A Comparative Analysis of Implementing 5G through Deep Learning

Mrinalini

, Kamlesh Kumar Singh

and Himanshu Katiyar

Department of Electronics Engineering, Amity University, Lucknow Campus, Uttar Pradesh, India

Department of Electronics Engineering, Sonbhadra, India

Keywords: MIMO, NOMA,DNN, GFDM

Abstract: Fifth Generation of Cellular Networks will give ubiquitous and wide reliable coverage as well as find its

applications in powering critical-mission, huge IoT deployments, and M2M Communications. These

utilisation need low latency and high capacity capable technology that can suggested as Generalized

Frequency Division Multiplexing due to its highly supportive physical structure for 5G. Deep Learning(DL)

is implemented to the large value of complex data of GFDM input Signal in order to analyse the

performance in terms of Bit Error Rate(BER) along with Signal to Noise Ratio(SNR).In this paper, two

different methods of DL is considered and compared for better designing and performance purpose. Various

methods of Deep Learning are analysed for technical advancement of 5G Cellular Network. This paper

consists analysis of different aspects of DL in 5G such as implementation in Massive MIMO, mm Wave

Communication and NOMA systems.

1 INTRODUCTION

During last decade, studies show that consumption

of data exchanged by users over internet is

increasing and this growth will enhance with

staggering factor in upcoming years. The reason

behind is the increase in population, application of

smartphones and highspeed broadband services.

Fourth Generation works on LTE is considered to be

one of the versatile technology, but nowadays a low

latency communication is needed. This led evolution

of 5G that supports IOT applications, M2M,

vehicular networks, tactile communications and

many more. Various modulation methods have been

proposed for 5G such as Orthogonal frequency

division multiplexing (OFDM), Universal Filter

Multicarrier Modulation (UFMC), Filter Bank

Multicarrier Modulation (FBMC) and Generalised

Frequency Division Multiplexing (GFDM).

Advanced LTE in OFDM had been utilized in 4G

but this method of multicarrier modulation suffers

from high Out-of-band (OOB), high peak to average

power ratio (PAPR) that makes it not suitable for 5G

and trending communication network. Generalized

frequency division multiplexing (GFDM) is

considered to be one of the most promising

technology for future emerging communication

network. Positive side of the GFDM are spectral

efficiency, low latency, block structure and

bandwidth efficient make it most suitable candidate

for 5G Cellular System. (Amirhossein, 2018) Deep

learning technique has off late gained significant

attention as it provides solution to bulk complex

with high performance in several fields like object

detection problems, language processing as well as

computer vision. Researchers have been

implementing DL to design various wireless

communication techniques as it has high potential in

various aspects like channel estimation, optimization

of performance as well as in multiples access.5G can

be implemented more practically in order to meet

new demands of forth coming cellular network by

designing DL- based Non-Orthogonal Multiple

Access (NOMA), DL- based massive MIMO and

DL-based mm Wave. This paper gives description of

GFDM with implementation of DL as GFDM is

considered to be one of the prominent waveforms

that fulfils vital challenges of upcoming wireless

networks. It has various advantages over OFDM

such as reduced OOB emission, bandwidth

efficiency and lowest latency. Several research had

showed that MIMO transmission unquestionable

spectral efficiency, so its application is compulsory

for 5G modulation candidate. In this paper, deep

convolutional neural network-based GFDM-IM

detector is compared to DL-aided GFDM detection

Mrinalini, ., Singh, K. and Katiyar, H.

A Comparative Analysis of Implementing 5G through Deep Learning.

DOI: 10.5220/0010564600003161

In Proceedings of the 3rd International Conference on Advanced Computing and Software Engineering (ICACSE 2021), pages 165-169

ISBN: 978-989-758-544-9

165

and demodulation scheme in terms of BER and SNR

for performance analysis. (G.Gui,2018)Several

applications of Deep Learning in wireless

communication are examined and best method for

implementation in 5G with GFDM is concluded.

This paper comprises of three sections in which

Section.2 explains the reason behind implantation of

deep learning in wireless communication, Section3

deals how DL is applied into channel estimation and

performance analysis of DL in 5G and Section 4

gives explanation about GFDM, its block structure

and how spectral efficiency is enhanced by using DL

scheme.

2 DEEP LEARNING IN

WIRELESS NETWORK

DL gives immense support to new demands of 5G

communication network. Three methods are

considered to be best one for practical

implementations with technical advancement.

Deep learning-based NOMA: NOMA has

potentials to enhance spectral efficiency as well as

system capacity, so it is assumed to be a 5G

technique but mobility in user results to complex

channel issues in each link. (B.H.Juang, 2018) Over

all operation of NOMA frameworks depends on

CSI. Therefore, there is a necessity of deep learning

for better NOMA performance.

Deep learning-based massive MIMO: Channel

State Information (CSI) has favourable gain effects

of massive MIMO. It is required to know about

channel estimation for getting perfect CSI. Deep

Learning is considered for designing frameworks

with utilization of spatial information of MIMO

systems. Deep Neural Network is a suitable

candidate that assures reconstruction of CSI for

enhanced performance of massive MIMO with

channel estimation. (Hongji, 2019)

Deep learning-based mm Wave: Millimetre Wave

communication is frequently coming into scenario

nowadays because of its low latency and ten-fold

enhancement in bandwidth as compared to present

wireless networks. (H.Huang, 2018) Deep Learning

approach can help in sensing mmWave with the help

of hybrid precoding in terms of matrices.

3 IMPLEMENTATION OF DL IN

CHANNEL ESTIMATION

Fig.1 is block structure of OFDM system with

implementation of Deep neural network (DNN) for

better performance in terms of channel estimation.

Pilot symbols are given into IFFT that converts

signal into time domain from frequency spectrum.

(Michailow, 2014) Cyclic prefix (CP) is added so

that interference (ISI) can be reduced. Received and

transmitted signals are produced into DNN for

minimizing the difference between both ends. In this

manner channel estimation with high resolution can

be achieved.

Figure 1: Channel Estimation DNN Architecture in

OFDM System

Performance Analysis of Deep Learning in 5G:

Deep Learning in NOMA framework plays

important role for optimization of CSI issues.

Typical NOMA system is considered where many

users are served by Base Station and several channel

conditions are present in communication link

individually. DNN is utilised for estimation of CSI

at user separately and to solve optimization

problems in encoding as well as signal detection.

There is a requirement of learning policy to put

DNN into NOMA framework. Present techniques

help in designing simulator for various channel

conditions like slow and fast fading channels. Every

simulation result in generation of data sequences

further CSI is produced by channel models

Modulated

ata Strea

S/P

DFT &

Pilot

Insertio

Paralle

Serial

insertio

Baseband

To R

hanne

Baseband

IFFT

IDF

Chest

Equatio

Serial

arallel

emoval

ICACSE 2021 - International Conference on Advanced Computing and Software Engineering

166

depending upon NOMA principles. Output data at

DNN are named as training data set.

Channel detection in real time is achieved by perfect

DNN framework depending on present input signals

and channel. Fig.2 is the graphical representation of

proposed DNN that supports Single Instruction and

Single Data Stream (SISD) algorithm. NOMA

frames along with pilot symbols are obtained for

generation of data sequences. (Mrinalini,2018)

Finally, channel vectors are produced after feeding

NOMA frames into channel model. Here, analysis of

250,000 is taken for training of DNN with 24 data

sequences present in each batch. From this graph it

can be seen that deep learning has proved as

effective technique for solving problems related to

power allocation and data detection in NOMA[9,10].

It provides high precision to each user linked with

base station.

Figure 2: Deep learning data rate at different CSI

Massive Multiple Input Multiple Output (MIMO) is

one of the suggested methods for coming

communication networks that deals with

enhancement in spectrum efficiency but increased

spatial complexity of system result in decrease in

performance. Deep learning is applied massive

MIMO for channel estimation problems. (R.Li

al.,2017) Fig.3 is the comparison analysis deep

learning scheme and spatial basis expansion model

(SBEM). There is transmission of unit signals in

different direction of massive MIMO in uplink.

Finally, received vectors as output is obtained that

have specific direction. In this MIMO analysis of we

transmit unit signals to the uplink massive MIMO

model in different direction and corresponding

received vectors are obtained. Samples taken for

analysis are 150,000 training dataset,20000 testing

dataset and 30000 validation datasets respectively.

Offline Learning method is processed to the network

for performance analysis of DOA estimation. It can

be clearly seen that BER shows 4 decibel better

performance in DL scheme as compared to other

methods.

BER plays a vital role in analysing performance

in terms of degradation and upgradation of signals.

Figure 3: Massive MIMO based DL scheme

4 GENERALIZED FREQUENCY

DIVISION MULTIPLEXING

(GFDM)

GFDM is a flexible multicarrier technique that is

one the best approach for 5G communications. This

robust technique has CP of low bandwidth that

makes GFDM more spectral efficient.(Nichola,

2014) GFDM is known for its block structure ‘N’

that consists of subcarriers(M) and subchannels(K).

OOB is reduced by applying circular convolution on

subcarriers. GFDM signal is derived by

superposition of all transmit signals given by

following expression:

x(n)=

∑∑

𝑑

𝑀1

𝑚0

𝑘1

𝑘0

k,m

(n) (1)

where gk,m(n) is a time as well as frequency shifted

version of a prototype filter g(n).

Fig.4 is block structure of GFDM in which g

k,m[n]

assumed to be a filter with N samples and k, m and n

are subcarrier, sub symbol, and time indices.

GFDM is flexible for TTI length that helps in taking

data to specific user making it most waveform for

low latency network. GFDM is composed of

adjustable filters present at transmitter end that helps

in reduction of OOB emissions. This structure deals

with the scheduling of many users in frequency and

time domain. This technique is free from high OOB

as well as PAPR values and it finds applications in

white space aggregation.

GFDM Signal is given by following expression:

k,m(n)

=g((n-mk)mod(N)exp(j2П(kn/N)) (2)

A Comparative Analysis of Implementing 5G through Deep Learning

167

where n is sampling index.

Figure 4: DL Based GFDM Transreceiver

Deep Learning is implemented in GFDM, above

proposed method is Joint Detection and

Demodulation JDD DL Scheme.(T.J. O’Shea,2017)

It is a process that contains two stages firstly coarse

detection obtained by using linear detectors like

Zero Forcing (ZF) and Minimum Mean Square Error

(MMSE) and second stage is application of neural

network for getting fine detection.(X.Gao, 2017) In

fig.5 performance analysis is done which has given

parameters training data has 160000 symbols,

GFDM symbol has 90000symbols and testing data

have 90000 symbols.

Figure 5: BER Performance of MMSE-JDD DL System

Figure 6: GFDM Samples

5 CONCLUSION

This paper deals with the explanation of deep

learning and its perspective as well as

implementation in 5G communication. It can be

concluded that DL is one of the prominent methods

to solve complex problems of future

communications. Frameworks based on MIMO and

NOMA is studied with performance analysis

description. BER and SNR analysis is studied that

shows better performance with implementation DL

in 5G. Therefore, conclusion can be drawn that this

method of optimization and analysis has positive

approach with solution to many complex problems

of 5G. Deep learning into 5G has given easy

solutions to the complexity of high frequency

networks. Designing of GFDM and receiving signals

with minimal loss was a tough task but

implementation of DL has helped out to meet

performance and efficiency demands of future

coming networks.

REFERENCES

5G Waveform Candidate Selection

D3.25GNOW_D3.2_v1.3.docx.

Amirhossein Mohammadian, Abbas Mohammadi and

Abdolali Abdipour, Mina Baghani, (2018) Spectral

Analysis of GFDM Modulated Signal under Nonlinear

Behavior of Power Amplifier,Electrical Engineering

and Systems Science ,Signal Processing.

El Gamal, H., & Damen, M. O. (2002). An algebraic

number theoretic framework for space-time coding in

Proceedings IEEE international symposium on

information theory.

G. Gui, H. Huang, Y. Song, and H. Sari, (2018) Deep

Learning for An Effective Non-Orthogonal Multiple

odulate

data

Stream

Time

requenc

Filtering

Insertion

indowin

Remova

State

Matchin

Filter

eshapin

In block

odulate

decoded

Stream

ICACSE 2021 - International Conference on Advanced Computing and Software Engineering

168

Access Scheme, IEEE Trans. Veh. Technol., vol. 67,

no. 9, pp. 8440-8450.

H. Huang, J. Yang, Y. Song, H. Huang, and G. Gui,

(2018) Deep Learning for Super-Resolution Channel

Estimation and DOA Estimation based Massive

MIMO System, IEEE Trans. Veh. Technol., vol. 67,

no. 9, pp. 8549-8560.

H. Ye, G. Y. Li, and B. H. Juang,(2018) Power of Deep

Learning for Channel Estimation and Signal Detection

in OFDM Systems, IEEE Wireless Commun. Lett.,

vol. 7, no. 1, pp. 114-117.

Hongji Huang, Song Guo, and Guan Gui, (2019) Deep

Learning for Physical-Layer 5G Wireless Techniques:

Opportunities, Challenges and Solutions,IEEE

Wireless Communication Magazine.

Michailow, N., Matthé, M., Gaspar, I. S., et al. (2014)

Generalized frequency division multiplexing for 5th

generation cellular networks, IEEE Transactions on

Communications, 62(9), 3045–3061.

Mrinalini, Kamlesh Kumar Singh (2018) Survey paper on

Multicarrier Modulation Techniques ,in 5

IEEE

International Conference on Electrical, Electronics and

Computer Engineering.

M. Mirahmadi, A. Al-Dweik, and A. Shami (2013) BER

reduction of OFDM based broadband communication

systems over multipath channels with impulsive noise,

IEEE Trans. Commun., vol. 61, no. 11, pp. 4602–

4615.

Maximilian, M., Luciano, L. M., Nicola, M., et al. (2015).

Widely linear estimation for space-timecoded GFDM

in low-latency applications. IEEE Transactions on

Communications, 63(11), 4501–4509.

Maximilian, M., Luciano, L. M., Ivan, G., et al. (2015).

Multi-user time-reversal STC-GFDMA for future

wireless networks. Journal on Wireless

Communications and Networking.

Nicola Michailow, Maximilian Matthé, Ivan Simões

Gaspar, Ainoa Navarro Caldevilla, Luciano Leonel

Mendes, Andreas Festag (2014) Generalized

Frequency Division Multiplexing for 5th Generation

Cellular Networks,IEEE Transactions on

Communications, vol. 62, no. 9.

R. Li et al., “Intelligent 5G: When Cellular Networks Meet

Artificial Intelligence,” IEEE Wireless Commun., vol.

24, no.

T. J. O’Shea and J. Hoydis,(2017) An introduction to deep

learning for the physical layer, IEEE Trans. Cognitive

Commun.and Networking, vol. 3, no. 4, pp. 563-575.

X. Gao, L. Dai, Y. Zhang, T. Xie, X. Dai, and Z. Wang,

(2017) “Fast Channel Tracking for Terahertz

Beamspace Massive MIMO Systems,” IEEE Trans.

Veh. Technol., vol. 66, no. 7, pp. 5689-5696

A Comparative Analysis of Implementing 5G through Deep Learning

169