In radar-based applications, Time Delay Estimation (TDE) is an essential criterion. Because of non-stationary behaviour, estimating the time delay between two turbulent signals is difficult. Existing delay estimation methods such as the cross correlation technique are restricted to stationary signals. The non-stationary signals are either fractal or periodic signal. The accuracy of this method is more reliable for fractal signals than for periodic signals. With a cost function at hand it is sensible to check whether the state correction results in a cost decrease in the first place, new parameter is optimized using Fuzzy Elephant Herding Optimization (FEHO). Further this paper incorporates ADAM based neural network (ADAM-NN) model for efficient time delay estimation. The study resulted in significant improvement upto 21.5% in estimating the time delay when compared with conventional methods.
Research Article
Development of Time Delay Estimation Algorithm Using Fuzzy Based Optimized Iterative Unscented Kalman Filter
https://doi.org/10.21203/rs.3.rs-920674/v1
This work is licensed under a CC BY 4.0 License
You are reading this latest preprint version
Continuous Wavelet Transform
ADAM Neural Network
Cancellation ratio
False alarm
The test of deciding the time delay between two got signals that came from a similar transmitter is called Time Delay Estimate (TDE). [1] Depending on the idea of the applications, TDE can be isolated into two classifications: Time of Arrival (TOA) assessment and Time Difference of Arrival (TDOA) estimation. Matching channels are utilized to gauge the time postponement to guarantee target acknowledgment by connecting the got reverberation with a reference signal known as TOA assessment. When there is no such reference signal, the gauge is regularly created by contrasting the signs got from at least two spatially far off radar sensors, an interaction known as TDOA. This work utilizes reenactments and fleeting recurrence portrayals to look at the issues related with TDE of coordinated separating based objective distinguishing proof.
Time-delay estimation is an important research aspect in the signal processing domain, and is used to determine the time offset of some signal compared with the referenced signal [3]. For example, one of the important applications of wireless sensor networks is localization, and to locate an object accurately in the wireless sensor networks, the measurement based on time-delay is the most ideal method. The lower the time resolution is, the more distance error it would make. The time-delay estimation's major goal is to increase the time resolution.
The wide-band signal has a stable performance to overcome the multi-path interference, so the estimation algorithms based on the wide-band signal are concentrating on the time-delay between two signals in single path. Algorithms based on the pattern matching are the most common for the time-delay estimation of wide-band signal, in which the matching degree of two signals is the function of the time-delay difference [4].
However, the time precision of all these algorithms are determined by sampling interval [2]. In order to improve the precision, two methods are incorporated, one is on the basis of the interpolation technology and the other is of the fractional delay filter technology. The wide-band signal has the capability to combat the multi-path interference, the algorithms based on wide-band signal assumes that the multi-path signals have been separated perfectly. The trend of these algorithms are concentrating on how to improve precision of the time-delay further which is affected by the sampling frequency [5].
Numerous researchers have made various ways to deal with manage the issue of TDE during the last two decades[9]. In moderate clamor, the summed up cross-connection approach for TDE is utilized to appraise the deferral between two sensors, wherein the delay that boosts cross-relationship across separated duplicates of the got signal is used. Sun et al. proposed refreshing the multichannel cross-connection coefficient approach and assessing time delay in many channels utilizing straight insertion methods. At the point when the info signal is white, Xu et al. concocted a versatile TDE approach dependent on the most un-mean square calculation for fixed and moving sources. Chen et al. fostered a LS calculation for assessing the appearance seasons of covering multipath echoes from a loud and deferred copy of a known sign, where two mistake capacities, genuine adequacy blunder capacity and complex plentifulness blunder work, are utilized to get worldwide minima for abundancy lessening and time delays for various multipaths. A TDE normal greatness contrast work was provided by Yuan et al (AMDF). The AMDF is a variation on autocorrelation investigation where, rather than relating the information signal at different postponements, a distinction signal is created between the deferred discourse and the first sign. TDE is performed utilizing an Autocorrelation Estimator (AE) without earlier data of the channel. The TDE strategies for Maximal Likelihood Estimation (MLE) and Alternative Estimation (AE) were assessed, and MLE beat AE in high sign to commotion proportion (SNR) circumstances, while AE outflanked MLE in low SNR settings. The signs are parted and the time delays are registered more than once in each sub-band, as indicated by Li et al., utilizing two unique adaption strategies that limit the means squared mistake (MSE). Luo et al. suggest utilizing a versatile time-defer assessment strategy that joins both the M-band Wavelet Transform (MDWT) and Projection Cross Correlation (PCC) to upgrade target acknowledgment in low sign to commotion proportion (SNR) or high clamor levels. The MDWT is utilized to create various duplicates of the signs at various sub-groups, which works on signal decorrelation, while PCC is utilized to decrease Gaussian clamor. With little expense, Akcakaya et al. presented a cross-uncertainty work (CAF) approach dependent on higher-request vagueness work (AF) for foreseeing time delay, Doppler shift, and Doppler pace of moving items in detached radar.
Timedelay calculation is a crucial research area in the signal processing domain, and it is commonly used to calculate the time offset of a signal as compared to reference signal [10]. Since wide-band signals have a strong ability to resolve multi-path interference, estimation algorithms based on wide-band signals focus on the time-delay between two signals in a single path. Pattern matching algorithms are the most popular for time-delay estimation of wide-band signals, in which the degree of matching between two signals is a function of the time. Standard approaches based on the Fast Fourier Transform (FFT) method is unable to split the signals when the difference in arrival time is smaller, necessitating the super-resolution of time-delay estimate of the narrow signal.
$${x}{k}\left(n\right)={\alpha }{k}s\left(n-{n}{k}\right)\text{exp}\left(j{{\Omega }}{k}n\right)+{\omega }_{k}\left(n\right)$$
1,
A sign's CWT can be represented as the link between that sign and any wavelet work known as the mother wavelet. For disturbance signal analysis, a few mother wavelets are available. The Morlet wavelet is the most appropriate mother wavelet for examining choppiness signals. In the study of local periodicity of signals, the Morlet wavelet has proven to be quite effective. The accompanying condition can be used to convey the Morlet wavelet:
$${\psi }\left(\text{x}\right)=\text{C}{\text{e}}^{-{\text{x}}^{2}/2}\text{c}\text{o}\text{s}\left(5\text{x}\right)$$
2
The CWT of a sign x(t) may be obtained by interpreting the Morlet wavelet on the time (t) hub by scale "b" with fixed y hub and expanding by "a" as shown in the following condition:
$$\text{W}\left(\text{a},\text{b}\right)=\frac{1}{\sqrt{\text{a}}}\underset{-{\infty }}{\overset{{\infty }}{\int }}\text{g}\left(\text{x}\right){{\psi }}^{\text{*}}\left[\frac{\text{x}-\text{b}}{\text{a}}\right]\text{d}\text{x}$$
3
The second stage in determining the CWT coefficients is to use Fast Fourier Transform to modify the chosen mother wavelet and sign in recurrence space (FFT). By repeating the FFT of the signs generate implications. By performing an inverse FFT on the enhanced results, CWT coefficients for the main scale are recorded. A few scales can be obtained by repeating these methods. The FFT may be calculated using:
$$\text{S}\left(\text{x}\right)=\sum _{\text{n}=0}^{\text{N}-1}\text{s}\left(\text{n}\right){\text{e}}^{-\text{j}\frac{2{\pi }}{\text{N}}\text{n}\text{x}}$$
4
The inverse FFT may be calculated using the following formula:
$$\text{S}\left(\text{n}\right)=\frac{1}{\text{N}}\sum _{\text{n}=0}^{\text{N}-1}\text{s}\left(\text{x}\right){\text{e}}^{-\text{j}\frac{2{\pi }}{\text{N}}\text{n}\text{x}}$$
5
where S(x) denotes the time domain signal and s(x) denotes the converted signal.
For assessing the last time delay from assessed delays between CWT scales, three factual strategies dependent on the mean of assessed delays were utilized[11]. Practically speaking, a bunch of time postponement and recurrence shift in one of the auxiliary channels is chosen as a source of perspective set. Notice that the distinctions in the time postponement (or recurrence shift) between various auxiliary signals, rather than the first run through deferrals (or recurrence shifts) in totally got signals, are of interest.
The reference signal s(n) is sine wave with a period of ωo rad/s and it is given by,
$$s\left(n\right)=\text{sin}\left({\omega }_{0}n\right)$$
7
The reference signal s(n) is also a sine wave with a recurrence of ωo rad/s that is tested with a T-second examining time. The secluded correlation signal r(n) is,
$$r\left(n\right)=\alpha s\left(n-d\right)+w\left(n\right)$$
8
Let s(n) and r(n) be the vector sorts of the reference and main signals, individually, in the two information windows. Since, the auxiliary signal is noisy, it is ideal for the channel to accomplish a more exact estimate of the unsure time delay.
The two signals were shifted with a fourth order type II Chebyshev band pass optical channel. The channel will have a tight pass data transmission of 0.01 rad/sec and a 1 dB lessening for the two pass frequencies, just as a stop band transfer speed of 0.22 radians and a 30 dB weakening for the low freqeuncies. The bandpass channel recurrence, ωc, is equivalent to the mathematical mean of the pass and stop frequencies, and reference signal recurrence, ωo. The bandpass channel produced by the exacting limited data transfer capacity condition was equipped for decreasing the information noise capacity to around 1.6% of its worth at the channel yield. This relates to a decreasing noise factor of 62.5, inferring a 18-dB expansion in SNR.
$${\sigma }_{y}^{2}=\frac{1}{2\pi }\underset{-\pi }{\overset{\pi }{\int }}{\sigma }_{w}^{2}{\left|H\left(w\right)\right|}^{2}d\omega$$
9
The bandpass filter's frequency response H(ω) was obtained using the MATLAB. The performance signal variance σy2 was measured using Matlab and found to be 0.016.
An artificial neural network (ANN) is a computer simulation of the human brain in its most basic form. A normal brain is capable of learning new ideas and adapting to an evolving world[6]. The most basic animal's brain has more computing power than the most modern robot. It aims to monitor not only the physical parts of the body, but more abstract tasks such as thought, visualizing, dreaming, imagining, studying, etc, which cannot be physically defined. Artificial intelligence is now beyond the capabilities of all the sophisticated computers.
Until adding the sign to NN, the bandpass channel is indicated with noisy signal and increases the SNR[14]. The reference signal s(n) is separated by a similar channel with a goal to have a similar impact as shifted signal. In any case, for the indistinct consistent time-postpone esteem D and the damping factor, the two signs are presently coordinated. The two shifted signals are standardized regarding the most elevated qualities they get to limit the impact of the damping component. The NN is shielded from being affected by shifting many changes of the signs because of different and obscure damping factors by normalizing the shifted signals. The standardization interaction has little impact on the sign to-noise ratio, which is a basic determinant of the exactness of the time-delay assessment[12]. The NN ought to have the option to give dependable time-defer estimations by adding the shifted and standardized motions toward it. The NN would give time-postpone estimations that are a long way from solid, if the desired signals are utilized with all things that are being equal. The corresponding outputs are
$${s}_{f}\left(n\right)=\sum _{k=0}^{n}h\left(k\right)s\left(n-k\right), n=0,\dots \dots N-1$$
10
$${r}_{f}\left(n\right)=\sum _{k=0}^{n}h\left(k\right)r\left(n-k\right), n=0,\dots \dots N-1$$
11
As a result of the bandpass channel's separating stage, the two shifted signals have a transient impact. The NN is unaffected by the transient impact on the two shifted signals in spite of the fact that the steady time delay between the two separated signals stays consistent and isn't altered by the separating step[13].
The NN gets the two separated and standardized signs in three particular manners. The two signs are connected together to frame the information vector in the initial step, known as the equal information structure. The single information structure is the third sort, which utilizes just the separated and standardized signal as the NN input. The NN input doesn't have the reference signal. The time delay is installed in the reference signal, which brings about this present strategy's support.
Around 1,000 informational collections with relating time-defer values were added to the ADAM-NN as testing models during the preparation cycle. The time-postpone esteem was produced indiscriminately for each preparation model by a uniform irregular generator. Expect real qualities going from 0.0 to 0.5 seconds for the time delay. The examining stretch was accepted to be T = 0.05 seconds in length. As a result, time delays going from 0 to 10 inspecting stretches are created. A uniform irregular generator was utilized to create the damping factor, which could be of worth between the range of 0.25 and 1. The mutilating noise applied to the deferred signal is Gaussian with standard deviation esteems which was chosen indiscriminately by a uniform arbitrary generator. The recommended NN learning subordinate time delay assessment technique is resolved and the model that is considered for the research work needs more memory to rehearse. Moreover, since the neural organization is enormous, the gadget would affect the time postpone assessment proficiency. For introductory preparation, the size and memory of the neural organization are more fundamental. Subsequently, the work begins with a prologue to ADAM NN.
In this paper, we present ADAM, a stochastic streamlining method that just requires first request slopes and burns-through little memory. The methodology utilizes assessments of the first and second snapshots of the inclinations to work out interesting versatile learning rates for particular boundaries; the name ADAM comes from Adaptive Moment Estimation.
Consider a scenario where a tracking radar is deployed with a frequency of 3GHz, pulse bandwidth is 20 MHz, pulse width is 1000µs, pulse repetition frequency is 1 kHz, antenna height is 75m, antenna tilt angle is 10° with horizontal polarization, probability of false alarm is 1e− 06, cell averaging constant false alarm rate is preferred with 200 reference cells with swerling 1 target model. This scenario is created using radar toolbox in MATLAB.
Figure-2 shows the most recent postpone an incentive for the intermittent sign, which was calculated using the mean of assessed delays, the most successive, and the largest defer esteem. The postpone error remained consistent at higher scales from 10 to 64. In any event, assessments based on the most frequent postponements are more precise than those based on the mean or the biggest. In addition, on a scale of 1 to 10, the assessment based on the most deferrals is superior to the one based on the most consistent postponements. Taking the mean worth in the lower scales area resulted in an incorrect deferral.
The shot at location for different items with changing reach goal is portrayed in Fig. 3. As the reach goal between the objectives ascends from 200 to 300 mW, the identification chance improves from 70–98% for single and two objective circumstances.
As shown in Fig. 3, the last defer an incentive for the intermittent sign was obtained by combining the mean of assessed delays, the most incessant, and the greatest postpone esteem. The postpone error remained consistent at higher scales from 8 to 48. However, an estimate based on the most frequent postponements is more precise than one based on the mean or the greatest. Furthermore, on a scale of 1 to 10, the assessment based on the most postponements is superior to the assessment based on the most consistent deferrals. Taking the mean worth in the lower scales area yielded an inaccurate postponement.
Table 4.2 Comparison of Various time delay estimation algorithms based on standard deviation
|
SD=0.1 |
Time Delay Estimation Algorithms |
||||
|
Time Delay (s) |
LS |
PCC |
AMDF |
MDWT |
TDE-ADAMNN |
|
0.1 |
62 |
57 |
54 |
41 |
34 |
|
0.11 |
68 |
69 |
67 |
57 |
42 |
|
0.12 |
79 |
74 |
71 |
59 |
49 |
|
0.13 |
84 |
81 |
78 |
62 |
51 |
|
0.14 |
91 |
86 |
81 |
67 |
57 |
|
0.15 |
94 |
88 |
87 |
71 |
62 |
|
SD=0.3 |
Time Delay Estimation Algorithms |
||||
|
Time Delay (s) |
LS |
PCC |
AMDF |
MDWT |
TDE-ADAMNN |
|
0.1 |
57 |
54 |
41 |
34 |
31 |
|
0.11 |
69 |
67 |
57 |
42 |
39 |
|
0.12 |
74 |
71 |
59 |
49 |
47 |
|
0.13 |
81 |
78 |
62 |
51 |
53 |
|
0.14 |
86 |
81 |
67 |
57 |
59 |
|
0.15 |
88 |
87 |
71 |
62 |
69 |
|
SD=0.5 |
Time Delay Estimation Algorithms |
||||
|
Time Delay (s) |
LS |
PCC |
AMDF |
MDWT |
TDE-ADAMNN |
|
0.1 |
34 |
31 |
30 |
29 |
27 |
|
0.11 |
42 |
39 |
33 |
30 |
29 |
|
0.12 |
49 |
47 |
41 |
37 |
31 |
|
0.13 |
51 |
53 |
50 |
46 |
42 |
|
0.14 |
57 |
59 |
54 |
48 |
46 |
|
0.15 |
62 |
69 |
62 |
52 |
50 |
The proposed research work estimated efficient time delay and conventional time delay of two non-stationary signals with different levels of SNR. The results infer that ADAM based NN model provides efficient time delay estimation and high probability of detection in the presence of false alarms. The results are compared with various time delay estimation techniques for various levels of standard deviation. The ADAM NN model provides a significant improvement in detection probability around 39%.
Ethics approval and consent to participate: Not applicable.
Funding: Not applicable.
Conflict of interests: The authors declare that they have no conflict of interests.
Informed Consent:Not applicable
Authors' contributions: All authors discussed the results and implications and commented on the manuscript at all stages. All authors read and approved the final manuscript for publication.
[1]R. E. Boucher and J. C. Hassab, "Analysis of discrete implementation of generalized cross correlator", IEEE Trans. Acoust. Speech Signal Processing, vol. ASSP-29, no. 3, pp. 609-611, June 1981.
[2] G. Su and M. Morf, “The signal subspace approach for multiple wide-band emitter location,” IEEE Trans. Acoust., Speech, Signal Processing, vol. ASSP-31, pp. 1502–1522, Dec. 1983.
[3] Jing Yi Lu, Dong Ye and Wen Ping Ma, “Time Delay Estimation based on Variational Mode Decomposition” Advances in Mechanical Engineering Vol.9(1) pp.1-6, 2017.
[4] U.K Singh, A.K Singh, V.Bhatia and A.K.Mishra,“EKF and UKF based Estimators for Radar System” Frontiers in Signal Processing Vol.1,August 2021.
[5] Meng Sun, Yide Wang, Cedric Le Bastard, Jingjing Pan and Yuehua Ding, “Signal Subspace Smoothing Technique for Time Delay Estimation using MUSIC algorithm” MDPI Sensors Vol.17 pp.1-12, 2017.
[6] Xiangzhi Li, Weijia Cui, Haiyun Xu, Bin Ba and Yankui Zhang,”A Novel Method for DOA and Time Delay Joint Estimation in Multipath OFDM Environment”, International Journal of Antennas and Propagation, pp.1-11, 2020.
[7] Hyung Rae Park and Jian Li, “A Frequency Domain SPICE approach to high resolution time delay estimation” IEEE Wireless Communication Letters Vol.7(3), pp.360-363, 2017.
[8] Huaicheng Yan, Xuping Zhou, Hao Zhang, Fuwen Yang, and Zheng-Guang Wu, “A Novel Sliding Mode Estimation for Microgrid Control with Communication Time Delays”, IEEE Transactions on Smart Grid, 2017.
[9] Lei Zou, Zidong Wang, Qing-Long Han and Donghua Zhou,” Moving Horizon Estimation for Networked Time-Delay Systems under Round-Robin Protocol”, IEEE Transactions on Automatic Control, 64(12), 5191–5198.
[10] Sun, M., Bastard, C. L., Wang, Y., Pan, J., & Pinel, N. (2018). A Modified Min-Norm for Time Delay and Interface Roughness Estimation by Ground Penetrating Radar: Experimental Results. IEEE Geoscience and Remote Sensing Letters, 1–4.
[11] Mass, S., & Laukner, M. (2020). Ultrasonic Time Delay Difference Estimation With Analytic Signals and a Model System. IEEE Transactions on Circuits and Systems II: Express Briefs, 67(10), 2234–2238.
[12] Rao, X.; Tao, H.H.; Su, J.; Xie, J.; Zhang, X.Y. Detection of Constant Radial Acceleration Weak Target viaIAR-FRFT. IEEE Trans. Aerospace Electr. Syst. 2015, 51, 3242–3253.
[13] Li, X.L.; Kong, L.J.; Cui, G.L.; Yi, W. A low complexity coherent integration method for maneuvering targetdetection. Dig. Signal Process. 2016, 49, 137–147.