Passive Optical Networks

Yaakov (J) Stein May 2007

and
Zvika Eitan
 Outline

PON benefits
PON architecture
Fiber optic basics
PON physical layer
PON user plane
PON control plane
PON benefits
 Why fiber ?
todays high datarate networks are all based on optical fiber
the reason is simple (examples for demonstration sake)
twisted copper pair(s)
8 Mbps @ 3 km, 1.5 Mbps @ 5.5 km (ADSL)
1 Gb @ 100 meters (802.3ab)
microwave
70 Mbps @ 30 km (WiMax)
coax
10 Mbps @ 3.6 km (10BROAD36)
30 Mbps @ 30 km (cable modem)
optical fiber
10 Mbps @ 2 km (10BASE-FL)
100 Mbps @ 400m (100BASE-FX)
1 Gbps @ 2km (1000BASE-LX)
10 Gbps @ 40 (80) km (10GBASE-E(Z)R)
40 Gbps @ 700 km [Nortel] or 3000 km [Verizon]
 Aside why is fiber better ?
attenuation per unit length
reasons for energy loss
copper: resistance, skin effect, radiation, coupling
fiber: internal scattering, imperfect total internal reflection
so fiber beats coax by about 2 orders of magnitude
e.g. 10 dB/km for thin coax at 50MHz, 0.15 dB/km =1550nm fiber

noise ingress and cross-talk

copper couples to all nearby conductors
no similar ingress mechanism for fiber

ground-potential, galvanic isolation, lightning protection

copper can be hard to handle and dangerous
no concerns for fiber
 Why not fiber ?
fiber beats all other technologies for speed and reach
but fiber has its own problems
harder to splice, repair, and need to handle carefully
regenerators and even amplifiers are problematic
more expensive to deploy than for copper
digital processing requires electronics
so need to convert back to electronics
we will call the converter an optical transceiver copper fiber
optical transceivers are expensive
switching easier with electronics (but possible with photonics)
so pure fiber networks are topologically limited:

point-to-point

rings
 Access network bottleneck
hard for end users to get high datarates because of the access bottleneck

local area networks

use copper cable
get high datarates over short distances

core networks
use fiber optics
get high datarate over long distances access core
small number of active network elements

access networks (first/last mile)

long distances LAN
so fiber would be the best choice
many network elements and large number of endpoints
if fiber is used then need multiple optical transceivers
so copper is the best choice
this severely limits the datarates
 Fiber To The Curb
Hybrid Fiber Coax and VDSL
switch/transceiver/miniDSLAM located at curb or in basement
need only 2 optical transceivers

but not pure optical solution

lower BW from transceiver to end users
need complex converter in constrained environment

N end users
core feeder fiber

copper

access network
 Fiber To The Premises
we can implement point-to-multipoint topology purely in optics
but we need a fiber (pair) to each end user

requires 2 N optical transceivers

complex and costly to maintain

N end users
core

access network
 An obvious solution
deploy intermediate switches
(active) switch located at curb or in basement
saves space at central office
need 2 N + 2 optical transceivers

N end users
core feeder fiber

fiber

access network
 The PON solution
another alternative - implement point-to-multipoint topology purely in optics
avoid costly optic-electronic conversions
use passive splitters no power needed, unlimited MTBF
only N+1 optical transceivers (minimum possible) !

access network

1:2 passive splitter

N end users
core
typically N=32
max defined 128
feeder fiber

1:4 passive splitter

 PON advantages
shared infrastructure translates to lower cost per customer
minimal number of optical transceivers
feeder fiber and transceiver costs divided by N customers
greenfield per-customer cost similar to UTP

passive splitters translate to lower cost

can be installed anywhere
no power needed
essentially unlimited MTBF

fiber data-rates can be upgraded as technology improves

initially 155 Mbps
then 622 Mbps
now 1.25 Gbps
soon 2.5 Gbps and higher
 PON
architecture
 Terminology
like every other field, PON technology has its own terminology
the CO head-end is called an OLT
ONUs are the CPE devices (sometimes called ONTs in ITU)
the entire fiber tree (incl. feeder, splitters, distribution fibers) is an ODN
all trees emanating from the same OLT form an OAN
downstream is from OLT to ONU (upstream is the opposite direction)

downstream
upstream
NNI
Optical Distribution Network
core Optical Network Units
splitter
Optical Line Terminal UNI

Optical Access Network Terminal Equipment

 PON types
many types of PONs have been defined
APON ATM PON
BPON Broadband PON
GPON Gigabit PON
EPON Ethernet PON
GEPON Gigabit Ethernet PON
CPON CDMA PON
WPON WDM PON
in this course we will focus on GPON and EPON (including GEPON)
with a touch of BPON thrown in for the flavor
 Bibliography
BPON is explained in ITU-T G.983.x
GPON is explained in ITU-T G.984.x
EPON is explained in IEEE 802.3-2005 clauses 64 and 65
(but other 802.3 clauses are also needed)
Warning
do not believe white papers from vendors
especially not with respect to GPON/EPON comparisons

GPON BPON EPON

 PON principles
(almost) all PON types obey the same basic principles
OLT and ONU consist of
Layer 2 (Ethernet MAC, ATM adapter, etc.)
optical transceiver using different s for transmit and receive
optionally: Wavelength Division Multiplexer

downstream transmission
OLT broadcasts data downstream to all ONUs in ODN
ONU captures data destined for its address, discards all other data
encryption needed to ensure privacy

upstream transmission
ONUs share bandwidth using Time Division Multiple Access
OLT manages the ONU timeslots
ranging is performed to determine ONU-OLT propagation time

additional functionality
Physical Layer OAM
Autodiscovery
Dynamic Bandwidth Allocation
 Why a new protocol ?

downstream
PON has a unique architecture
upstream
(broadcast) point-to-multipoint in DS direction

(multiple access) multipoint-to-point in US direction

contrast that with, for example

Ethernet - multipoint-to-multipoint

ATM - point-to-point

This means that existing protocols

do not provide all the needed functionality
e.g. receive filtering, ranging, security, BW allocation
 (multi)point - to - (multi)point

Multipoint-to-multipoint Ethernet avoids collisions

by CSMA/CD
This can't work for multipoint-to-point US PON
since ONUs don't see each other
And the OLT can't arbitrate without adding a roundtrip time

Point-to-point ATM can send data in the open

although trusted intermediate switches see all data
customer switches only receive their own data
This can't work for point-to-multipoint DS PON
since all ONUs see all DS data
 PON encapsulation
The majority of PON traffic is Ethernet

So EPON enthusiasts say

use EPON - it's just Ethernet
That's true by definition -
anything in 802.3 is Ethernet
and EPON is defined in clauses 64 and 65 of 802.3-2005

But don't be fooled - all PON methods encapsulate MAC frames

EPON and GPON differ in the contents of the header
EPON hides the new header inside the GbE preamble
GPON can also carry non-Ethernet payloads

PON header DA SA T data FCS

 BPON history
1995 : 7 operators (BT, FT, NTT, ) and a few vendors form
Full Service Access Network Initiative
to provide business customers with multiservice broadband offering
Obvious choices were ATM (multiservice) and PON (inexpensive)
which when merged became APON
1996 : name changed to BPON to avoid too close association with ATM
1997 : FSAN proposed BPON to ITU SG15
1998 : BPON became G.983
G.982 : PON requirements and definitions
G.983.1 : 155 Mbps BPON
G.983.2 : management and control interface
G.983.3 : WDM for additional services
G.983.4 : DBA
G.983.5 : enhanced survivability
G.983.1 amd 1 : 622 Mbps rate
G.983.1 amd 2 : 1244 Mbps rate

 EPON history
2001: IEEE 802 LMSC WG accepts
Ethernet in the First Mile Project Authorization Request
becomes EFM task force (largest 802 task force ever formed)

EFM task force had 4 tracks

DSL (now in clauses 61, 62, 63)
Ethernet OAM (now clause 57)
Optics (now in clauses 58, 59, 60, 65)
P2MP (now clause 64)

2002 : liaison activity with ITU to agree upon wavelength allocations

2003 : WG ballot

2004 : full standard

2005: new 802.3 version with EFM clauses

 GPON history
2001 : FSAN initiated work on extension of BPON to > 1 Gbps

Although GPON is an extension of BPON technology

and reuses much of G.983 (e.g. linecode, rates, band-plan, OAM)
decision was not to be backward compatible with BPON

2001 : GFP developed (approved 2003)

2003 : GPON became G.984

G.984.1 : GPON general characteristics
G.984.2 : Physical Media Dependent layer
G.984.3 : Transmission Convergence layer
G.984.4 : management and control interface
Fiber optics - basics
 Total Internal Reflection
in Step-Index Multimode Fiber

= sin1 (n2/n1) t = Propagation

Time
t Vacuum: n=1,
V =c/n t=3.336ns/m
t Water : n=1.33, t=4.446ns/m
t = Ln/c
Types of Optical Fiber

Popular Fiber
Sizes

Multimode Graded-
Index Fiber

Single-mode
Fiber
 Optical Loss versus Wavelength

Click to edit Master text styles

Second level
Third level
Fourth level
 Sources of Dispersion

Total
Dispersion
Multimode Chromatic
Dispersion Dispersio
n

Material
Dispersio
n
 Multimode Dispersion

1 0 1 1 1 11

Dispersion limits bandwidth in optical

fiber
 Graded-index Dispersion

1 0 11 1 0 1
 Single-Mode Dispersion

1 0 11 1 0 1

In SM the limit bandwidth is caused by chromatic

dispersion.
 System Design Consideration

How to calculate
For bandwidth?
a 1.25 Gb/s we need a BW of 0.7 BitRate =
1.143ns
Tc = Dmat * *L

For Laser 1550nm Fabry Perot

Tc = (20ps/nm * km) * 5nm * 15km =
1.5ns
For Laser 1550nm DFB
Tc = (20ps/nm * km) * 0.2nm * 60km = 0.24n
Material Dispersion (Dmat)
 Spectral Characteristics

LASER/laser diode: Light Amplification by Stimulated Emission of Radiation. Done of the wide range of
devices that generates light by that principle. Laser light is directional, covers a narrow range of
wavelengths, and is more coherent than ordinary light. Semiconductor diode lasers are the standard light
sources in fiber optic systems. Lasers emit light by stimulated emission .
Laser Optical Power Output vs. Forward Current

Laser
 Light Detectors

PIN DIODES (PD)

- Operation simular to LEDs, but in reverse, photon are converted to electrons
- Simple, relatively low- cost
- Limited in sensitivity and operating range
- Used for lower- speed or short distance applications

AVALANCHE PHOTODIODES (APD)

- Use more complex design and higher operating voltage than PIN diodes
to produce amplification effect
- Significantly more sensitive than PIN diodes
- More complex design increases cost
- Used for long-haul/higher bit rate systems
Wavelength-Division Multiplexing
WDM Duplexing
 Basic Configuration of PON

OLT = Optical Line Termination

ONU = Optical Network Unit
BMCDR = Burst Mode Clock Data Recovery
Typical PON Configuration and Optical Packets
Eye diagram of ONU transceiver
in burst mode operation
Burst-Mode Transmitter in ONU
OLT Burst-Mode Receiver
Burst-Mode CDR
 Sampling

Ideal sampling instant

Hysteresis

Superimposed interference

Ideal, error-free transmission

Transceiver Block Diagram
Optical Splitters
Optical Protection Switch
Optical Splitter
 Budget Calculations

LB = PS - PO
LB = Link Budget
PS = Sensitivity
PO = Output Power

Example: GPON 1310nm

Power: 0dbm Single-mode fiber
Sensitivity: -23dbm } Link Budget: 23db
 Typical Range Calculation

Assume:
Optical loss = 0.35 db/km
Connector Loss = 2dB Range Budget: ~11Km
Splitter Insertion Loss 1X32 = 17dB
Relationship between transmission distance
and number of splits
GbE Fiber Optic Characteristics
PON physical layer
 allocations - G.983.1
Upstream and downstream directions need about the same bandwidth
US serves N customers, so it needs N times the BW of each customer
but each customer can only transmit 1/N of the time

In APON and early BPON work it was decided that 100 nm was needed

Where should these bands be placed for best results?

In the second and third windows !

Upstream 1260 - 1360 nm (1310 50) second window

Downstream 1480 - 1580 nm (1530 50) third window

US DS
1200 nm 1300 nm 1400 nm 1500 nm 1600 nm
 allocations - G.983.3
Afterwards it became clear that there was a need for additional DS bands
Pressing needs were broadcast video and data
Where could these new DS bands be placed ?

At about the same time G.694.2 defined 20 nm CWDM bands

these were made possible because of new inexpensive hardware
(uncooled Distributed Feedback Lasers)

One of the CWDM bands was 1490 10 nm 1270 1490 1630

same bottom as the G.983.1 DS

So it was decided to use this band as the G.983.3 DS

and leave the US unchanged
guard
available
US DS

1200 nm 1300 nm 1400 nm 1500 nm 1600 nm

 allocations - final
US DS
1200 nm 1300 nm 1400 nm 1500 nm 1600 nm

The G.983.3 band-plan was incorporated into GPON

and via liaison activity into EPON
and is now the universally accepted xPON band-plan

US 1260-1360 nm (1310 50)

DS 1480-1500 nm (1490 10)
enhancement bands:
video 1550 - 1560 nm (see ITU-T J.185/J.186)
digital 1539-1565 nm
 Data rates (for now )
PON DS (Mbps) US (Mbps)
BPON 155.52 155.52
622.08 155.52
Amd 1
622.08 622.08
Amd 2 1244.16 155.52
1244.16 622.08
1244.16 155.52
1244.16 622.08
GPON 1244.16 1244.16
2488.32 155.52
2488.32 622.08
2488.32 1244.16
2488.32 2488.32
EPON 1250*
* only 1G/10G usable due to1250*
linecode

10GEPONwork
in10312.5*
progress 10312.5*
 Reach and splits
Reach and the number of ONUs supported are contradictory design goals

In addition to physical reach derived from optical budget

there is logical reach limited by protocol concerns (e.g. ranging protocol)
and differential reach (distance between nearest and farthest ONUs)

The number of ONUs supported depends not only on the number of splits
but also on the addressing scheme

BPON called for 20 km and 32-64 ONUs

GPON allows 64-128 splits and the reach is usually 20 km

but there is a low-cost 10 km mode (using Fabry-Perot laser diodes in ONUs)
and a long physical reach 60 km mode with 20 km differential reach

EPON allows 16-256 splits (originally designed for link budget of 24 dB, but now 30 dB)
and has 10 km and 20 km Physical Media Dependent sublayers
 Line codes
BPON and GPON use a simple NRZ linecode (high is 1 and low is 0)
An I.432-style scrambling operation is applied to payload (not to PON overhead)
Preferable to conventional scrambler because no error propagation
each standard and each direction use different LFSRs
LFSR initialized with all ones
LFSR sequence is XOR'ed with data before transmission

EPON uses the 802.3z (1000BASE-X) line code - 8B/10B

Every 8 data bits are converted into 10 bits before transmission
DC removal and timing recovery ensured by mapping
Special function codes (e.g. idle, start_of_packet, end_of_packet, etc)
However, 1000 Mbps is expanded to 1250 Mbps
10GbE uses a different linecode - 64B/66B
 FEC
G984.3 clause 13 and 802.3-2005 subclause 65.2.3
define an optional G.709-style Reed-Solomon code

Use (255,239,8) systematic RS code designed for submarine fiber (G.975)

to every 239 data bytes add 16 parity bytes to make 255 byte FEC block

Up to 8 byte errors can be corrected

Improves power budget by over 3 dB,
allowing increased reach or additional splits

Use of FEC is negotiated between OLT and ONU

Since code is systematic
can use in environment where some ONUs do not support FEC

In GPON FEC frames are aligned with PON frames

In EPON FEC frames are marked using K-codes
(and need 8B10B decode - FEC - 8B10B encode)
 More physical layer problems

Near-far problem
OLT needs to know signal strength to set decision threshold
If large distance between near/far ONUs, then very different attenuations
If radically different received signal strength can't use a single threshold
EPON: measure received power of ONU at beginning of burst
GPON: OLT feedback to ONUs to properly set transmit power

Burst laser problem

Spontaneous emission noise from nearby ONU lasers causes interference
Electrically shut ONU laser off when not transmitting
But lasers have long warm-up time
and ONU lasers must stabilize quickly after being turned on
 US timing diagram
How does the ONU US transmission appear to the OLT ?

grant grant
inter-ONU
guard
data data
lock

lock
laser laser laser laser
turn-on turn-off turn-on turn-off

Notes:
GPON - ONU reports turn-on and turn-off times to OLT
ONU preamble length set by OLT
EPON - long lock time as need to Automatic Gain Control and Clock/Data Recovery
long inter-ONU guard due to AGC-reset
Ethernet preamble is part of data
PON User plane
 How does it work?
ONU stores client data in large buffers (ingress queues)
ONU sends a high-speed burst upon receiving a grant/allocation
Ranging must be performed for ONU to transmit at the right time
DBA - OLT allocates BW according to ONU queue levels
OLT identifies ONU traffic by label
OLT extracts traffic units and passes to network

OLT receives traffic from network and encapsulates into PON frames
OLT prefixes with ONU label and broadcasts
ONU receives all packets and filters according to label
ONU extracts traffic units and passes to client
 Labels
In an ODN there is 1 OLT, but many ONUs
ONUs must somehow be labeled for
OLT to identify the destination ONU
ONU to identify itself as the source
EPON assigns a single label Logical Link ID to each ONU (15b)
GPON has several levels of labels
ONU_ID (1B) (1B)
Transmission-CONTainer (AKA Alloc_ID) (12b) (can be >1 T-CONT per ONU)
For ATM mode

VPI VC

VCI VC
VC
ONU T-CONT VP VC
For GEM mode PON VP

Port_ID (12b) (12b)
ONU T-CONT Port
Port
 DS GPON format
GPON Transmission Convergence frames are always 125 sec long
19440 bytes / frame for 1244.16 rate
38880 bytes / frame for 2488.32 rate
Each GTC frame consists of Physical Control Block downstream + payload
PCBd contains sync, OAM, DBA info, etc.
payload may have ATM and GEM partitions (either one or both)

GTC frame scrambled 125 sec

PCBd payload PCBd payload PCBd payload

PSync (4B) Ident (4B) PLOAMd (13B) BIP (1B) ATM GEM
partition partition
PLend (4B) PLend (4B) US BW map (N*8B)
 GPON payloads
GTC payload potentially has 2 sections:
ATM partition (Alen * 53 bytes in length)
GEM partition (now preferred method)
PCBd ATM cell ATM cell ATM cell GEM frame GEM frame GEM frame
ATM partition
Alen (12 bits) is specified in the PCBd
Alen specifies the number of 53B cells in the ATM partition
if Alen=0 then no ATM partition
if Alen=payload length / 53 then no GEM partition
ATM cells are aligned to GTC frame
ONUs accept ATM cells based on VPI in ATM header

GEM partition
Unlike ATM cells, GEM delineated frames may have any length
Any number of GEM frames may be contained in the GEM partition
ONUs accept GEM frames based on 12b Port-ID in GEM header
 GPON Encapsulation Mode
A common complaint against BPON was inefficiency due to ATM cell tax
GEM is similar to ATM
constant-size HEC-protected header
but avoids large overhead by allowing variable length frames
GEM is generic any packet type (and even TDM) supported
GEM supports fragmentation and reassembly
GEM is based on GFP, and the header contains the following fields:
Payload Length Indicator - payload length in Bytes
Port ID - identifies the target ONU
Payload Type Indicator (GEM OAM, congestion/fragmentation indication)
Header Error Correction field (BCH(39,12,2) code+ 1b even parity)
The GEM header is XOR'ed with B6AB31E055 before transmission

PLI Port ID PTI HEC payload fragment

(12b) (12b) (3b) (13b) (L Bytes)
5B
 Ethernet / TDM over GEM
When transporting Ethernet traffic over GEM:
only MAC frame is encapsulated (no preamble, SFD, EFD)
MAC frame may be fragmented (see next slide)

Ethernet over GEM

PLI ID PTI HEC DA SA T data FCS

When transporting TDM traffic over GEM:

TDM input buffer polled every 125 sec.
PLI bytes of TDM are inserted into payload field
length of TDM fragment may vary by 1 Byte due to frequency offset
round-trip latency bounded by 3 msec.
TDM over GEM
PLI ID PTI HEC PLI Bytes of TDM
 GEM fragmentation
GEM can fragment its payload
For example
unfragmented Ethernet frame
PLI ID PTI=001 HEC DA SA T data FCS
fragmented Ethernet frame
PLI ID PTI=000 HEC DA SA T data1

PLI ID PTI=001 HEC data2 FCS

GEM fragments payloads for either of two reasons:

GEM frame may not straddle GTC frame
PCBd ATM partition GEM frame GEM frag 1 PCBd ATM partition GEM frag 2 GEM frame

GEM frame may be pre-empted for delay-sensitive data

PCBdATM partition urgent frame large frag 1 PCBd ATM partition urgent frame large frag 2
 PCBd
We saw that the PCBd is

PSync Ident PLOAMd BIP PLend PLend US BW map

(4B) (4B) (13B) (1B) (4B) (4B) (N*8B)
B6AB31E0
PSync - fixed pattern used by ONU to located start of GTC frame
Ident - MSB indicates if FEC is used, 30 LSBs are superframe counter
PLOAMd - carries OAM, ranging, alerts, activation messages, etc.
BIP - SONET/SDH-style Bit Interleaved Parity of all bytes since last BIP
PLend (transmitted twice for robustness) -
Blen - 12 MSB are length of BW map in units of 8 Bytes
Alen - Next 12 bits are length of ATM partition in cells
CRC - final 8 bits are CRC over Blen and Alen
US BW map - array of Blen 8B structures granting BW to US flow
will discuss later (DBA)
 GPON US considerations
GTC fames are still 125 sec long, but shared amongst ONUs

Each ONU transmits a burst of data

using timing acquired by locking onto OLT signal
according to time allocation sent by OLT in BWmap

there may be multiple allocations to single ONU

OLT computes DBA by monitoring traffic status (buffers)
of ONUs and knowing priorities
at power level requested by OLT (3 levels)

this enables OLT to use avalanche photodiodes which are
sensitive to high power bursts
leaving a guard time from previous ONU's transmission
prefixing a preamble to enable OLT to acquire power and phase
identifying itself (ONU-ID) in addition to traffic IDs (VPI, Port-ID)
scrambling data (but not preamble/delimiter)
 US GPON format
4 different US overhead types:
Physical Layer Overhead upstream
always sent by ONU when taking over from another ONU
contains preamble and delimiter (lengths set by OLT in PLOAMd)
BIP (1B), ONU-ID (1B), and Indication of real-time status (1B)
PLOAM upstream (13B) - messaging with PLOAMd
Power Levelling Sequence upstream (120B)
used during power-set and power-change to help set ONU
power so that OLT sees similar power from all ONUs
Dynamic Bandwidth Report upstream
sends traffic status to OLT in order to enable DBA computation

if all OH types are present:

PLOu PLOAMd PLSu DBRu payload
 US allocation example
DS frame
PCBd payload

BWmap Alloc-ID SStart SStop Alloc-ID SStart Sstop Alloc-ID SStart SStop

US frame

preamble guard scrambled

+ time
delimiter

BWmap sent by OLT to ONUs is a list of

ONU allocation IDs
flags (not shown above) tell if use FEC, which US OHs to use, etc.
start and stop times (16b fields, in Bytes from beginning of US frame)
 EPON format
EPON operation is based on the Ethernet MAC
and EPON frames are based on GbE frames
but extensions are needed
clause 64 - MultiPoint Control Protocol PDUs
this is the control protocol implementing the required logic
clause 65 - point-to-point emulation (reconciliation)
this makes the EPON look like a point-to-point link

and EPON MACs have some special constraints

instead of CSMA/CD they transmit when granted

time through MAC stack must be constant ( 16 bit durations)

accurate local time must be maintained

 EPON header
Standard Ethernet starts with an essentially content-free 8B preamble
7B of alternating ones and zeros 10101010
1B of SFD 10101011

In order to hide the new PON header

EPON overwrites some of the preamble bytes
10101010 10101010 10101010 10101010 10101010 10101010 10101010 10101011

10101010 10101010 10101011 10101010 10101010 LLID LLID CRC

LLID field contains

MODE (1b)

always 0 for ONU

0 for OLT unicast, 1 for OLT multicast/broadcast
actual Logical Link ID (15b)

Identifies registered ONUs
7FFF for broadcast

CRC protects from SLD (byte 3) through LLID (byte 7)

 MPC PDU format
MultiPoint Control Protocol frames are untagged MAC frames
with the same format as PAUSE frames

DA SA L/T Opcode timestamp data / RES / pad FCS

Ethertype = 8808
Opcodes (2B) - presently defined:
GATE/REPORT/REGISTER_REQ/REGISTER/REGISTER_ACK
Timestamp is 32b, 16 ns resolution
conveys the sender's time at time of MPCPDU transmission
Data field is needed for some messages
 Security
DS traffic is broadcast to all ONUs, so encryption is essential
easy for a malicious user to reprogram ONU to capture desired frames
US traffic not seen by other ONUs, so encryption is not needed
do not take fiber-tappers into account

EPON does not provide any standard encryption method

can supplement with IPsec or MACsec
many vendors have added proprietary AES-based mechanisms
in China special China Telecom encryption algorithm

BPON used a mechanism called churning

Churning was a low cost hardware solution (24b key)
with several security flaws
engine was linear - simple known-text attack
24b key turned out to be derivable in 512 tries
So G.983.3 added AES support - now used in GPON
 GPON encryption
OLT encrypts using AES-128 in counter mode
Only payload is encrypted (not ATM or GEM headers)
Encryption blocks aligned to GTC frame
Counter is shared by OLT and all ONUs
46b = 16b intra-frame + 30 bits inter-frame
intra-frame counter increments every 4 data bytes

reset to zero at beginning of DS GTC frame

OLT and each ONU must agree on a unique symmetric key

OLT asks ONU for a password (in PLOAMd)
ONU sends password US in the clear (in PLOAMu)
key sent 3 times for robustness
OLT informs ONU of precise time to start using new key
 QoS - EPON
Many PON applications require high QoS (e.g. IPTV)

EPON leaves QoS to higher layers

VLAN tags
P bits or DiffServ DSCP

In addition, there is a crucial difference between LLID and Port-ID

there is always 1 LLID per ONU
there is 1 Port-ID per input port - there may be many per ONU
this makes port-based QoS simple to implement at PON layer

RT EF BE
GPON
 QoS - GPON
GPON treats QoS explicitly
constant length frames facilitate QoS for time-sensitive applications
5 types of Transmission CONTainers

type 1 - fixed BW

type 2 - assured BW

type 3 - allocated BW + non-assured BW
type 4 - best effort

type 5 - superset of all of the above

GEM adds several PON-layer QoS features

fragmentation enables pre-emption of large low-priority frames
PLI - explicit packet length can be used by queuing algorithms
PTI bits carry congestion indications
PON control plane
 Principles
GPON uses PLOAMd and PLOAMu as control channel
PLOAM are incorporated in regular (data-carrying) frames
Standard ITU control mechanism

EPON uses MPCP PDUs

Standard IEEE control mechanism
EPON control model - OLT is master, ONU is slave
OLT sends GATE PDUs DS to ONU
ONU sends REPORT PDUs US to OLT
 Ranging

Upstream traffic is TDMA

Were all ONUs equidistant, and were all to have a common clock
then each would simply transmit in its assigned timeslot
But otherwise the signals will overlap

To eliminate overlap
guard times left between timeslots
each ONU transmits with the proper delay to avoid overlap
delay computed during a ranging process
 Ranging background
In order for the ONU to transmit at the correct time
the delay between ONU transmission and OLT reception
needs to be known (explicitly or implicitly)
Need to assign an equalization-delay

The more accurately it is known

the smaller the guard time that needs to be left
and thus the higher the efficiency

Assumptions behind the ranging methods used:

can not assume US delay is equal to DS delay

delays are not constant

due to temperature changes and component aging
GPON: ONUs not time synchronized accurately enough
EPON: ONUs are accurately time synchronized (std contains jitter masks)
with time offset by OLT-ONU propagation time
 GPON ranging method
Two types of ranging
initial ranging

only performed at ONU boot-up or upon ONU discovery

must be performed before ONU transmits first time
continuous ranging
performed continuously to compensate for delay changes
OLT initiates coarse ranging by stopping allocations to all other ONUs
thus when new ONU transmits, it will be in the clear
OLT instructs the new ONU to transmit (via PLOAMd)
OLT measures phase of ONU burst in GTC frame
OLT sends equalization delay to ONU (in PLOAMd)
During normal operation OLT monitors ONU burst phase
If drift is detected OLT sends new equalization delay to ONU (in PLOAMd)
 EPON ranging method
All ONUs are synchronized to absolute time (wall-clock)
When an ONU receives an MPCPDU from OLT
it sets its clock according to the OLT's timestamp
When the OLT receives an MPCPDU in response to its MPCPDU
it computes a "round-trip time" RTT (without handling times)
it informs the ONU of RTT, which is used to compute transmit delay

OLT sends MPCPDU ONU receives MPCPDU ONU sends MPCPDU OLT receives MPCPDU
Timestamp = T0 Sets clock to T0 Timestamp = T1 RTT = T2 - T1

time
OLT time T0 T2
ONU time T0 T1
RTT = (T2-T0) - (T1-T0) = T2-T1
OLT compensates all grants by RTT before sending
Either ONU or OLT can detect that timestamp drift exceeds threshold
 Autodiscovery
OLT needs to know with which ONUs it is communicating
This can be established via NMS
but even then need to setup physical layer parameters

PONs employ autodiscovery mechanism to automate

discovery of existence of ONU
acquisition of identity
allocation of identifier
acquisition of ONU capabilities
measure physical layer parameters
agree on parameters (e.g. watchdog timers)

Autodiscovery procedures are complex (and uninteresting)

so we will only mention highlights
 GPON autodiscovery
Every ONU has an 8B serial number (4B vendor code + 4B SN)
SN of ONUs in OAN may be configured by NMS, or
SN may be learnt from ONU in discovery phase
ONU activation may be triggered by
Operator command
Periodic polling by OLT
OLT searching for previously operational ONU
G.984.3 differentiates between three cases:
cold PON / cold ONU
warm PON / cold ONU
warm PON / warm ONU
Main steps in procedure:
ONU sets power based on DS message
OLT sends a Serial_Number request to all unregistered ONUs
ONU responds
OLT assigns 1B ONU-ID and sends to ONU
ranging is performed
ONU is operational
 EPON autodiscovery
OLT periodically transmits DISCOVERY GATE messages
ONU waits for DISCOVERY GATE to be broadcast by OLT
DISCOVERY GATE message defines discovery window

start time and duration
ONU transmits REGISTER_REQ PDU using random offset in window
OLT receives request
registers ONU

assigns LLID

bonds MAC to LLID

performs ranging computation
OLT sends REGISTER to ONU
OLT sends standard GATE to ONU
ONU responds with REGISTER_ACK
ONU goes into operational mode - waits for grants
 Failure recovery
PONs must be able to handle various failure states

GPON
if ONU detects LOS or LOF it goes into POPUP state

it stops sending traffic US

OLT detects LOS for ONU
if there is a pre-ranged backup fiber then switch-over

EPON
during normal operation ONU REPORTs reset OLT's watchdog timer
similarly, OLT must send GATES periodically (even if empty ones)
if OLT's watchdog timer for ONU times out

ONU is deregistered
 Dynamic Bandwidth Allocation
MANs and WANs have relatively stationary BW requirements
due to aggregation of large number of sources
But each ONU in a PON may serve only 1 or a small number of users
So BW required is highly variable
It would be inefficient to statically assign the same BW to each ONU
So PONs assign dynamically BW according to need
The need can be discovered
by passively observing the traffic from the ONU
by ONU sending reports as to state of its ingress queues
The goals of a Dynamic Bandwidth Allocation algorithm are
maximum fiber BW utilization
fairness and respect of priority
minimum delay introduced
 GPON DBA
DBA is at the T-CONT level, not port or VC/VP
GPON can use traffic monitoring (passive) or status reporting (active)
There are three different status reporting methods
status in PLOu - one bit for each T-CONT type
piggy-back reports in DBRu - 3 different formats:
quantity of data waiting in buffers,
separation of data with peak and sustained rate tokens
nonlinear coding of data according to T-CONT type and tokens
ONU report in DBA payload - select T-CONT states
OLT may use any DBA algorithm
OLT sends allocations in US BW map
 EPON DBA
OLT sends GATE messages to ONUs
GATE message
DA SA 8808 Opcode=0002 timestamp Ngrants/flags grants
flags include DISCOVERY and Force_Report
Force_Report tells the ONU to issue a report

REPORT message
DA SA 8808 Opcode=0003 timestamp Nqueue_sets Reports

Reports represent the length of each queue at time of report

OLT may use any algorithm to decide how to send the following grants