Content uploaded by Will Serrano
Author content
All content in this area was uploaded by Will Serrano on Dec 31, 2020
Content may be subject to copyright.
1
The Blockchain Random Neural Network for cybersecure IoT and 5G
infrastructure in Smart Cities
Will Serrano
The Bartlett, University College London, London United Kingdom
w.serrano@ucl.ac.uk
Abstract
5G promises much faster Internet transmission rates at minimum latencies with indoor and outdoor coverage in Smart Cities. 5G
could potentially replace traditional Wi-Fi for network connectivity and Bluetooth technology for geolocation with a seamless
radio coverage and network backbone therefore accelerating new services such as the Internet of Things (IoT). Although Wi-Fi 6
is already in the market designed for IoT applications. New Smart City applications based on Big Data will depend on 5G as a
mobile Internet service provider therefore eliminating the need to deploy additional private network infrastructure or mobile
networks. The benefits due to the expanded network access and enhanced connectivity between devices also intrinsically increase
cybersecurity risks. Cyber attackers will be provided with additional digital targets; in addition, wireless a nd mobile network
including the access channel infrastructure will be shared between independent services. To address these cybersecurity issues,
this article presents the Blockchain Random Neural Network for Cybersecurity applications in a holistic digital a nd physical
cybersecurity user and channel authentication methods. The user identity is kept secret as the neural weights codify the user
information, although in case of cybersecurity breach, the confidential identity can be mined and the a ttacker identified . The
proposed method therefore enables a decentralized authentication method. The validation results prove that the addition of the
Blockchain Random Neural Network provides a user access control algorithm with increased cybersecurity resilience and
decentralized user access and connectivity.
Keywords ∙ Random neural network ∙ Blockchain ∙ Distributed ledger technology ∙ Cybersecurity ∙ Internet of things 5G ∙ Sma rt
Cities
1 Introduction
There are many definitions of Smart Cities with numerous
approaches that cover their implementation, the key common
concept is that Smart Cities must adapt to its different user needs
and requirements. Functionality must be ready to use and services
sha ll be “invisible” to Sm a rt City users. There is not a unique
Smart City driver, although technology in Smart Cities facilitates
numerous Human to Human (H2H), Machine to Machine (M2M)
and Human to Machine (H2M) combination of both services and
applications. The main Smart City enabler is the Internet of Things
(IoT), an ecosystem of interconnected devices, sensors and servers
that collect, exchange and process Big Data. In the IoT, things a re
objects of the physical world (physical things) that can be sensed
or objects of the information world (virtual things) that can be
digitalized; both objects can be identified, integrated into a digita l
model and information transmitted via sensors and wired or
wireless communication networks (International
Telecommunications, 2012). The IoT is the natural technical
evolution of cloud computing, edge computing, decentralization,
transmission networks and communication protocols. IoT
technologies in Smart Cities enable intelligent decision making
and higher level services that facilitate the provision of a dvanced
functionalities; these innovations present numerous a dvantages
including the optimization of its assets, energy usage and
maintenance.
5G will be an innovative mobile solution for Smart Cities that
consists of very high carrier frequencies with great bandwidth
(mmWave), extreme node and device densities (ultra-
densification) and large numbers of MIMO antennas (m assive
multiple-input multiple-output) (Palattella et al., 2016). Potential
new 5G applications in Smart Cities are divided into Enhanced
Mobile Broadband, Ultra Reliable Low Latency Communications
and Massive Machine Type Communications. 5G will be also
highly integrative: the connection to a 5G air interface and
spectrum together with LTE and WiFi will enable global, reliab le,
scalable,
available and cost-efficient connectivity solutions that provide
high rate coverage and a seamless user experience. These two
features are considered as a potentially key driver for the
expansion of a global IoT into Smart Cities (Andrews et al., 2014).
To achieve these additional requirements supported by high data
rates, extremely low latency, and a significant improvement in
users’ perceived Quality of Service (QoS); the core network of 5G
2
will have to reach higher levels of flexibility and intelligence to
increase node capacity where energy and cost efficiencies will
become even more critical considerations (Agiwal et al., 2016).
Big Data will expand current Smart City services with data
analytics and data correlations to enable informed decisions by
citizens and decision makers (Joshva Devadas et a l., 2019). The
IoT cloud will be key to successfully orchestrate the connection of
an increasing number of smart devices, manage the transmission of
larger volumes of information, store the Big data, make data
operations, data analytics and finally visualise key insights to
Smart City users (Jeet Kaur et al., 2019). Big Data will solve
numerous Smart City issues, mostly related to asset management,
proactive maintenance, energy optimization, usage prediction
where information can be accessed at mobile phone level via a
common Smart City apps.
1.1 Research motivation
These higher levels of network integration and comprehensive data
connectivity between smart devices enabled by the IoT and 5G are
provided at a cybersecurity cost. The benefits of additional
services and performance also intrinsically increase cybersecurity
vulnerabilities as cyber attackers are provided with a greater
network access and additional digital targets (Andrea et al., 20 15;
Deogirikar and Vidhate, 2017; Granjal et al., 2015). Key
cybersecurity risks that 5G incorporates include: 1) the
dependency to a single 5G service provider, mobile users will not
be supported with a back-up infrastructure such as Wi-Fi in ca se
5G service providers do not deliver their agreed Service Level
Agreements. 2) Single access to user information, 5G rogue nodes
could entirely collect user metadata such as geolocation and
personal data. In addition, IoT also includes key cybersecurity
vulnerabilities: 1) Unsecured IoT devices can be orchestrated to
launch larger and greater Distributed Denial of Services (DDoS)
attacks. 2) Easier reach to cloud and edge computing servers via
the additional IoT gateways. 3) IoT solutions are normally
provided with dedicated connections to the dedicated cloud from
the service provider, this additional decentralized connectivity
enables further backdoors access and spyware.
1.2 Research Proposal
Recent Blockchain solutions enable the digitalisation of contracts
as it provides authentication between parties and the encryption of
information that gradually increments while it is processed in a
decentralized network such as the IoT (Huh et al., 2017). To
address the increased cybersecurity risk of 5G in the IoT for Smart
City applications and services such as rogue 5G nodes; this article
proposes a holistic digital and physical cybersecurity user
authentication method based on the Blockchain Random Neural
Network. The Blockchain Random Neural Network configuratio n
has analogue biological properties as the Blockchain where
neurons are gradually incremented and chained through synapses
as variable user access credentials increase. In addition,
information is stored and codified in decentralized neural networks
weights. The Blockchain Random Neural Network solution is
equivalent to the Blockchain with the same properties: user
identity authentication, data encryption and decentralization where
user access credentials are gradually incremented and learned. The
presented application can be generalized to emulate an
Authentication, Authorization and Accounting (AAA) server
where user access information is encrypted in the neural weights
and stored decentralized servers. The main advantage of this
research proposal is the biological simplicity of the solution
however it suffers high computational cost when the neurons
increase.
1.3. Research structure
Distributed Ledger Technologies in the IoT, 5G, IoT, Neural
Networks applications in Cryptography and Blockchain solutions
in Cybersecurity, Software Defined Networks and Privacy-
Preserving Machine Learning related work is described in Section
2. The Blockchain Random Neural Network mathematical m odel
presents in Section 3 a decentralized solution that emulates the
Blockchain validation process: mining the input neurons until the
neural network solution. Two models of the presented Blockchain
algorithm applied to Cybersecurity in Smart Cities are proposed in
Section 4. The user access authentication covers holistically its
digital access through the seven OSI layers and its physical
identity, such as passport ID, before is permitted access to IoT
network resources. The method forces the user to be physically
authenticated before access to the IoT network is allowed,
therefore cybersecurity is increased by reducing the likelihood of
criminal network access. The user digital OSI layer identification
that includes its MAC and IP address in addition to its physical
identification such as biometrics generate the Private Key, whereas
there is no need for a Public Key. The 5G node identity
authentication model follows a similar approach; 5G nodes are
authenticated before they are enabled into the Smart City where
the associated International Mobile Subscriber Identity (IMSI)
provides the Private Key. Experimental results in Section 5
demonstrate that the additional Blockchain Random Neural
Network provides increased cybersecurity resilience and
decentralized confidentiality to user access, connectivity and
authentication. The main conclusions presented in Section 6 share
that the user identity is kept secret codified in the neural weights
although in case of cybersecurity breach the identity can be mined
and the attacker identified.
2 Research background
There are already security frameworks that integrate Blo ckchain
technology with smart devices for Smart City applications (Biswas
and Muthukkumarasamy, 2016); the solution is structured in a
physical layer formed of sensors and actuators that ga ther a nd
transmit data to the next protocol layer. The communications layer
provides different data transmission methods such as Wi-Fi, 5G,
Ethernet and Bluetooth where Blockchain protocols are inserted to
enable the security and privacy of data. The database layer consists
on the distributed ledger that stores the records from the physical
layer; these records are classified as permissionless or pub lic and
permissioned or private. Finally, the interface layer consists of the
several smart applications such as smart parking, smart home,
smart health that integrate with each other for effective decisions.
An Intelligent Transport System based on the Blockchain is also
implemented via several layers (Yuan and Wang, 2016): the
physical layer covers the field assets of the solution, the data layer
specifies the chained data blocks and the related Blockchain
techniques such as asymmetric encryption, time stamping, ha sh
algorithms and Merkle trees. The network layer provides the
3
mechanisms of distributed peer to peer networking, data
forwarding and verification. The consensus layer covers the
agreement algorithms such as proof of work, proof of stake or
proof of movement. The incentive layer inclu des the economic
reward into the Blockchains and specifies its issuance and
allocation methods. The contract layer contains scrips, algorithms
and smart contracts which are self-verifying, self-executing a nd
self-enforcing; this layer is activated with the static data stored in
the Blockchain. Lastly, the application layer packages services and
applications such as ride sharing and logistics.
2.1. Distributed Ledger Technologies in the IoT
Distributed Ledger Technologies (DLT) and the services they
provide are the solution for the IoT management of connected
devices, privacy and reliability issues. (Arslan et al., 2020). DLT
applications include tracking of connected devices, enabling the
transaction process and verification while keeping data semi-
anonymous. In addition, due to its decentralized architecture, DLT
also avoids single points of failure and provides mass scale
reliability for the stored and communicated content. Th e current
centralized IoT architecture faces several issues as all operations of
all the devices in the network are computed using a single server.
This creates a single point of failure while creating the server as a
target of security and privacy attacks (Atlam and Wills, 2019 ).
DTL provides solutions to these drawbacks such as lack of
maturity, scalability, latency and interoperability in a ddition to
legal and regulatory challenges. IoT requirements such as security,
privacy, identity management, M2M transactions traceability and
provenance could be addressed with DTL features based on
decentralized information, cryptographic security, transparency
and tamper proof features (Zhu et al., 2019). Key issues of
adopting DLT in the IoT is the scalability due to overheads from
metadata, such as digital signatures, and computing la tency that
prevents DLTs on real time applications (Elsts et al., 2018).
Alternatives such as IOTA can be considered although the Proof or
Work is still not feasible in IoT devices characterised by low
computing power and low battery. Another obstacle to consider for
DTL in wireless IoT is the high amount of the data in relation to
the small payloads of the Wireless protocols (Danzi et al., 2020).
DTL technologies have been designed and optimised for mostly
upload connections whereas its application in Wireless solutions
requires uplink and download link traffic. Finally DLT solutions
do not support the required service differentiation demanded by
IoT services with high reliability requirements. To address some of
the issues of IoT information security, a lightweight authentication
mechanism in cloud-based IoT environment is proposed that
enables authenticated users to access the data of IoT sensors
remotely (Wazid et al., 2020). The algorithm is based on one-way
cryptographic hash functions along with bitwise XOR operations
and a fuzzy extractor mechanism employed at the use r ’s e n d f o r
local biometric verification. Similarly, a lightweight Block chain -
enabled RFID-based authentication protocol is used for supp ly
chains in 5G mobile edge computing environments (Jangirala et
al., 2020). The solution supports the required bandwidth to secure
real-time data about goods in transit within supply chains. Another
authentication scheme for the exchange of in formation via the
cloud is based on two connections: authentication between a user
and a cloud server and authentication between an IoT device and a
cloud server (Challa et al., 2020). Both entities must first mutually
authenticate each other with the help of a trusted authority; once
the authentication is successful, they can establish a session key
for their future secure communication. A comprehensive analysis
explores cutting-edge IoT technologies though the fundamentals,
principles, architectures, applications, challenges, and promises of
the Internet of Things (Firouzi et al., 2020). The study also covers
details of IoT and its underlying technologies from embedded
systems to cloud computing, big data analytic and the ability of
IoT within healthcare for personalized and tailored content in a
cost-effective health delivery. Similar to Blockchain, Directed
Acyclic Graphs (DAG) also store data transactions which are
represented as nodes. Transactions are directed linked to one or
several other newer transactions in topological ordering without
loops (Liu et al., 2020). DAGs do not have blocks therefore there
is not a mining process. Every new transaction refers to its pa rent
transactions, signs their hashes validating the DAG, a nd finally
includes the hashes in the new transaction. Although Blockchains
possess better immutability than DAGs; these perform better at
handling a large number of transactions with a greater scalability
than Blockchains.
2.2. 5G
5G will support a wide range of industry business models and use
cases that address motivations from different stakeholders such a s
mobile network operators, infrastructure, cloud service providers
and tenants (Adam and Ping, 2018); these diverse devices and
applications will lead to several cybersecurity requirements, in
addition to performance requirements from future application s as
well as regulatory compliance with Service Level Agreements in
multi-tenant cloud environments. SHIELD is a novel design an d
development cybersecurity framework that offers Security as a
Service in 5G (Katsianis et al., 2018). SHIELD framework
manages network functions and Software Defined Networks
(SDNs) for virtualization and dynamic placement of virtual
network security, Big Data analytics for real-time incident
detection and mitigation, trusted computing attestation techniques
for securing both the infrastructure and its services. Future 5G
cellular networks will facilitate the physical system
communications through different technologies such as device to
device communications that need to be protected from
eavesdropping, especially with the large amount of traffic that will
constantly flow through the network (Atat et al., 2017). The
connectivity of numerous stand alone IoT systems thro ugh the
Internet or 5G Networks introduces numerous cybersecurity
challenges as sensitive information is prone to be exposed
(Mozzaquatro et al., 2018); an ontological analysis improves IoT
cybersecurity by delivering security services adapted to dif ferent
threats. The next wireless strategy is centered on 5G and IoT
functionality (Chang et al., 2017); while the confidence of the
future mobile technology will help innovate governments,
workforce, industry and social media, the key threat to future
mobile wireless development is the growing concern of its
anomaly detection. Fog and mobile edge computin g will pla y a
key role in the upcoming 5G mobile networks to support
decentralized applications, data analytics and self network
management by using a highly distributed computing model
(Fernandez et al., 2018). User centric cybersecurity solutions in 5G
networks apply Deep Learning techniques to detect network
anomalies (Fernandez et al., 2018). The method requires the
collection, process and analysis of large a mount of data tra ffic
from a great number of network connections in real time and in an
autonomous way. The protection of the 5G is key for its success as
system security is the foundation for 5G development (Lu and Da -
Xu, 2018). The significant factors of the security model a re the
protection and integration of heterogeneous smart devices with
4
Information Communication Technologies (ICT) where m obile
cloud computing leverages 5G mobile technologies. With the
development of 5G, defensive mitigations a gainst threats from
wireless communications based on mobile cloud computing have
been playing a notable role in the Web security domain and
Intrusion Detection Systems (IDS) (Gai et al., 2016).
2.3. Internet of Things
The evolution of the IoT technology started from Machine to
Machine to connect machines and devices, Interconnec tions of
Things that connect any physical or virtual object and finally Web
of Things that enables the collaboration between people and
objects (Lee et al., 2013). The IoT is formed of three layers (Jing et
al., 2014): sensor, transportation and application that sim ilar as
traditional networks, also present security issues and integration
challenges. Because physical, virtual and user private informatio n
is captured, transmitted and shared by the IoT sensors, the
enforcement of security and privacy policies shall include
cybersecurity threats on data confidentiality and authentication,
access control within the IoT network, identity management,
privacy and trust among users and things (Roman et a l., 2011).
The dynamic IoT is formed by heterogeneous technologies that
provide innovative services in various application domains wh ich
shall meet flexible security and privacy requirements. Traditional
security mitigations cannot be directly applied to the IoT due th e
different standards, communication protocols and scalability issues
due to of the greater number of interconnected devices (Sica ri et
al., 2015). An important challenge for supporting diverse
multimedia applications in the IoT is the heterogeneity of security
in wired and wireless sensor and transmission networks that
requires a balance between flexibility and efficiency (Zhou and
Chao, 2011). Secure and Safe Internet of Things (SerIoT)
improves the information and physical security of different
operational IoT application platforms in a holistic and cross
layered manner (Gelenbe et al., 2018). SerIoT covers areas such as
mobile telephony, networked health systems, the Internet of
Things, Smart Cities, smart transportation systems, supply chain s
and industrial informatics (Domanska et al., 2018).
2.4. Neural networks and cryptography
A linear scheme based on the Perceptron problem, or NP problem,
is used in zero-knowledge interactive identification protocols
(Pointcheval, 1994). Three and five pass identification protocols
perform similar to other smart cards applications where an
identification process between the subject and the authoriser
requires approximately 6 Kbytes of data to obtain the standard
security level of 10 6 with a very secure secret key. Two
multilayer neural networks are trained on their mutual output bits
with discrete weights to achieve a synchronization that can be
applied to a secret encryption key exchange over a public channel
(Kinzel and Kanter, 2002). The two trusted parties chose the secret
initial weight vectors, agree on a public sequence of input vectors
and exchange public bits where for each transmission, a new secret
key has to be agreed without any previous secret information
stored. The above key exchange protocol can be broken by three
different cryptanalytic attacks (Klimov et al., 2002). The genetic
attack simulates a large population of neural networks with the
same structure as the two parties and trains them with the same
inputs where networks whose outputs mimic those of the two
parties stay and multiply while unsuccessful networks die. The
geometric attack exploits the neural configuration of the attacker
network based on the similarity to the unknown network although
it provides a different output for a given input. Finally, the
probabilistic attack considers the complexity to predict the position
of a point in a bounded multidimensional box after several random
transitions instead of predicting accurately its original position.
Feed forward neural networks are applied as a n encryption and
decryption algorithm with a permanently changing key (Volna et
al., 2012). The method consists on the division of an input
message into 6-bit data sets, no predetermined number of neurons
in the hidden layer and an output layer that is formed of six
neurons, one for each bit. The network is trained on the binary
representation of the ASCII symbols. A two-sta ge multilayered
neural network also provides a cryptography solution (Yay ık and
Kutlu, 2014): The neural network of the first stage generates based
pseudo random numbers used parameters to initia lize the input
values, initial network weights and the number of neurons within
the hidden layers. The neural network of the second stage encrypts
information based on the non-linearity of the model; each block is
formed of seven bits that represent the ASCII values where digits
are mixed with other blocks with a random number of iterations.
This random number is obtained as a parameter from the first stage
neural network. A high level review of neural networks
applications to different cryptography elements (Schmidt et a l.,
2008) covers private key protocols such as key exchange protocol,
private and public keys. In addition, the review also includes
visual cryptography that transmit secret images across a public
network, hidden communications within redundant bits inserted in
the transmission of the message, design of random generators and
finally digital watermarking, the inclusion of a signature into the
media in order to provide authentication and content protection.
2.5. Blockchain and security
Currently; there is a great research effort in Blockchain algorithms
applied to security applications. A punishment method that
includes adversary behaviours is based on the Blockchain action
ledger that eliminates the incentive to attack edge servers and
mobile devices in the edge network (Xu et al., 2017). The
interactions between a mobile device and an edge server are
modelled as a Blockchain ledge where mobile devices either
transmit a request to the server in order to obtain real-time service
or launch an attack against the server for illegitimate a ccess. On
the other hand, the server also decides to permit the request from
the mobile device or attack it. An ISO/IEC 15408-2 compliant
security auditing system is generated from a Blockchain network
embedded into the backbone communication architecture (Cha and
Yeh, 2018). The Blockchain auditing system covers
authentication, access control, block query, de-identification,
encryption and key management. A conceptual model fuses
Blockchains and cloud computing over three deployment modes
(Gai et al., 2018): Cloud over Blockchain, Blockchain over Cloud
and Mixed Blockchain-Cloud. A Blockchain consensus model
implements IoT security (Gupta et al., 2018) for devices that
present constrained resources. The method applies a distributed
voting system for power and bandwidth requirements to prevent
Denial of Service attacks. A Blockchain mechanism continuously
evaluates legitimate presence of user in valid IoT zones without
user intervention (Agrawal et al., 2018). The solution addresses the
lack of trust between IoT entities, removes single points of failure
and predicts the user next zone visit via the use of IoT crypto
tokens and Markov prediction models. A secure mutual
authentication with an exhaustive access control system for
industry 4.0 is proposed based on a Blockchain (Lin et al., 2018).
5
The presented solution provides anonymous authentication,
auditability and confidentiality based on integrated attribute
signature, multi-receivers encryption and a message authentication
code. A decentralized Ethereum Blockchain enables smart
contracts to survey data from the Industrial Internet of things
(IIoT) to the cloud (Fan et al., 2020). The Ethereum method
provides four security design features: the selection of current
Blockchain nonce as a random seed, the writing of the auditing
rules into the Blockchain, the use of the block number on the
Ethereum as the security timestamp and the requirement of
payment of ether as deposit.
2.6. Software Defined Networks
Software Defined Networks (SDN) detach the forwarding process
of network packets in the hardware (data plane) from the routing
process in the software (control plane). This approach enables a
cloud based network management with similar advantages:
efficient network configuration, performance and monitoring
based on remote access. An overview of current SDN technologies
for IoT applications (Bera et al., 2017) is based on edge
networking, access, core and data center networking. The key
issues of the integration of SDN with IoT ranges from wireless
mobility, adequate policy enforcement, independent hardware
platform, and practical deployments. An IoT architecture combines
the feature of SDN for traffic control and resource management
with fog computing for low and predictable latency (To movic et
al., 2017) to address the key intrinsic IoT issue based on low
power low bandwidth devices. The functionality of fog
orchestration is delegated to an SDN controller in order to achieve
higher efficiency whereas some SDN controller tasks are delegated
to the fog node for higher scalability. In addition, a Blockchain
based distributed cloud architecture with an SDN enable controller
and fog nodes at the edge of the network is proposed for IoT
applications (Kumar et al., 2018). The model provides low-cost,
secure and on-demand access to the computing infrastructure in an
IoT network where the exchange tokens register key actions such
as the performance of a computation, the transfer of a file or the
provision of a set of data. A cooperative IoT security via SDN
reduces the risk of Cybersecurity vulnerabilities in the IoT by
orchestrating defense mechanisms that share the attacking
information with peer controllers and block the attacks (Grigoryan
et al., 2018). A secure framework that acts as an intrusion
detection system for IoT based on SDN generalizes the integration
of SDN and IoT based on Deep Learning (Dawoud et al., 2018).
The anomaly detection module uses Restricted Boltzmann
Machines (RBM) to measure the data loss between the input and
the reconstructed record. A detection approach that applies
Bayesian models detects malicious devices inside an SDN using
trust computation (Meng et al., 2018). After identifying untruthful
devices by means of a Bayesian inference approach, the SDN
controller updates its routing tables and direct traffic to bypass
malicious points.
2.7. Privacy-Preserving machine learning
Machine Learning and Artificial Intelligence can be ap plied for
Privacy-Preserving user data. A framework protects the privacy
requirement of different data providers within the cloud (Li et a l.,
2018). The method is based on a public-key encryption with a
double decryption algorithm that encrypt their data sets with
different public keys. For a successful implementation of machine
learning systems for privacy concerns, the knowledge gap between
Machine Learning and privacy communities must be addressed
(Al-Rubaie and Chang, 2019). Key issues of Machine Learning for
privacy preserving are the abundance of new methods, the
scalability of the algorithms in terms of processing and
communication costs, the security assumptions a nd the policies
that specify the rules or privacy guarantees. Machine Learning as a
Service trains and deploys machine learning models on cloud
providers’ infrastructure however these services require
accessibility to the raw data which is often privacy sensitive and
may generate potential security and privacy risks (Hesamifard et
al., 2018). CryptoDL is a client-server cloud method that applies
neural network algorithms to encrypt data enabling a data sharing
service without revealing sensitive data. Privacy-preserving
distributed Machine Learning for permissioned Blockchain
networks deploys learning models without data centralization
while addressing the privacy, security and performance issues of
the traditional models (Kim et al., 2019). The model treats any
private learning algorithm as a non-deterministic function where
an effective error-based aggregation rule is defined to prevent
attacks that try to deteriorate the accuracy of the model.
3 The Blockchain Random Neural Network
Blockchain (Nakamoto, 2008) is based on cryptographic concepts
which can be applied similarly by the use of Neural Networks.
Information, or Big Data, in the Blockchain is contained in blocks
that also include a timestamp, the number of attempts to mine the
block and the previous block hash. Decentralized miners then
calculate the hash of the current block to validate it. Smart City
information contained in the Blockchain consists of tra nsactions
which are authenticated by a signature that uses the Smart City
asset private key, transaction origin, destination and value (Fig. 1).
Hash
Time Stamp
Transactions
Iterations
Previous Hash
Hash
Time Stamp
Transactions
Iterations
Previous Hash
Hash
Time Stamp
Transactions
Iterations
Previous Hash
Block n-1 Block n Block n+1
Transaction
From
To
Data
Signature
Signature = Function (Private Key,
From, To, Value)
Verify Signature = Function (Public Key,
Signature, From, To, Value)
Fig. 1. Blockchain Model
3.1. The Random Neural Network
The Random Neural Network (Gelenbe, 1989, 1993a, 1993b) is a
spiked recurrent stochastic model. The main analytical properties
are the “product form” and the existence of a unique network
steady state solution. The Random Neural Network model
represents more closely how signals are transmitted in many
biological neural networks where they travel as spikes rather than
as fixed analogue signals.
3.1.1. Definition
The Random Neural Network consists on n-neurons. The state of
the n neuron network at time t is represented by the vector of non -
negative integers k(t) = [k1(t), … ki(t)] where ki(t) is the potential
6
of neuron i at time t. Neurons interact with each other by
interchanging signals in the form of spikes of unit amplitude:
• A positive spike is interpreted as excitation signal because it
increases by one unit the potential of the receiving neuron m,
km(t+) = km(t) + 1;
• A negative spike is interpreted as inhibition signal decreasing
by one unit the potential of the receiving neuron m, km(t+) =
km(t) – 1, or has no effect if the potential is already zero, km(t) =
0.
Each neuron accumulates signals and it will fire if its potential is
positive. Firing will occur at random and spikes will be sent out at
rate r(i) with independent, identically and exponentially distributed
inter-spike intervals:
• Positive spikes will go out to neuron j with probability p+(i,j) as
excitatory signals;
• Negative spikes with probability p-(i,j) as inhibitory signals.
3.1.2. Principles
A neuron may send spikes out of the network with probability d(i).
We have:
di+p+i,j+p-i,j=1 for 1 ≤ i ≤ n
n
j=1
(1)
Neuron potential decreases by one unit when the neuron fires
either an excitatory spike or an inhibitory spike (Fig. 2). External
(or exogenous) excitatory or inhibitory signals to neuron i will
arrive a t rates Λ(i), λ(i) respectively by stationary Poisson
processes.
x1w+(j,i): excitatory
network weights
x2
xL
z1
z2
zM
y1
y2
yN
Input Layer Hidden Layer Output Layer
w-(j,i): inhibitory
network weights
Λ1
λ1
Λ2
λ2
ΛL
λL
Λ: External
excitatory signal
λ: External
inhibitory signal
External signals
i1
i2
iL
Fig. 2.The Random Neural Network structure
The Random Neural Network weight parameters w+(j,i) and w-(j,i)
are the non-negative rate of excitatory and inhibitory spike
emission respectively from neuron i to neuron j:
w+j,i=r(i)p+(i,j)≥0
w-j,i=r(i)p-(i,j)≥0
(2)
Information in this model is transmitted by the rate or frequency at
which spikes travel. Each neuron i, if it is excited, behaves as a
frequency modulator emitting spikes at rate w(i,j) = w+(i,j) + w-
(i,j) to neuron j. Spikes will be emitted at exponentially distributed
random intervals (Fig. 3). Each neuron acts as a non-linear
frequency demodulator transforming the incoming excitatory a nd
inhibitory spikes into potential.
qi
λi
Λiqiri(p+i,1+p-i,1)
qiri(p+i,k+p-i,k)
q1r1(p+1,i+p-1,i)
qkrk(p+k,i+p-k,i)di
w+j,i = rip+i,j
w-j,i = rip-i,j
Non negative rates spike emission
Excitatory / Inhibitory
Fig. 3.The Random Neural Network model
This network model has a product form solutio n; th e n e t w or k’ s
stationary probability distribution can be represented as the
product of the marginal probabilities of the state of each neuron.
3.1.3. Theorem
Let’s define the proba bility distribution of the network sta te as
p(k,t) = Prob[k(t) = k)] and the marginal probability a neuron i is
excited at time t as qi(t) = Prob[ki(t)>0]. The stationary probability
distribution p(k) = limt →∞p(k,t) and qi = limt→∞qi(t) where k(t) is a
continuous time Markov chain that satisfies Chapman-
Kolmogorov equations.
Let’s define:
qi= λ+(i)
ri+λ-(i)
(3)
ri=w+i,j+w-i,j for 1 ≤ i ≤n
n
j=1
(4)
where the λ+(i), λ-(i) for i=1,…,n sa tisfy the system o f n onlin ea r
simultaneous equations:
λ+i=qjr(j)p+(j,i)+Λ(i)
n
j=1
(5)
λ-i=qjr(j)p-(j,i)+λ(i)
n
j=1
(6)
If a nonnegative solution {λ+(i),λ-(i)} exists to the equations (3),
(5) and (6) that meets qi < 1 then:
pk=1-qiqi
ki
n
i=1
7
(7)
The network will be stable if a value qi < 1 can be found. The
average potential at a neuron i is qi/[1-qi] and the rate of emission
of spikes from neuron i in steady state is qir(i). If we h a ve λ+(i) >
[r(i) + λ-(i)] for any neuron means that the neuron is unstable or
saturated; this implies that it is constantly excited in steady state
and its rate of excitatory and inhibitory spike emission r(i) to
another neuron j will be r(i)p+(i,j) and r(i)p-(i,j) respectively.
3.2 The Blockchain Random Neural Network
The Random Neural Network with Blockchain configuration
consists of L Input Neurons, M hidden neurons and N output
neurons. Information in this model is contained within the
networks weights w+(j,i) and w-(j,i) rather than neurons xL, zM, yN
themselves.
• I = (ΛL, λL), a variable L-dimensional input vector I Є [-1,1]L
represents the pair of excitatory and inhibitory signals entering
each input neuron respectively; where scalar L values range
1<L< ∞;
• X = (x1, x2, … , xL), a variable L-dimensional vector X Є [0,1]L
represents the input state qL for the neuron L; where scalar L
va lues ra nge 1<L< ∞;
• Z = (z1, z2, … , zM), a M-dimensional vector Z Є [0,1]M that
represents the hidden neuron state qM for the neuron M; where
scalar M values range 1<M< ∞;
• Y = (y1, y2, … , yN), a N-dimensiona l vector Y Є [0,1]N that
represents the neuron output state qN for the neuron N; where
scalar N values range 1<N< ∞;
• w+(j,i) is the (L+M+N) x (L+M+N) matrix of weights that
represents from the excitatory spike emission from neuron i to
neuron j; where i Є [xL, zM, yN] and j Є [xL, zM, yN];
• w-(j,i) is the (L+M+N) x (L+M+N) matrix of weights that
represents from the inhibitory spike emission from neu ron i to
neuron j; where i Є [xL, zM, yN] and j Є [xL, zM, yN].
The key concept of the Blockchain Random Neural Network
model is that the neuron vector sizes, L, M and N are variable
rather than fixed. The input layer X corresponds the user
incremental data; the hidden layer Z are the neural v alues of the
chain and the output layer Y codifies the user Private Key (Fig. 4).
Neurons or blocks are iteratively added where the value of the
additional neurons consists on both the value of the additional
information and the value of previous neurons therefore forming a
neural chain.
Validation 1
V(1) Validation 2
V(2) Validation t
V(t)
Data Neuron
Link Neuron
Key Neuron
Fig. 4.The Blockchain Random Neural Network
3.2.1. The Blockchain Random Neural Network Model.
The Blockchain Random Neural Network model is ba sed on the
main concepts shown on Figure 5:
• Private key, yN;
• Validation, V(t) and Data D;
• Neural Chain Network and Mining;
• Decentralized information, w+(j,i) and w-(j,i).
Data 1
d1
No Mining
User
yN
Decentralized Neural Network
w+(j,i) w-(j,i)
Data 2
d2
Mines xL and zM
Validation 1
V(1)
User
yN
User
yN
Validation 2
V(2) Validation t
V(t)
Data t
dt
Mines xL and zM
Fig. 5.The Blockchain Random Neural Network Model
3.2.2. Private Key.
The private key Y = (y1, y2, … , yN) consists on the user or
application digital credentials specifically assigned to them. This
includes biometrics or any other cryptographic algorithm such as
Advanced Encryption Standard (AES) 256-bit cipher. The private
key is presented by the user, or the application, every tim e new
data is to be inserted therefore requiring user credentials
validation.
3.2.3. Validation and Data.
The neural network blockchain model defines Validation and Data
as:
• Validation, V(t) = {V(1), V(2), … V(t)} as a va riable
accumulative vector where t is the validation stage;
• Data, D = {d1, d2, … dt} as a set of t I-vectors where do = (eo1,
eo2, … eoI) and eo are the I different dimensions for o=1,2, ... t.
The first Validation V(1) has associated an input state X = x I tha t
corresponds d1 representing the user data. The output state Y = yN
corresponds to the user Private Key and the hidden la yer Z = z M
corresponds to the value of the neural chain that will be inserted in
the input layer for the next transaction. The calculated neural
network weights w+(j,i) and w-(j,i) are stored in the decentralized
network and retrieved in the mining process. The second
Validation V(2) has associated an input state X = xI that
corresponds to the user data d1 for the first Validation V(1), the
chain, or the value of the hidden layer zM and the additional user
data d2. The output state Y = yN still corresponds the user Priva te
Key and the hidden layer Z = zM corresponds to the value of th e
neural chain for the next transaction. This process iterates as more
user data is inserted. The neural chain can be formed from the
values of the entire hidden layer neurons, a selection of neurons, or
any other combination to avoid the reverse engineering of the
Private Key from the stored neural weights.
3.2.4. Neural Chain Network and Mining.
Data is validated or mined by calculating the outputs of the
Random Neural Network using the transmitted network weighs,
w+(j,i) and w-(j,i) at variable random inputs X = xI, o r following
any other method. This process emulates the traditional
8
Blockchain Proof of Work mining process where miners ra ce to
find a target hash. The Random Neural Network with Blockchain
configuration is mined when an Input I is found that delivers an
output Y with an error Ek lesser than a threshold T for the retrieved
user neural network weights w+(j,i) and w-(j,i):
Ek=1
2y'n-yn2
N
n=1
<T
(1)
where Ek is the minimum error or threshold, y’n is the output of the
Random Neural Network with mining or random input X = xI a nd
yn is the user or application Private Key. The mining complexity
can be tuned by adjusting Ek. Once the solution is found, or mined,
the user or application data can be processed. The potential va lue
of the neural hidden layer Z = zM is added to form the neural chain
as the input of the next transaction along with the user new data.
Finally, the proposed method calculates the Random Neural
Network with gradient descent learning algorithm for the new pair
(I, Y) where the new generated neural network weights w+(j,i) and
w-(j,i) are stored in the decentralized network. The mining process
increases on complexity as more user data is inserted.
3.2.5. Decentralised Information.
The user data I is crypted in the neural network weights w+(j,i) and
w-(j,i) are stored in the decentralized network rather than
distributing its data I directly. The user data is decrypted within the
mining process when the user presents its biometric key therefore
making secure to store information in a decentralized system. The
neural network weights expand as more verification data is
inserted creating an adaptable method.
4 The Blockchain Random Neural Network in Smart Cities
4.1 Internet of Things Cybersecurity Model
The Internet of Things Cybersecurity model as Smart City
application based in the Blockchain Random Neural Network is
described in this section. The key concepts for this roaming service
are represented on Figure 6:
Area 1
v1
No Mining
User
yN
Decentralized Network
w+(j,i) w-(j,i)
Area 2
v2
Mines xL and zM
Roaming 1
R(1)
User
yN
User
yN
Roaming 2
R(2) Roaming 3
R(3)
Area 3
v3
Mines xL and zM
Area t
vt
Mines xL and zM
User
yN
Roaming t
R(t)
Fig. 6. The Internet of Things Random Neural Network Model
The private key Y = (y1, y2, … , yN) consists on the user digital
AAA authentication credentials that covers the seven layers of the
OSI model and physical information such as a passport, biometrics
or both. The private key is presented by the user every time its
credentials require verification from the accepting roaming node
(Table 1).
Table 1. Private Key
Private Key
Bits
Reference
Type
Interface
y8
72
User
Physical
Passport - Biometrics
y7
16
Web
Digital
OSI Layer 7
y6
16
Middleware
OSI Layer 6
y5
16
Socket
OSI Layer 5
y4
16
Port
OSI Layer 4
y3
32
IP
OSI Layer 3
y2
48
MAC
OSI Layer 2
y1
16
Bit
OSI Layer 1
The first Roaming R(1) has associated an input state X = xI which
corresponds to v1 and represents the user verification data. The
output state Y = yN corresponds to the user Priva te Key and the
hidden layer Z = zM corresponds to the value of the neural ch ain
that will be inserted in the input layer for the next roaming. The
second Roaming R(2) has associated an input state X = xI wh ich
corresponds to the user verification data v1 for the first Roaming
R(1), the chain (or the value of the hidden layer zM) and the
additional user data v2. The output state Y = yN still co rresponds
the user Private Key and the hidden layer Z = zM corresponds to
the value of the neural chain for the next transaction.
The first Roaming R(1) calculates the Random Neural Network
neural weights with an Ek < Y for the input data I = (ΛL, λL) a nd
the user private key Y = yN. The calculated neural network weights
w+(j,i) and w-(j,i) are stored in the decentralized network and are
retrieved in the mining process. After the first Roaming; the user
requires to be validated at each additional Roaming with its private
key. Its verification data is validated and verification data v t f rom
Roaming R(t) are added to the user profile.
4.2 5G Cybersecurity model
The 5G Cybersecurity model as Smart City applicatio n based in
the Blockchain Random Neural Network is described in this
section. The key concepts for this authentication service are
represented on Figure 7:
9
Floor 1
v1
No Mining
Decentralized Network
w+(j,i) w-(j,i)
Floor 2
v2
Mines xL and zM
Authentication 1
A(1) Authentication 2
A(2) Authentication 3
A(3)
Floor 3
v3
Mines xL and zM
Floor t
vt
Mines xL and zM
Intelligent Building
IMSI yN
Authentication t
A(t)
Fig. 7. 5G Blockchain Random Neural Network Model
The private key Y = (y1, y2, … , yN) consists on the Intelligent
Building International Mobile Subscriber Identity (IMSI). The
private key is presented by the Intelligent Building within the
Smart City every time a 5G node becomes operational and its
identity credential requires authorization and verification from the
Intelligent Building in order to enable its digital channel (Table 2)
for IoT services.
Table 2. IMSI Private Key yN
Identifier
Code
Digits
Bits
Name
Key
IMSI
MCC
3
12
Mobile Country Code
y3
MNC
3
12
Mobile Network Code
y2
MSIN
9
36
Mobile Subscription
Identification Number
y1
The first Authentication A(1) corresponds to the 5G node identity
data v1. The output layer of the Blockchain neural network is
assigned to the Intelligent Building Private Key and the hidden
layer contains the value of the neural chain that will be inserted in
the input layer for the next authentication. The second
Authentication A(2) has associated the 5G node identity data for
the first Authentication A(1), the chain, or the value of the hidden
layer zM and the additional 5G node identity data v2. The output
state Y = yN still corresponds the Intelligent Building Private Key.
This process iterates as more 5G node identity data is inserted in a
staged enabling process. After the first Authentication; the 5G
node requires to be validated at each additional Authentication
with the Intelligent Building private key. The 5G node verification
data is validated and verification data vt from Authentication A(t)
are added to the Intelligent Building information.
5 Smart City Blockchain Random Neural Network Validation
This section proposes a practical validation of the Blockchain
Random Neural Network model in the Smart City that covers the
Internet of Things and 5G. The software used in the experiments is
the network simulator Omnet ++ with Java. In both validations the
Blockchain Random Neural Network is trained based on an in put
state X = xI that corresponds the node Vt and the output state Y =
yN represents the user Private Key YN.
5.1 Blockchain Cybersecurity in IoT
The three independent experiments will emulate a 1) Bluetooth
network with roaming validation of MAC a ddresses, 2) WLAN
network with roaming validation of MAC and IP addresses and 3)
LTE Mobile network with a roaming validation of MAC a nd I P
and user Passport (Table 3).
Table 3. Neural Blockchain in IoT and Cybersecurity Validation – Node
values
Variable
Bluetooth Master
Wireless LAN
Access Point
Mobile LTE
Base Station
Use
Room
Floor
Building
Campus
City
Country
Coverage
10m
3140m2
100m
0.314km2
1km
31.4km2
Layer
MAC
MAC-IP
MAC-IP
PASSPORT
Node Vt
48 bits
01-23-45-67-89-XX
48+32 bits
192.168.11.XX
48+32 bits
N/A
User yN
48 bits
01-23-45-67-89-AB
48+32 bits
192.168.11.11
48+32+72 bits
VGD12345F
The user is assigned a private key yN that requires validation
before is allowed to transmit. When the user travels through the
space, the credential private key is validated by the roaming node.
The decentralized system retrieves the neural weights associated to
the private key, mines the block, adds the node code and stores
back the neural network weights in the decentralized system. Th is
validation considers mining as the selection of random neuron
values until Ek < T. When the user roams, the private key is
presented and the information of the node (MAC and IP address) vt
is added to the neural chain once it is mined (Fig. 8).
User yN={MAC,IP, PASSPORT}
R(1)
Area 10
V10={MAC10,IP10}
R(2) R(3) R(4) R(5)
Area 9
V9={MAC9,IP9}Area 8
V8={MAC8,IP8}Area 7
V7={MAC7,IP7}Area 6
V6={MAC6,IP6}
Area 5
V5={MAC5,IP5}
Area 4
V4={MAC4,IP4}
Area 3
V3={MAC3,IP3}
Area 2
V2={MAC2,IP2}
Area 1
V1={MAC1,IP1}
User yN={MAC,IP, PASSPORT}
R(10) R(9) R(8) R(7) R(6)
Fig. 8. Neural Blockchain in IoT - Simulation Model
Each bit is codified as a neuron however rather than the binary 0 -
1, neuron potential is codified as 0.25 – 0.75 (Fig. 9); this
approach removes overfitting in the learning algorithm as neurons
only represent binary values.
0.25
0.75
1
0
bit Neuron Potential
1.0
0.5
0.0
Fig. 9. Neural Blockchain in IoT – Neural Codification
10
The simulations are run 100 times for a Bluetooth MAC Network
(Table 4, Table 5). The information shown is the number of
iterations the Random Neuron Network with Blockchain
configuration requires to achieve a Learning Error < 1.0E-10, the
error Ek, the learning and mining time, the number of iterations to
mine the Blockchain to to achieve a Mining Error Ek < 1 .0E-0 5
and the number of neurons for each layer: input xL, hidden zM and
output yN.
Table 4. Bluetooth MAC Simulation – Learning
Roaming
Learning
Iteration
Learning
Time (s)
Learning
Error
Number of Neurons
(xL, zM, yN)
1
233.00
136.26
9.96E-11
48-4-48
2
190.52
124.47
9.46E-11
100-4-48
3
171.42
108.48
9.40E-11
152-4-48
4
160.90
112.29
9.18E-11
204-4-48
5
150.67
101.04
9.12E-11
256-4-48
6
144.31
93.33
9.43E-11
308-4-48
7
140.00
94.47
9.47E-11
360-4-48
8
137.00
104.99
9.06E-11
412-4-48
9
132.99
116.77
8.75E-11
464-4-48
10
131.00
99.98
9.30E-11
516-4-48
Table 5. Bluetooth MAC Simulation – Mining
Roaming
Mining
Iteration
Mining
Time (ms)
Mining
Error Ek
Number of Neurons
(xL, zM, yN)
1
36.22
1.811
2.48E-06
48-4-48
2
24.88
2.49
3.30E-06
100-4-48
3
56.32
10.70
3.01E-06
152-4-48
4
758.37
227.51
3.97E-06
204-4-48
5
109.10
45.822
3.64E-06
256-4-48
6
116.59
68.79
3.13E-06
308-4-48
7
354.38
287.05
3.09E-06
360-4-48
8
2134.31
2326.39
3.70E-06
412-4-48
9
12.13
18.0737
3.56E-06
464-4-48
10
141.99
282.56
3.26E-06
516-4-48
With four neurons in the hidden layer, the number of learning
iterations gradually decreases while the number of input neurons
increases due to the additional information added when roaming
between nodes. The results for the mining iteration are not as
linear as expected because mining is performed using random
values (Fig. 10, Fig. 11). Mining is easier in some Roaming stages
when it would have been expected harder as the number of
neurons increases.
Fig. 10. Bluetooth MAC Simulation – Learning
Fig. 11. Bluetooth MAC Simulation – Mining
The learning and mining time follow their respective number of
iterations although the processing time per iteration increases at
each step as the solution of the Block-chain Neural network
equations increase in complexity (Fig. 12).
11
Fig. 12. Bluetooth MAC Simulation – Mining Time per Iteration
The Learning Threshold (LT) and Mining Threshold (MT)
parameters tune the Blockchain Random Neural Network
performance, or number of iterations, as shown on Ta ble 6 a nd
Table 7. As expected, the higher accuracy in the learning stage
correlates with higher learning iterations and lower number of
mining iterations as the network is trained with more accurate
values.
Table 6. Bluetooth MAC Simulation – Network Parameters – Learming
Iteration
Roaming
LT
Learning Iteration
MT=E-03
MT=E-04
MT=E-05
1
1.00E-05
1.00E-10
118.00
233.00
118.00
233.00
118.00
233.00
2
1.00E-05
1.00E-10
97.30
190.70
97.32
190.58
97.35
190.68
3
1.00E-05
1.00E-10
88.03
171.58
88.02
171.48
88.03
171.58
4
1.00E-05
1.00E-10
83.00
160.88
83.00
160.91
83.00
160.92
5
1.00E-05
1.00E-10
78.00
150.69
78.00
150.77
78.00
150.72
6
1.00E-05
1.00E-10
75.00
144.54
75.00
144.55
75.00
144.53
7
1.00E-05
1.00E-10
73.00
140.00
73.00
140.01
73.00
140.00
8
1.00E-05
1.00E-10
71.13
137.00
71.06
137.00
71.14
137.00
9
1.00E-05
1.00E-10
69.00
132.99
69.00
133.00
69.00
133.00
10
1.00E-05
1.00E-10
68.69
131.00
68.69
131.00
68.74
131.00
Table 7. Bluetooth MAC Simulation – Network Parameters – Mining
Iteration
Roaming
LT
Mining Iteration
MT=E-03
MT=E-04
MT=E-05
1
1.00E-05
1.00 E-10
6.08
4.39
13.80
11.92
63.81
35.18
2
1.00E-05
1.00E-10
3.22
3.26
10.88
10.71
65.10
26.24
3
1.00E-05
1.00E-10
5.56
4.73
19.94
17.77
115.85
52.78
4
1.00E-05
1.00E-10
33.36
33.86
190.97
242.29
1645.15
704.04
5
1.00E-05
1.00E-10
4.98
4.09
28.42
34.75
203.37
99.85
6
1.00E-05
1.00E-10
4.59
3.97
35.33
35.32
291.99
147.27
7
1.00E-05
1.00E-10
5.23
4.47
137.24
72.15
833.42
416.99
8
1.00E-05
1.00E-10
7.00
6.74
555.00
306.91
6431.53
2614.85
9
1.00E-05
1.00E-10
1.05
1.04
3.64
3.21
27.12
13.78
10
1.00E-05
1.00E-10
1.34
1.29
24.02
23.11
290.39
134.65
Table 8 and Table 9 show the Learning and Mining Errors; the
Blockchain Random Neural Network adjusts to the to the Learning
Threshold and Mining Threshold.
Table 8. Bluetooth MAC Simulation – Network Parameters – Learning
Error
Roaming
LT
Learning Error
MT=E-03
MT=E-04
MT=E-05
1
1.00E-05
1.00E-10
9.49E-06
9.96E-11
9.49E-06
9.96E-11
9.49E-06
9.96E-11
2
1.00E-05
1.00E-10
9.55E-06
9.34E-11
9.51E-06
9.40E-11
9.51E-06
9.34E-11
3
1.00E-05
1.00E-10
9.65E-06
9.30E-11
9.69E-06
9.36E-11
9.65E-06
9.31E-11
4
1.00E-05
1.00E-10
9.44E-06
9.22E-11
9.46E-06
9.21E-11
9.46E-06
9.21E-11
5
1.00E-05
1.00E-10
9.41E-06
9.10E-11
9.38E-06
9.01E-11
9.42E-06
9.08E-11
6
1.00E-05
1.00E-10
9.41E-06
9.18E-11
9.44E-06
9.18E-11
9.44E-06
9.21E-11
7
1.00E-05
1.00E-10
9.19E-06
9.46E-11
9.21E-06
9.45E-11
9.20E-06
9.43E-11
8
1.00E-05
1.00E-10
9.65E-06
9.08E-11
9.76E-06
9.09E-11
9.65E-06
9.06E-11
9
1.00E-05
1.00E-10
9.70E-06
8.75E-11
9.69E-06
8.74E-11
9.72E-06
8.71E-11
10
1.00E-05
1.00E-10
8.88E-06
9.33E-11
8.88E-06
9.30E-11
8.81E-06
9.29E-11
Table 9. Bluetooth MAC Simulation – Network Parameters – Mining
Error
Roaming
LT
Mining Error
MT=E-03
MT=E-04
MT=E-05
1
1.00E-05
1.00E-10
4.07E-04
3.03E-04
3.23E-05
3.21E-05
8.45E-06
3.00E-06
2
1.00E-05
1.00E-10
3.40E-04
2.89E-04
3.87E-05
3.36E-05
8.42E-06
3.23E-06
3
1.00E-05
1.00E-10
3.45E-04
3.74E-04
3.88E-05
3.08E-05
8.41E-06
3.18E-06
4
1.00E-05
1.00E-10
4.74E-04
4.50E-04
4.33E-05
3.76E-05
8.12E-06
3.22E-06
5
1.00E-05
1.00E-10
4.44E-04
3.97E-04
4.32E-05
3.65E-05
8.27E-06
2.72E-06
6
1.00E-05
1.00E-10
4.81E-04
5.30E-04
3.85E-05
3.43E-05
8.11E-06
3.40E-06
7
1.00E-05
1.00E-10
6.00E-04
5.81E-04
4.65E-05
4.30E-05
7.93E-06
2.95E-06
8
1.00E-05
1.00E-10
7.01E-04
6.62E-04
5.57E-05
4.83E-05
8.23E-06
3.75E-06
9
1.00E-05
1.00E-10
2.34E-04
2.43E-04
4.43E-05
3.79E-05
8.19E-06
3.53E-06
10
1.00E-05
1.00E-10
5.26E-04
4.70E-04
5.26E-05
5.15E-05
7.83E-06
3.41E-06
Figure 13 shows the number of iterations on Logarithm scale for a
Mining Threshold MT=1.00E-03. The lowest Learning Threshold
implies a reduced number of Mining iterations.
12
LT=1.00E-05
MT= 1.00E-03
LT=1.00E-10
MT= 1.00E-03
Fig. 13. Number of Mining Iterations MT=1.00E-03
Figure 14 shows the number of iterations on Logarithm scale for a
Mining Threshold MT=1.00E-04. Similar to the previous
validation, the lowest Learning Threshold implies a reduced
number of Mining iterations.
LT=1.00E-05
MT= 1.00E-04
LT=1.00E-10
MT= 1.00E-04
Fig. 14. Number of Mining Iterations MT=1.00E-03
Figure 15 shows the number of iterations on Logarithm scale for a
Mining Threshold MT=1.00E-05. Following the previous
validations, the lowest Learning Threshold implies a reduced
number of Mining iterations.
LT=1.00E-05
MT= 1.00E-05
LT=1.00E-10
MT= 1.00E-05
Fig. 15. Number of Mining Iterations MT=1.00E-03
Figure 16 shows the number of iterations on Logarithm scale for a
Learning Threshold MT=1.00E-10. As expected, a reduced Mining
Threshold implies a lower number of Mining Iterations.
LT=1.00E-10
MT= 1.00E-05
LT=1.00E-10
MT= 1.00E-04
LT=1.00E-10
MT= 1.00E-03
Fig. 16. Number of Mining Iterations LT=1.00E-10
The Blockchain Random Neural Network algorithm shall detect
tampering to be effective (Table 10, Table 11) where Δ represent s
the number of tampered bits.
Table 10. Bluetooth Simulation – Tampering Error
Roaming
Bluetooth - MAC
Error
Δ=0.0
Error
Δ=1.0
Neurons
(xL, zM, yN)
1
9.96E-11
1.31E-03
48-4-48
2
9.49E-11
1.50E-04
100-4-48
3
9.18E-11
4.32E-05
152-4-48
4
9.38E-11
1.95E-05
204-4-48
5
9.01E-11
8.54E-06
256-4-48
6
9.07E-11
4.28E-06
308-4-48
7
9.49E-11
2.56E-06
360-4-48
8
9.33E-11
1.71E-06
412-4-48
9
8.93E-11
9.00E-07
464-4-48
10
9.59E-11
7.22E-07
516-4-48
Table 11. WLAN Simulation – Tampering Error
Roaming
WLAN - IP
Error
Δ=0.0
Error
Δ=1.0
Neurons
(xL, zM, yN)
1
9.28E-11
1.18E-03
80-4-80
2
9.57E-11
1.63E-04
164-4-80
3
9.36E-11
5.21E-05
248-4-80
4
9.50E-11
2.23E-05
332-4-80
5
9.21E-11
1.02E-05
416-4-80
6
9.36E-11
5.22E-06
500-4-80
7
9.12E-11
3.07E-06
584-4-80
8
9.28E-11
2.10E-06
668-4-80
9
9.25E-11
1.15E-06
752-4-80
10
9.55E-11
7.81E-07
836-4-80
The effects of tampering the Neural Blockchain (Fig. 17) is
detected by the learning algorithm even when the tampered values
only differ in a bit, Δ=1.0, although this error reduces with a n
incrementing roaming as the number of neurons increases. Both
Bluetooth and WLAN Networks perform similarly.
13
Fig. 17. Bluetooth and WLAN Simulation – Tampering Error
5.2 Blockchain Cybersecurity in 5G
The experiment will emulate a gradual implementation of a 5G
network in an Intelligent Building were 5G nodes are increasingly
enabled into the Smart Building for IoT services (Table 12).
Table 12. Neural Blockchain in 5G and Cybersecurity Validation – Node
values
Simulation
Application
Cell
Coverage
5G Node Identity
Vt
Building
IMSI YN
Five
5G Nodes
Intelligent
Building
Pico
100 m
0.314km2
60 bits
234-151-
234512340-4
60 bits
234-151-
234512351
A 5G Network generic Performance Specification is shown on
Table 13:
Table 13. 5G Network Performance Specification
Data Rate
Frequency
Latency
Modulation
Cell
20 Gbit/s
3.5 GHz -
26 GHz
<1 ms
Non-Orthogonal
Multiple Access
Pico
Small
The Intelligent Building is assigned to an IMSI private key yN
whereas the gradual addition of 5G nodes require validation before
they are enabled to transmit IoT information into the Intelligent
Building. When an 5G node is ready to become operational, its
identity is validated by the Intelligent Building. The decentralized
system retrieves the neural weights associated to the private key,
mines the block, adds the node code and stores back the neural
network weights in the decentralized system (Fig. 18). When more
5G nodes are ready to become operational, the Intelligent Building
uses its IMSI private key yN and the 5G node identity vt is added to
the neural chain once it is mined.
V1 5G Node 1
0.25
0.75
1
0
bit Neuron Potential
1.0
0.5
0.0
Simulation Model – 5G Deployment Stages Bit - Neuron Codification
Intelligent
Building
IMSI yN V2 5G Node 2
V3 5G Node 3
V4 5G Node 4
V5 5G Node 5
A(1)
A(2)
A(3)
A(4)
A(5)
Fig. 18. Neural Blockchain in 5G Cybersecurity validation
The simulations are run 100 times for a five 5G node Network.
The information shown in Table 14 is the number of learning
iterations and time the Random Neuron Network with Blockchain
configuration requires to achieve a Learning Error < 1.0E-10.
Table 14. 5G Simulation – Learning
Variable
5G
Node 1
5G
Node 2
5G
Node 3
5G
Node 4
5G
Node 5
Learning
Iteration
267.00
219.73
194.76
178.91
167.22
Learning
Time (s)
43.548
36.488
34.504
34.294
36.741
Standard
Deviation
0.00
0.51
0.49
0.49
0.42
95%
CR
0.00
0.10
0.10
0.09
0.08
Learning
Error
9.50E-11
9.48E-11
9.38E-11
9.29E-11
9.45E-11
Standard
Deviation
5.20E-26
3.06E-12
3.36E-12
3.36E-12
4.28E-12
95%
CR
1.02E-26
5.99E-13
6.58E-13
6.59E-13
8.39E-13
Neurons
(xL, zM, yN)
60-4-60
124-4-60
188-4-60
252-4-60
316-4-60
Table 15 shows the error Ek, the number of iterations and time to
mine the Blockchain and the number of neurons for each la yer;
input xL, hidden zM and output yN for a mining threshold of 1.00E-
02.
Table 15. 5G Simulation – Mining
Variable
5G
Node 1
5G
Node 2
5G
Node 3
5G
Node 4
5G
Node 5
Mining
Iteration
1.07E+02
1.02E+04
1.04E+05
6.32E+05
5.70E+05
Mining
Time (s)
7.49E+00
1.63E+03
3.02E+04
2.84E+05
3.82E+05
Standard
Deviation
1.05E+02
9.83E+03
9.58E+04
7.88E+05
6.30E+05
95%
CR
2.07E+01
1.93E+03
1.88E+04
1.54E+05
1.24E+05
Mining
Error
4.54E-03
6.63E-03
8.03E-03
8.47E-03
8.87E-03
Standard
Deviation
3.15E-03
2.55E-03
1.72E-03
1.33E-03
1.02E-03
95%
CR
6.17E-04
5.00E-04
3.36E-04
2.60E-04
2.00E-04
14
Neurons
(xL, zM, yN)
60-4-60
124-4-60
188-4-60
252-4-60
316-4-60
With four neurons in the hidden layer, the number of learning
iterations gradually decreases while the number of input neurons
increases due to the additional information added activatin g 5G
nodes. The results for the mining iteration are not as linear as
expected because mining is performed using random values (Fig.
19, Fig. 20). Mining is easier in the final authentication stage when
it would have been expected harder as the number of neurons
increases.
Fig. 19. 5G Network Simulation – Learning
Fig. 20. 5G Network Simulation – Mining
Similar to the previous validation, the learning and mining time
follows their respective number of iterations although the
processing time per iteration increases at each step as the solution
of the Blockchain Neural network equations increase in
complexity (Fig 21).
Fig. 21. 5G Network Simulation – Mining Time per Iteration
The Blockchain Random Neural Network algorithm must detect
5G rogue nodes to be effective (Table 16); Δ represents the
number of bit changes in the Node identity vt for different values
hidden neurons ZM.
Table 16. 5G Network Simulation – Rogue Node Tampering Error
Variable
5G
Node 1
5G
Node 2
5G
Node 3
5G
Node 4
5G
Node 5
Neurons
(xL, zM, yN)
60-4-60
124-4-60
188-4-60
252-4-60
316-4-60
Error Δ=0.0
Error Δ=1.0
9.50E-11
1.83E-03
9.84E-11
2.43E-04
9.01E-11
6.76E-05
8.91E-11
2.58E-05
8.88E-11
1.57E-05
15
Neurons
(xL, zM, yN)
60-60-60
124-60-
60
188-60-
60
252-60-
60
316-60-60
Error Δ=0.0
Error Δ=1.0
9.97E-11
1.53E-03
9.93E-11
1.40E-04
9.93E-11
3.57E-05
9.95E-11
1.30E-05
9.95E-11
6.08E-06
Neurons
(xL, zM, yN)
60-60-60
124-124-
60
188-188-
60
252-252-
60
316-316-
60
Error Δ=0.0
Error Δ=1.0
9.97E-11
1.53E-03
9.98E-11
1.40E-04
9.98E-11
3.78E-05
9.99E-11
1.48E-05
9.99E-11
4.54E-06
The addition of a 5G rogue node into the Intelligent Building is
detected by the learning algorithm Neural Blockchain (Fig. 22)
even when the identity value only differs in a bit, Δ=1.0.
Fig. 22. 5G Network Simulation – Rogue Node Tampering Error
The simulation results show that the increment of neurons in the
hidden layer requires additional learning iterations. However, th is
increment is not reflected in higher accuracy to detect input errors
or 5G rogue nodes, therefore ZM=4 is the most optimum
configuration for this model.
6 Conclusions
This article has proposed the application of the Blockchain
Random Neural Network in IoT and 5G infrastructure f or Smart
Cities where neurons are gradually incremented as user validation
data increases. This configuration provides the proposed algorithm
with the same properties as the Blockchain: security and
decentralization with the same validation process: mining the input
neurons until the neural network solution is found. The presented
model can be generalized to any Authentication, Authorization and
Accounting (AAA) server.
The Random Neural Network in Blockchain configuration has
been applied to a 1) an IoT AAA server that covers the digital
seven layers of the OSI Model and the physical user cred entials
such as passport or biometrics, 2) a 5G node authentication process
in Intelligent Buildings. Experimental results show that
Blockchain applications can be successfully implemented in Smart
City infrastructure using neural networks. Mining effort is
gradually increased as user information expands where data is
encrypted in a decentralized network. In addition, validation
results demonstrate that rogue users or cyber attackers can be
detected and identified.
Future work will include the validation of the Blockchain with
other neural networks such as Long Short-Term Memory to
compare the mining results. The expansion of the Roaming a nd
Authentication stages will assess the mining effects in a longer
term. The balance between the number of neurons that f orm the
neural chain versus new user data will be assessed in terms of
learning, mining iterations and the detection of rogue nodes
tampering the network.
Credit author statement
Dr Will Serrano is the only author for this article.
Declaration of competing interest
The authors declare that they have no known competing financial
interests or personal relationships that could have appeared to
influence the work reported in this paper.
References
Adam, I., Ping, J., 2018. Framework for security event
management in 5G. Int. Conf. Availability, Reliab. Secur. 51, 1–
10.
Agiwal, M., Roy, A., Saxena, N., 2016. Next generation 5G
wireless networks: a comprehensive survey. IEEE Commun. Surv.
Tutorials 18 (3), 1617–1655.
Agrawal, R., Verma, P., Sonanis, R., Goel, U., De, A., Anirudh, S.,
Shekhar, S., 2018. Continuous security in IoT using blockchain.
IEEE Int. Conf. Acoust. Speech Signal Process. 6423–6427.
Al-Rubaie, M., Chang, J., 2019. Privacy preserving machine
learning: threats and solutions. IEEE Secur. Priv. Mag., IEEE
Secur. Priv. 17 (2), 49–58.
Andrea, I., Chrysostomou, C., Hadjichristofi, G., 2015. Internet of
things: security vulnerabilities and challenges. In: IEEE
Symposium on Computers and Communication, pp. 180–187.
Andrews, J., Buzzi, S., Choi, W., Hanly, S., Lozano, A., Soong,
A., Zhang, J., 2014. What will 5G Be? IEEE J. Sel. Area.
Commun. 32 (6), 1065–1082.
Arslan, S., Jurdak, R., Jelitto, J., Krishnamachari, B., 2020.
Advancements in distributed ledger technology for Internet of
Things. Internet Things 9 (100114), 1–3.
16
Atat, R., Liu, L., Chen, H., Wu, J., Li, H., Yi, Y., 2017. Enabling
cyber-physical communication in 5G cellular networks:
challenges, spatial spectrum sensing, and cyber-security. IET
Cyber-Phys. Syst.: Theor. Appl. 2 (1), 49–54.
Atlam, H., Wills, G., 2019. Chapter three - intersections between
IoT and distributed ledger. Adv. Comput. 115., 73–113.
Bera, S., Misra, S., Vasilakos, A., 2017. Software-defined
networking for Internet of things: a survey. IEEE Internet Things J
4 (6), 1–15.
Biswas, K., Muthukkumarasamy, V., 2016. Securing smart Cities
using blockchain technology. In: International Co nference High
Performance Computing and Communications/Smart City /Data
Science and Systems, pp. 1392–1393.
Cha, S.-C., Yeh, K.-H., 2018. An ISO/IEC 15408-2 compliant
security auditing system with blockchain technology. In: IEEE
Conference on Communications and Network Security, pp. 1–2.
Challa, S., Kumar, A., Gope, P., Kumar, N., Wu, F., Vasilakos, A.,
2020. Design and analysis of authenticated key agreement scheme
in cloud-assisted cyber–physical systems. Future Generat. Comput.
Syst. 108, 1267–1286.
Chang, E., Gottwalt, F., Zhang, Y., 2017. Cyber situational
awareness for CPS, 5G and IoT. Front. Electron. Technol. 433,
147–161.
Danzi, P., Kalor, A., Sorensen, R., Hagelskjaer, A., Nguyen, L.,
Stefanovic, C., Popovski, P., 2020. Communication aspects of the
integration of wireless IoT devices with distributed ledger
technology. IEEE Netw. 34 (1), 47–53.
Dawoud, A., Shahristani, S., Chun, R., 2018. Deep learning and
software-defined networks: towards secure IoT architecture.
Internet Things 3–4, 82–89.
Deogirikar, J., Vidhate, A., 2017. Security Attacks in IoT: A
Survey. IEEE International Conference on IoT in Social, Mo bile,
Analytics and Cloud, pp. 32–37.
Domanska, J., Nowak, M., Nowak, S., Czachorski, T., 2018.
European cybersecurity research and the SerIoT project. In:
International Symposium on Computer and Information Sciences,
pp. 66–173.
Elsts, A., Mitskas, E., Oikonomou, G., 2018. Distributed ledger
technology and the Internet of things: a feasibility study.
Workshop Blockchain-enabled Netw. Sens. Syst. Nov. 7–12.
Fan, K., Bao, Z., Liu, M., Vasilakos, A., Dredas, W. Shi, 2020.
Decentralized, reliable and efficient remote outsourced data
auditing scheme with blockchain smart contract for industrial. IoT
Future Generat. Comput. Syst. 110, 665–674.
Fernandez, L., Huertas, A., Manuel Gil, C., Garcia, F., Ma rtinez,
G., 2018. Dynamic management of a deep learning-based anomaly
detection system for 5G networks. J. Ambient Intell. Humanized
Comput. 1–15.
Fernandez, L., Perales, A., Garcia, F., Gil, M., Martinez, G., 2018.
A self-adaptive deep learning-based system for anomaly detection
in 5G networks. IEEE Access 6, 7700–7712.
Firouzi, F., Chakrabarty, K., Nassif, S., 2020. Intelligent Internet
of Things: from Device to Fog and Cloud, pp. 1–647.
Gai, K., Qiu, M., Tao, L., Zhu, Y., 2016. Intrusion detection
techniques for mobile cloud computing in heterogeneous 5G.
Secur. Commun. Network. 9, 3049–3058.
Gai, K., Raymond, K.-K., Zhu, L., 2018. Blockchain-enabled
reengineering of cloud datacenters. IEEE Cloud Comput. 5 (6),
21–25.
Gelenbe, E., 1989. Random neural networks with negative and
positive signals and product form solution. Neural Comput. 1,
502–510.
Gelenbe, E., 1993a. Learning in the recurrent random neural
network. Neural Comput. 5., 154–164.
Gelenbe, E., 1993b. G-networks with triggered customer
movement. J. Appl. Probab. 30, 742–748.
Gelenbe, E., Domanska, J., Czachorski, T., Drosou, A., Tzovaras,
D., 2018. Security for Internet of things: the SerIoT project. IEEE
international symposium on networks. Comput. Commun. 1–5.
Granjal, J., Monteiro, E., Sa Silva, J., 2015. Security for the
Internet of things: a survey of existing protocols and open research
issues. IEEE Commun. Surv. Tutorials 17 (3), 1294–1312.
Grigoryan, G., Liu, Y., Njilla, L., Kamhoua, C., Kwiat, K., 2018.
Enabling cooperative IoT security via software defined networks
(SDN). IEEE Int. Conf. Commun. 1–10.
Gupta, Y., Shorey, R., Kulkarni, D., Tew, J., 2018. The
applicability of blockchain in the Internet of Things. In: IEEE
International Conference on Communication Systems & Networks,
pp. 561–564.
Hesamifard, E., Takabi, H., Ghasemi, M., Wright, R., 2018.
Privacy-preserving machine learning as a service. Proc. Priv.
Enhanc. Technol. 3, 123–142.
Huh, S., Cho, S., Kim, S., 2017. Managing IoT devices using
Blockchain platform. In: International Conference on Advanced
Communication Technology, pp. 464–467.
International Telecommunication Union, 2012. Overview of the
Internet of Things. Y.2060, pp. 1–22.
Jangirala, S., Kumar, A., Vasilakos, A., 2020. Designing secure
lightweight blockchain-enabled RFID-based authentication
protocol for supply chains in 5G mobile edge computing
environment. IEEE Trans. Ind. Inf. 16 (11), 7081–7093.
Jeet Kaur, M., Riaz, S., Mushtaq, A., 2019. Cyber-Physical Cloud
Computing Systems and Internet of Everything. Principles of
Internet of Things (IoT) Ecosystem: Insight Paradigm, pp . 201–
227.
Jing, Q., Vasilakos, A., Wan, J., Lu, J., Qiu, D., 2014. Security of
the Internet of things: perspectives and challenges. Wireless
Network 20, 2481–2501.
Joshva Devadas, T., Thayammal, S., Ramprakash, A., 2019. IoT
Data Management, Data Aggregation and Dissemination.
17
Principles of Internet of Things (IoT) Ecosystem: Insight
Paradigm, pp. 385–411.
Katsianis, D., Neokosmidis, I., Pastor, A., Jacquin, L., Gardikis,
G., 2018. Factors influencing market adoption a nd evolution of
NFV/SDN cybersecurity solutions. Evidence from SHIELD
project. In: European Conference on Networks and
Communications, pp. 261–265.
Kim, H., Kim, S., Hwang, J., Seo, C., 2019. Efficient privacy-
preserving machine learning for blockchain network. IEEE Access
7, 136481–136495.
Kinzel, W., Kanter, I., 2002. Interacting Neural Networks and
Cryptography Secure exchange of information by synchronization
of neural networks. Adv. Solid State Phys. 42, 383–391.
Klimov, A., Mityagin, A., Shamir, A., 2002. Analysis of neural
cryptography. In: International Conference on the Theory and
Application of Cryptology and Information Security, 2501, pp.
288–298.
Kumar, P., Chen, M., Hyuk, J., 2018. A software defined fog node
based distributed blockchain cloud architecture for IoT. IEEE
Access 6, 115–124.
Lee, G., Crespi, N., Choi, J., Boussard, M., 2013. Internet of
things. Telecommunication services evolution. LNCS 7768, 25 7–
282.
Li, P., Li, T., Ye, H., Li, J., Chen, X., Xiang, Y., 2018. Privacy-