Almost every installation of cognitive IEEE-802.X wireless
communication standards [1] which are designed and
implemented for Supervisory Control and Data Acquisition
(SCADA) based Digital Wireless Wide Area Networks
(SCADA-DWWAN) including Industrial Control Network
standards (DCS-Net) suffer from both co-channel interference
(CCI) as well as bandwidth congestion for most of the multiple
user stations or multi-access orthogonal transceiver stations
[1], [16], [12]. Undiversified relaying, manual discrepancies
in the network resource management as well as
channel equalization errors lead to congestion and latency in
secondary channels [1], which causes disruption and
biasing in the utilization of the primary channel
bandwidth, leading to spectrum losses [16]. Full-duplex
switched-mode transmissions [11], [15] suffer the worst
bandwidth discrepancy [1], [15] which occurs due to relay-
buffer losses, bandwidth congestion and latency [1], [11],
[15]. Field studies [1], [7] and simulation tests [16] show
that ‘proper sensing and allocation’ of sub-band resources
in the Physical layer, Data-Link layer as well as Network
layer of the different ‘secondary station users’ in each of the
multiple access primary channels (PHY,MAC,FTP)
using ‘cooperative’ and ‘adaptive’ spectrum sensing and
sharing is likely to help mitigate the spectrum deficiencies
caused by congestion which will collaboratively enhance the
stability and fidelity of the network system. Although
conventional artificial intelligence based models including
machine learning prototypes [1], [16] as well as deep
learning neural network prototypes [16] and numerous
convolutional neural network prototypes [16] have been
designed, proposed and virtually tested in real time
simulations, yet static convergence errors as well as
dynamic decision errors show high rates of false-alarm and
heuristic lags especially ranging over high frequency
operational bands (350Mhz - 480Mhz) [1], [2]. Since
industrial supervisory and control networks mostly focus
and rely upon precise controllability as well as high
observation accuracy, hence Artificial Neural Networks
(ANN’s) utilizing adaptive learning, training and decision
models are potentially likely to provide reliable solutions
[1][16]. In our study, we have tested two popular ADALINE
(Adaptive Linear Neuron) [6], [9], [13] based neural
network models for application in ‘spectrum sensing’ [5] as
well as ‘spectrum allocation’ [12], [15] which are
operationally semi-supervised, dynamic decision models and
Adaptively Equalized Bandwidth Optimization Model using SCADA-
DWWAN based Neural Network
1PRIYANKO RAJ MUDIAR, 2KANDARPA KUMAR SARMA, 3NIKOS MASTORAKIS
Abstract: Artificial training and learning algorithms, enhanced with semi-supervised or self-supervised feature extraction capacities,
employ adaptive decision optimization models. These are often favored over complex deep learning algorithms for achieving better
controllability and ease of observation, lower complexity in simulating, building or designing and virtual prototyping of automatic
network resource management (ANRM) standards. An Adaptive Linear Neuron type Artificial Neural Network (ADALINE-ANN)
which is based on multi-tapered machine learning approach has been simulated in a virtual Supervisory Control and Data Acquisition
(SCADA) framework integrated with a Distributed Control System (SCADA/DCS-Net). The system has been virtually simulated
considering an adaptively equalized learning and decision approach which utilizes Markov Trained-Steepest Gradient Descent
(HMM-SGD) based machine learning model employing Kalman optimization. Affinity clustering is employed for spectrum sensing
by extracting the Constellation Nyquist Bands from an M-Quadrature Amplitude Modulation (QAM) orthogonal signal undergoing
AWGN and Rayleigh fading as well as co-channel interference (CCI), and ensemble analysis using Channel State-Space plots are
used for optimal spectrum allocation in an Adaptive Orthogonal Frequency Division Multiple Access (Adaptive-OFDMA) layout.
It has been done by implementing an adaptively equalized Automatic Repeat Request (ARQ) pipelining model which utilizes
minimum least square error (MLSE) minimization model. The objective is to improve the bandwidth allocation and usage ensuring
most minimum spectrum wastage or loss. Successive Interference Cancellation (SIC) has been implemented to minimize static
buffer and interference loss. Thus, spectrum loss due to latency and jitter which occurs from bandwidth congestion is minimized by
improving the network resource tracking and allocation. It results in improved and stable bandwidth equalization.
Keywords: 256-Quaternary Amplitude Modulation(256-QAM), ADALINE, Kalman-Least Mean Square Algorithm (Kalman-
LMS), Kalman optimized Kernel Recursive Least Square Algorithm (Kalman-KRLS), Affinity Propagation Clustering (AP-
Clustering), Maximum Correntropy, ANOVEE, Eigen-Plot, Fourier-Bessel Transform, Huang-Hilbert Transform, Hidden Markov
Model (HMM), Steepest Gradient Descent (SGD), Adaptive Automatic Repeat Request (Hybrid-ARQ), Maximal Ratio
Combination Channel Diversity (MRCD), Successive Interference Cancellation (SIC).
Received: May 24, 2021. Revised: July 6, 2022. Accepted: August 8, 2022. Published: September 19, 2022.
1. Introduction
WSEAS TRANSACTIONS on ELECTRONICS
DOI: 10.37394/232017.2022.13.14
Priyanko Raj Mudiar, Kandarpa Kumar Sarma, Nikos Mastorakis
E-ISSN: 2415-1513
107
Volume 13, 2022
1, 2 Department of Electronics & Communication Engineering Gauhati University, Guwahati, INDIA
3Technical University, Sofia, BULGARIA.
have been formulated and simulated in an
IEEE-802.22x SCADA-DWWAN based virtual transmission
environment [1]. A novel Hidden Markov type stochastic
gradient based training model [3] has been infused upon
the standard “Widrow-Hoff Learning Method”
implemented over Steepest Gradient Descent / Ascent (or
Delta-learning) Rule using tuned kernel widths [13],
thereby providing high dynamic learning adaptability
as well as improved convergence during semi-supervised
regression. To ensure good controllability and equalised
error interpolation, two novel Kalman optimization models
[6], [9] have been implemented upon a Least Mean Square
(LMS) based spectrum sensing neuron [5] as well as a
Kernelized Recursive Least Square (KRLS) based spectrum
decisive neuron [12], [15]. An “affinity detection
and clustering” based feature extraction approach [2], [4] is
utilized for the sensing “or extraction” of Sub-Band Nyquist
clusters from a spatial Constellation sequence [10] carrying 8-
bit / 256-Quaternary Amplitude Modulation signals over an
orthogonal multiple access transmission channel [11], [12],
[15]. Applying a maximally correlative entropy equalization
[17], [13] concept, or “maximum-correntropy” condition,
an adaptive channel diversity [11], [12], [15] based relaying
and pipelining model is implemented which trains and
decides the sub-channel spectrum allocation margins
(or “Shannon sub-bands”) according to an Eigen-space
MLSE (Minimum Least Square Estimation) minimization
approach [8], [17], known as “Perron-Frobenius” model [17].
Thus, channel detection is achieved by identifying the
cumulative entropies as well as mass ensemble distributions
(CMF) in terms of some frequency convolved Energy
Spectrum Distributions (ESD) by sequentially
identifying and approximating the envelope side-band levels
using Fourier Bessel Transforms and Huang-Hilbert
Transforms [14] upon the sequentially identified Nyquist
clusters with our proposed Markov-ADALINE type KRLS
Neuron. An adaptively equalized routing prototype employs an
Automatic Repeat Request (ARQ) based switch [15], [17],
[22] over an adaptive orthogonal frequency division multiple
access (OFDMA) virtual channel [7], [11], [15] for dedicated
downlink relaying by estimating and aggregating the Maximal
Signal-to-Noise Ratio (SNR) for each orthogonal sub-user
links with a proposed ANN approach [3] against the
minimum Bit Error Ratio (BER) paths by normalizing and
mixing the Maximal sub-channel SNR Gains [7], [15]
over each route using a Maximal Ratio Combination
approach. A semi-supervised approach has been
formulated and proposed by authors [11], [12] by
updating the cost transition vectors using some ergodic
Lagrangian determinants. This concept is being virtually
implemented by our proposed Steepest Gradient Ascent/
Descent based Hidden Markov Determinants (or
Liapunov State Minimization Determinants, “LSMD”) with
respect to the ergodically but asymmetrically distributed
“Perron-Frobenius” Eigen State Space Reduction (Eigen-SSR)
Model [3], [17].
Finally, generic training and learning performances have
been obtained, visualized and then virtually tested with
reference to the overall improvement in bandwidth equalization
[17], [20] as well as spectrum sensing [8] over the
entire multiple access transmission channel [1], [15].
In this work, we report the design of an ANN based
spectrum sensing and allocation system designed for a virtual
SCADA-DCS framework. The design considers an adaptive
Automatic Repeat Request (ARQ) mechanism in support of an
Adaptive OFDMA network layout, as depicted in the block
diagram in Fig. 1. Furthermore, the system uses Successive
Interference Cancellation (SCI) which provides improved
interference cancellation for bandwidth optimization [12].
2. Proposed Work
Fig. 1. Block Diagram for Kalman-SGD/SGA ADALINE
2.1 Spectrum Sensing
Constellation sequences which are obtained from a
synchronous rake receiver channelling 8-bit octal words
encoded over 256-Quaternary Amplitude Modulation (256-
QAM) carrier signals are fed to a Kalman-optimized Least
Mean Square ADALINE (LMSE) [6], to obtain the stochastic
weights identifying each constellation cluster sequence using
Steepest Gradient Descent rule (SGD), as shown in the block
diagram in Fig. 1. A novel Hidden Markov training model [3]
is implemented to optimize the sequencing and interpolation
speeds as well as minimize the convergence errors while
sampling the SGD training samples (or batches) for
improvising the convergence speeds and regression efficiency
of the SGD model. Thus, the ‘Nyquist clusters’ are identified
and extracted from the constellation sequences by employing a
novel Affinity Propagation [2], [4] clustering method
which detects the maximum availability measures (or
maximum similarity measure between auto-correlation
and auto-covariance) of each sequences between each
constellation points [10] in order to identify and
‘automatically link the constellation points’ into ‘groups’
of separate sequential clusters or ‘common correntropy
symbols’ carrying similar information, known as ‘Nyquist-
Bands’. By implementing an interpolative Fourier-Bessel
Transform upon the identified and extracted Nyquist-clusters
followed by Hilbert Transform using the novel Markov
Trained SGD model under Gaussian Q-function limits, the
gradient of the individual energy spectrum densities
(ESD) or entropy transition levels corresponding to
each of the ‘ergodically identified’ or ‘sensed’ clusters
are being detected and are converged to identify the
number of ‘channel transitions’ corresponding to each of the
discrete energy levels in terms of cumulative mass function
distribution (CMF). This detected ensemble
WSEAS TRANSACTIONS on ELECTRONICS
DOI: 10.37394/232017.2022.13.14
Priyanko Raj Mudiar, Kandarpa Kumar Sarma, Nikos Mastorakis
E-ISSN: 2415-1513
108
Volume 13, 2022
distribution corresponds to the aggregate Secondary User sub-
bands against each Primary Channel User.
Fig. 2. Generic logic flow model of ADALINE
Let us consider the standard logic flow model of our
proposed ADALINE Neuron as depicted in Fig.2.The Widrow-
Hoff interpolation levels obtained for a parabolic sample signal
implementing Steepest Gradient Descent (SGD) can be
approximated as given in Fig.3. Let us consider the general
MQAM signal entropy function as given by,-
Φ
Fig. 3. General pattern of Newtonian Interpolation for SGA/SGD
Where,
is the Minimum amplitude or lowest entropy
signal and ( , )
are the Eigen co-factors of any random
signal constellation point determined according to the
respective coordinate location (Euclidean distance from each
local centroids) . Let us consider a norm L = , for any M-
coded Orthogonal QAM Signal (here, M = 256).
The local centroids corresponding to each individual
constellation Nyquist Cluster can be determined with the ith
message points cardinally located as the binary tuple vector
( , ), where Eigen co-factors are an element of the
{LxL} matrix given by ,-
( , ) =
(2)
Let us consider the following typical S-T Windowed
Constellation [16] , as shown in Fig. 4, standard for any M-
QAM Signal undergoing AWGN Noise and Rayleigh Fading,
given by authors [16], for (xm, yn) tuples , where h = (x, y)
denotes the I/Q coordinates of the 256-QAM / 8-bit Signal , -
Fig. 4. S/T Windowed Constellation for MQAM Signal [16]
The constellation In-Phase / Quadrature (I/Q) coordinates
can be expressed as given by authors [10] , -
(xm , yn ) = [(2n-1) +1- ].d + [2.mod(m , 2) -1]. (
)
-[2(m -1) + 1 - ]. (
) (3)
where , d
is the cluster point Euclidean distance from local
centroids; and (m, n)
[1, …… , ] ;
2.2 Spectrum Allocation
Corresponding to the CMF Levels, the signals extracted
from each of the detected 256-clusters are being weighted by a
Maximum Likelihood Sequence Estimator (MLSE) according
to the individual aggregate ratios between Signal Power to
Noise Power ratios (SNR) as well as Bit Error Rates (BER) for
each ‘identified’ cluster [2], [8], [7]. Then, a clocked
register phase-normalizes each of the serial cluster
packets by co-phasing them (using a sequential clock
synchronous phasing circuit) and feeding them serially
to an ARQ assembly controlled by a Kernelized [9], [13]
Recursive Least Square (KRLS) ANN. Further, 20-virtual
users comprising of 16-secondary users and 4-primary
channel users have been implemented for parallel signal
pipelining. The optimum transmission and reception
capacity corresponding to each channel sub-user is
determined using the delay information with respect to the
maximal diversity ratios, or SNR gains mixed with
minimum BER paths [7], [15], which is identified and
approximated from the envelope energies (ensemble
entropies) by the Eigen-Space Minimization [17][8]
model using HMM determinants [3], prior to each
acknowledgement and non-acknowledgement key (ACK/
NACK) [11][12]. Thus, using sequential/regressive training of
the switching rates using a novel ‘Lagrangian-
Determinant’ [12] for a Markov-trained Hybrid-ARQ [3]
[17], average Shannon Bands (sub-bands) are being
determined corresponding to each primary/secondary
station groups to achieve optimized sharing of the
bandwidth while ensuring minimum buffer wastage as well
as minimum critical latency. Hilbert frequency transforms [14]
are used and learning curves are characterized for
evaluating the performance and fidelity of our proposed
architecture.
3. Proposed Methodology and Design
WSEAS TRANSACTIONS on ELECTRONICS
DOI: 10.37394/232017.2022.13.14
Priyanko Raj Mudiar, Kandarpa Kumar Sarma, Nikos Mastorakis
E-ISSN: 2415-1513
109
Volume 13, 2022
Thus, the similarity matrix, for each constellation point can
be determined as [2][4]:
ϒ(m,n) = ║(xm , xn )2 – (ym , yn)2║ (4)
Let, tuple vector : hk = T
We know for LMS-SGD :
= wTN (n)* hN (n) (5)
ѐk (n) = hk (n) -
(6)
(n+1) = (n) - α ѐk *(n) hN (n) (7)
Where, is the weight vector,
is the
equalizer learning output, ѐk (n) is the iteration step error,
hk (n) is the desired asymptote (local maxima of the SGD)
and α
is the convergence ratio or step size factor for an
Ergodic distribution, and N
size of input constellation (m,n)
The weight update function taken for normalized batches is,-
(n+1) = (n) –
α
= (n) + α ѐk (n)
(8)
Let
denote the state covariance matrix of
the weight transition vector, where λ is the eigenvalue of the
state covariance matrix and denote the input correlation
matrix. Convolving Kalman Filter rule in LMS / KRLS
Widrow-Hoff Condition, we derive the posterior Kalman state
space model as given below ,-
α(n) =-hk T(n) . (9)
(n+1) = λ +λ
(10)
= λ2
- + (n) (11)
=
(12)
= hk T(n).hk (n) (13)
To estimate the Affinity Correlation margins in order to
iteratively determine the Minimum Entropy distances between
each constellation propagation path, let us first derive the
reduced correntropy determinants for Kalman-LMS/KRLS
traced AP/SGD as we obtain the adaptive training state model,-
(n)= ϒk (n)
- hk-1(n).(n) (14)
=
+ (15)
(16)
The recursive hyper-parameters or gradient interpolation
determinants for affinity propagation clustering [4] to “extract”
the constellation correntropy clusters for MQAM Const.n are,-
Responsibility / Availability margins : (r
a) , -
(17)
Where, :[0,1] is the exponential damping fraction known
as the “regression factor”. Bigger values of lowers the
possibility of oscillation as well error in the iteration process,
and also kernelizes the epoch points. Here, denotes the
responsibility margin and denotes the availability margin.
The next step is to identify the ensemble densities (or
Energy Spectrum Density) for each identified clusters in terms
of their band distribution, and then determining their
cumulative Gaussian entropy levels for channel estimation. In
order to improve convergence speeds as well as dynamic
efficiency, authors [3] have proposed and implemented a
Markov Trained, Eigen-Space induction model upon the
conventional Kalman-LMS approach in order to “dynamically
reduce” the training batch sizes during each LMS iteration.
For an Ergodic model, Markov State model is given as [3], -
(ϒ) = (ϒ) – ϒ.[f’( (ϒ)) + (ϒ)
Provided, =
+ ѐk (n) (18)
Where, is the regression factor ;
is the state
convergence factor, (n) in HMM Model space. ϒ
is the
similarity determinant obtained by taking the Wronskian
Determinant for each .
For any epoch nodal point, along the principal trajectory,
(ϒ) : k, the Markovian STM (State Transition
Model) is normally given as [3] ,-
(ϒ) =
(k)
Thus the Most Likelihood Estimator (MLE) for Markovian
STM masked over any orthogonal signal is given [3] as,-
( - ) =
=
(i,j) = Ϙ ()
To find the Eigen Space Model, authors [3] have provided
Liapunov determinants to determine any general Minimization
Model for Markovian MLSE Interpolation which can be used
identically in place of Lagrangian Correctors [11][12][15] for
predicting and correcting decision margins in terms of both
LMS and KRLS Markovian approximation as given by ,-
( = -
( - )
( =
(19)
Thus, the Maximum Correntropy condition described in terms
of Eigen-Space minimization criterion is derived by us
according to the ODE-MSE ensemble distribution given by
authors [3] as , -
WSEAS TRANSACTIONS on ELECTRONICS
DOI: 10.37394/232017.2022.13.14
Priyanko Raj Mudiar, Kandarpa Kumar Sarma, Nikos Mastorakis
E-ISSN: 2415-1513
110
Volume 13, 2022
Applying Wronskian Space model for input correlation matrix ,
, the minimization model in term of co-variance margin
}can be further expressed as , -
* W
() = .η) W (
+ (20)
And the weight update condition according to the “Perron-
Frobenius” condition [17][8] can be obtained as , -
- .η + + .η (21)
To apply maximum correntropy decision model in a Kalman-
KRLS [9][13] ADALINE model, the following standard
algorithm is considered . Here, f(
) corresponds to a
transformation function like Fourier-Bessel Transform as well
as Huang-Hilbert Transform [14] for Gaussian Q-determinant.
Upper and lower bounds are determined according to
maximum MSE Probabilities as derived by authors [3] .
The experimental works are carried out as per the flow
logic as depicted in Fig. 5. The schematic is implemented using
Matlab-R/2022(a) compiler with a VeriLog assembly design.
Fig. 7. AP Clustering based on Maximum Correntropy Margins
iterative minimization of the Euclidean Norm (Sec.III). For
convenience, only an S/T windowed [16] portion is displayed.
Fig. 8. Energy Spectrum Density and Entropy Level Detection
The Gaussian Energy Spectrum Density (ESD) extraction,
as shown in Fig. 8, has been determined using a Second Order
Initialize
= , =
= wTN (n)* hN (n)
ѐk (n) = hk (n) -
(n+1) = (n) –
α
= (n) + α ѐk (n)
Iterate from
0 to
XCorrLen [,(μ:η)] , Set : P = μT (n).μ(n)H
L( =
( )
=
Subtend / Update
(n)= ϒk (n)
- hk-1(n).(n)
(n+1) = λ +λ
Approximate
= λ2
- + (n)
=
Iterate and Update Cost Function (J(*))
∆(λ,μ) =
;
J(λ,μ) =
∆(λ,μ)
Determine Markovian Convergence
∆ J(λ,μ)
= -exp (
) (
)
Reset
(n)= ϒk (n)
- hk-1(n).(n)
=
+
4. Simulation Results
4.1 Transmitted and Received Signal Constellation
Fig. 6. Received Constellation Scatter plot for Rake Receiver
A 256-QAM/8-bit digital, orthogonally encoded signal is
transmitted over a Rayleigh Fading channel with Active White
Gaussian Noise (AWGN) margin of +10.4dB, as depicted in
Fig. 6. Baud speed is set up to 2.2Mbps
(IEEE-802.22x standard) [1], [16] , and channel capacity is
14.2Mbps with a total channel bandwidth capacity of
15Mhz [1]. Spatial constellation plot is obtained by a rake
receiver assembled with a synchronous PLL-VCO decoder.
4.2 Nyquist Cluster Estimation for 256-QAM
Scatter Plot
Considering the simulation plot shown in Fig. 7, sequential
sub-band cluster nodes are recursively “approximated” through
4.3 Spectrum Ensemble and Channel Entropy
Detection
WSEAS TRANSACTIONS on ELECTRONICS
DOI: 10.37394/232017.2022.13.14
Priyanko Raj Mudiar, Kandarpa Kumar Sarma, Nikos Mastorakis
E-ISSN: 2415-1513
111
Volume 13, 2022
(a)
(b)
Fig. 9. Eigen-Space Plot for Spectral Estimation (MLSE)
Fig. 10. ARQ Sub-band allocation using MLSE Minimization
Considering the weight nodal decisions taken by the Eigen-
Space Entropy MLSE plot, which is used for learning and
minimizing the spectral divergence in the previous subsection,
the Kernelized Recursive network updates and iterates the
Automatic Repeat Request (ARQ) switching delays according
to the corresponding Markovian trained sequences as shown in
Fig. 5. Schematic layout of Markov Trained Kalman-LMS/KRLS Adaptive Spectrum Equalizer
Fourier-Bessel Transform sequentially traced over each
identified “Nyquist clusters” as described in Section-(III).
Upper bound and lower bound ensemble margins have been
determined using ODE mass function limits as described
by authors [3]. Thus, applying MLSE-Quantization [8]m
[17] and Huang-Hilbert norms [14] upon the “extracted”
ESD, entropy transition levels have been approximated
and extracted iteratively.
4.4 Eigen-Space Minimation and Markovian
Decision Nodes
Considering the simulation diagrams depicted in Fig. 9, a
real time graphical visualization of the proposed
“Perron-Frobenius” MLSE maximization approach [17]m
[8] has been given as explained in Section-III. Diagram_(a)
visualizes the minimum spectral dispersion condition
corresponding to Eigen Spectral gaps for the determined
ESD and stochastically identified CMF Gaussian
Entropy levels. Diagram_(b) characterizes the
conditional Central Node decision taken recursively by
the Markovian SGD/SGA model for optimum channel
estimation and diversity channeling as described in the next
subsection.
4.5 KRLS-ARQ switching for Maximal Ratio
Combination
WSEAS TRANSACTIONS on ELECTRONICS
DOI: 10.37394/232017.2022.13.14
Priyanko Raj Mudiar, Kandarpa Kumar Sarma, Nikos Mastorakis
E-ISSN: 2415-1513
112
Volume 13, 2022
Fig. 10. The adaptive cost function [9], [13] determined
for kernel batch learning models supervises the relaying
speeds for each OFDMA sub-bands (or Shannon-Bands) . The
“optimum” relaying speed used for packet-buffer
transmissions in each orthogonally multiplexed sub-
channel is determined recursively by the KRLS-ARQ
engine considering a Maximal Ratio Combination [15]
condition to equalize the SNR gains while minimizing (or
normalizing) the BER levels for each sub-channel. Thus,
bandwidth equalization is achieved by regressively
approximating the channel diversity determinants.
Using Huang-Hilbert Transform [14] over the sequentially
transmitted buffer links, we thus obtain the overall
Instantaneous Bandwidth Distribution which are artificially
identified for the Virtually Simulated SCADA-OFDMA
Transmission Channel [1] .
Fig. 11. Bandwidth Distribution for Manual Transmission
Fig. 12. Bandwidth Distribution for Supervised Adaptive
Transmission using ANN
The given plots describe the spectrum utilization over the
virtual IEEE-802.22x OFDMA adaptive channel. Fig. 11 plots
the Bandwidth coverage for regular transmissions for the
SCADA-DCS standard. Fig. 12 plots the same bandwidth
distribution for spectrum equalized channel. We observe from
comparison, that an aggregate marginal coverage of ~94.8% is
achieved after bandwidth equalization, as compared to the
~48.8% coverage for a maximum bandwidth allotment of
15Mhz with a maximum capacity of 14.2Mbps, with buffer
speeds of 2.8Mbps. Thus, an absolute bandwidth improvement
of nearly 50% has been achieved in the allotted channel
capacity. It can also be observed that normalization precision
of the individual sub-channel SNR and BER ratios effect the
Shannon-Band’s allocation especially in a “reflexive” manner,
5. Performance Evaluation
5.1 Retransmitted Signal Improvement by
ANN Equalization
as over-normalizing or under-normalizing distorts the
switching diversity decision.
5.2 Spectrum Equalization Margins and Mean
Square Errors
Considering the given plots, the individual
bandwidth equalization margins with respect to the Mean
Square Errors (MSE) traced over the entire tracking trajectory
has been visualized. It can be observed that cascading
stochastic induction in SGA/SGD based deterministic models
do not hamper the regular performance of the Delta-Learning
method. By stochastic gradient induction through Markov
Training approach, the trained batch sizes reduce considerably
which assures a large improvement in the dynamic
responsiveness of conventional SGA/SGD models, as well as
increasing the convergence speeds for random inputs which
may or may not be causal in nature. This makes the infusion
approach suitable for application in real time signal detection
as well as channel allocations with observable controllability
under unpredictable transmission environments.
WSEAS TRANSACTIONS on ELECTRONICS
DOI: 10.37394/232017.2022.13.14
Priyanko Raj Mudiar, Kandarpa Kumar Sarma, Nikos Mastorakis
E-ISSN: 2415-1513
113
Volume 13, 2022
Fig. 13. Equalization Margins for Kalman-LMS/KRLS
Fig. 13, determines the equalization margins for Kalman-
LMS ADALINE (shown in “green” fonts) as well as Kalman-
KRLS ADALINE (shown in “blue” fonts). The increased
regression rates for Kalman-KRLS SGD can be understood
from the MSE levels during nodal transitions as given below.
Fig. 14. Mean Square Error Distribution for Kalman-LMS/KRLS
Fig. 14, determines the Learning Accuracies in terms
of MSE convergence patterns for Kalman-LMS
ADALINE (shown in “green” fonts) as well as Kalman-
KRLS (shown in “blue” fonts). We can observe that a
considerably high degree of over-learning and over-training
situation can be seen for KRLS-SGD in comparison to
the ergodically transient convergence pattern observed for
Kalman-LMS SGD. Thus, dynamic adaptability is
maintained in cluster detection while discreteness in the
regression pattern can be ascertained for diversity
allocation, provided the cost functions are iterated properly
with a proper selection of minimization rates. Improper
selection of minimization rates can severely destabilize the
tracking performances as well.
[1]IEEE Std C37.1TM-20017, “IEEE Standard for SCADA and
Automation Systems” 2018R-TM @ IEEE Power and Energy
Society, IEEE, USA, 2018, pp. 19-21.
[2]Li, Qiang & Shen, Dong & Wang, Fei. (2016) , “MQAM
Modulation Recognition Based on AP Clustering Method”,
MATEC Web of Conferences. 44. 01002.
10.1051/matecconf/20164401002.DOI:10.1051/matecconf/201
64401002 PubINC:January 2016, MATEC Web of
Conferences 44:0100
[3]G. G. Yin and V. Krishnamurthy, "LMS algorithms for
tracking slow Markov chains with applications to hidden
Markov estimation and adaptive multiuser detection," in IEEE
Transactions on Information Theory, vol. 51, no. 7, pp. 2475-
2490, July 2005, doi: 10.1109/TIT.2005.850075.
[4]X. Liu, M. Yin, J. Luo and W. Chen, "An improved affinity
propagation clustering algorithm for large-scale data sets,"
2013 Ninth International Conference on Natural Computation
(ICNC), 2013, pp. 894-899, doi: 10.1109/ICNC.2013.6818103.
[5]Imen Nasr , Sofiane Cherif , Published in: 2012 IEEE
Vehicular Technology Conference (VTC Fall) Date of
Conference: 3-6 Sept. 2012 Date Added to IEEE Xplore: 31
December 2012 ISBN/ISSN Information: INSPEC Accession
Number: 13226663 DOI: 10.1109/VTCFall.2012.6399369
Pp.: “A Novel Adaptive Fusion Scheme for Cooperative
Spectrum Sensing” Publisher: IEEE Conference
Location: Quebec City, QC, Canada
[6]P. A. C. Lopes and J. B. Gerald, "New Normalized LMS
Algorithms Based on the Kalman Filter" , 2007 IEEE
International Symposium on Circuits and Systems, 2007, pp.
117-120, DOI:10.1109/ISCAS.2007.378235. Published in:
2007 IEEE International Symposium on Circuits and Systems
INSPEC Accession Number: 9538973 DOI:
10.1109/ISCAS.2007.378235 Print ISBN : 1-4244-0920-9
CD:1-4244-0921-7 Print ISSN: 0271-4302 Electronic ISSN:
2158-1525
6. Conclusion
In this paper, we have implemented and simulated a novel
Stochastic Markov infusion concept over a Kalman optimized
Steepest Gradient Ascent / Descent based Adaptive Artificial
Neural Network to virtually implement a network resource
allocation based operation in a SCADA-DWWAN based
transmission standard (IEEE-802.22x) using OSI-4.0
Architecture. Learning performances have been observed with
respect to the improvement in bandwidth allocation though
spectrum equalization approach over an OFDMA-ARQ based
adaptive multiplexed channel. Results show dynamic
adaptability, improved efficiency as well as large reduction of
the training complexity through stochastic infusion in decisive
ADALINE approaches.
Acknowledgement
The author would like to confer his solemn gratitude to the
Department of Electronics and Communication Engineering
(ECE), under Gauhati University for providing technical
support and guidance towards the completion of this
simulation based project.
References
[7]
[8] Amit Baghel , Monika Kumari , Published in: 2020 IEEE
International Conference for Innovation in Technology
(INOCON) Date of Conference: 6-8 Nov. 2020 Date Added to IEEE
Xplore: 01 January 2021 ISBN Information:
INSPEC_Accession Number: 20334148 DOI:
10.1109/INOCON50539.2020.9298447 Publisher: IEEE Conference
Location: Bangaluru, India Pp. : “Spectrum Sensing using
Analysis of Variance, Eigen, and Energy
(ANOVEE)” .
[9] Façanha, T.S., Barreto, G.A. & Costa Filho, J.T. A Novel
Kalman Filter Formulation for Improving Tracking
Performance of the Extended Kernel RLS. Circuits Syst Signal
Process, Springer Pub. 40, 1397–1419 (2021).
https://doi.org/10.1007/s00034-020-01533-4
[10] K. N. Pappi, A. S. Lioumpas and G. K. Karagiannidis, "θ-QAM: A
parametric quadrature amplitude modulation family
and its performance in AWGN and fading channels," in IEEE
Transactions on Communications, vol. 58, no. 4, pp. 1014-1019,
April 2010, doi: 10.1109/TCOMM.2010.04.080552.
[11] S. Saraç and Ü. Aygölü, "An ARQ-based protocol for
cooperative spectrum sharing in underlay cognitive radio," 2016
24th Signal Processing and Communication Application Conference
(SIU), 2016, pp. 405-408, doi: 10.1109/SIU.2016.7495763.
[12] Y. Chen, H. C. Lee, J. You and S. Wei, "Less Complexity
Successive Interference Cancellation for OFDM System," 2007 IEEE
18th International Symposium on Personal, Indoor and Mobile
Radio Communications, 2007, pp. 1-5, doi: 10.1109/
PIMRC.2007.4394648.
[13] Aquino, M.B.L., F. Guimarães, J.P., Linhares, L.L.S. et
al. Performance evaluation of the maximum complex
correntropy criterion with adaptive kernel width
update. EURASIPJ. Adv. Signal Process. 2019, 53(2019).
Springer Open. EURASIP https://doi.org/10.1186/s13634-019-0652-2
[14] Y. Liu, H. An and S. Bian, "Hilbert-Huang Transform and the
Application," 2020 IEEE International Conference on Artificial
Intelligence and Information Systems (ICAIIS), 2020, pp. 534-539,
doi: 10.1109/ICAIIS49377.2020.9194944.
[15] T. Kumagai, M. Mizoguchi, T. Onizawa, H. Takanashi and M.
Morikura, "A maximal ratio combining frequency diversity ARQ
scheme for OFDM signals," Ninth IEEE International Symposium
on Personal, Indoor and Mobile Radio Communications
(Cat. No.98TH8361), 1998, pp. 528-532
vol.2, doi: 10.1109/PIMRC.1998.734257.
[16] Solanki, Surendra & Dehalwar, Vasudev & Choudhary,
Jaytrilok. (2021). Deep Learning for Spectrum Sensing in
Cognitive Radio. Symmetry. 13. 147. 10.3390/sym13010147.
[17] S. U. Pillai, T. Suel and Seunghun Cha, "The Perron-Frobenius
theorem: some of its applications," in IEEE Signal Processing
Magazine, vol. 22, no. 2, pp. 62-75, March 2005, doi:
10.1109/MSP.2005.1406483.
WSEAS TRANSACTIONS on ELECTRONICS
DOI: 10.37394/232017.2022.13.14
E-ISSN: 2415-1513
114
Volume 13, 2022
[7]Jian Hua.Lu, K.B Letaif, M.L Liou, J.C.I-Chuang, “M-PSK
and M-QAM BER computation using signal-space concepts in
OFDMA Channel Switched ARQ” Published in: IEEE
Transactions on Communications ( Volume: 47, Issue: 2, Feb
2012) INSPEC Accession Number: 6212385 DOI:
10.1109/26.752121 Publisher : IEEE (RC)
Priyanko Raj Mudiar, Kandarpa Kumar Sarma, Nikos Mastorakis
Conflicts of Interest
The author(s) declare no potential conflicts of interest concerning the research,
authorship, or publication of this article.
Contribution of individual authors to the creation of a scientific
article (ghostwriting policy)
The author(s) contributed in the present research, at all stages from
the formulation of the problem to the final findings and
solution.
Sources of funding for research presented in a scientific article
or scientific article itself
No funding was received for conducting this study.
Creative Commons Attribution License 4.0
(Attribution 4.0 International, CC BY 4.0)
This article is published under the terms of the
Creative Commons Attribution License 4.0
https://creativecommons.org/licenses/by/4.0/deed.en_US
