-domain machine learning-based pulse detection method, comprises the following steps:
configuring a two-channel receiver hardware with ultra-wideband printed vivaldi antennas;
implementing a detection algorithm in three levels, including a first, second, and third level of computing steps;
in the first level, summing the FFT data of all channels bin-wise to obtain a single vector;
in the second level, computing the phased difference among two phase spectrums for all combinations of channels and providing the variance output to a variance block to exploit the signal's spread across multiple bins in a given intercept;
in the third level, processing the phase difference using the phase spectrum of the FFT between consecutive frames of the same channel and providing the output to a statistical block for the final detection;
detecting the final signal by applying a weighted sum to the three levels of processed data.

In summary, the present invention discloses a multi-domain machine learning-based pulse detection system and method that leverages the Phase, Time, and Frequency properties of the signal for accurate detection and classification of pulses. The system employs a two-channel receiver hardware and a detection algorithm implemented in three levels of processing. The first level involves the summation of FFT data to obtain a single vector representing the magnitude spectrum of the received signals. The second level computes the phased difference among two phase spectrums for all combinations of channels, exploiting the signal's spread across multiple bins. The third level processes the phase difference using the phase spectrum between consecutive frames of the same channel, enabling detection of pulsed signals spread across time.

The system and method provide enhanced pulse detection capabilities, including low Signal-to-Noise Ratio (SNR) pulse detection with high probability of detection and low false alarm rates. The utilization of machine learning techniques for training the detection algorithm allows adaptability to various signal categories. The exploitation of available domain-wise information, such as spatial diversity, frequency, and time, enhances processing gain and overall detection performance. The system and method are applicable to random burst emissions without periodicity and can be implemented in real-time with minimal resource requirements. Additionally, the system can estimate time of arrival, frequency, bandwidth, and pulse width, providing valuable insights into the detected pulses. Furthermore, the system can identify closely spaced emitters in frequency with sufficient angular separation, facilitating accurate detection and classification of multiple simultaneous pulse signals.

Objectives:
To provide a pulse detection method and system that utilizes machine learning techniques to effectively classify signals based on their Phase, Time, and Frequency properties.
To reduce false alarms in pulse detection by exploiting the signal characteristics in multiple domains.
To enable range extension of existing ELINT and COMINT systems without requiring additional hardware upgrades.
To offer real-time signal processing capabilities using FPGA implementation for efficient and accurate pulse detection.

Applications:
Electronic Intelligence (ELINT) systems: The invention can be used in ELINT systems to accurately detect and classify pulses in various frequency bands, allowing for enhanced signal intelligence gathering.
Communication Intelligence (COMINT) systems: The system can be employed in COMINT systems to identify and analyze pulse signals, facilitating improved communication interception and analysis.
Radar systems: The invention can be utilized in radar systems for robust detection of radar pulses, enabling better target detection and tracking.
Wireless communication systems: The system can enhance pulse detection in wireless communication systems, improving the overall reliability and performance of the network.

Advantages:
Improved detection accuracy: The multi-domain approach, along with machine learning techniques, enhances the system's ability to accurately detect pulses in low signal-to-noise ratio (SNR) conditions.
Reduced false alarms: By considering the Phase, Time, and Frequency characteristics of the signal, the system significantly reduces false alarms and improves the reliability of pulse detection.
Compatibility with existing systems: The system can be seamlessly integrated into existing ELINT and COMINT systems, eliminating the need for extensive hardware modifications.
Real-time processing: The FPGA implementation enables real-time signal processing, allowing for efficient and timely pulse detection.
Enhanced range extension: The method and system extend the range capabilities of the existing systems, enabling detection and classification of pulses from a greater distance.
Versatile applications: The system can be applied to various domains such as ELINT, COMINT, radar systems, and wireless communication systems, offering flexibility and adaptability to different environments and signal types.

Brief Description of the Drawings

Fig. 1 shows a block diagram representing system hardware setup wherein an ultra wide band printed vivaldi antenna is used for transmission of test signal and two channel receiver hardware with receive antenna inside a dome, in accordance with an exemplary embodiment of the present invention.

Fig. 2 shows block diagram of the implemented system showing the major signal processing and statistical blocks, in accordance with an exemplary embodiment of the present invention;

Fig. 3 shows a visual representation of the magnitude spectrum obtained from 2000 frames, each consisting of 128 samples;

Fig.4 shows the phased difference variance plot. It illustrates a clear separation between the signal and noise floor, enabling high probability of detection even with a fixed threshold. This demonstrates the effectiveness of the system in accurately identifying pulse signals;

Fig. 5 depicts the weighted sum plot of the variance and magnitude outputs. The weighted sum combines the advantages of both the variance of phase difference across channels and the variance of phase difference across frames. The plot represents the final output used for signal detection.

Detailed Description of the Invention

The present invention discloses a multi-domain machine learning-based pulse detection method and system that utilizes the Phase, Time, and Frequency properties of the signal for accurate classification and detection of pulses. The system is designed to operate in challenging environments, where reliable detection and classification of pulse signals amidst high levels of background noise are crucial. The following detailed description, along with reference to the accompanying figures and experimental results, provides a comprehensive understanding of the invention.

In accordance with the exemplary embodiment of the present invention, the multi-domain machine learning-based pulse detection system, comprises:
a two-channel receiver hardware configured to receive signals in the frequency range of 400 to 6000 MHz;
a detection algorithm implemented in three levels, including a first, second, and third level of computing steps;
in the first level, summing the Fast Fourier Transform (FFT) data of all channels bin-wise to obtain a single vector;
in the second level, computing the phased difference among two phase spectrums for all combinations of channels and providing the variance output to a variance block to exploit the signal's spread across multiple bins in a given intercept;
in the third level, processing the phase difference using the phase spectrum of the FFT between consecutive frames of the same channel and providing the output to a statistical block for the final detection;
detecting the final signal by applying a weighted sum to the three levels of processed data.

In accordance with the exemplary embodiment of the present invention, the system can achieve low Signal-to-Noise Ratio (SNR) pulse detection with high probability of detection and low false alarm, even at SNR conditions as low as -12 dB.

In accordance with the exemplary embodiment of the present invention, the system comprises a data storage module configured to store the received signals and corresponding detection results for subsequent analysis and system optimization.

In accordance with the exemplary embodiment of the present invention, wherein the detection algorithm of the systemin the second level includes adaptive thresholding to dynamically adjust the variance threshold based on the signal characteristics and noise level.

In accordance with the exemplary embodiment of the present invention, wherein the detection algorithm of the system in the third level includes outlier detection techniques to identify and mitigate the impact of anomalous signals or interference.

In accordance with the exemplary embodiment of the present invention, wherein the two-channel receiver hardware of the system includes configurable RF front-end modules to accommodate different frequency bands and signal types.

In accordance with the exemplary embodiment of the present invention, the system comprises of a user interface module configured to display the detected signals, statistical analysis results, and system performance metrics in a user-friendly manner.

In accordance with the exemplary embodiment of the present invention, wherein the machine learning algorithm of the system used for training the detection algorithm includes deep learning techniques such as convolutional neural networks or recurrent neural networks.

In accordance with the exemplary embodiment of the present invention, the system can detect and classifying different types of pulse signals, including single pulses, multi-pulses, and pulse trains, based on their distinctive characteristics in the multi-domain analysis.

In accordance with the exemplary embodiment of the present invention, the system comprises of a signal localization module configured to estimate the direction of arrival (DOA) of detected pulse signals using techniques such as beam forming or angle of arrival (AOA) estimation.

In accordance with the exemplary embodiment of the present invention, wherein the system can adapt to dynamic signal environments by continuously updating the detection algorithm and weights based on real-time feedback and performance evaluation.

In accordance with the exemplary embodiment of the present invention, wherein the detection algorithm of the system includes preprocessing steps such as signal filtering, signal normalization, or signal enhancement to improve the accuracy of pulse detection in challenging scenarios.

In accordance with the exemplary embodiment of the present invention, the system involves a multi-domain machine learning-based pulse detection method, comprising steps of:
configuring a two-channel receiver hardware with ultra-wideband printed vivaldi antennas;
implementing a detection algorithm in three levels, including a first, second, and third level of computing steps;
in the first level, summing the FFT data of all channels bin-wise to obtain a single vector;
in the second level, computing the phased difference among two phase spectrums for all combinations of channels and providing the variance output to a variance block to exploit the signal's spread across multiple bins in a given intercept;
in the third level, processing the phase difference using the phase spectrum of the FFT between consecutive frames of the same channel and providing the output to a statistical block for the final detection;
detecting the final signal by applying a weighted sum to the three levels of processed data.

In accordance with the exemplary embodiment of the present invention, the system utilizing machine learning techniques for training the detection algorithm with simulated signal categories, enables improved accuracy and adaptability to different signal types.

In accordance with the exemplary embodiment of the present invention, the method utilizing all available domain-wise information, including spatial diversity, frequency, and time, to exploit processing gain using Fourier transform and correlation advantages, thereby enhancing the detection performance.

In accordance with the exemplary embodiment of the present invention, the method is applicable to random burst emissions without demonstrating any periodicity, enabling efficient detection and classification of pulsed signals with varying characteristics.

In accordance with the exemplary embodiment of the present invention, the method is suitable for real-time implementation, requiring minimal resources to enable the upgrade of existing systems with the proposed algorithm, minimizing the need for extensive hardware modifications.

In accordance with the exemplary embodiment of the present invention, the method is capable of estimating time of arrival, frequency, bandwidth, and pulse width with respect to signal quality, in comparison to traditional pulse detection techniques, providing valuable insights into the detected pulses.

In accordance with the exemplary embodiment of the present invention, the method is capable of identifying closely spaced emitters in frequency with sufficient angular separation, and vice versa, facilitating accurate detection and classification of multiple simultaneous pulse signals.

COMPONENTS AND OPERATIONAL FLOW
In accordance with the exemplary embodiment of the present invention, the system consists of the following components and operational flow:
Two-channel Receiver Hardware: The system incorporates a two-channel receiver hardware that is configured to receive signals in the frequency range of 400 to 6000 MHz.
Detection Algorithm: The detection algorithm is implemented in three levels of computing steps. In the first level, the Fast Fourier Transform (FFT) data of all channels is summed bin-wise to obtain a single vector representing the magnitude spectrum of the received signals. In the second level, the phased difference among two phase spectrums is computed for all combinations of channels. The variance output from this computation is provided to a variance block, which exploits the signal's spread across multiple bins in a given intercept. In the third level, the phase difference between consecutive frames of the same channel is processed using the phase spectrum obtained from the FFT. The output of this processing is provided to a statistical block for the final detection. The final signal is detected by applying a weighted sum to the three levels of processed data.
Achieving Low Signal-to-Noise Ratio (SNR) Pulse Detection: The system can achieve low SNR pulse detection with high probability of detection and low false alarm, even at SNR conditions as low as -12 dB.
Multi-domain Machine Learning-based Pulse Detection Method: The system utilizes machine learning techniques for training the detection algorithm with simulated signal categories. This enables improved accuracy and adaptability to different signal types.
Exploitation of Domain-wise Information: The method utilizes all available domain-wise information, including spatial diversity, frequency, and time, to exploit processing gain using Fourier transform and correlation advantages. This enhances the detection performance of the system.
Applicability to Random Burst Emissions: The method is applicable to random burst emissions without demonstrating any periodicity. This enables efficient detection and classification of pulsed signals with varying characteristics.
Real-time Implementation and Minimal Resource Requirements: The method is suitable for real-time implementation and requires minimal resources. It enables the upgrade of existing systems with the proposed algorithm, minimizing the need for extensive hardware modifications.
Estimation of Signal Parameters: The method is capable of estimating various signal parameters such as time of arrival, frequency, bandwidth, and pulse width with respect to signal quality. This provides valuable insights into the detected pulses, compared to traditional pulse detection techniques.
Identification of Closely Spaced Emitters: The method is capable of identifying closely spaced emitters in frequency with sufficient angular separation, and vice versa. This facilitates accurate detection and classification of multiple simultaneous pulse signals.

In accordance with the exemplary embodiment of the present invention, the system specifications include a frequency range of 400 to 6000 MHz, a two-channel configuration, ultra-wideband printed vivaldi antennas, receiver sensitivity of -110 dBm, a system dynamic range of 60 dB, instantaneous bandwidth of 100 MHz, and options for different FFT points (128/256/512/1024/2048).

In accordance with the exemplary embodiment of the present invention, the main components of the hardware include ultra-wideband antennas, a limiter + LNA, a wideband filter bank, and the two-channel receive hardware with sync signal generation. The system operation involves the generation of control signals for the wideband switch filter bank and the bypass amplifier based on the configured frequency. The received data is processed inside an FPGA using the proposed detection algorithm.

Now specifically referring to the drawings,
Fig. 1 illustrates the exemplary embodiment disclosing a block diagram representing system hardware setup (100), consisting of a two-channel receiver hardware (110) with ultra-wideband printed vivaldi antennas and a dome enclosure to protect the receive antenna from external interference. The system configuration enables the reception of signals for further processing and analysis. The received signal is converted into complex baseband data. The system configuration allows the reception of signals in the frequency range of 400 to 6000 MHz.

Fig. 2 depicts the exemplary embodiment system complete algorithm, consisting of three levels of processing (120, 130, 140), each designed to exploit different aspects of the signal. The first level (120) involves summing the FFT data of all channels bin-wise to obtain a single vector representing the magnitude spectrum of the received signals. The second level (130) focuses on computing the phased difference among two phase spectrums for all combinations of channels and providing the variance output to a variance block (135) to exploit the signal's spread across multiple bins in a given intercept. The third level (140) involves processing the phase difference using the phase spectrum of the FFT between consecutive frames of the same channel and providing the output to a statistical block (145) for the final detection.

As shown in Fig. 2, the complete algorithm consists of three levels (120, 130, 140) of processing, each designed to exploit different aspects of the signal.

In the first level (120) of processing, the FFT data of all channels is summed together bin-wise to obtain a single vector representing the magnitude spectrum of the received signals. This processing step provides a comprehensive representation of the signal power present in different frequency bins.

The second level (130) of processing focuses on the phase spectrum of the received signals. It involves computing the phased difference among two phase spectrums for all combinations of channels. This phase difference is then provided to a variance block, which exploits the property of the signal spread across multiple bins in a given intercept. By analyzing the variance output, the system can distinguish signals coming from a fixed direction, which exhibit minimal variance in phase difference among the channels, from additive white Gaussian noise (AWGN), which shows higher variance. This level of processing enhances the system's ability to detect and differentiate signals of interest from background noise.

In the third level (140) of processing, the phase difference among consecutive frames of the same channel is calculated using the phase spectrum obtained from the FFT. This phase difference is then provided to a statistical block, which further analyzes the variation of phase difference across frames. The processing in this level takes advantage of the property of the signal spread across time, where signals coming from a fixed direction exhibit consistent phase differences over time. By examining the variance of phase differences across frames, the system can detect the presence of pulsed signals, both narrowband and wideband.

The final signal detection is performed by applying a weighted sum to the three levels of processed data. The weights assigned to each level can be optimized using machine learning techniques, ensuring the best combination for achieving high probability of detection and low false alarm rates. The weighted sum provides a comprehensive evaluation of the signals, considering their magnitude, phase differences across channels, and phase differences across frames.

A threshold is applied to the weighted sum of the magnitude output, phase difference variance output across channels, and phase difference variance output across frames, as given in the equation below:

Final signal=(W1*Magnitude out)+(W2*phase difference variance out across channels)
+ (W3*Phase difference variance out across the frames)

The weights W1, W2, and W3 can be optimized using machine learning techniques.

Experimental results further demonstrate the effectiveness of the multi-domain machine learning-based pulse detection system and method. Experiment 1 involved setting up the system with specific parameters and evaluating its performance under controlled conditions. Table 1 presents the system specifications used in Experiment 1, including the ADC sampling rate, number of receive channels, transmit waveform characteristics, receiver signal SNR, FFT points, and variance window size.

Table 1: Experiment 1 System Specifications
Below are the specifications of the experimental setup
S. No Parameter Value
1. ADC sampling rate 30.72MHz
2. No of receive channels 2
3. Transmit wave form Pulsed FMCW waveform with bandwidth 5 MHz
4. Receiver signal SNR 0 dB
5. FFT points 128
6. Variance window size 8

Experiment 2 involved generating test signals with varying pulse widths and SNR conditions to evaluate the performance of the system. Table 2 presents the results of Experiment 2, indicating the probability of detection, false alarm rate, and other relevant metrics under different SNR conditions.

Table 2:
Below table gives the Pd and Pfa for different SNR conditions.
Pulse width given(usec) PRI given (usec) SNR(dB) False alarms Pulse miss Correctly detected pulses Probability of detection % False alarm rate
2 100 19.08 0 0 10000 100 0
2 100 14.65 0 0 10000 100 0
2 100 11.07 0 0 10000 100 0
2 100 8.058 0 0 10000 100 0
2 100 5.94 0 0 10000 100 0
2 100 3.21 0 0 10000 100 0
2 100 2.6 0 0 10000 100 0
2 100 1.82 0 0 10000 100 0
2 100 1.003 0 97 9903 99.03 0
2 100 -0.0137 0 97 9903 99.03 0
2 100 -2.589 4 97 9903 99.03 4.037e-4
2 100 -8.2 6 97 9903 99.03 6.055e-4
2 100 -10.88 5 97 9903 99.03 5.046e-4
2 100 -12.16 53 100 9903 99.03 0.0053
2 100 -14.16 51 1250 8750 87.5 0.00579
2 100 -16.16 75 1442 8558 85.58 0.00868
2 100 -18.16 26 2596 7404 74.04 0.00349

The results clearly demonstrate the high probability of detection achieved even at negative SNR conditions, with minimal false alarm rates.

Fig. 3 shows a visual representation of the magnitude spectrum obtained from 2000 frames, each consisting of 128 samples. Fig. 3 illustrates the output of the detection algorithm for Experiment 1. It can be observed that there is a small margin of variation between the signal and noise floor when using a fixed threshold or adaptive threshold. Achieving 100% signal interception without any pulse loss is not feasible in this case.

Fig. 4 presents the phased difference variance plot, which showcases a clear separation between the signal and noise floor, enabling high probability of detection even with a fixed threshold. This demonstrates the effectiveness of the system in accurately identifying pulse signals.

Fig. 5 depicts the weighted sum plot of the variance and magnitude outputs. The weighted sum combines the advantages of both the variance of phase difference across channels and the variance of phase difference across frames. The plot represents the final output used for signal detection.

The present invention offers several advantages in various applications. The system's machine learning-based approach enhances pulse detection accuracy, reducing false alarms and improving reliability. The system is compatible with existing ELINT and COMINT systems, eliminating the need for extensive hardware upgrades. Additionally, the FPGA-based real-time signal processing implementation enables efficient and timely pulse detection. The invention can be applied in domains such as ELINT, COMINT, radar systems, and wireless communication systems, providing versatile applications and improving overall system performance.
, Claims:CLAIMS
We Claim
1. A multi-domain machine learning-based pulse detection system (100), comprising:
a. a two-channel receiver hardware (110) configured to receive signals;
b. a detection algorithm implemented in three levels, including a first (120), second (130), and third (140) level of computing steps;
c. in the first level (120), summing the Fast Fourier Transform (FFT) data of all channels bin-wise to obtain a single vector;
d. in the second level (130), computing the phased difference among two phase spectrums for all combinations of channels and providing the variance output to a variance block (135) to exploit the signal's spread across multiple bins in a given intercept;
e. in the third level (140), processing the phase difference using the phase spectrum of the FFT between consecutive frames of the same channel and providing the output to a statistical block (145) for the final detection;
f. detecting the final signal by applying a weighted sum to the three levels of processed data.

2. The system (100)as claimed in claim 1, capable of achieving low Signal-to-Noise Ratio (SNR) pulse detection with high probability of detection and low false alarm, even at SNR conditions as low as -12 dB.

3. The system (100)as claimed in claim 1, comprises a data storage module configured to store the received signals and corresponding detection results for subsequent analysis and system optimization.

4. The system (100)as claimed in claim 1, wherein the detection algorithm in the second level (130) includes adaptive thresholding to dynamically adjust the variance threshold based on the signal characteristics and noise level.

5. The system (100) as claimed in claim 1, wherein the detection algorithm in the third level (140) includes outlier detection techniques to identify and mitigate the impact of anomalous signals or interference.

6. The system (100) as claimed in claim 1, wherein the two-channel receiver hardware includes configurable RF front-end modules to accommodate different frequency bands and signal types.

7. The system (100) as claimed in claim 1, comprises a user interface module configured to display the detected signals, statistical analysis results, and system performance metrics in a user-friendly manner.

8. The system (100) as claimed in claim 1, wherein the machine learning algorithm used for training the detection algorithm includes deep learning techniques such as convolutional neural networks or recurrent neural networks.

9. The system (100) as claimed in claim 1, capable of detecting and classifying different types of pulse signals, including single pulses, multi-pulses, and pulse trains, based on their distinctive characteristics in the multi-domain analysis.

10. The system (100) as claimed in claim 1, comprises a signal localization module configured to estimate the direction of arrival (DOA) of detected pulse signals using techniques such as beam forming or angle of arrival (AOA) estimation.

11. The system (100) as claimed in claim 1, capable of adapting to dynamic signal environments by continuously updating the detection algorithm and weights based on real-time feedback and performance evaluation.

12. The system (100) as claimed in claim 1, wherein the detection algorithm includes preprocessing steps such as signal filtering, signal normalization, or signal enhancement to improve the accuracy of pulse detection in challenging scenarios.

13. A multi-domain machine learning-based pulse detection method, comprising steps of:
a. configuring a two-channel receiver hardware (110) with ultra-wideband printed Vivaldi antennas;
b. implementing a detection algorithm in three levels, including a first (120), second (130), and third (140) level of computing steps;
c. in the first level (120), summing the FFT data of all channels bin-wise to obtain a single vector;
d. in the second level (130), computing the phased difference among two phase spectrums for all combinations of channels and providing the variance output to a variance block (135) to exploit the signal's spread across multiple bins in a given intercept;
e. in the third level (140), processing the phase difference using the phase spectrum of the FFT between consecutive frames of the same channel and providing the output to a statistical block (145) for the final detection;
f. detecting the final signal by applying a weighted sum to the three levels of processed data.

14. The method as claimed in claim 13, utilizes machine learning techniques for training the detection algorithm with simulated signal categories.

15. The method as claimed in claim 13, utilizes all available domain-wise information to exploit spatial diversity, frequency, and time for processing gain using Fourier transform and correlation advantage.

16. The method as claimed in claim 13, applicable to random burst emissions without demonstrating any periodicity.

17. The method as claimed in claim 13, suitable for real-time implementation, requiring minimal resources to enable the upgrade of existing systems with the proposed algorithm.

18. The method as claimed in claim 13, capable of estimating time of arrival, frequency, bandwidth, and pulse width with respect to signal quality, in comparison to traditional pulse detection techniques.

19. The method as claimed in claim 13, capable of identifying closely spaced emitters in frequency with sufficient angular separation, and vice versa.