This full text paper was peer-reviewed at the direction of IEEE Instrumentation and Measurement Society prior to the

acceptance and publication.

The Implementation of IEEE 1588 Clock

Synchronization Protocol Based on FPGA

Hongtao Yin, Ping Fu*, Jiaqing Qiao, Yuping Li

Automatic test and control institute
Harbin Institute of Technology
Harbin, China
[email protected]

Abstract—IEEE 1588 defines a precision time protocol, which mechanism, PTP is more appropriate for precise measurement
is widely used in distributed test and measurement systems. It is and control in a local area network (LAN) [10-11]. PTP provides
very important to capture the timestamps of the location that can a mean by slave clocks, which can estimate their offset from
affect the synchronization accuracy seriously in the master clock to ensure every node in the distributed system,
synchronization process. In this paper, we proposed a method to achieve the precise synchronization [12]. PTP implementations
capture the timestamps based on specialized hardware Field typically have a clock servo which uses a series of time offset
Programmable Gate Array (FPGA) between the physical layer estimates coordinate the local slave clock with the reference
and MAC layer. We designed IEEE 1588 message detection master clock [13]. Precise time coordination with PTP relies on
module and frequency compensation clock to detect IEEE 1588
precise estimates of the send and receive times (time stamps) of
message and record the timestamps, respectively. This method
can eliminate the delay jitter which is caused by the network
messages exchanged between the master and slaves.
protocol stack to improve the synchronization accuracy. The test There is an open resource software Precision Time Protocol
experiments results show that 97.76% of the synchronization daemon (PTPd), which is a pure software implementation of
deviation is located within ±40nS. PTP [14] and can only achieve the precision of the time
coordination between networked test and measurement systems
Keywords—IEEE 1588; FPGA; Synchronization on the order of microseconds [15]. It can capture the timestamps
in the application layers of the network protocol stack, which
I. INTRODUCTION introduces large degrees of non-determinism in the time stamp
The capacity of many control applications in distributed latencies, known as jitter [16]. In order to achieve high precision
systems has been greatly enhanced such as measurement and time stamps, it should draw the support of specialized hardware
control system, network communication systems, certain interfaces in the physical layer of the network. Figure.1
telecommunication systems, and et al[1-2]. High-speed data illustrates the various places in a network communication
transmission between network nodes could not achieve high protocol where time stamping can occur. The synchronization
performance in measurement and control systems without the precision is affected by the location of time stamps
accuracy and precision of time synchronization because the significantly.
network has the character of uncertainty and lack of real time.
The system needs a unified reference time to coordinate with
other subsystems.
The network time synchronization technology, Network
Time Protocol (NTP) was proposed to solve the problem of
lack real time in the network. The moderate achievable
accuracy of NTP is the range of millisecond which is not
sufficient for measurement and control systems[3]. It needs
synchronized clocks for distributed and correlated data
collection[4-5]. Synchronized clocks are used to coordinate
medium access in real-time networks or highly dynamic
applications such as time trigger control in measurement and
control systems[6-8]. Accuracies in the millisecond or even
microsecond range are often desired and node number would
be hundreds. To meet these constraints, the Precision Time
Protocol (PTP) was designed.
PTP, formerly IEEE 1588, is normally preferred in Fig. 1. Possible locations and relevant implementations for time stamping in
Ethernet-based local networks, mainly for measurement and a network-communication protocol.
control networks[9]. Compared with NTP synchronization

978-1-5386-2222-3/18/$31.00 ©2018 IEEE

Authorized licensed use limited to: Xian Jiaotong University. Downloaded on January 22,2021 at 06:43:42 UTC from IEEE Xplore. Restrictions apply.
 The rest of this paper is organized as follows. Section II is a message to master clock and record the sent time Ts2. Master
brief introduction of the principle of IEEE 1588 (PTP). Section clock will record the received time Tm2 when received the
III provides an overview of the PTP implementation with the Delay_Req message. Then master clock sends Tm2 in the
support of specialized hardware FPGA. Section IV describes Delay_Resp messages to the slave clock. Slave clock can
the logic design in FPGA. The software design in the calculate the delay from slave clock to master clock
implementation is described in Section V. Section VI presents (slave_to_master_delay ) based on Tm2 and Ts2, namely:
test experiments results, followed by the conclusion in section
VII. slave_to_master_delay =Ts 2 -Tm 2 (3)

Assume that when master clock received Delay_Req

II. IEEE 1588 CLOCK SYNCHRONIZATION PRINCIPLE message, the time deviation between master clock and slave
The time synchronization is defined in IEEE 1588V2. IEEE clock is Offset2, then:
1588 follows a master-slave model and determines the only
one master clock, which owns the best stability, accuracy, and Tm 2 = Ts 2 − Offset2 + slave_to_master_delay (4)
can stand for the standard time source in the system, through
the best master clock algorithm. IEEE 1588 synchronizes clock Based on the hypothesis that networks delays and offset
nodes in the networked system by broadcasting 4 basic PTP between master clock and slave clock are symmetric, namely:
messages, which are Sync message, Delay_Req message, slave _ to _ master _ delay = master _ to _ slave _ delay
Follow_Up message and Delay_Resp message, based on user (5)
datagram protocol (UDP). The slave clock calculates offset = one_way_delay
from master clock and one_way_delay between master clock
Offset1 = Offset2 = Offset (6)
and slave clock with receipt and transmission timestamps of
those PTP messages. Then according to equation(1)-(6), we can draw:

one_way_delay =
(Ts1
− Tm1 ) + (Tm 2 − Ts 2 )
(7)
2

Offset = (Ts1 − Tm1 ) − one_way_delay (8)

Then the slave clock can be synchronized to the master

clock with the offset.

A. Maintaining the Integrity of the Specifications

The template is used to format your paper and style the text.
All margins, column widths, line spaces, and text fonts are
prescribed; please do not alter them. You may note
peculiarities. For example, the head margin in this template
measures proportionately more than is customary. This
measurement and others are deliberate, using specifications
Fig. 2. Principle of Time Synchronization. that anticipate your paper as one part of the entire proceedings,
and not as an independent document. Please do not revise any
Figure 2 illustrates the method of time synchronization, Tm of the current designations.
stands for the master clock time and Ts stands for the slave
clock time. Master clock sends Sync message at each
synchronization cycle (typically 2S) and records the Sync III. FPGA-BASED PROTOCOL IMPLEMENTATION
message transmission time Tm1. Slave clock records the In Section II, the synchronization algorithm is proposed
accepted time Ts1 when received Sync message. Then master based on two assumptions, Offset1 is equal to Offset2 and
clock sends Tm1 in the Follow_Up messages to the slave clock. master_to_salve_delay is equal to slave_to_master_delay. In
Slave clock can calculate the delay from master clock to slave fact, Offset1 and Offset2 are not equal which is caused by the
clock (master_to_slave_delay ) according to Tm1 and Ts1. instability of crystal. In this paper, the effect brought by the
instability of crystal has been reduced by the frequency
master_to_slave_delay =Tm1 -Ts1 (1) compensation clock module. The master_to_salve_delay is not
equal to slave_to_master_delay because the asymmetry of the
We suppose that the time deviation between master clock network transmission delays which includes delay jitter in the
and slave clock is Offset1 when slave clock received Sync protocol stack and in the network component.
message, then:
Message transmission delay includes the delay in physical
Ts1 = Tm1 + Offset1 + master_to_slave_delay (2) layer and communication stack. The delay in the physical layer
without storage transponder is generally stable but the delay in
In order to calculate the amount of delay from the slave the communication stack is not stable. If the location where the
clock to master clock, slave clock needs to send Delay_Req timestamp is captured is closer to the physical layer, the

Authorized licensed use limited to: Xian Jiaotong University. Downloaded on January 22,2021 at 06:43:42 UTC from IEEE Xplore. Restrictions apply.
synchronization accuracy will be higher. In this paper, we
captured the timestamp between MAC layer and physical layer
with the help of FPGA. So the message transmission delay is
only caused by the physical network transmission, and it can
achieve sub-microsecond synchronization accuracy.
In our hardware design, an ARM processor is used as the
processing core and a FPGA is used to capture timestamps and
detect PTP messages, the hardware part would complete a
frequency adjustable real time clock, detect PTP messages,
record or insert timestamps. The software part runs on the
ARM platform equipped with Linux operation system to
manage the process of synchronization, calculate the offset and
adjust the hardware clock in FPGA, etc.

ARM
MAC Layer

Fig. 4. Overall structure of the FPGA module.

FPGA The receive/send message detection module can detect all

Message Detection the messages transmitted through the MII interface. When the
Module messages module detect the PTP messages, receive message
module detection module produces Rrdtime signal to read
received time in the clock module. Then the PTP messages and
received timestamps are stored in the message register FIFO.
Oscillato Receive message module generates receive message interrupt
Clock Module signal R0. Send message detection module produces a Trdtime
r
signal to read sent time in the clock module. Then the PTP
messages and sent timestamps are stored in the message
register FIFO. Send message module generates send message
Physical layer interrupt signal R1. When interrupt module detects signal R0 or
R1, it will generate an interrupt signal IRQ to inform ARM to
Fig. 3. Overall structure of the FPGA-based PTP implementation. read the messages and timestamps.

The structure of our model is shown in Figure 3. FPGA is A. Receive/send Messages Module
connected to the independent media Interface (MII) which is
Receive/send messages module are designed to detect and
between the Ethernet physical layer (PHY) and MAC layer it
store the PTP messages which are transmitted on the MII.
can detect every message passing through the MII interface.
Because the two modules are almost the same, we only
There are two modules in FPGA which are Message Detection
introduce the receive messages module. The detection module
Module and Clock Module. When the message detection
realizes PTP messages detection by listening and analyzing the
module detects the PTP messages, FPGA will record the
received and sent messages on the MII. The detection module
received or sent time as the timestamps and store the PTP
detects whether the message is PTP message by comparing
messages in First Input First Output register (FIFO). Then
only the following fields in the transmitting messages.
FPGA generates an interrupt signal to inform ARM to fetch the
timestamps. After gets the timestamps, the ARM calculates the (1) Type: this field explains the protocol type of the
offset and FreqCompValue using the timestamps and then Ethernet data packet. PTP adopts UDP/IP protocol, and its field
sends them to FPGA to adjust the hardware clock. ARM and should be IP.
FPGA realize data communication through the external bus
interface (EBI). (2) IP version: PTP messages use IPV4 version.
(3) Protocol: PTP messages use UDP protocol, so this
IV. THE LOGIC DESIGN OF THE FPGA field should be UDP.
The logic design is accomplished in the FPGA, it is mainly (4) Destination Port Number: PTP messages are
responsible for capturing the timestamps, detecting the PTP transmitted by event port, which port number is 319.
messages and generating a frequency adjustable real-time clock.
If it is PTP Sync message or Delay_Req message, the
The overall structure of the FPGA module is shown in Figure 4:
receive message module generates Rrdtime signal to read the
clock time as the timestamp and stores the message and
timestamp in FIFO1. FIFO1 will be cleared if an error occurred

Authorized licensed use limited to: Xian Jiaotong University. Downloaded on January 22,2021 at 06:43:42 UTC from IEEE Xplore. Restrictions apply.
during the message transmission, otherwise the message and counter value adds an increment in the next oscillator cycle,
timestamp in FIFO1 will be stored in FIFO2. Then the receive otherwise, the clock counter value keeps unchanged. The
message module generates interrupt signal R0. The module number of the clock counter value increases in unit time is the
design is shown in Figure 5. count frequency FreqCnt which is decided by FreqCompValue
and FreqOsc.
FreqOsc (9)
FreqDivRatio =
FreqCnt
2q (10)
FreqCompValue =
FreqDivRatio
Fig. 5. Receive message module.
Through formula (9) and (10), we can observe that when
the actual frequency changes, we must adjust FreqDivRation
B. The clock module by adjusting FreqCompValue to keep the FreqCnt unchanged.
The function of the clock module is to generate an
adjustable frequency real-time clock. The structure of the real- V. SOFTWARE DESIGN
time clock is shown in Figure 6.
The software runs based on the UDP/IP protocol. We
download the software to ARM platform which is implemented
in Linux application programs. The software is designed
according to IEEE 1588 and runs the protocol engine while the
system is powered on, which will always be running unless a
system error occurs. The function definition and interface of
every unit is shown in figure 7.

Fig. 6. Receive message module.

It is often to use a counter register to implement a clock,

which is driven by active crystal oscillator or passive crystal Fig. 7. The overall structure of the software.
oscillator. The character of these oscillators is low accuracy,
easy to be effected by the environment, low long-term stability, The software overall includes two parts, the application
and the frequency is not adjustable. As time passing and program and kernel program. The application program and
ambient temperature changing, the frequency of the crystal will kernel program communicate during the three modules. The
changes, called crystal’s time drift and temperature drift. With interface between the kernel and FPGA is the FPGA driver
the help of frequency compensation technique, the crystal module. So the protocol engine can access FPGA through Net
frequency can get dynamic compensation, thus the frequency Layer module, Clock Servo module and FGPA Driver module.
can be kept in a stable value. The relatively poor performance
crystal can also be successfully used in high-precision clock In the main flow chart, as shown in Figure 8, we will
synchronization system. introduce the function of every module definitely.

The frequency compensation clock is composed of a clock

counter, an accumulator, and an addend register. The function
of accumulator is to divide frequency. The addend register is to
save the frequency compensation value FreqCompValue which
id got from ARM. These three parts are driven directly by
crystal which frequency is FreqOsc. During each oscillator
cycle, the value in the accumulator adds FreqCompVlue which
is stored in the addend register, and the result is written back to
the accumulator. If the add action causes overflow, the clock

Authorized licensed use limited to: Xian Jiaotong University. Downloaded on January 22,2021 at 06:43:42 UTC from IEEE Xplore. Restrictions apply.
 Fig. 9. Main flowchart of software implementation.

In the process of the IEEE 1588 implementation, the

location where the timestamps are captured is critical. The
closer the location distances from the bottom of the protocol
stack, the smaller the impact of the protocol stack on the
synchronization accuracy is. In order to avoid the effect
brought by the network, the master clock and slave clock are
connected by a crossover cable directly. In this test platform,
we take two methods respectively to implement IEEE 1588.
One method is the FPGA-based implementation designed in
this paper. In this method the timestamps are captured between
the physical layer and the MAC layer. The other method is the
pure software implementation and the timestamps are captured
Fig. 8. Main flowchart of software implementation.
in the application layer. The test results are shown in Figure 10.
The clock will be initialized first to give the variable
original value when the protocol engine runs. Then the clock
goes to listening state. If it does not receive synchronization
message in the certain time when it is in the listening state, it
will state as master clock. If it receives the synchronization
message, it will do BMC algorithm using the received message
to define the clock state. The format of the message is defined
in Msg Packer module. When the clock is in master state, it
sends synchronization message. If the clock is in slave state, it
receives the synchronization message from the master clock.
When FPGA detects the PTP messages, the Time Stamp
module will record the timestamps in the 1588 Clock module.
Then FPGA generate an interrupt signal to inform the software
part to fetch the timestamp. Net Layer module would send,
receive messages, and read the timestamp. Then the Clock
Servo module calculates the offset from the salve clock to the Fig. 10. The effect of protocol stack.
master clock to implement time synchronization using the
timestamps. According to the timestamps, the Clock Servo In Figure 10, we can observe that the accuracy of the
module calculates and writes it to 1588 Clock module in FPGA software-based implementation is on the order of microsecond,
to adjust the clock frequency. but in the FPGA-based implementation, 97.76% of the errors
stand within ±40nS. When we adopt the software-based
VI. TEST RESULTS implementation, the time stamp is captured in the top of the
stack. In the CPU, one interrupt signal may take the CPU
IEEE 1588 implements clock synchronization by
several microseconds to process it and one task scheduler may
transmitting packets with network. Not only the clock itself,
deprive the process of the packet encapsulation or analysis.
but also any part related to the packets transmitting in the
Thus, the uncertainty of the time spent in the packet
network can be a factor that affects the synchronization
encapsulation and resolution process in the protocol stack is
accuracy. So there are many factors affecting the
very strong. In the FPGA-based implementation, the timestamp
synchronization accuracy, such as the delay produced by
is captured between the MAC and the physical layer. The
protocol stack, the stability of the crystal, the delay in the
propagation velocity of the electrical signal in the twisted pair
switch, and the state of the network, etc. In this paper, we
cable is about 2x108m/S, the signal propagation delay is about
mainly analyze the factor of the delay produced by protocol
5nS/m. In the case where the propagation distance is 5m, the
stack.
signal propagation delay is approximately 25nS. Assuming that
In order to analyze the factor how to affect the the uncertainty of the signal propagation velocity is 1%, the
synchronization result, we build the experiment environment, uncertainty of the signal propagation delay is 0.25nS. Thus, as
as shown in Figure 9. we can see from the figure, the precision of FPGA-based
implementation is higher than the software-based
implementation.

Authorized licensed use limited to: Xian Jiaotong University. Downloaded on January 22,2021 at 06:43:42 UTC from IEEE Xplore. Restrictions apply.
 VII. CONCLUSION Handbook, Part II (Industrial Information Technology), Section 7
(Integration Technologies). Boca Raton, FL: CRC Press, 2005, 106:1-
In this paper, we design an IEEE 1588 standard 15.
synchronization structure based on FPGA. The timestamps can [7] H. Hansson, M. Nolin. Real-time systems [M]. in The Industrial
be captured between the physical layer and MAC layer. It can Information Technology Handbook, Part II (Industrial Information
eliminate the delay jitter which is caused by the network Technology), Section 6 (Real-Time Embedded Systems). Boca Raton,
FL: CRC Press, 2005, 81:1-28.
protocol stack to improve the synchronization accuracy. The
deviation of the IEEE 1588 clock synchronization can be [8] P. Ferrari, A. Flammini. IEEE 1588-based synchronization system for a
displacement sensor network [J]. IEEE Trans. Instrument. Meas., 2008,
limited into ±40nS with a rate of 97.76%, when the master 57:254-260.
clock and slave clock are connected by a crossover cable [9] D. L. Mills. Internet time synchronization: The network time protocol
directly. [J]. IEEE Trans. Commun., 1991, 39(10):1482-1493.
[10] Network Time Protocol (version 3): Specification, Implementation and
REFERENCES Analysis, 1992, RFC 1305.
[11] Kendall Correll, Nick Barendt. Design Considerations for Software Only
Implementations of the IEEE 1588 Precision Time Protocol [J].
[1] R. Subrahmanyan. Timing recovery for IEEE 1588 applications in Automation instrumentation, 2005, 26(4): 1-3.
elecommunications [J]. IEEE Trans. Instrum. Meas., 2009, 58(6):1858-
[12] Lee Barford, Jeff Burch. Fourier Analysis From Networked
1868.
Measurements Using Time Synchronization [J]. IEEE Transaction on
[2] J. Han, D. K. Jeong. Practical considerations in the design and Instrumentation and Measurement, 2007, 56(5): 1601-1604.
implementation of time synchronization systems using IEEE 1588 [J].
[13] Lee, Edward. Timeliness in Networked Embedded Systems [J]. IEEE
IEEE Commun. Mag., 2009, 47(11):164-170.
Computer, 2005, 38(7): 85-87.
[3] D. L. Mills. Internet time synchronization: The network time protocol
[14] Hller R., Santer T.. Embedded Sync UTC and IEEE1588 clock
[J]. IEEE Trans. Commun., 1991, 39(10):1482-1493.
synchronization for industrial Ethernet [J]. Proceedings of Emerging
[4] P. Loschmidt, G. Gaderer. White Rabbit- Sensor/actuator protocol for Technologies and Factory Automation, 2003, 1: 422-426.
the CERN LHC particle accelerator [C]. in Proc. 2009 IEEE Sensors
[15] Dirk S. Mohl. IEEE 1588 Precise Time Synchronization as the Basis for
Conf., 2009:781-786.
Real Time Applications in Automation [R]. Industrial Networking
[5] T. Sauter. Fieldbus systems-Embedded networks for automation [M]. in Solutions, 2003: 110-113.
Networked Embedded Systems Handbook. Boca Raton, FL: CRC Press,
[16] Roland Holler, Georg Gaderer. Chip Design Building Blocks for
2009, 20:1-64.
Precision Clock Synchronization [R]. IEEE 1588 Annual, 2005: 78-79.
[6] R. W. Brennan, J. H. Christensen. Holonic manufacturing systems: A
technical overview [M]. in The Industrial Information Technology

Authorized licensed use limited to: Xian Jiaotong University. Downloaded on January 22,2021 at 06:43:42 UTC from IEEE Xplore. Restrictions apply.