You are on page 1of 9

10G EPON vs.

XG-PON1 efficiency

Pamela Begovic, Nasuf Hadziahmetovic, Darijo Raca


Faculty of Electrical Engineering, University of Sarajevo
Sarajevo, Bosnia and Herzegovina
pamela.begovic@etf.unsa.ba

Abstract - 10G EPON and XG-PON1 are both recently When it comes to FTTx networks, a large specter of
standardized but emerging technologies, ready to replace their solutions has been developed, so far. These solutions are based
less capacitive predecessors, 1G EPON and GPON respectively. on different architecture, types of network elements, multiple
Both technologies offer row bit rate of 10 Gbps on downlink, but access techniques and data layer protocols. At the moment, the
are based on significantly different signal processing principles most attractive and most usable solutions are those based on
and protocol stacks characteristics. According to these passive link components between central office and output
principles, different technologies introduce different amount and point on users side, known as passive optical networks (PON).
types of transmission overhead, which must be transmitted along These networks are the simplest and cheapest to implement [1],
with the useful information (in order to provide network
which is why they represent around 85% of installed FTTx
functionality). While these overheads drastically reduce real
available bit rates to the end users, it is of the great interest to
networks today in the world [3]. Therefore, these networks and
estimate these overheads with high accuracy. Accordingly, in this their newest standard versions are the subject of this paper
paper, entire overhead (including PHY and MAC layers analysis.
overheads) introduced by 10G EPON and XG-PON1 technologies Chronologically speaking, after a few less successful
is estimated through their corresponding protocol stack analysis. solutions, ITU-T and IEEE in 2004 have both individually
Based on these estimations, final conclusions about 10G EPON announced PON networks standards, called GPON (ITU-T
and XG-PON1 efficiency and thus maximum real achievable data
G.984) and EPON (IEEE 802.3ah) respectively [4]. They
rates are derived. This enables network operators to make
provide maximum theoretical rates up to 1.25/1.25 Gbps
appropriate decision when choosing between the two concurrent
technologies.
(EPON) and 2.48/2.48 Gbps (GPON) [5]. However, not entire
bandwidth is available to the end user. It is shared among
Keywords - NGA, 10G EPON, XG-PON1, protocol stack, multiple users (up to 32 or even 64) and moreover, it is used for
bandwith efficiency, transmission overhead transmission of system information (transmission overhead)
whose amount depends on type of a PON standard. Because of
that overhead, GPON and EPON efficiencies are 94% and 73%
I. INTRODUCTION
respectively (for maximum raw data rates available) [4].
The growth of high bit-rate services, such as TV over IP Consequently, these technologies provide around 30 Mbps
(IPTV and High Definition IPTV, which require up to 20 (EPON) and 70 Mbps (GPON) per user (assuming 32 users
Mbps), Web-TV (e.g. YouTube) and also the increasing sharing the total bandwidth) [4]. Accordingly, real available
popularity of social networks (resulting in a multimedia content data rates per user are different, depending on chosen
exchange) required from current telecommunications networks technology, and moreover, they are both drastically reduced in
a migration towards architectures and technologies capable of respect to theoretical rates. Bearing in mind the above
ensuring ever higher bandwidth [1]. These migrations had to be mentioned, since 2006, first IEEE then ITU-T also, estimated
made in all sections of the network, especially in the access that in time to come, available rates will become insufficient
part where the bandwidth limitations bottled the development for providing desired quality of service. During that period,
of the next-generation networks [1]. both organizations started working on standardizing NGA
(Next Generation Access) optical access networks upon a PON
In order to overcome this issue, a significant number of
fiber infrastructure.
different access wideband technologies have been developed,
based on existing, newly-built cable infrastructure or the use of In 2009, a 10G EPON standard (IEEE 802.3av) was
wireless media. However, popular technologies like xDSL, announced, followed by XG-PON1 (ITU-T G.987) standard in
cable modem networks as well as existing wireless 2010. 10G EPON and XG-PON1 enable shared raw bit rates up
technologies are not capable of providing enough bandwidth to 10 Gbps [6]. However, similarly to previously standardized
for emerging services [2]. Among all currently existing PON solutions, 10 Gbps bit rates are just theoretical and must
technologies, only optical medium has capacities to actually be used for multiple users data transmissions together with
deal with future needs [3]. Therefore, in long terms, optics in transmission overheads [7]. While both NGA systems features
access segments (under the generic term FTTx - Fiber To The introduce a number of different transmission overheads, real
x), represent the only solution which supports large bandwidth available bit rates are reduced differently, depending on
capacities, required by communications networks [3]. technology [7].

978-963-8111-77-7
Accordingly, in the process of access network planning it is EPON is completely compatible with its predecessor, which
necessary to perform objective assessment of available rates makes system upgrade very simple and provides coexistence of
offered by each of these technologies to their users, in terms of both solutions over the same optical distribution network
their efficiency factor [8]. Reflecting the above mentioned, a (ODN) [14]. This has made 10G EPON technology very
detailed overview analysis of NGA solutions (10G EPON and attractive for operators already having installed EPON systems
XG-PON1, that have been standardized up to the moment of of former generation.
writing) are given in this paper, focusing on comparing
efficiency and assuming the implementation protocols as close Similarly, FSAN/ITU-T in 2007 has also started to work on
as possible to MAC protocols. NGA PON solution [6], but in two different directions. The
first group (NG-PON1) included systems that could coexist
Similar efficiency comparison or only the results have been with GPON on the very same ODN infrastructure. The second
presented for EPON and GPON technologies in [5], [4], [8] group (NG-PON2) included all other systems that either
and [9]. As far as NGA solutions are concerned, technologies required a different ODN or that required technologies that
mentioned are just recently standardized, so up until now, not were not to be available in close future [15].
many papers dealing with these issues have been published.
Concluding from a survey of literature available, there are few In 2008, FSAN group gathered proposals for NG-PON1
and among seven different propositions choose XG-PON1, as it
papers ([7], [10], [11]) dealing with only specific overhead
segments separately and for each technology individually. seemed to be the most appropriate candidate for starting phase
According to the knowledge of the authors, not one paper that of standardization development [15]. As a result, the solution
gives comparison efficiency analysis of both solutions (with has been standardized in June 2010, as a part of ITU-T G.987
overall overhead estimation concluded with MAC layer of standard. It offers asymmetric bandwidth of 10/2.5 Gbps, uses
protocol stack) has been published. Hence, this paper should one wavelength in each direction and can support between 32
then represent a contribution to this field and provide objective and 256 users on each ODN [16]. These networks are,
insight into a real performance of rival solutions to the future according to request, compatible with previous GPON
operators. networks generation, which gives them a complete
competitiveness against IEEE solution.
Paper is organized as follows: Section II contains a short
overview of NGA solutions standardized to this day (10G Moreover, within the NG-PON1 group, XG-PON2
EPON and XG-PON1) and those in perspective. Concepts of (offering 10 Gbps symmetric bandwidth) was also
systems functioning through identifying a transmission recommended for further development, and its standardization
overhead introduced by 10G EPON and XG-PON1 is expected towards the end of 2011/2012 [16].
technologies are given in Section III and Section IV However, when more bandwidth required than can possibly
respectively. Section V contains an assessment of performed be provided by 10G EPON or NG-PON1, and when PTP
efficiency and real user rates in respect to scenario suggested, (point-to-point) fiber is not an option, wavelength division
for each of the solutions. A brief discussion of analysis results PON (WDM-PON) may be the answer. Although WDM PON
and conclusion is given in last section of the paper. solutions are still under standardization, they are going to offer
rates up to 40 Gbps and symmetric bandwidth, but they are also
II. CURRENT STATUS OF NEXT-GENERATION PON going to exhibit a higher cost per line, compared to 10G EPON
or NG-PON1 [6].
A large step forward has been made with an introduction of
EPON and GPON technologies in terms of available capacities, Accordingly, up to moment of writing, just 10G EPON and
especially compared to other access network technologies like XG-PON1 have been standardized as NGA solution, which
DSL or cable modem networks. However, with the makes them only candidates for comparative analysis within
development of more demanding services (such as 3DTV, this paper. Moreover, both technologies offer the same 10 Gbps
multiple full HDTVs per household etc. [12]), these capacities downstream bandwidth, but are built on quite different
will soon be insufficient for providing a desired quality of principles. Therefore it is of great importance to compare
service to end user. These trends were first recognized in 2006, objectively their technical characteristics and consequent
when IEEE and ITU started preparing optical access NGA efficiency. This enables network operators, which are planning
network standards. to implement FTTx solution within their access networks, to
make appropriate choice between these two concurrent
After several years of preparation, IEEE at the end of the technologies. In the sections that follow, a comparison
year 2009, finally published standard named 802.3av defining efficiency analysis of 10G EPON and XG-PON1 solutions are
the specifications for 10G EPON networks. At the moment of given. Their aim is to provide conditions for maximum
announcing, standard assured a transmission through rates available rates per user estimation and represent a starting point
higher than all other access technologies, providing asymmetric in the process of PON networks planning.
transmission bandwidth of 10/1 Gbps and symmetric, equal to
10/10 Gbps [13]. As its reference for the optical link loss
budgets, the 802.3av specification uses a split ratio of either III. 10G EPON OVERHEAD IDENTIFICATION
1:16 (i.e., 16 optical network units (ONUs) on a single PON Each TDMA PON network (EPON, GPON, XG-PON1),
connecting to one optical line terminal (OLT) interface) or therefore 10G EPON network as well, has an identical
1:32, but in practice, larger split ratios such as 1:64 or 1:128 topology which can be described as multipoint-to-point for
can also be used [14]. It is also important to mention that 10G upstream, and point-to-multipoint for downstream, because of
the character of PON architecture. Accordingly, all system
users, represented as ONU units, share a single transmission
channel towards central device (OLT located in central office).
Thus it is necessary to allocate its resources between all
existing ONU units.
Figure 1. a) Ethernet frame format; b) 10G EPON frame format [17]
The upstream transmission occurs in burst mode, since
ONUs must only enable their lasers to transmit a burst of data According to Fig. 1, header contains IFG (Inter Frame Gap)
and disable them immediately after that [7]. According to their periods, introduced in order to prepare devices for receiving
traffic demands, multiple (N) ONU units take turn transmitting next frame. After that, a preamble field comes, whose content
in their dynamically allocated time slots, which occur almost has been slightly modified in 802.3av standard in relation to
periodically. original 802.3 field. It consists of LLID (Logical Link ID), start
On the other side, downstream link transmission is of LLID delimiter (SLD) and appropriate CRC part. LLID field
broadcast. All traffic is transmitted to every ONU unit, but only carries information about EPON device address and is used to
one having the corresponding address will pass the traffic emulate a PTP medium. It relies on tagging of the Ethernet
through upper OSI model layers. frames, with tags unique for each ONU [9], thus differing from
original preamble field used for Ethernet frame
Therefore, it is obvious that data related to time slots synchronization. Before MAC layer, an additional addressing
allocation, devices addressing etc., are beside user data, scheme is useful because in downstream direction, even though
necessary to transmit through a PON system. Moreover, for all frames are delivered to each ONU, only the ONU with
transmissions full functionality (as in any other network), appropriate LLID pass the frame to its MAC layer [9]. Other
pieces of information used for synchronization, encoding etc., fields of each Ethernet frame contain MAC addresses of
also need to be transmitted. Apparently, all mentioned occupies receiver (DA) and transmitter (SA), after which comes a field
a part of overall links capacity, directly influencing maximum indicating length/type (L/T) of Ethernet frame. Ethernet
transmission rate available to users. Consequently, shared links payload optionally contains from 46 to 1500 bytes (Npl),
rate is being significantly decreased compared to theoretical followed by 4 bytes of FCS (Frame Check Sequence) code for
rate of 10.3125 Gbps (recommended by standard for potential errors detection, caused due to the transmission [9].
symmetrical 10G EPON systems [11]). In order to evaluate real Length of payload depends on momentary available amount of
data transmission rates available to users, it is of great data being transmitted within the frame, causing the overhead
importance to perform an assessment of overall overhead. That (as a consequence of Ethernet encapsulation) to become 2.47%
overhead is introduced in order to provide system functioning out of 45.24% of total amount of data being transferred to a
conditions and is necessary to identify all elements leading to lower layer from the MAC layer. Accordingly, the share of this
bandwidth efficiency decrease in both transmit directions. overhead in total amount of transmitted data, for Npl bits within
Similarly to other actual technologies, concept of 10G each Ethernet payload, can be expressed as aEt = 38/(38 +
EPON networks is based on redirecting system information Npl)100%.
through 802.3av layers of reference model (bearing various However, while many frames coming from (or to) each
types of frames in mind). Therefore, in order to identify the ONU with different payload's length are forwarded together to
overhead, it is necessary to identify all encapsulation elements lower level, the value of Ethernet frame overhead depends on
and control messages exchange, through referent model layers specific distribution of packet sizes [7]. Thus, if different
and according to the standard mentioned. Moreover, bearing in payload length, marked as Npl1, Npl2,...,Npln, appear with
mind that 802.3av standard speficies only two lower layers of probabilities p1, p2, ..., pn, average Ethernet encapsulation
ISO-OSI model (PHY and MAC layer), the data within overhad share in total amount of transmitted data can be
payload of second layer are considered as usage data. Under expresed as aEt = 38/(38 + p1Npl1 + p2Npl2 ++ pnNpln)100%.
these circumstances, in 10G EPON, three different types of
overhead can be identified: Ethernet framing overhead, control B. Line coding, FEC and burst mode overhead
message overhead and line coding, FEC and burst mode
overhead. Accordingly, in the rest of this Section, 10G EPON In accordance with 10G EPON concept, all buffered
overhead identification of appropriate ISO-OSI model is Ethernet frames are forwarded to lower ISO-OSI layer for each
presented, from the highest standardized level to the bottom ONU individually. Additional data processing and
layer of it. encapsulation is there performed. However, since downstream
and upstream transmissions feel different topologies, their
overheads differ and must be treated separately.
A. Ehternet encapsulation overhead
10G EPON technology has inherited data encapsulation 1) Downstream line coding and FEC overhead: In the
concept within Ethernet frame on a data link layer, from its downstream direction (from the OLT to ONUs), Ethernet
predecessor EPON [14]. Accordingly, looking from a data packets transmitted by the OLT, pass through a 1N passive
layer of ISO-OSI model, a typical Ethernet frame header is splitter and reach each ONU [9]. However, prior transmitting,
added on Ethernet payload (which is considered as usage data buffered set of Ethernet frames within OLT passes to next
within this analysis). lower layer, or more precisely, to line coder. The line coding
for 10G EPON is 64B/66B thus reducing line coding overhead
(aLC_d) from 20% in its precursor EPON to just 3.03%.
After line coder, a self-synchronous scrambler is used on After line coding, scrambling and FEC (identically as in
the block payload in order to help randomize the payload bits. downstream direction) are performed. However, since 10G
However, scrambler doesnt introduce any overhead and it EPON specification does not allow shortened codewords in
doesnt affect bandwidth efficiency in any way. FEC coding structure, upstream transmission slot must be
extended to include integer number of codewords [7]. That is
Afterwards, scrambled bits enter the forward error while variable size Ethernet frames may not be able to
correction (FEC) coder. 10G EPON specification includes a completely occupy the granted timeslot, and according to [7],
mandatory FEC based on Reed-Solomon (255, 223) channel half of one codeword payload (composed of 27 66 bit blocks)
encoding [7]. Accordingly, FEC coding process is in these remains unused during actual transmission by each OLT.
networks preformed as follows: for data streams at the output Consequently, so called codeword quantization overhead
of the scrambler, FEC encoder accumulates 27 of 66 bit blocks
occurs, being equal to aCQO = (0.51782N)/(10.3125109T)
to form the basis of an FEC codeword. First, it removes the
redundant first bit of each block (which is guaranteed to be the 100%. It is also important to emphasize that codeword
complement of the second bit) [15]. It then adds 29 padding quantization overhead is specific only for upstream
bits (binary 0) to the 27 blocks (65 bits each) to form the 223- transmission, assuming that in downstream direction there will
byte payload portion of an FEC codeword which, combined always be enough traffic to fill-in entire codeword.
with 223-byte payload, form the 255-byte Reed-Solomon Opposite to downstream direction, FEC encoded data can
codeword [18]. After the Reed-Solomon codeword has been not be directly transmitted to media. While at the moment, just
computed, the FEC encoder constructs the transmittable FEC one ONU can transmit, it has to turn on its laser first and turn it
codeword with the original sequence of 27 66 bit blocks off after the transmission. It means that additional overhead,
(including the redundant first bit, but not including the 29 known as burst mode overhead, must be introduced within each
padding bits). Moreover, while 32 parity bytes can be treated as ONU transmission.
four groups of 64 bits, FEC coder also adds additional 2 bits to
each of these blocks, to remain compatibility with line coding
principles [14].
At the end, it is important to emphasize that 10G EPON is
stream based, so it makes estimation of FEC overhead
independent on packet size distribution on upper layers [18]. Figure 2. Internal structure of the upstream channel data burst in 10G EPON
Accordingly, for every 27 64B/66B blocks of transmitted data, [7]
FEC encoder adds 4 64B/66B blocks of parity, resulting in
aFEC_d = 12.9% FEC overhead share, in total amount of According to Fig. 2, before starting a transmission from
downstream data. Finally, in downstream direction, created 31 each ONU, some amount of time (known as laser on time -
66 bit blocks are sent directly to the media for transmission TON) needed for the laser to turn on, must be reserved.
[14]. Afterwards, it is also necessary to reserve some time for gain
control (TAGC) and for performing clock and data recovery
2) Upstream Line coding, FEC and burst mode overhead:
(TCDR). Depending on the quality and implementation of the
In the upstream direction (from ONUs to the OLT), each ONU
OLT burst-mode receiver, these values might vary for different
transmits in its allocated time slot. These time slots for each
manufacturers. In [19], proposed values for each of them equal
ONU occur almost periodically and the period in which all N
200 ns. Moreover, before the start of FEC data portion,
ONU units (within existing ODN) are served, is called cyclic
additional burst delimiter (BD) field, containing 66, bit must be
period T. Each time slot within that cyclic period is capable of
added. This field is used to enable OLT to reliably identify the
carrying several Ethernet packets. ONU must buffer frames
beginning of an incoming data burst, even in the presence of bit
received from a subscriber until its time slot arrives [9]. When
errors [7]. After the transmission of FEC coded data (for each
its time slot arrives, the ONU ''burst'' all stored frames at the
ONU unit), end of burst (EOB) delimiter (composed of 66 bits)
full channel. However, before transmitting, additional
indicates to the OLT receiver that the end of the data burst from
overhead must be added.
the given ONU is also necessary [7]. Finally, after
Similarly to downstream direction, in upstream, line coding transmission, amount of time, named laser off period (TOFF),
64B/66B is also performed. But, opposite to downstream, in must be reserved in order to enable laser to turn off. TON and
upstream direction two 66 bit blocks with idle code, at the TOFF depend primarily on ONU laser diode and the driver
beginning of the FEC protected portion of every burst (for each circuit [7] and in 10G EPON, these times usually equal to 64 ns
ONU) must be added [7]. These idle blocks are used to [19]. Although the IEEE 802.3av specification does not specify
synchronize the descrambler and to provide packet delineation any guard band between two consecutive upstream bursts, laser
at the MAC layer of the OLT. As this overhead (called idle- on and laser off periods of two consecutive bursts may overlap.
prefix overhead) is present in every upstream burst, it is easy to Thus it is recommended to ensure a gap between these burst,
estimate the total bandwidth it consumes [7]. Accordingly, named guard band period (TGB). Accoridng to [7], its value is
assuming that raw transmission rate equals to 10.3125 Gbps, equal to 81.25 ns. Consequently burst mode overhead
within each cyclic period T, the total of 10.3125109T bits can (assuming downstream row bit rate R = 10.3125109 bps) now
be transmitted. While idle-prefix overhead is composed of two can be calculated as: aBM = (LON + TAGC + TCDR + LOFF + BD/R
66 blocks for each ONU unit, for N of them, it can be + EOB/R + TGB)N/T100%.
calculated as: aIP = (266N)/(10.3125109T)100%.
However, in the upstream transmission, guard band and physical layer; the second is the framing layer (FS), which does
burst model overheads are not line coded and FEC protected. the main work of transmission convergence (that is, the control
Therefore, the upstream line coding and FEC overhead is less of the PON TDMA system); the third is the client adaptation
than in downstream direction, and can be calculated as aLC_u = layer (CAS), which carries user signals over the XG-PON
3.03(100 aBM)/100% and aFEC_u = 12.9(100 aBM)/100%. system [15]. Since downstream and upstream transmissions
have different topologies and thus related transmission
C. Control overhead principles, frame formats through identified sublayers differ in
different transmission directions.
To support a time slot allocation by the OLT, the multi-
point control protocol (MPCP) is being developed by the
802.3ah task force, and taken out from within 10G EPON A. Downstream XG-PON1 overhead
standard. This protocol relies on two Ethernet messages: GATE In XG-PON1 networks, downstream transmission occurs in
and REPORT, which are both MAC control frames, type 88-08 125 s PHY frames. These frames are used as transport
[9]. A GATE message is sent from the OLT to an ONU and it structures for FEC encoded blocks of data coming from higher
is used to assign a transmission timeslot. A REPORT message layer, together with physical synchronization overhead (PSBd).
is used by an ONU to convey its local conditions (such as
buffer occupancy, etc.) to the OLT and to help the OLT make
intelligent allocation decisions [9]. In view of that, for the
purpose of time slots allocation, during each cyclic period at
least one pair of these messages must be exchanged between
OLT and each ONU. To simplify the calculations, according to
[7], it can be assumed that each GATE message carries a single
bandwidth grant which includes time for transmission of a
REPORT message back to the OLT. Since each MAC control
frame type 88-08 contains 84 bytes in total, the control
overhead is therefore aC = (672N)/(10.3125109T)100% in
both downstream and upstream.
Beside the above components, there is also discovery
overhead in both, upstream and downstream. Discovery Figure 3. Downstream XG-PON1 layer organization [15]
overhead presents the time needed for process of discovering
new ONU units. It is usually repeated periodically every 1 In view of that, supposing transmission rate of 9.95328
second, and last for 300 s [9]. During that time, no other Gbps, defined within G.987 standard for downstream direction
transmission is allowed, thus introducing additional overhead [15], total of 155520 bytes within each 125 s PHY frames can
known as discovery overhead, equals to aD = 0.03%. be transmitted. However, 24 of these bytes must be reserved for
PSBd overhead (Fig. 3). In downstream direction, PSBd
IV. XG-PON1 OVERHEAD IDENTIFICATION overhead is composed of following: 8 bytes set to fixed
framing pattern, used to enable receiver to find the beginning
XG-PON1 presents, up to this point, the only standardized
of each PHY frame (PSync), 8 bytes of Superframe Counter,
version of NG-PON solutions. It emerged as a successor of
which provides a much longer scale time reference and 8 bytes
GPON standard, allowing transmission of 10 Gbps in
carrying PON identity (PON-ID) [15]. Therefore total of
downstream and 2.5 Gbps in upstream direction. Thus, it
155496 bytes within each PHY frame can be used for carrying
inherited GPON concept, introducing some novelties (for
FEC codewords. XG-PON1 uses Reed-Solomon (248, 216)
example, higher security level, both in upstream and
FEC correction code in downstream, and accordingly, within
downstream and more heavily structured protocol stack etc.
each PHY frame, the total of 627 codewords can be
[15]).
transmitted. Since 32 bytes in each of these 627 codewords
Being a representative of TDMA PON, XG-PON1 represent FEC redundancy, PAS downstream overhead can be
recognizes the same topologies as its competitor 10G EPON, calculated as aPAS_d = (24+32627)/155520 100% = 12.92%.
including broadcast transmission in downstream and dynamic
time slot allocation in upstream directions. But opposite to 10G However, before sending to FEC encoder, the group of
EPON, no control and discovery message must be transmitted 135432 bytes is formed on FS layer. Thus, XGTC frame is
through the network. In purpose of ONU discovery and time composed of CAS PDU and XG-PON transmission
slot allocation, special encapsulation overhead is being convergence downstream header (XGTC he.), as shown in Fig.
introduced within transmitted frame formats, in both 3. XGTC header consists of three parts. The first is the fixed
transmission directions. size part of the header, which carries the lengths of the next
two parts of the header (11 bits of bandwidth map (BWmap)
As a result, the entire XG-PON1 overhead can be identified length and 8 bits of PLOAM count), and is protected with 13
through its protocol stacks frames encapsulation overhead bits header error correction (HEC) code [20]. The second part
identification. The appropriate protocol stack, defined within is the bandwidth map (BWmap). It carries N 8-bayts upstream
ITU-T G.987.3 standard, is structured with three distinct bandwidth allocations to the ONUs (or more precisely,
sublayers: the first is the physical (PHY) adaptation layer transmission containers (T-CONTs) within each ONU) and
(PAS), which handles the unique issues of the XG-PON defines time slots allocation scheme for upstream traffic.
Therefore, after the bandwidth allocation identification (Alloc-
ID), each of these N fields carries start time and payload length
information for each transmission (regarding to T-CONTs
within ONUs). Moreover, it carries information about necessity
of transmitting DBRu and PLOAMu fields in appropriate
upstream transmission within the allocated time slot, FWI
(Forced Wakeup Indicator) and BProfile (Burst Profile) fields,
all protected with 13-bits HEC [20]. Finally, the third part
carries physical layer operations administration and
management (PLOAM) messages to the ONUs on the PON.
Opposite to GPON, in XG-PON1, zero, one or more (P)
PLOAM messages (consists of 48 bytes each) can be
transmitted in downstream direction [20]. Furthermore, just one Figure 4. Upstream XG-PON1 layer organization [15]
of these messages can be broadcast and just one message can
be transmitted per each ONU. Following this header, the However, upstream PHY frame has a few differences from
remainder of the downstream carries the payload. According to the downstream. The first is that the physical synchronization
above mentioned, FS downstream overhead can be calculated block (PSBu) contains additional preamble and delimiter
as: aFS_d = (4 + N8 + P48)/155520 100%. patterns, composed of at least 160 and 32 bit intervals (or
maximum 1856 and 64 bit intervals) [21]. Moreover, to prevent
Previously mentioned XGTC payload also consists of the upstream transmissions to collide with each other, OLT builds
set of higher layers units, known as XGEM (XG-PON BWmap, allowing suitable guard time between upstream burst
Encapsulation Method) frames. XGEM frame is composed of from different ONUs. Minimum guard band duration is 64 bit
header and payload. Further, XGEM header consists of 8 bytes: intervals, and maximum, 128 [21]. The second is that the
payload length indicator (PLI), indicating the length in bytes of payload is not of a fixed size, and for this reason, the number of
an SDU or SDU fragment in the XGEM payload; key index FEC codewords in each burst is variable. In addition, to reduce
related to encryption method; XGEM Port ID which indicates the loss of bandwidth for odd-sized bursts, the last codeword in
the port to which the frame belongs; LF, presenting last frame the burst can be shortened to fit the available amount of time in
indicator, (used because XG-PON allows fragmentation of the burst [15]. Finally, in upstream direction Reed-Solomon
SDU) [20]. However, since typical Ethernet traffic is treated (248, 232) is used, differently from correction code in
within the 10G EPON overhead identification, in order to make downstream direction.
these two solutions comparable, Ethernet traffic is also going to
be treated within XG-PON1 overhead identification. So it is of However, if we assume 125 s to be duration of upstream
great importance to describe how Ethernet frames can be PHY frame (also assumed within the standard G.987.3 [20]),
mapped into a XGEM frame. Thereby, it must be stressed that taking upstream data rate to be 2.48832 Gbps [15], maximum
each Ethernet frame can be mapped only into a single XGEM of 38880 bytes can be transmitted within each upstream PHY
and one XGEM frame may not encapsulate more than just one frame. These frames are further transmitted as a series of
Ethernet frame [20]. Bearing in mind the importance of relatively short PHY bursts, coming from different ONUs.
Ethernet traffic in todays networks, within XG-PON1 Thus, while minimum 32 bytes must be reserved for PSBu
standard, Ethernet frames are carried directly to XGEM frame and guard time, in each 125 s PHY frame, 156 full codewords
payload. However, some of its parts (IFG and preamble fields and one shortened (also with 16 bytes of redundancy) can be
refer to Fig. 1) are previously discarded, which additionally transmitted. Consequently, PHY upstream overhead is equal to
reduces XGEM overhead. Finally, in respect to Ethernet
minimum aPAS_u = (32+15716)/38880100% = 6.54%. If more
payload (which is, according to Section 3.A, considered as
than one PHY bursts must be transmitted within one PHY
users data), only 8 bytes of XGEM header and overhead
frame, overhead is eventually increased for few percent (for
composed of Ethernet source and destination MAC address,
example, assuming 32 PHY bursts within 125 s PHY frame,
field indication Ethernet frame type and appropriate FCS (18
overhead increases to 8.9%). However, while XG-PON1,
bytes in total) must be transmitted. Moreover, while XGEM
opposite to 10G EPON, allows shortened code words, there is
payload length must be divisible with four (for length bigger
no quantization overhead, presented within 10G EPON
then 8 bytes), additional 0 to 3 padding bytes must also be
technology, in upstream direction.
added [20]. Taking average value to be 2 padding bytes, and
using principles proposed in Section 3.A, the following can be According to Fig. 4, it can be concluded that each PHY
concluded: if different payload length, marked as Npl1, burst is composed of few XGTC payloads and AO (Allocation
Npl2,...,Npln, appear with probabilities p1, p2, ..., pn,, average Overhead) fields related to them, XGTC header (X.H), and
XGEM encapsulation overhad can be expressed as aXGEM = XGTC trailer (X.T). XGTC header is composed of ONU-ID
28/(28 + p1Npl1 + p2Npl2 ++ pnNpln)100%. (that contains the unique ONU-ID of the ONU that is
transmitting the burst), Ind field, providing fast unsolicited
B. Upstream XG-PON1 overhead signaling if the ONU status and appropriate HEC, which are all
The upstream of the XG-PON1 is burst-transmission together, occupy 4 bytes [19]. Additionally, it also carries 0 or
oriented, and is, similarly to downstream direction, at PAS 48 PLOAMu bytes, in respect to PLOAMu flag in allocation
layer composed of PSBu and FEC data fields. structure, within appropriate downstream BWmap. Moreover,
related to each T-CONT within ONUs, the presence of
allocation overhead (AO) is controlled by the OLT, using
DBRu flag of the corresponding allocation structure within the
BWmap. It caries buffer status report, allocated with a specific
Alloc-ID. XGTC trailer is 4 bytes long and represents
interleaved parity computed over entire XGTC frame. So, if we
assume M PHY bursts to be within 125 s PHY frame, half of
the PLOAMu fields equal to 1 (within appropriate BWmap)
and NT-CONT of different transmission containers within each
ONU for which is necessary to transmit AO, upstream FS
overhead can be calculated as: aFS_u = (8M + 48/2M + 4MNT-
CONT)/38880100%.
Figure 6. 10G EPON uspteram overhead
However, similarly to downstream transition, XGTC
payload is composed of identically structured XGEM frames,
thus making XGEM overhead calculation identical in both B. XG-PON1 overhead calculation
upstream and downstream directions. Relying upon transmission principles described in Section
IV, entire XG-PON1 overhead now can be calculated.
V. 10G EPON AND XG-PON1 EFFICIENCY COMPARISON Downstream and upstream PAS overheads are already
estimated in Section IV. Beside, for traffic distribution, tree-
According to transmission principles both, for 10G EPON
modal distribution identical to 10G EPON can be adopted,
and XG-PON1, presented in Section III and Section IV, it now
enabling us to calculate also a XGEM overhead. However, in
became possible to calculate entire overheads for proposed
order to estimate entire overhead, it is also necessary to make
scenarios, and according to these overheads, to estimate system
some assumptions about the number BWmap allocation
efficiency and maximum real available data rates.
structures and PLOAMd messages within each XGTC
downstream frame, together with the assumption about number
A. 10G EPON overhead calculation of PHY bursts within each 125 s frame. Assuming that, within
In order to calculate 10G EPON efficiency, some each 125 s download frame, one broadcast PLOAMd message
assumptions must be made: the assumption about number of is going to be sent together with unicast messages for one T-
ONUs within existing PON, the assumption about cyclic period CONT within each of NONU (NONU is total number of ONU
duration, and moreover, about traffic distribution. Since units within existing PON), P becomes equal to 1 + NONU.
(according to IEEE 802.3av) 1:32 split ratio is recommended, Moreover, supposing that in each 125 s frame, two T-CONT
the number of 32 ONUs is going to be treated within this within each ONU is allowed to transmit, N becomes equal to
calculation; according to literature ([5], [7], [9], [10]) the most 2NONU and downstream FS overhead can easily be calculated.
commonly used cyclic periods are 1 ms, and 2 ms, thus within Furthermore, it is of great importance to emphasize that,
this paper, different scenarios, including 1 ms and 2 ms are opposite to 10G EPON, maximum value of number N, in
going to be treated. Moreover, when it comes to traffic general, can be calculated as the sum of all T-CONTs (usually
distribution, simplified tree-modal distribution is taken from equals to 5) within all ONUs (where T-CONT represents a
[5]. It assumes appearance of 64, 500 and 1500 bytes payload group of logical connections that appear as a single entity for
lengths, appearing with probability of 0.6, 0.2 and 0.2, the purpose of upstream bandwidth assignment within a PON
respectively. Thereby, 1500 byte is the maximum Ethernet system [21]).
limit, whereas 60% of transmitted packages with average size
of 500 bytes are ACK packages [5]. However, assumed traffic However, when an upstream PHY burst belongs to
model is adopted within this first stage of research, while in upstream PHY frame, as long as this burst is specified in
further analysis, more realistic traffic models should be BWmap transmitted with appropriate downstream PHY frame
considered. [20], M is equal to N. If we assume one half of the bursts within
the frame having PLOAMu flag to be equal to 1, and for one
In view of that, the results of 10G EPON overhead half of T-CONTs within the burst, DBRu flag to be equal to 1
calculation, for different scenarios and different transmission (NT-CONT = 1), it can finally become possible to calculate the
directions, are presented in Fig. 5 and Fig. 6. last part of transmission overhead. The results of above XG-
PON1 overhead calculations are given in Figure 7.

Figure 5. 10G EPON downstream overhead


Figure 7. XG-PON1 downstream and upstream overheads
At the very end of these calculations, the effects of adopted these values have the tendency to equate. However, available
presumptions on final result must also be discussed. As long as users rates are comparable only in downstream direction, and
10G EPON concerns, duration of cyclic period doesnt affect are pretty the same for both technologies, not depending on
much of the results, both in downstream and upstream (making number of ONUs. While in upstream direction, XG-PON1
the improvement of less then 1% in upstream direction, even standard specifies four time smaller rates than 10G EPON, the
under the assumption of 8 ms duration per cyclic period). impact of efficiency differences are negligible in respect to
Moreover, because the same traffic distribution is proposed in these standardized raw bit rate differences.
both 10G EPON and XG-PON1, its influence on comparative
results are negligible. On the other side, assumptions made VI. CONCLUSION
about XG-PON1 overhead calculations dont significantly
affect downstream transmission overhead (which can be The aim of this paper was to compare the 10G EPON and
improved for maximum 1% under the other initial conditions), XG-PON1 technologies, mainly in terms of MAC efficiency,
but they have more effect on upstream transmission overhead relying on their overhead identification. Accordingly, within
(of about 3% difference). However, within this theoretical this paper, all elements which introduce transmission overhead
analysis, it is not possible to test the accuracy of these were estimated. This enabled us to calculate overall efficiency
assumptions, and they must be tested in real network and real available data rate per user, for both technologies.
environments. Accordingly, it is shown that XG-PON1 has slightly improved
efficiency performance, achieved mostly by smaller FEC and
C. Comparation of overhead between 10G EPON and XG- Ethernet encapsulation overhead. Bearing in mind obtained
results, 10G EPON (with 10 Gbps in downstream) is
PON1
competitively superior to XG-PON1, providing almost the
According to results previously gained, it finally became same data rate per user in downstream and four time higher
possible to estimate efficiency and real available data rates per rates in upstream.
user, in both 10G EPON and XG-PON1 networks.
Still, in order to obtain more precise result, including the
If there should be 32 ONUs within each network, 10G effects of FEC correction codes on overall data rates in real
EPON efficiency is around 76% in downstream and 74% in traffic conditions, it is recommended to perform appropriate
upstream direction. Consequently real available rates per user simulation, or even obtain the result in real networks
are 245/238 Mbps. In XG-PON1, these efficiencies are 80% environment. However, according to authors best knowledge,
and 82% respectively, thus giving the rates per user equal to up to this date, no paper dealing with XG-PON1 efficiency and
249/64 Mbps. The comparative results of efficiencies, for 10G EPON and XG-PON1 efficiency comparison is published.
different numbers of ONU units, in downstream and upstream, Thus, even though transmission overheads decrease theoretical
are presented in Fig. 8 and Fig. 9. rates even for almost 40% (for large number of ONUs), they
have great influence on real available data rates and their
estimate is of great interest in the process of network planning.
Finally, this paper represents a contribution to future operators
efforts and it provides an objective insight into a real
performance of rival solutions.

REFERENCES
[1] M. Giuntini, J. Morabito, A. Valenti, F. Matera, V. Carrozzo, S. Di
Bartolo: ''Integration Of Optical Telecommunications and Radio Access
Networks to Assure Quality of Service'', 12th IEEE International
Figure 8. 10G EPON and XG-PON1 downstream efficiencies Conference on Transparent Optical Networks (ICTON), Germany, July
2010
[2] Glen Kramer, Gery Pesavento: ''Ethernet Passive Optical Networks
(EPON): Building a Next-Generation Optical Access Networks'', IEEE
Comunication Magazine, Vol. 40, pp. 66 - 73, Februar 2002
[3] N. Gici, N. Hadiahmetovi, Z. Aljievi: ''Possibilities of FTTH
technology implementation for Telco operators'', Proceedings of VIII
International Symposium on Telecommunications (BIHTEL 2010),
Bosnia and Herzegovina, November 2010
[4] Skaljo, E.; Hodzic, M.; Bektas, I.: ''Migration from G(E)PON to
NGPON'', International Conference on Ultra Modern
Telecommunications & Workshops (ICUMT '09), October 2009
[5] T. Orphanoudakis, H.-C. Leligou, E. Kosmatos, J. D. Angelopoulos:
''Performance evaluation of GPON vs EPON for multi-service access'',
International Journal of Communication Systems, Vol. 22, No. 2, pp.
Figure 9. 10G EPON and XG-PON1 upstream efficiencies 187 - 202, John Wiley and Sons Ltd. Chichester, UK, February 2009
[6] Chen Ling, Stefan Dahlfort, Dave Hood: ''Evolution of PON: 10G-PON
According to Fig. 8 and Fig. 9, it can be concluded that, and WDM-PON'', IEEE Communications and Photonics Conference and
when smaller number of ONU units, XG-PON1 efficiency is Exhibition (ACP), China, December 2010
slightly higher than 10G EPON, but with increase of ONUs,
[7] Rajesh Roy, Marek Hajduczenia, Glen Kramer, and Henrique J. A. da
Silva: ''10G-EPON Efficiency'', 3rd IEEE International Symposium on
Advanced Networks and Telecommunication Systems (ANTS), India,
2009
[8] FlexLight Networks: ''GPON - The Next Big Thing in Optical Access
Networks'', March 2004. godine
[9] E. Daido, T. Inoue, Y. Kawanishi, K. Yamazaki, A. Yoshimura, S. Shiba
S. Kouyama: ''Development of Asymmetric 10G-EPON System'',
Sumito Electric Information and Communications Tecnical Review,
2009
[10] Glen Kramer: ''What is Next for Ethernet PON?'', 5th International
Conference on Optical Internet (COIN 2006), Korea, 2006
[11] S. Kouyama: ''Development of Asymmetric 10G-EPON System'',
Sumito Electric Information and Communications Tecnical Review,
2009
[12] Jain, S., Effenberger, F., Szabo, A., Zhishan Feng, Forcucci, A., Wei
Guo, Yuanqiu Luo, Mapes, R., Yixin Zhang, O'Byrne, V.: ''Worlds First
XG-PON Field Trial'', Jurnal of Lightwave Technology, Vol. 29, No. 4,
pp. 524 - 528, February 2011
[13] Keiji Tanaka, Akira Agata, and Yukio Horiuchi: ''IEEE 802.3av 10G-
EPON Standardization and Its Research and Development Status'',
Journal of Lightwave Technology, Vol. 28, No. 4, February 2010
[14] Steve Gorshe, Jeff Mandin: ''Introduction to IEEE 802.3av 10 Gbps
Ehernet Passive Optical Networks (10G EPON)'', White Paper MC-
Sierra, Inc., No. 1, January 2009
[15] Frank Effenberger: ''The XG-PON System: Cost Effective 10 Gb/s
Access'', Journal of Lightwave technology, Vol. 29., No. 4., February
2011
[16] eljko Popovic: ''Evolution from GPON to Next Generation PONs'',
Ericsson Magazine, February 2010
[17] J. Santos, P. Inacio, J. Gomes, A. Arsenio, J. Pires, M. Freire, P.
Monteiro: ''Improving the Upstream Channel Efficiency of Passive
Optical Networks via Frame Aggregation and Compression'', 7th
Conference on Telecomminications ConfTele, 2009
[18] IEEE Computer Society: ''IEEE standard 802.3av-2009'', 2009
[19] Glen Kramer: ''10G-EPON: Drivers, Challenges and Solutions'', ECOC
2009
[20] ITU-T, G.987.3: ''10-Gigabit-capable passive optical networks (XG-
PON): Transmission convergence (TC) specification'', October 2010
[21] ITU-T, G.987.2: ''10-Gigabit-capable passive optical networks (XG-
PON): Physical media dependent (PMD) layer specifications'', October
2010

You might also like