IEEE Circuits and Systems Magazine - Q2 2020 - 13

metering infrastructure, new communications and synchronization requirements will surely emerge. Synchronizing data in a fully integrated manner is expected to
become a challenging requirement in this domain, as a
variety of wired and wireless communications technologies will be used in those applications.
Other Applications: Many other types of applications
exist or are emerging that require tight synchronization
of devices [4], [11], such as those IoT applications that
need highly accurate time-based devices to be deployed
in large-scale. For instance, synchronized sensors are
important in law enforcement, military and crisis applications [9]. Accurate timing is a key element in such
applications since the processing of collected information would be significantly impaired if timestamps are
not accurate. Financial industry, where PTP is currently
used, is another example that time synchronization is
critical since inaccuracy timing can lead to disastrous
outcome [10].
E. Clock Synchronization in Network Standards
The task of clock synchronization can be carried out by
either hardware or software, depending on required precision (from millisecond to nanosecond) and -geographic
spread of the distributed system (from a few meters to
thousands of kilometers). Moreover, cost is an important
factor as well. In actual implementation, work needs to
be done according to protocols defined in industrial standards. The most popular ones are Network Time Protocol (NTP), Precision Time Protocol (PTP, IEEE 1588) and
Sync-E (Synchronous Ethernet).
NTP is one of the oldest protocols that is mostly
known due to the fact that it is used in the Internet
[22], [26], [27]. It is a highly robust protocol, especially
for distributed unreliable networks. It is widely deployed throughout all branches of the Internet. It is
generally regarded as the state of the art among time
synchronization protocols. Its synchronization accuracy can reach to the order of a few milliseconds in
public Internet and to sub-millisecond level in local
area networks. NTP is cost-effective since it does not
require any specific hardware. It is also scalable due
to its hierarchical structure. Compared to the solution of using a GNSS or atomic clock at every node, it
is very cost effective. Its primary drawbacks are the
extra communication overhead associated with the
request-response messages and the relatively low synchronization accuracy.
PTP is a master-slave protocol for delivering highly
accurate time over local area networks [20], [28]-[30].
It enables precise synchronization of clocks over heterogeneous systems with accuracy in the order of microsecond to sub-microsecond range. The supported
SECOND QUARTER 2020

protocols are UDP, DeviceNet, ControlNet, and PROFINET. PTP targets at large scale distributed applications
requiring microsecond synchronization, such as smart
grid, industrial automation and telecom systems. Unlike
NTP that does not require any special device between
slave and master nodes, PTP needs dedicated hardware
at intermediate nodes for the introduction of transparent and peer-to-peer clocks. It is worth to mention that
PTP provides a method to correct the rate of slave clock.
This is intended for syntonization. In current practices,
however, this method of syntonization is only applied
on time register (the logical clock). In this work, it will
be used to correct the frequency of clock hardware, as
will be discussed later.
Sync-E is based on and is compatible to its TDM predecessor SONET/SDH. They share the common architecture of synchronization, clock specifications, and
network elements (e.g. PRC and SSU). Synchronization
networks can be built as mix of Sync-E and SDH/SONET
elements. Sync-E and SDH/SONET share the principle of
Synchronization Status Message (SSM) and it allows interworking [21].
F. Key Components in Clock
Synchronization Algorithms
In centralized systems, the problems of mutual exclusion and inter-task communication are generally
solved using methods of semaphores and monitors.
For -distributed systems connected through a network,
the situation is much complex. They typically deal with
issues of multiple concurrent computation threads,
interconnections for inter-thread communication and
globally shared state that all individual computers cooperatively maintain. Synchronized clocks provide an
internal consistency in flow-of-time for all the involved
nodes. It can help ease the implementation of the aforementioned issues. For this reason, during past decades,
much research has been conducted in developing algorithms for seeking a common notion of time in fault-tolerant distributed systems. They can be software, hardware or hybrid -solutions.
From performance point of view, clock synchronization algorithms can be classified as deterministic, probabilistic and statistical. Deterministic algorithms assume
an upper bound on transmission delays and guarantee
a maximum difference between any two simultaneous clock readings. Probabilistic algorithms guarantee
a maximum deviation between synchronized clocks.
At any time, a clock knows if it is synchronized or not
with the other. But there is a non-zero probability that
a clock can become out of synchronization when things
go wrong. Statistical algorithms assume that the expectation and standard deviation of the delay -distributions
IEEE CIRCUITS AND SYSTEMS MAGAZINE

IEEE Circuits and Systems Magazine - Q2 2020

Table of Contents for the Digital Edition of IEEE Circuits and Systems Magazine - Q2 2020