Professional Documents
Culture Documents
Performance Analysis of Support Vector Machine Learning Based
Performance Analysis of Support Vector Machine Learning Based
Performance Analysis of Support Vector Machine Learning Based
com
ScienceDirect
Available online at www.sciencedirect.com
Available online at www.sciencedirect.com
Procedia Computer Science 00 (2022) 000–000 www.elsevier.com/locate/procedia
ScienceDirect
ScienceDirect
Procedia Computer Science 218 (2023) 2776–2785
Procedia Computer Science 00 (2022) 000–000 www.elsevier.com/locate/procedia
5G is designed to provide the knowledge and transmission technology industry keeping up a variation of
approaching services with different conditions [1]. It covers the magnified connectivity of vehicle-to-everyone
transmission, uRLLC, IoT, and various time specification applications. The 5G will supply advances in technological
matrices as excessive output, supporting the latest high request application and improving the uses of high radio
frequency bands [2]. The technologies are used in 5G networks, Coordinate Multipoint (CoMP), massive-MIMO,
mmwave, UDSC, dense multiple-radio access technology (Multi-RAT), M2M communication, D2D communication,
cloud computing, network virtualization, etc. [3]. 5G NR is 5th generation accessing RAN, also known as New Radio,
which has been advised to the worldwide standards of 5th generation and completely under the 3G Partnership Project.
5G New Radio supports both Time Division Duplex (TDD) for higher and Frequency Division Duplex (FDD) for
lower frequency bands. It contains a dynamic TDD that assigns mobile slots based on the amount of data traffic. 5G
is the next generation technology addressing the large-scale application of mobile communication [4]. The fifth
generation provides a limitless broadband experience to the users. 5G technology can assist critical machine
communication with strong reliability and fast association. The first action for the 5G mobile cellular communication
is available in release 15 of 3GPP, where essentially focus was fulfilling the need of mobile operator’s needs in terms
of high ultra-high broadband services [5]. In 2019, the upcoming 3GPP Release 16 incorporates new characteristics
of the internet of things (IoT), and ultra-reliable and low latency communication (URLLC).
5G is designed to provide the knowledge and transmission technology industry keeping up a variation of
approaching services with different conditions. It covers the magnified connectivity vehicle-to-everyone transmission,
uRLLC, IoT, and various time specification applications [6]. The 5G is the latest wireless technology, which provides
the main aim promises of extending the A.I. and machine learning welfares in the radio network together with the
end-users. The 5G will supply advances in technological matrices as excessive output, supporting the latest high
request application and improving the uses of high radio frequency bands [7]. The technologies are used in 5G
networks, Coordinate Multipoint (CoMP), massive-MIMO, mm-wave, UDSC, dense multiple-radio access
technology (Multi-RAT), M2M communication, D2D communication, cloud computing, network virtualization,
carrier aggregation. CA is used in this paper and extended the throughput of 5G technology by using CA with Machine
Learning [8].
Carrier Aggregation is a method of increasing bandwidth while maintaining the channel's capacity. Both FDD and
TDD are supported by CA. Before the CA, data was transported on a single component carrier channel called the
principal component carrier channel [9]. A primary cell is another name for the corresponding cell. This is the state
before the Carrier Aggregation. Carrier Aggregation in LTE-A increases capacity by combining one or more carriers
to offer high bandwidth of up to 100 MHz. To build a 100MHz channel, a single channel has a bandwidth of
1.4,3,5,10,15 and up to 20MHz [10]. It uses about 5 channels for data transmission, one of which is a fixed major
component carrier channel and the other four are secondary component carrier channels. The principal component
carrier oversees all RRC and NAS (Non-Access Stratum) procedures, as well as the removal and addition of secondary
component carriers [11]. Carrier aggregation helps to meet high data rate demand while also improving coverage for
medium data rates by employing lower order modulation and coding rates that need less transmission power and
interference [12].
Machine learning is a combination of different techniques that perform on a computer and can learn, operate
automate, and provides the best solution using a huge dataset that cannot be used consistently by humans [13].
Machine learning provides the techniques for a wireless network to learn the complex network and collect knowledge
by operating with data [14]. Without Artificial Intelligence and Machine Learning, network service providers will be
unable to effectively be providing future generation services with its numerous requirements. The machine learning
approach is useful for determining future scenarios, it learns the network based on online data/offline data, calculate
the system performance on relevant parameters, and suggests or adjusts the network scenarios for better performance
[15]. Machine learning used different kinds of algorithms for various types of problems and tackles each problem
2778 Samiksha Mathur et al. / Procedia Computer Science 218 (2023) 2776–2785
simultaneously with different types of learning models for each stage [16]. For future generation networks, machine
learning concepts will be very efficient to predict the possible resource allocation in a network for users to achieve a
high data rate in different mobility scenarios [17]. Hence, the optimal scenario can be built automatically without the
direct participation of the system engineer or network services provider. This will overcome the computational
complexity of the system [18]. Various ML techniques can be used for observing the optimal resource allocation for
heterogeneous networks, which are categorized as:
• Supervised machine learning
• Unsupervised machine learning
• Reinforcement machine learning
• Deep learning
In this section, an SVM-based machine learning approach for enhancing the total throughput for the 5G NR cellular
system is designed [19]. The carrier aggregation aggregated the carrier component for the user whose channel quality
indicator block error rate is less than 10% otherwise its works as an ordinary system. The eNodeB of the MAC layer
sends the reference signal to each connected user [20].
The connected users decode this reference signal and compute the channel rate and BLER. The eNodeB gets this
signal and allocates the modulation and channel decoding scheme for each user [21]. The users with low channel
quality are assigned with a lower modulation scheme which degrades the system capacity and users’ throughputs. A
carrier aggregation system is used to enhance the throughput of these low signal quality users. In an ordinary system,
before activation of the carrier aggregation, there must follow a procedure to activate the coordination among the
eNodeB [22]. It was so complex and need more synchronization amongst the carrier component. The proposed system
Samiksha Mathur et al. / Procedia Computer Science 218 (2023) 2776–2785 2779
Proposed Algorithm:
Input: Number of Resource Block, UE, TTI
1. Deploy NB users, Generate Channel
2. Compute SINR for all deployed Users in the ROI
3. Compute BLER and UE Throughput
4. Generate the Channel Quality Indicator-feedback matrix
5. Split CQI data into 80% and 20% for training and testing purposes
6. Train data with SVM-Linear, SVM-RGB, SVM-Sigmoid Machine Learning function
7. Compute Error Rate
8. if
9. Error rate < Threshold (T)
10. Allocate resource block to UE through MAC layer
11. Schedule the UE until TTI and UE requests left
12. else
13. Initiate Carrier Aggregation
14. Aggregate free carrier component {primary carrier component and secondary carrier component}
15. Allocate aggregate carrier RB to UE until TTI and UE request left
16. Repeat step 4 to 14 until free RB, TTI, and UE request are left.
17. Output: Resource block allocated matrix.
The primary level of initialization was dependent on 3 sector cell model with an antenna pattern. This sector scenario
is a 120-degree based 3 directional antenna that enhances the system's cell edge throughput. The transmission channel
is most prior and was given with the standards of 5G NR is Spatial Channel model which is illustrated in simulation
scenario/system architecture in figure 2, for computing the shadow fading, antenna gain array, base station based sub-
path, power at the nth path, direct path, and the total path by applying the radio network parameters and simulation
parameters given in table 1.
2780 Samiksha Mathur et al. / Procedia Computer Science 218 (2023) 2776–2785
In this section, the simulation results are described as concerns the edge UE throughput, average UE throughput, peak
UE throughput, average cell throughput, spectral efficiency, and fairness index with the proposed SVM linear, SVM
sigmoid, and SVM RFB Carrier Aggregation initiates resource allocation approach. The key performance indicator
illustrated in table 2, are analyzed with increments in the number of users and transmission time interval (TTI). The
proposed resource allocation techniques compared with the ordinary round robin (RR), best channel quality indicator
(BCQI), fractional frequency reuse (FFR), carrier aggregation (CA), carrier aggregation with MU-MIMO, and carrier
aggregation with Beam forming approach.
Figure 4 is a result of the average cell throughput against the number of transmission time intervals increasing from
10 to 50. The standard CA, as seen in the figure, has underperformed the MU-MIMO with CA, Beamforming with
CA, proposed SVM with CA, and outperformed the round robin, best channel quality indicator, fractional frequency
reuse technique. It achieved the highest average cell throughput 0.69 Mbps at 10 TTI. Proposed SVM with CA
achieved 0.73 Mbps, 0.745 Mbps, and 0.75 Mbps with Linear, RBF, and Sigmoid functions respectively. BCQI and
FFR techniques performed similarly. Figure 5 shows the user peak UE throughput contrasted with the number of
transmission time intervals (TTI). The graphs represent output in three patterns all ML-based resource allocation was
performed the same way and achieved 0.41 Mbps to 0.44 Mbps, Round robin and BQI performed identical way
achieving 0.34 Mbps to 0.38 Mbps. The fractional frequency reuse (FFR), carrier aggregation (CA), carrier
Samiksha Mathur et al. / Procedia Computer Science 218 (2023) 2776–2785 2781
Average UE Throughput ∑𝑛𝑛𝐾𝐾=1 Tk Average UE Throughput is the overall packets received by each
𝑇𝑇avg = user in a specific time.
𝑛𝑛Ti
(Mbps)
Fairness Index [∑𝑛𝑛𝑘𝑘=1 𝑇𝑇𝑘𝑘 ]2 The fairness index measures the "equality" of allocations.
𝐽𝐽(𝑇𝑇) =
(Jain fairness) 𝑛𝑛[∑𝑛𝑛𝑘𝑘=1 𝑇𝑇𝑘𝑘2 ]
Spectral Efficiency ∑𝑛𝑛𝐾𝐾=1 Tk Spectral utilization specifies the rate at which data is being
𝑆𝑆 = transmitted over the available bandwidth.
𝐵𝐵𝐵𝐵
(Bit/cu)
Fig. 4 Average Cell Throughputs V/s TTI Fig. 5 Peak Throughputs V/S TTI
Figure 6 is of Average UE throughput of users with the expanded transmission time interval. The output SVM with
RBF and sigmoid resource allocation outperform among all and attains 0.0745 Mbps and 0.0751 Mbps average UE
throughput respectively. SVM with linear underperforms the RBF and Sigmoid function. The ordinary CA and CA
with beamforming perform similarly. Round robin scheduling techniques perform worst in all and it achieved the
maximum of 0.064 Mbps UE average throughputs at 20 TTI after that declined. Figure 7 show the fairness index of
the users in the scheme when the number of transmission time interval increased all resource scheduling techniques
initial increase the fairness among the user after that declined. Surprisingly, the entire machine learning techniques
work in the same manner and all non-machine learning works identically. Figure 8 illustrates the spectral efficiency
and transmission time interval. As the transmission time interval is increased from 10 to 50 all resource scheduling
techniques whether it is machine learning or non-machine learning all decreased the bit used by the spectrum. The
highest performance is shown in the SVM with sigmoid, after that SVM with RBF and SVM with linear, maintain
2782 Samiksha Mathur et al. / Procedia Computer Science 218 (2023) 2776–2785
their efficiency from 0.354 bit/cu to 0.399 bit/cu. The lowest gain showed in round robin efficiency at 0.287 bit/cu at
50 TTI.
0.076
R
BCQ
R
IFFR 0.50
0.074 C
CA with MU-
A
CA with Beam
MIMO
0.072 CA with SVM
Forming 0.45
CA with SVM
Linear
CA with SVM
RBF
Sigmoid RR
0.070 BCQI
Fairness Index
0.40 FFR
CA
CA with MU-MIMO
0.068 CA with Beam Forming
CA with SVM Linear
0.35 CA with SVM RBF
CA with SVM Sigmoid
0.066
0.064 0.30
0.062
0.25
0.060
0.20
10 20 30 40 50 10 20 30 40 50
TTI TTI
Fig. 6 Average UE Throughputs V/S TTI Fig. 7 Fairness Index V/S TTI
10 20 30 40 50 10 20 30 40 50
Fig. 8 Spectral Efficiency V/S TTI Fig. 9 Average Cell Throughputs VS No. of Users
Figure 9 shows the average cell throughput against the enhancement in the number of users. The output of each
resource scheduling technique is not common to one another. The behavior of all SVM machine learning techniques
outperformed and achieved 0.799 Mbps to 0.83Mbps. The initial value of SVM with linear machine model maintains
its output constant at 20 transmissions time interval after that declined. The SVM with sigmoid outperforms among
all. SVM linear is more output as compared to the carrier aggregation, carrier aggregation with MU-MIMO, and
carrier aggregation with beamforming schemes. Round robin and BCQI underperform all techniques and attain the
maximum of 0.61 Mbps and 0.63 Mbps respectively.
Samiksha Mathur et al. / Procedia Computer Science 218 (2023) 2776–2785 2783
0.40 0.04
0
0.38
0.36 0.03
5
0.34
0.32
0.03
0
1 20 3 40 50
1 2 3 4 5
0 0
No. of Users 0 0 No.0 of Users
0 0
Fig. 10 Peak UE Throughputs VS No. of Users Fig. 11 Average UE Throughputs VS No. of Users
Figure 10 shows the user peak UE throughput versus the number of increments in user of carrier aggregation, carrier
aggregation with RR, BCQI, MU-MIMO, carrier aggregation with Beamforming, and Carrier Aggregation with SVM
machine learning. The output of each resource scheduling technique is similar to one another It clearly shows in the
graph that SVM with sigmoid, RBF, and linear model is attaining the highest peak throughputs in all the scheduling
techniques. The highest peak UE throughput is 0.485 Mbps achieved by SVM sigmoid with 20 users. Beamforming
with carrier aggregation attains higher gain among Carrier aggregation with MU-MIMO and ordinary carrier
aggregation. FFR is attaining the higher gain among the best channel quality indicator and round robin. In the figure,
it is clearly illustrated that when the number of users increases all techniques initially increased their peak UE
throughputs after decreases in a similar fashion.
Figure 11 is for the Average UE throughput of users with the enlargement in the number of users in the cell. Here the
result of every procedure is equivalent to one another. Initially from 10 to 20 users, all resources scheme increased
gradually. As the number of users are enlarged from 30 to 50 all the scheduling techniques declined in their average
UE throughputs. Round robin, BCQI, and FFR output decrease drastically. The behavior of all SVM machine learning
techniques outperformed. The initial value of SVM with linear machine model maintains its output constant at 20
transmissions time interval after that declined, and achieved maximum 0.051 Mbps Average UE throughputs with
SVM sigmoid with 20 users in a cell.
Figure 12 show the fairness index of the users in layout, when the number of users increased all resource scheduling
techniques initial increase the fairness among the user after that declined all scheduling techniques. SVM with
sigmoid scheduling techniques achieves the highest fairness, its fair shares the 55% of users to all resources. SVM
with RBF and SVM with linear model fair share the resources below the SVM sigmoid and above the carrier
aggregation, carrier aggregation with MU-MIMO, carrier aggregation with beam forming, BCQI, and RR. Carrier
aggregation with MUMIMO and carrier aggregation with beamforming perform in a similar manner and fair share
the resources of 50% and 52% respectively. Round robin showed 41% fairness which is lesser among all.
Figure 13 illustrates the spectral efficiency and number of users in the cell. As the number of users grows from 10 to
30 all resource scheduling techniques whether it is machine learning or non-machine learning all increased their
spectral efficiency after that declined. The highest performance is shown in the SVM with sigmoid i.e., 0.48 bit/cu,
after that SVM with RBF i.e., 046 bit/cu, and SVM with liner i.e. 045 bit/cu. SVM with RBF maintain their spectral
efficiency constant.
2784 Samiksha Mathur et al. / Procedia Computer Science 218 (2023) 2776–2785
0.48
RR
BCQI
0.55
FFR
CA
0.46
CA with MU-MIMO RR
CA with Beam Forming BCQI
CA with SVM Linear 0.44 FFR
CA
0.36
0.40
0.34
0.32
0.35
0.30
10 20 30 40 50 10 20 30 40 50
Fig. 12 Fairness Index VS No. of Users Fig. 13 Spectral Efficiency VS No. Of Users
5. Conclusion
5G mobile communication is meant to deliver a high data rate, low latency, and multiuser connectivity for the wireless
communication environment. To achieve the above promise and provides a better experience to the user resource
allocation plays an important role. The MAC layer and physical layer of 5G NR radio architecture are responsible to
assign efficient resources to the users of 5G. The MAC scheduler schedules the resources for the users. There are
different types of resource scheduling schemes in the MAC layer. In this paper, the Support vector-based machine
learning approach is designed for enhancing the carrier throughputs for the UE who’s not received the required data
from the networks. The proposed approach predicts the channel state information from the physical layer and initiates
the carrier aggregation component selector to aggregate the carrier component. The proposed work is based on SVM
with linear, SVM with RBF, and SVM with sigmoid hyper plane function. The result shows that the designed scheme
achieves higher throughput as compared to the traditional approaches. The performance of the scheduled system with
average cell throughputs, spectral efficiency, fairness index, peak throughputs, and average UE throughputs is
evaluated. From all the evaluation parameters and metrics, our proposed approach outperforms and assigns the
essential bandwidth and resources to the sufferer users. As result, the performance of the designed scheme is simulated
by increasing the number of TTI and number of users. TTI and the number of users is from 10 to 50 in increasing
order. It is found that the SVM with linear, SVM with RBF and SVM sigmoid achieved the highest average cell
throughput at 0.73 Mbps, 0.74 Mbps, and 0.75 Mbps respectively in the case of transmission time interval, while they
achieved average cell throughputs of 0.79 Mbps, 0.81 Mbps and 0.83 Mbps in SVM with linear, RBF and Sigmoid
function when increasing in user per cells. In the case of peak throughputs, SVM achieved 0.448 Mbps to 0.453 Mbps.
The ordinary carrier aggregation performs less than 0.39 Mbps peak throughputs. Machine learning-based spectral
efficiency and fairness are the highest among all resource scheduling techniques achieving up to 0.48 bit/cu and a
55% fair share of the users’ essential resources. The proposed approach full the current and future requirements of 5G
resource allocation requirements.
References
[1] Sun, Z., Yin, C., & Yue, G. (2016) “Reduced-complexity proportional fair scheduling for OFDMA systems.” in IEEE international conference
on communications, circuits, and systems: 1221–1225.
[2] Noliya A., Kumar S. (2020) “Performance Analysis of Resource Scheduling Techniques in Homogeneous and Heterogeneous Small Cell LTEA
Networks.” Wireless Pers Commun 112: 2393–2422.
[3] R Sharma, Y Chaba, Y Singh. (2010) “Analysis of Security Protocols in Wireless Sensor Network.” Int. J. Advanced Networking and
Applications 02(03): 707-713.
[4] S. C. Wang, W. S. Hsiung, K. Q. Yan, and Y. T. Tsai. (2019) “Optimal agreement achievement in a fog computing based IoT.” Journal of
Internet Technology 20(6): 1767-1779.
[5] Y. Sun, M. Peng, and H. V. Poor. (2018) “A distributed approach to improving spectral efficiency in uplink device-to-device-enabled cloud
radio access networks.” IEEE Transactions on Communications 66(12): 6511-6526.
Samiksha Mathur et al. / Procedia Computer Science 218 (2023) 2776–2785 2785
[17] N. Jinaporn, S. Armour, and A. Doufexi. (2019) “Performance Evaluation on Resource Allocation with Carrier Aggregation in LTE Cellular
Networks.” in 2019 IEEE 90th Vehicular Technology Conference (VTC2019-Fall), Honolulu, HI, USA: 1–5.
[18] R. Chataut and R. Akl. (2020) “Massive MIMO systems for 5G and beyond networks— overview, recent trends, challenges, and future research
direction.” Sensors 20(10): 1–35.
[19] Y. Xu and G. Li. (2019) “Optimal and Robust Interference Efficiency Maximization for Multicell Heterogeneous Networks.” IEEE Access 7:
102406–102416.
[20] J. Ghosh. (2019) “Interrelationship between energy efficiency and spectral efficiency in 20 cognitive femtocell networks: A survey.” Pervasive
and Mobile Computing 59: 101066.
[21] W. Xu, R. Qiu, and J. Cheng. (2018) “Fair optimal resource allocation in cognitive radio networks with co-channel interference mitigation.”
IEEE Access 6: 37418-37429.
[22] Y. Sun, M. Peng, and H. V. Poor. (2018) “A distributed approach to improving spectral efficiency in uplink device-to-device-enabled cloud
radio access networks.” IEEE Transactions on Communications 66(12): 6511-6526.
[23] Lukmanhakim Sukeran, M. H., Al-Hareth Zyoud, M. M., Ahmad, S. H., Amelia Wong, M. D., & Islam, R. (2014) “Performance evaluation of
LTE scheduling techniques for heterogeneous traffic and different mobility scenarios.” Springer Advanced Computer and Communication
Engineering Technology 315: 141–150.
[24] Schwarz, S., Mehlfuhrer, C., & Rupp, M. (2011) “Throughput maximizing multiuser scheduling with adjustable fairness.” In IEEE international
conference on communications (ICC), Kyoto, Japan.
[25] Schwarz, S., Mehlführer, C., & Rupp, M. (2010) “Low complexity approximate maximum throughput scheduling for LTE.” In Conference
record of the forty fourth Asilomar conference on signals, systems and computers, Pacifc Grove, CA, USA: 1563–1569.
[26] Alqahtani, S. A., Alhassany, M. (2013) “Performance modeling and evaluation of novel scheduling algorithm for LTE networks.” In IEEE 12th
international symposium on network computing and applications, Cambridge, MA, USA: 101–105.