1. Introduction
In recent years, mobile devices have become widespread: cell phones, laptops, smartphones,
and tablets. This has opened up new opportunities for the developers of network solutions [
1
].
One of the areas of development of network technologies for mobile devices is the Internet of
Things.
The current direction of development of communication networks is the concept of the
Internet of Things. The main task of the Internet of Things is to create a single network that
includes objects of the information (virtual) and physical (real) worlds and will ensure the
interaction of objects with each other.
The technological base of the first stage of development of the Internet of Things is
allpervasive (wireless) sensor networks, which are widely used in the modern world in almost all
spheres of life, due to their low cost, rapid deployment and eficiency.
The implementation of the concept of the Internet of Things is expressed in the
penetration of telecommunication technologies into all spheres of human activity. Currently, this is
reflected in the expansion of the field of application of wireless sensor networks.
The concept of the Internet of Things includes data exchange between devices (M2M), sensor
networks, and self-organizing networks of mobile devices (MANET). This article provides an
overview of the main methods for calculating the topology of self-organizing networks.
The task of improving the quality of data transmission in self-organizing networks of
mobile devices can be solved by diferent methods: the method of retransmission request (ARQ)
[
2
], the method of redundant coding (FEC) [
3, 4, 5, 6, 7, 8, 9
], the method of Network Coding
[
10
]. One of the approaches to data transmission in self-organized networks is the use of the
method of superimposed networks (P2P) [
11, 12
]. P2P protocol sets the rules for streaming
data between nodes [
13, 14, 15, 16, 17
]. Streaming data is transmitted between the nodes of the
superimposed network along the routes selected by the underlying protocols [
18, 19
].
Controlling the data transmission process will avoid congested areas in the network, increase the
throughput and improve the reliability of the network as a whole [
20, 21, 22, 23, 24, 25
].
Overlay networks rely on tree and multi-link structures [
26, 27, 28
]. To improve network reliability,
some researchers [
29
] use diferent types of multipath redundancy, such as “routing braids”,
which demonstrate improved reliability and stability in self-organizing networks. The
environmental sector demonstrates particular interest in IoT, where modern air quality monitoring
systems can be built using sensor networks [
30, 31, 32, 33, 34, 35
]. Also as a part of complex
diagnostic systems of energy facilities [
36, 37, 38, 39, 40
] which build on hierarchical structures,
IoT can be used in energy sector.
2. Analysis of routing protocols in IoT systems
Self-organizing networks are an alternative to infrastructure networks. In such a network, each
node in the network can act as a router. The possibility for each node to leave the network
or connect to it will lead to the fact that an important issue in the organization of the
selforganizing network is the choice of a routing protocol. The routing protocols developed are
classified according to the approach to update the network topology information into reactive,
proactive, and hybrid [
41
]. Figure 1 shows the classification of routing protocols.
The reactive approach to routing involves constructing routes as they are needed. When a
connection to a network node is attempted, a complete enumeration of all options is performed
and the best route to it is found according to the routing metric. This route is used as long as
there is a connection to the destination.
With a proactive approach, the network topology must be monitored and updated at specific
intervals. Proactive protocols update the network topology with periodic queries. Protocols
belonging to this group may use diferent numbers of databases with information about the
network topology and diferent ways of keeping this information up to date. The proactive
approach relies on keeping track of the network topology, so nodes are constantly exchanging
messages, which can lead to higher power consumption compared to the reactive approach.
On the other hand, a node in a network using the reactive approach has to wait for the route
variants to be enumerated, which can afect the transmission speed in networks with changing
topologies. The hybrid approach involves combining reactive and proactive approaches within
the same network. The best route between network nodes is selected based on metrics: number
of routing steps, ETX, ETT, Air Time Link, etc. Metrics can take into account information from
the physical, data link, and network layers of the OSI model.
2.1. Proactive protocols
The OLSR (Optimized Link State Routing) protocol is proactive and oriented for use in large
networks with a high density of nodes. Each node uses HELLO broadcast messages that are
transmitted at regular intervals to nodes within one routing step. After receiving the HELLO,
the destination node tries to establish a two-way connection with the sender node. The
number of control messages in OLSR is reduced due to the MPR (Multipoint Relays) approach [
42
].
In OLSRv2, the exchange of control messages in the network has become more eficient, and
the message form itself has been standardized and simplified. OLSR interacts with the
network layer by managing routing tables and using IP addresses for packet transmission. The
B.A.T.M.A.N. protocol (Better Approach To Mobile Ad hoc Networks) also uses a proactive
approach [
43
], in which all nodes produce an Originator Message (OGM) broadcast. An OGM
contains an originator address, a recipient address, and a unique sequential number. Each
neighboring node changes the recipient address to its own and sends the message back to the
originator. OGM messages do not include any additional information such as QoS metrics and
routing tables. The B.A.T.M.A.N. protocol. Has lower non-productive costs in networks with
more nodes than the OLSR protocol. One of the first proactive protocols was DSDV
(Destination Sequenced Distance Vector), developed in 1994 [
44
]. Its main feature was the addition of an
ordinal number field in control messages because this bypassed the problem of loops between
nodes in the network (Loop free) since each node now knew whether its information about
the network topology was obsolete. DSDV proved inefective in large networks with rapidly
changing topologies but influenced the development of other protocols, such as AODV.
2.2. Reactive protocols
Reactive DSR (Dynamic Source Routing) protocol uses a special DSR Options Header Format
that can be added to any packet and contains the route from source to destination node [
45
].
A node can perform a route discovery process to the desired node (Route Discovery) using
broadcast messages. The Route Maintenance process is to monitor the link-layer notifications.
If a link-layer notification is accepted or node requests are left unanswered, the discovery
process is repeated. Disadvantages and advantages of DSR include its reactivity, which reduces
the cost of sending control messages but makes it necessary to bufer packets for the duration
of route discovery. Besides, the special header format can lead to a large header for small
payloads, reducing the eficiency of the network. Further evolution of the reactive approach was
the AODV protocol [
46
]. Instead of relying on the transmission of voluminous headers, AODV
reintroduced routing tables that accumulated all the information about the network topology
as messages were received from other nodes. To avoid looping, two sequence numbers were
introduced, one for the source and one for the destination, allowing you to track the novelty
of topology information as you use the route from the destination to the source. The use of
AODV is recommended for networks of 10 to 1000 mobile nodes. The main goal of its
development was to reduce the cost of sending control messages and to improve the scalability and
performance of the network. Another protocol based on DSR was reactive SrcRR [
47
]. Its main
diference from DSR was the use of an ETX metric, which was measured by periodic
broadcasts to neighboring nodes, and the total ETX of its parts was used for the entire route. Also,
SrcRR was independent of the network layer and could use MAC addresses to find the path.
Microsoft developed and patented the LQSR (Link Quality Source Routing) protocol, which is
also based on DSR [
48
]. It is implemented between the link layer and the network layer using
a virtual network adapter, allowing it to handle multiple physical connections at once. The
LQSR header is located between the Ethernet header and the frame payload. Each node, as in
SrcRR, measures the QoS metric to neighboring nodes, propagates this information through
the network, and it is taken into account in selecting the best path to the destination. Guided
by the rule that the shortest path does not mean the best path, LQSR allows the use of three
QoS metrics: ETX, RTT, and PktPair. VNF sees network latency as a critical attribute for
reliability, availability, and QoS requirements by most researchers. By automating and elastically
allocating resources, these enhanced service oferings are implemented [
49
].
2.3. Hybrid protocols
The hybrid approach enables the use of reactive and proactive approaches within a single
network. It is used in 802.11s to provide WMN support at the link layer [
50, 51
]. In previous 802.11
family standards, there was no way to obtain link-layer QoS metrics. For the QoS metric to be
more accurate, it should be obtained at a lower layer of the OSI network model. As a default
protocol in the standard 802.11s recommended hybrid HWMP (Hybrid Wireless Mesh
Protocol), and the optional protocol can act as OLSR. The reactive approach is implemented based
on AODV (Ad hoc On-demand Distance Vector). In this case, the node looks for the best route
as needed, taking into account QoS metrics. Using a proactive approach, a root node (Root) is
assigned to a WMN that polls nodes at intervals, thus updating the network map. The
connected node, can contact the root node and get information about the routes to all nodes in
the network. Both approaches can be used separately or simultaneously in the same network
(figure 2).
The hybrid approach has been used not only in HWMP but also in earlier protocols, such
as the HSLS (Hazy Sighted Link State) Routing Protocol. Intending to reduce non-productive
costs, HLSL controls the interval at which network topology information is updated to reduce
the number of control messages [
52
]. If the route is obsolete, HLSL begins to operate in
reactive mode. The lack of up-to-date network topology information is a major drawback of this
protocol. Another hybrid protocol for WMN is Babel. Based on the ideas of DSDV, AODV,
and the Cisco EIGRP (Enhanced Interior Gateway Protocol), Babel takes a proactive approach
and is aimed at working in networks with mobile nodes [
53
]. It allows the implementation of
diferent QoS metrics, although by default it uses ETX. Reactive mode is used in Babel if no
route from a node is suitable for reliable packet transmission. In a hybrid protocol, ZRP (Zone
Routing Protocol) node applies proactive route lookup within a certain section of the network
and reactive outside of it [
54
]. The FSR (Fish-eye State Routing Protocol) protocol is
characterized by the fact that the accuracy of the network topology information decreases with distance
from the node [
55
].
3. Analysis of data transmission methods and correction of transmission errors
3.1. Retransmission request method
Many diferent methods are used to recover lost and corrupted data. In self-organizing
networks, the network topology and transmission environment change rapidly, it is very dificult
to ensure reliable communication, to overcome the high mobility of nodes and external
interference. Therefore, many packets are received with errors, which means that error correction
methods play a significant role in data transmission processes [
56
].
In the transmission error correction method with a repeated request (ARQ), data reception
acknowledgment messages (ACK) are used for reliable data transmission. For example, if the
source has not received an acknowledgment from the recipient within a certain time interval
(timeout), it will retransmit until it receives an ACK message. The ARQ method relies on sum
and sequence number check fields in each packet header to detect corrupted and retransmitted
data. The retransmission request is used in the Stop-and-Wait ARQ, Go-Back-N ARQ, and
Selective-Repeat ARQ methods.
The Stop-and-Wait ARQ and Selective-Repeat ARQ methods are used in the 802.11 families
of standards at the data link layer of the OSI model, and all three approaches are used in various
transport layer protocols. The methods difer in the size of the transmit window and the receive
window. The Stop-and-Wait ARQ method starts a timer for each packet that is sent, and the
source waits until the ACK message arrives. If the ACK message has not been received and the
timeout has already expired, the source will repeat the packet. Thus, the interaction between
source and destination occurs from packet to packet.
The Go-Back-N ARQ method is more eficient than Stop-and-Wait ARQ. With this approach,
the source transmits several packets at a time and stores them in a bufer until it receives a
group ACK message. After the timeout expires, the source repeats all packets for which no
ACK message arrived. The Stop-and-Wait ARQ and Go-Back-N ARQ methods are very similar,
but they use diferent transmit window sizes.
In the Selective-Repeat ARQ method, the source transmits several packets at once but waits
for an individual acknowledgment for each packet. The receive and transmit windows sizes are
the same, and the destination can receive and store packets received in any order. The source
repeats those packets for which the timeout has expired.
The ARQ method can improve connection reliability but is not suitable for use in video
broadcasting because of the large and unstable delay.
3.2. Direct error correction method
The FEC method adds redundancy to the data being sent, which allows the addressee to
detect and correct errors without a second request from the source, and the maximum number
of bits recovered depends on the code used. The FEC method is usually implemented at the
physical layer and is responsible for correcting errors caused by interference in the
communication channel. The application layer FEC method uses Reed Solomon codes or BCX codes. By
introducing redundancy, this code can detect and correct bit errors in transmission. But the
introduction of redundancy reduces the eficiency of communication channels if the transmission
is error-free. Therefore, an adaptive noise coding method has been developed [
57, 58, 59, 60
].
This method allows controlling redundancy at byte or packet-level [
61, 62
], using video
characteristics or quality of service metrics, such as information fragment delivery ratio.
3.3. Network coding method
One method that is very similar to the FEC method is network coding (NC) [
63
]. In the NC
method, data are encoded by intermediate nodes. The self-organizing network provides new
opportunities for the implementation of the NC method. Today, the NC method for reliable
video data transmission is most often implemented based on random linear network coding
(RLCN). The source node groups the data into generations encodes each generation with SLSC
and writes the encoding coeficients in the header of each packet. In the NC method,
redundancy can be controlled: k line-independent packets of a given generation are required to
decode all packets of the same generation. Thus, more encoded packets can be transmitted in
unreliable transmission channels. SLSC has advantages over other coding methods (e.g.,
fountain code, block code) in that it makes the handling of encoded packets more flexible, reduces
the network delay for video transmission, eliminates the transmission of identical packets, and
uses the bandwidth more eficiently. The SLSC method can be used in conjunction with the
ARQ method to prevent the loss of the entire packet generation. Packets needed for decoding
the current generation can be requested from neighboring nodes to obtain k line-independent
packets of each generation. The SLSC method has much in common with FEC coding, but FEC
is implemented only on the destination node, while the SK method is also implemented on the
intermediate nodes. FEC and SK methods can work simultaneously without additional
modifications, but more complex hybrid solutions are possible. The SK method can improve network
reliability and resilience when used in conjunction with the multipath redundancy method.
3.4. Multipath redundancy method
The multipath redundancy method is to transfer data from the source to the destination via
multiple routes. This method of routing has diferent goals: to distribute the load on the
network routes or to increase the bandwidth and reliability of the network as a whole [
64, 65
].
Applying the multipath redundancy method, it is possible to get rid of congested sections in
the network or simultaneously deliver streaming data via multiple routes. However, a
selforganizing network is usually congested at the section between the source and the destination,
and its network topology is inconstant and therefore requires recalculation of independent
routes. It is because of this that multipath redundancy is more often used to improve network
reliability. Multipath redundancy can be provided by a superimposed network [
66, 67
]. The
simplest structure of a superimposed network is a single-layer tree. The root of such a tree is the
source node. The short existence time of a connection between nodes imposes restrictions on
the application of this structure in a self-organizing network. A multi-layer structure can also
be used in superimposed networks. This structure is more resilient to user outages, the load
is evenly distributed throughout the network, and does not require centralized coordination,
both during normal network operation and during emergencies. The structure is adapted both
for single-source transmissions and to provide transmissions from multiple sources. The
disadvantage of this structure is that the networks built on their basis are more complex than their
counterparts. This entails that to maintain this structure, a large volume of control messages
are transferred between the nodes of the network. This can significantly limit the applicability
of such a structure to nodes that vary considerably in self-organizing networks. The
multilayer tree structure seeks to eliminate two major drawbacks of an overlapping network with
a single-tree structure [
68
]. First, in single-tree structures, the few closest nodes are loaded
much more heavily than the rest of the network, since the “leaves” of the tree have not been
involved in content transmission. Secondly, the disconnection of these highly stressed nodes
leads to mass switching of network users looking for a new data source and a new connection
point. Node outages could lead to the degradation of streaming data quality. In a multilayer
tree, each node must stream data in multiple trees with a common root, distributing
complimentary content. Such a structure ensures that all content is not lost if one of the trees loses
connectivity and better utilizes the available resources of each node in the network. Source
S distributes streaming data to all nodes at once, but some video (e.g., every third fragment)
can be transmitted along the intended paths between the nodes themselves. Thus, instead of a
single tree, we consider three single-layer trees at once in the case of a multilayer tree with a
multilayer coeficient equal to three [
69
]. To improve network reliability, some algorithms use
diferent types of multipath redundancy, such as multipath “braid” routing. This move allows
the use of multiple routes instead of a single route to apply the SC method, which improves
the reliability and robustness of self-organizing networks [
29
]. The multilayer tree structure
helps to overcome packet loss in multicast in case of multiple node outage [
70
]. But the use
of multiple paths can increase the unproductive cost of forwarding data, so more research on
this method is needed. The multilayer tree structure can be combined with SC, FEC, and ARQ
methods to improve data transmission eficiency [71].