See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/3242731

Development of GPON Upstream Physical-Media-Dependent Prototypes

Article  in  Journal of Lightwave Technology · December 2004

DOI: 10.1109/JLT.2004.836767 · Source: IEEE Xplore

CITATIONS READS
62 1,606

8 authors, including:

Xing-Zhi Qiu Peter Ossieur

Ghent University Ghent University
150 PUBLICATIONS   1,211 CITATIONS    142 PUBLICATIONS   1,200 CITATIONS   

SEE PROFILE SEE PROFILE

Johan Bauwelinck Berry de Vos

Ghent University Vrije Universiteit Amsterdam
315 PUBLICATIONS   2,386 CITATIONS    15 PUBLICATIONS   155 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

ATTO: Ultra High Capacity Wireless Networking View project

Burst-Mode Receiver View project

All content following this page was uploaded by Johan Bauwelinck on 31 January 2017.

The user has requested enhancement of the downloaded file.

2498 JOURNAL OF LIGHTWAVE TECHNOLOGY, VOL. 22, NO. 11, NOVEMBER 2004

Development of GPON Upstream

Physical-Media-Dependent Prototypes
Xing-Zhi Qiu, Member, IEEE, Peter Ossieur, Student Member, IEEE, Johan Bauwelinck, Student Member, IEEE,
Yanchun Yi, Dieter Verhulst, Student Member, IEEE, Jan Vandewege, Member, IEEE, Benoit De Vos, and Paolo Solina

Abstract—This paper presents three new gigabit-capable pas- services with guaranteed quality of service (QoS) and with very
sive optical network (GPON) physical-media-dependent (PMD) high bandwidth and transport efficiency [6], [7].
prototypes: a burst-mode optical transmitter, an avalanche pho- Fig. 1 illustrates the GPON access system. A continuous
todiode/transimpedance amplifier (APD-TIA), and a burst-mode
optical receiver. With these, point-to-multipoint (P2MP) upstream downlink in the wavelength band of 1480–1500 nm carries
transmission can be realized in a high-performance GPON at 1.25 Gb/s time-division-multiplexed (TDM) data from a single
1.25 Gb/s. Performance measurements on the new burst-mode optical line terminator (OLT) toward multiple optical network
upstream PMD modules comply with GPON uplink simulations. units (ONUs) or optical network terminations (ONTs). A
The laser transmitter can quickly set and stabilize the launched burst-mode link in the 1310-nm window collects all ONU/ONT
optical power level over a wide temperature range with better
than 1-dB accuracy. A burst-mode receiver sensitivity of 32.8 upstream traffic toward the OLT as variable-length packets at a
dBm (BER = 10 10 ) is measured, combined with a dynamic 1.25-Gb/s aggregate rate, in a P2MP time-division multiple-ac-
range of 23 dB at a fixed APD avalanche gain of 6. Full compliance cess (TDMA) scheme. The paper focuses on this uplink, which
is achieved with the recently approved ITU-T Recommendation is difficult to design due to the bursty nature of the multitalker
G.984.2 supporting an innovative overall power-leveling mecha- traffic.
nism.
Recently, the FSAN study group, a forum for the world’s
Index Terms—Burst-mode, gigabit-capable passive optical net- leading telecommunications service providers and equipment
work (GPON), optical access network, physical-media-dependent suppliers to work towards a common goal of truly broad-band
layer, receiver, transmitter.
access networks, initiated GPON network standardization via
recommendations for the GPON physical-media-dependent
I. INTRODUCTION (PMD) layer and the transmission convergence (TC) layer.
Both have now been approved by the International Telecommu-
T HE passive optical network (PON) technology is based
on a passive star fiber network and offers a cost-effective
optical access solution with point-to-multipoint (P2MP) nature.
nication Union—Telecommunication Standardization Sector
(ITU-T) and ratified as ITU-T Recommendation G.984.2 [8]
With rapidly growing customer bandwidth requirements and and G.984.3 [9], respectively.
proliferation of bandwidth in metro networks, broad-band The paper presents an overview of the technology require-
passive optical networks (BPONs) [1]–[4] and the emerging ments and specifications of the key GPON PMD building blocks
gigabit-capable passive optical networks (GPONs) are expected in Section II. Section III illustrates the design of a generic burst-
to prevail as the leading optical access technology eliminating mode optical transmitter, followed by the design of a high per-
the bandwidth bottleneck in the last mile. The full-services formance dc-coupled burst-mode optical receiver in Section IV.
access networks (FSAN) GPON can provide high-band- The back-to-back GPON uplink modeling and its results are de-
width services to customers following different fiber-to-the- scribed in Section V. Finally, uplink burst-mode experiments are
premises/cabinet/building/home/user (FTTx) scenarios. presented in Section VI.
A symmetric 1.25-Gb/s GPON system optimized for vari-
able-length packet transmission is currently under development II. FSAN GPON PMD PROTOTYPES
within the framework of the European IST project GIANT (gi- A. Burst-Mode Upstream PMD Building Blocks
gapon access network) [5]. GIANT will demonstrate efficient
Fig. 2 depicts the GPON physical layer as a set of PMD
gigabit transport for “triple play” suites of voice, video, and data
building blocks. The 1.25-Gb/s upstream transmitter (US-TX)
mainly contains the burst-mode laser diode driver (BM-LDD),
Manuscript received December 15, 2003; revised June 15, 2004. The burst- while the upstream receiver (US-RX) comprises the avalanche
mode chip design work was supported in part by the Flemish Government under
Research Contract 010019 IWT Sympathi and in part by Alcatel Bell and STMi- photodiode/transimpedance amplifier (APD-TIA) and the burst-
croelectronics. The uplink building blocks simulations, development, and inte- mode receiver (BM-RX). Table I lists the GPON PMD layer
gration were supported by in part by the European Commission under Research Class B key specifications in the upstream direction as defined in
Contract IST-2001-34523 GIANT and in part by Alcatel Bell.
X.-Z. Qiu, P. Ossieur, J. Bauwelinck, Y. Yi, D. Verhulst, and J. Vandewege ITU-T Recommendation G984.2. The optical distribution net-
are with the Department of INTEC, Ghent University, B-9000 Gent, Belgium work (ODN) consists of passive optical elements such as split-
(e-mail: [email protected]). ters, fibers, connectors, and splices forming an optical path.
B. De Vos is with Alcatel Bell, B-2018 Antwerp, Belgium.
P. Solina is with Telecom Italia Laboratory, Turin 10148, Italy. Three classes (Class A, B, C) are specified with a different ODN
Digital Object Identifier 10.1109/JLT.2004.836767 attenuation range of 5–20, 10–25, 15–30 dB, respectively. No
0733-8724/04$20.00 © 2004 IEEE
QIU et al.: DEVELOPMENT OF GPON UPSTREAM PMD PROTOTYPES 2499

Fig. 1. GPON network architecture for FTTx scenarios. From a single OLT at an access node, it connects a maximum of 32 ONUs/ONTs at the customer’s
premises via shared media of the ODN, which mainly contains a maximum of 20-km fiber and one or more passive optical splitters.

Fig. 2. GPON PMD functional building blocks consist of a downstream transmitter (DS-TX) and an upstream receiver (US-RX) at the OLT; a downstream receiver
(DS-RX) and an upstream transmitter (US-TX) at the ONT. The US-TX contains a laser diode and a burst-mode laser diode driver (BM-LDD), and the US-RX
contains an avalanche photodiode/transimpedance amplifer (APD-TIA) and a burst-mode receiver (BM-RX). A burst-mode clock-phase alignment (BM-CPA) is
also developed, whose detailed design is not included in this paper.

GPON 1.25-Gb/s upstream chip set supporting Class B ODN is splitting factor and long reach) but also a wide dynamic range
available on the open market at the time of this writing. (long/short optical network paths and different splitting fac-
After extensive research, three burst-mode chips were spec- tors). These requirements are aggravated by the combined
ified to be designed in 0.35- m SiGe BiCMOS technology: tolerances on all-optical and elctrooptic (EO) components.
a 1244.16-Mb/s BM-LDD with fast and accurate digital auto- As the combined 7 28 dBm is a very demanding speci-
matic power control (APC), a high sensitivity, and wide dy- fication for 1244-Mb/s burst-mode operation, an innovative
namic range APD-TIA receiver front end, and a burst-mode overall power-leveling mechanism (PLM) was proposed for
receiver (BM-RX) chip for fast but accurate signal recovery. standardization [1], [10]. The PLM was adopted in the ITU-T
This GPON burst-mode chip set was designed and tested suc- G984.2 as an optional PMD-layer implementation. Today, a
cessfully at the INTEC (information technology) Department p-i-n photodiode-based RX at the OLT can obtain a minimum
of Ghent University, Gent, Belgium, within the Flemish IWT RX sensitivity of 24 dBm, which is suitable for Class A
project SYMPATHI (symmetrical PON at high bit rate) [10]. operation only, with 5–20-dB ODN loss. Therefore, the OLT
Currently 1244-Mb/s burst-mode US-TXs and the US-RXs are RX requirements in Table I dictate the use of an APD at 1244
integrated into the GIANT GPON laboratory demonstrator [11]. Mb/s for Class B operation (10–25-dB ODN loss), to reach a
This research contributed to the FSAN efforts toward ITU-T minimum RX sensitivity of 28 dBm.
standardization via Alcatel Bell and Telecom Italia Lab, and was Although an increase of the avalanche gain or multiplica-
perfectly in line with the ITU-T GPON standardization progress tion factor of the APD by proper biasing can improve the
[12]. RX sensitivity, overload figures may deteriorate as strong sig-
nals result in duty-cycle distortion or saturation of the APD-TIA
B. Overall Power Leveling Mechanism output. Optimizing the APD factor for achieving both high
Table I specifies a minimum OLT RX overload of 7 dBm sensitivity and wide dynamic range is not straightforward. Most
combined with a minimum RX sensitivity of 28 dBm. A commercial APD-TIA modules are designed for point-to-point
flexible network deployment requires high sensitivity (high (P2P) transmission in continuous-wave (CW) operation. High-
2500 JOURNAL OF LIGHTWAVE TECHNOLOGY, VOL. 22, NO. 11, NOVEMBER 2004

TABLE I
KEY PMD PARAMETERS OF GPON CLASS B 1244-Mb/s UPSTREAM

Fig. 3. In the worst case, the dynamic range requirement for the US-RX can
be as high as 21 dB. By implementing the PLM, the input level of the strong
packet P1 can be decreased by X dB (X = 3–6), and the input level of
the weak packet P2 can be increased by Y Y
dB ( = 3–6). Overall, PLM
alleviates overload of the US-RX caused by the strong packet P1 and reduces
the burst-mode sensitivity penalty for the weak packet P2 caused by the “tail”
of the strong packet P1 as the guard time is limited to 4 B (see Section IV
for the details). The PLM can improve the uplink burst bit-error-rate (BBER)
performance, especially in the worst case shown in this figure.

sensitivity figures are obtained at a higher , while range. When an ONU receives the message to change from one
overload is mostly defined with a smaller , or improved by a mode to another, it sets its emitted power within the range of
slow automatic gain control (AGC) loop. The fast succession of the new mode and then resumes sending upstream data. (For
upstream bursts, however, separated by a short guard time of 4 the detailed PLM procedure, one can refer to G984.2, App. II.)
B (25.6 ns at 1244 Mb/s), do not allow for a change of APD gain A first benefit of the proposed PLM is a reduction of the dy-
in between bursts. It requires quite some time for stabilizing the namic range requirement at the OLT receiver with 5–6 dB (from
APD gain after adjustment. Moreover, no slow AGC is possible, 21 to 15 dB) as shown in Fig. 3. Another advantage is that it in-
as a sudden overload of the TIA could happen when a high APD creases the laser lifetime and reduces the power consumption of
gain is set and a nearby ONU (strong packet P1) starts talking. ONUs working in mode 1 and/or mode 2. It also reduces pos-
Nonlinearity would cause a “tail” effect and have an impact on sibly strong optical reflections due to nearby ONUs.
the receiver sensitivity, in case the strong packet is followed by
a weak packet (P2) emitted from a far-end ONU (Fig. 3). There- III. DESIGN OF A GPON ONU US-TX
fore, the dynamic range specification of 21 dB must be achieved
at a fixed APD gain and without a slow AGC loop. Symmetric A. US-TX Requirements
gain clamping, however, can be used to extend the RX dynamic Where performance requirements dominate for the OLT-RX,
range. a single RX serving all ONUs, the US-TX located in each sub-
To further relax the 21-dB dynamic range specification of scriber ONU is very important in terms of system cost and com-
the OLT receiver, the transmit power level of the ONUs ex- patibility. The US-TX must provide gated laser bias and drive
periencing a low ODN loss should be reduced. A PLM was currents, programmable between 1 and 160 mA in total, over
conceived operating each ONU in three discrete output power a temperature range of 40 to 85 C, as required for outdoor
modes, with the following mean launched power: 1) normal- operation. The currents must be digitally set to avoid trimming
mode minimum and maximum 2 3 dBm, as stated in elements on the printed circuit board (PCB), reducing calibra-
Table I; 2) mode 1 normal 3 dB; and 3) mode 2 tion costs. A dc-coupled interface between the driver and the
normal 6 dB. The PLM mode can be set locally via a serial laser diode is needed. As the maximum TX enable and disable
peripheral interface (SPI). time is limited to 16 b, or 12.8 ns at 1244 Mb/s, (specified in
The PLM implementation also requires functionalities ITU-T Recommendation G.984.2), whereas the launched data
belonging to the TC layer, such as the ONU capability to after scrambling may have 72 consecutive identical digits, no
increase/reduce the transmit power on the basis of downstream single time constant for an ac-coupled circuit can meet both
messages sent by the OLT. The US-RX at the OLT measures specifications. Moreover, ac-coupling would require heavy line
the received average power and compares it with two threshold coding of the data and a much longer guard time in between
voltages ( and ). The OLT then decides whether the bursts. In contrast to GPON, Ethernet PON (EPON) systems do
incoming optical signal is too low or too high or within the use ac coupling at the expense of more than a 20% throughput
QIU et al.: DEVELOPMENT OF GPON UPSTREAM PMD PROTOTYPES 2501

Fig. 4. Simplified functional block diagram of the GPON ONU US-TX and its interfaces to the network termination (NT) digital application-specified integrated
circuit (ASIC).

decrease due to 8b10b line coding and of considerable loss in laser characteristics, their temperature dependence, and the
transport efficiency and interactivity due to an extended inter- nonlinear relationship between laser current and optical power,
packet gap (IPG). both the laser modulation current and the bias current must be
The multiple upstream access requires laser power level stabi- regulated according to the optical power and extinction-ratio
lization within the short time slots allocated to a specific ONU: requirements specified in Table I.
a US-TX must not send upstream light in timing windows allo- The laser driver has two differential pairs, each powered by
cated to other ONUs, as this would disturb the upstream traffic a current source and a current digital–analog converter (IDAC),
of operational services. The time needed to level a US-TX must for independent setting of the laser bias current I-bias and mod-
be tightly restricted, and fast initialization of ONUs must be per- ulation current I-mod. Both of them can be set quickly up to
formed after power-on or first connection to the network with a 80 mA with a resolution of 0.1 mA, providing a total drive cur-
minimum of control signals [13]. This is an important require- rent up to 160 mA. The currents are generated by custom-made
ment when developing high-split-ratio GPON systems, as fast 10-b IDACs designed for BM-LDD outdoor operation over a
network recovery after, for example, power failure is needed. On 40–85 C temperature range [14].
the other hand, accurate tracking of slow laser-power-level drift In contrast to commercially available CW laser drivers, a
is also needed during the data transmission to keep the launched BM-LDD cannot regulate the emitted power by means of a slow
power variation small, thus decreasing the dynamic range re- averaging measurement. Due to the bursty nature of the data,
quirement at the OLT US-RX side. there is no stable average power available. Optical power stabi-
lization is only possible when transmitting, i.e., when a short
B. BM-LDD Design Challenges time slot is allocated to a specific subscriber. Fast but accu-
Fig. 4 shows the architecture of the US-TX. It mainly con- rate optical level monitoring circuitry and an APC algorithm
tains a high-speed laser diode and a generic and intelligent were designed based on current-mode circuits [15], to overcome
BM-LDD chip. Laser diodes are not really the limiting element the drawbacks of voltage-mode implementations at higher bit
for 1244-Mb/s upstream transmission, but the device choice rates. The optical-level monitoring circuit contains current mir-
has a strong impact on the ONU cost. The BM-LDD consists of rors and comparators. An active-input current mirror reduces
a laser driver stage, optical level monitoring, pattern detection, the impact of the parasitic capacitance of the laser back
a dual-mode (fast/slow) digital APC algorithm, and SPI inter- facet monitor photodiode (PD), which is the main speed-lim-
face logic. The BM-LDD chip was developed in a 0.35- m iting factor of the level monitoring circuitry. The active-input
SiGe BiCMOS process. CMOS processes with even shorter current mirror produces two copies of the photocurrent for the
submicrometer gates require lower supply voltages and cannot “1” and “0” level measurement, respectively [14]. During trans-
provide the dc coupling, because the laser diode voltage drop mission, two current comparators compare the monitor current
can be 1.6 V in the worst case, whereas the transient voltage with two reference currents (Iref-mod and Iref-bias) cor-
drop caused by parasitic inductances also has to be taken into responding to the desired “1” and “0” launched optical power.
account. During transmission, the laser diode must be biased Each reference current needs only one calibration at room tem-
above its threshold to reduce the turn-on delay and to limit perature and is digitally set via an SPI interface to avoid trim-
the duty-cycle distortion. Due to the wide spread of individual ming elements on the PCB.
2502 JOURNAL OF LIGHTWAVE TECHNOLOGY, VOL. 22, NO. 11, NOVEMBER 2004

The measurements are only valid after the transmission of a The average signal level may vary 21 dB in the worst case
sufficient number of successive “1”s and “0”s. A pattern detec- from packet to packet due to the following contributions:
tion block scans the incoming data when the BM-LDD is en- 1) up to 15 dB of differential attenuation in the ODN
abled by a burst envelope signal (TH in Fig. 4) and searches for (10–25 dB for Class B);
consecutive strings of “1”s or “0”s with a given (programmable) 2) up to 5-dB tolerance on the mean launched power of an
length. The detection of a suitable string in the data stream en- ONU (from 2 dBm to 3 dBm);
ables the level monitoring and the APC control. Due to this 3) a 1-dB optical path penalty over the ODN, which is com-
self-detection technique, no time-critical signal such as a pre- pensated by an increase of the minimum receiver sensi-
amble envelope or an arming signal is required for APC, as is tivity.
the case in most published burst-mode laser drivers [16], [17].
Successive “1”s and “0”s can be programmed in a power lev- The combination of 21-dB-level variation, a short guard time
eling sequence upstream (PLSu) field with a maximum of 120 (4 B), and a maximum of 72 consecutive identical bits within a
B in length, and in a physical layer operation administration packet also requires the US-RX to be dc coupled. It is impos-
and maintenance upstream (PLOAMu) field. Both are specified sible to choose a right time constant(s) without spoiling the ini-
in G984.3 for power control measurements by an ONU. The tial conditions at the start of a new packet. This would result in
PLOAMu, part of GPON overheads, is up to a length of 13 B, severe signal distortion and burst-mode penalty [20]. If ac cou-
where 10 B of “data” field can be used as a laser control field for pling would be used between the APD-TIA and the BM-RX,
accurately tracking laser power drift during data transmission, the high-pass filtering nature of the ac-coupled circuit would re-
performing a slow APC. ject the low-frequency contents in the data payload [21], [22].
During each allocation period, according to the OLT control, Indeed any mechanism containing the memory of a preceding
the ONU can transmit either the PLSu during an initialization packet, such as ac coupling, slow dc offset compensation, and
phase or the PLOAMu during data transmission for regulating slow AGC can hardly be employed in the GPON US-RX. How-
its launched power. This supports a fast APC algorithm based on ever, dc coupling implies the presence of dc offsets, which may
a binary-like search with maximum power level protection [13]. drift due to temperature variations. As dc offset can become a
The I-bias and I-mod can start from loosely specified but safe limiting factor in obtaining high sensitivity, a BM-RX should
preset values, and there is no need for the storage of a number be able to automatically correct such offsets [23]. Moreover,
of calibration values in lookup tables as some burst-mode laser the BM-RX requires fast but accurate threshold setting on in-
drivers require. The digital APC algorithm [18] quickly and ac- dividual incoming packets to perform dynamic-level detection
curately adjusts the IDAC settings of I-bias and I-mod until the and amplitude recovery. The BM-RX must quickly extract the
level errors are small enough, after which a “level-OK” signal is decision threshold within a preamble length of a few bytes (e.g.,
generated and sent to the NT digital application-specified inte- 3 B in this case) at the beginning of each packet [8].
grated circuit (ASIC). This US-TX can drive most commercial
gigabit laser diodes with a wide range of laser back-facet capaci- B. US-RX Design Challenges
tance (from 2 to 15 pF), photodiode responsivity (from 25
to 1400 A/mW) and laser slope efficiency in almost any cir- The main functions and design challenges of the US-RX are
cumstances. No laborious calibration of the laser characteristic illustrated in Fig. 5. High sensitivity, wide dynamic range, and
curve is needed, and no off-chip component needs to be trimmed fast response are important figures of merit. A 3-dB improve-
for different laser types. There are no time-critical external con- ment on the sensitivity can increase the splitting ratio by a factor
trol signals to initiate a regulation cycle [19]. The BM-LDD chip of 2, which almost doubles the number of subscribers that can
is intelligent and easily programmed via an SPI interface, in con- be connected to the network at little extra cost. A large dynamic
trast with many burst-mode laser drivers published so far [16], range guarantees a long logical reach and increases the network
[17]. To support the PLM, the launched optical power of the flexibility. A high RX sensitivity can be obtained by applying
US-TX is adjustable in a 6-dB range. During initialization, an a higher APD factor, for example, . This, however,
ONU first sets its power from low power mode 2. The minimum does not increase the dynamic range or power budget propor-
mean power required for guaranteeing a 10-dB extinction ratio tionally. A strong optical signal emitted from a nearby ONU ex-
is specified at 5.5 dBm (mode 2). After level measurements periencing a minimum ODN loss would saturate the APD-TIA
performed at the BM-RX, an increase of 3 dB (mode 1) or 6 dB when the APD gain is set to ten. This will result in an over-
(normal mode) can be requested by the line termination (LT). load of the TIA or a severe duty-cycle distortion of the detected
signal. A tradeoff must be made to set an appropriate APD gain
IV. DESIGN OF A GPON OLT US-RX so that the combined requirement of RX sensitivity and dynamic
range can be met. This optimum gain further depends upon the
A. US-RX Requirements transimpedance gain of the preamplifier and the input voltage
The US-RX located in the OLT mainly consists of an sensitivity of the BM-RX. Indeed, in a traditional optical re-
APD-TIA module and a BM-RX chip as shown in Fig. 5. It ceiver, the input voltage sensitivity of the postamplifier is usu-
receives optical packets from all active subscribers in very fast ally much better than the sensitivity of the preamplifier, such that
succession, with varying signal level and phase from packet to this preamplifier solely determines the complete receiver sensi-
packet. The packets are interleaved with a guard time of 4 B at tivity. A BM-RX, however, has a minimum voltage sensitivity
1244 Mb/s (or 25.6 ns) as specified in G984.2. below which the fast dc-coupled threshold extraction circuitry
QIU et al.: DEVELOPMENT OF GPON UPSTREAM PMD PROTOTYPES 2503

Fig. 5. Design challenges for a high sensitivity and a wide dynamic range US-RX.

will not operate properly anymore. For a fixed differential tran-

simpedance gain of 2 k and a differential input voltage sen-
sitivity of the BM-RX of 20 mV, simulation and experimental
results show that an APD gain of around 5 to 6 is a good
compromise to meet the requirements as given in Table I, [24].
It is clear that other optimum values will result from different
values of the transimpedance gain. Hence, one cannot decouple
optimization of the avalanche gain from the transimpedance
gain due to the limitation of a minimum input voltage sensi-
tivity of the BM-RX itself. Note that this value is quite low
when compared with conventional optical receivers. The reason
for this is that at a high avalanche gain, the dynamic range of the
receiver is limited by the tail following a strong packet, which
needs to be decayed sufficiently within the guard time of 25.6 ns
before a following, possibly much weaker, packet can be han-
0
Fig. 6. Simulation result in case of a strong packet ( 11 dBm with 2-dB
dled.
The APD-TIA converts the photocurrent into a differential
0
margin in case of PLM) followed by a weak packet ( 29.5 dBm, almost not
visible) as handled by the BM-RX. The upper trace was the input current of the
input signal for the BM-RX. The BM-RX mainly contains auto- TIA, and the lower traces were plotted at the differential outputs of the BM-RX.
matic threshold extraction circuitry and limiting amplifiers with The level difference of 19.5 dB is 4.5 dB larger than the required 15 dB listed
in Table I yielding ample system margin).
symmetric gain clamping and dc-offset compensation as shown
in Fig. 5. The “level measure PLM” block was designed to mea-
sure the level of the incoming signal by comparing it with two blocks. Modeling has been performed using three different
threshold voltages that are programmable. Fig. 5 gives a simpli- methods, each method having its own advantages.
fied view of the BM-RX chip and its interfaces toward the LT First, transistor-level models were used to validate the eye
digital ASIC and the clock-phase alignment (CPA). However, diagrams of the transmitter, and in particular the laser driver.
the design of the CPA chip is not included in this paper. The Indeed, the required speed and accuracy necessitates a detailed
BM-RX threshold extraction requires both a positive and nega- modeling of the interface between laser diode and driver, for
tive peak detection circuit to extract the amplitude information example, with regard to bonding wires. On the receiver side,
(a “1” and a “0” level) at the beginning of each packet and to careful modeling of any memory effects that could destroy a
generate an offset bias for the limiting amplifier stages. Due to succeeding packet is needed. For example, Fig. 6 shows the
the short guard time, the BM-RX needs an active reset to erase simulated differential outputs of the BM-RX (lower traces) to-
the threshold of a preceding packet. gether with the input current of the TIA (upper trace) in the
worst case of a weak packet (almost not visible) preceded by
a strong packet.
V. UPSTREAM LINK MODELING AND SPECIFICATION
In a second step, abstract models of the transistor-level cir-
Optical transmission system simulations are widely used cuits of both the BM-LDD and BM-RX were combined with
for high-capacity core networks. However, how to accurately detailed descriptions of the laser diode, the fiber plant, and the
model fast burst-mode P2MP transmission is still a challenging avalanche diode. In this way, eye diagrams are produced that
research topic. Intensive study, modeling, and simulations have can be used to quickly evaluate the influence of several system
been performed for GPON upstream burst-mode PMD building parameters, such as the extinction ratio at the US-TX and the
2504 JOURNAL OF LIGHTWAVE TECHNOLOGY, VOL. 22, NO. 11, NOVEMBER 2004

Fig. 7. Simulated eye diagram at the BM-RX output. The launched output
0
power of the US-TX was set to 5 dBm with an extinction ratio of 10.5 dB, a
0
20-km G.652 fiber was included, and the BM-RX input level was 28.5 dBm, Fig. 8. Measured BER on the first-version prototypes (US-TX-version1
with an APD gain of 6 at 1244 Mb/s. The input referred noise current of the TIA and US-RX-version1) and simulated BER curves versus the received average
was 230 nA , and its differential transimpedance was 2 k
. optical power at 1244 Mb/s with an APD gain of 6. The APD was a Mitsubishi
PD8933. The input-referred noise current of the TIA was 230 nA , and
its differential transimpedance was 2 k
. The decision threshold is halfway
APD gain at the US-RX, on the GPON physical-layer perfor- from the peak eye opening, and the sample moment was optimized for best
mance quantified by parameters such as differential range, split-
0
sensitivity. The transmitter optical power was set to a 3.0-dBm average, with
an optical extinction ratio of 10 dB.
ting factor, and feeder length. As an example, Fig. 7 shows the
simulated optical uplink end-to-end eye diagram for a received
packets launched by ONUs. The recommended allocations of
average optical power of 28.5 dBm with an APD gain 6
the physical-layer overhead specified by G984.2 are illustrated
at 1244 Mb/s.
in Fig. 9. The mandatory total length of overhead at 1244 Mb/s
In a last refinement, a detailed description of the mechanics
is 12 B (promoted for standardization [1] and adopted in the
giving rise to bit errors in the receiver is needed. The cal-
ITU-T G984.2) and consists of guard time (mandatory 32 b),
culation of the error probability for the US-RX needs to be
preamble time (44 b), and delimiter time (20 b). The length
split into two parts—the first part corresponding to the am-
of the guard time is determined by the laser turn-on/turn-off
plitude extraction performed by the BM-RX, and the second
time, time shifts caused by 1) slight variations of the fiber
part corresponding to the clock-phase extraction performed
delay, 2) the APD and transistor discharge time, and 3) the
by the CPA. The sensitivity penalty for burst-mode receivers
fiber propagation delay equalization granularity determined
using APDs has been analyzed in depth [24]. The analysis
by a time ranging process. The preamble can be split into
takes into account detailed APD statistics, additive Gaussian
two parts, a so-called threshold determination field (TDF) for
noise, intersymbol interference, and dc offsets in the receiver
amplitude recovery and the CPA field for clock-phase recovery,
channel. The penalty was calculated via comparison of bit-error
both of which are programmable under the OLT’s control. As
rates (BERs), obtained using numerical integration, both in
explained in [24] and [25], quick extraction of the decision
continuous-mode and burst-mode operation. It shows that
threshold and clock phase from a short preamble at the start
dc offsets and finite extinction ratios can easily dominate
of each packet results in a sensitivity penalty. Both penalties
the penalty due to the noise-corrupted threshold. Important
depend in a complex way upon the length of the TDF and
guidelines were given to the design of high-sensitivity and
CPA field, respectively. An optimum distribution between both
wide-dynamic-range BM-RXs. Fig. 8 shows a calculated BER
should be found to maximize the performance of the GPON.
curve compared with a measured BER curve obtained from the
A combination of measurements and simulation will be used
first-version 1244-Mb/s burst-mode prototypes (US-TX-ver-
for further study to find such an optimum. Finally, note that
sion1 and US-RX-version1), which were not yet optimized for
many topics, such as reflections from transmitters close to the
best performance. At a BER of , it can be observed that
receiver and tails occurring after a strong burst in the receiver,
the measured curve and the calculated curve differ by about 0.4
need further investigation in detail.
dB. This difference can be attributed to a tail within the US-RX
and to small optical power fluctuations (in the range of 0.2 dBm
VI. UPLINK BURST-MODE EXPERIMENTAL RESULTS
with a time constant of several seconds) of the BM-TX (which
uses an uncooled laser diode). To evaluate physical-layer uplink performances such as the
As described in [25], the error probability of the signal BM-RX sensitivity, dynamic range, and BBER, a back-to-back
leaving the BM-RX can be combined with a model for the uplink containing one US-TX, an APD-TIA, and a BM-RX
clock-phase extraction, giving rise to a final burst bit-error rate was established and tested at INTEC [26], [27]. Fig. 10 shows
(BBER). Some physical-layer upstream burst-mode overhead the measured eye diagram of the first-version burst-mode
is added, for the US-RX to correctly receive the upstream transmitter prototype (US-TX-version1) with a distributed
QIU et al.: DEVELOPMENT OF GPON UPSTREAM PMD PROTOTYPES 2505

Fig. 9. Specification of upstream burst-mode overhead (96 b in total). It consists of guard time (mandatory 32 b), preamble time (44 b), and delimiter time (20 b).

Fig. 10. US-TX eye diagram (left) was measured using a fourth-order Thompson filter at 1244 Mb/s (mean optical power = 02.5 dBm,
bias power =0
13.5 dBm). The measured turn-on and turn-off times are shown on the right side.

feedback (DFB) laser diode (Mitsubishi FU-445SDF). The margin for the tracking error of the PON optics. At the time
mean and bias optical power was set to 2.5 dBm and 13.5 of this writing, this is the first GPON US-TX prototype to be
dBm, respectively. The test results show that the eye diagram published supporting the PLM.
of the US-TX-version1 falls well into the mask specified in the Two US-TXs (version 1 prototypes) were subsequently con-
ITU-T G.984.2. The measured burst turn-on time is 6 pre-bias figured to send a strong packet P1 followed by a weak packet P2
bits, and the turn-off takes about 13 b; both meet the PMD toward the US-RX (version 1 prototype), emulating a worst-case
specification (TX enable/disable time 16 b) as illustrated in condition, as shown in Figs. 3 and 6. The US-TX2 (P2) was
Fig. 9. The launched optical power tolerance of the US-TX was set to 3-dBm mean optical power with an extinction ratio of
tested over the full temperature range (from 40 to 85 C) [14]. 10 dB. The test result is shown in Fig. 11, where the upper and
In order to support PLM, three different mean optical powers lower trace represent the output of the APD-TIA and output
of 5.5, 2.5, and 0.5 dBm, respectively, were set together of the BM-RX-version1, respectively; the avalanche gain
with a fixed bias level of 13.5 dBm during the temperature was set to 6; the strong and weak packets received in the APD
tests. A maximum total optical power variation of 0.8 dB has have an input average power of 10.0 and 29.5 dBm, respec-
been achieved including the tracking error of the photodiode, tively. The transmitted payload was pseudorandom bit sequence
the tolerance of the reference current, and the offset variations (PRBS) with a packet length of 19.52 kB; the guard time
of the high-speed monitoring circuits, which is much better between the two packets was 4 B, or 25.6 ns, followed by a pre-
than the 5-dB tolerance specified in G.984.2, and leaves ample amble of 12 b of “1” and 12 b of “0” for the RX amplitude re-
2506 JOURNAL OF LIGHTWAVE TECHNOLOGY, VOL. 22, NO. 11, NOVEMBER 2004

Fig. 12. BER performance at 1244 Mb/s. The input signal of the BM-RX
consisted of packets, separated with a guard time of 25.6 ns. Each packet
contained a preamble of 24 b, followed by a pattern of 20 times “10” and a
PRBS sequence 2 0 1 with a length of 128 000 b.

Fig. 11. Strong packet followed by a weak packet emulates the worst-case
condition. The upper trace was plotted at the output of the APD-TIA, and the
lower trace was plotted at the output of the BM-RX1. The strong signal was
0 0
measured at 10 dBm, and the weak signal was 29.5 dBm; the APD gain
was set to 6.

covery; 20 successive “10” patterns were included for the CPA

and delimiter field, as illustrated in Fig. 9. From Fig. 11, one can
see that the weak packet at an input optical power of 29.5 dBm
is well recovered, though the bandwidth was limited for the “10”
pattern (in front of the data payload) due to capacitive loading of
the external components required for the first-version BM-RX
chip, which was not yet optimized for best performance.
BERs were measured using packets for which the preamble
is composed of 24 b followed by a pattern of 20 times “10” and
a PRBS sequence with a length of 128 000 b; the guard
time was 25.6 ns. Only bit errors within the PRBS sequences Fig. 13. Back-to-back burst-mode BER performance at 1244 Mb/s. The
were taken into account. The average optical power of the 0
optimized US-RX sensitivity of 32.8 dBm was measured ( M
= 6), and its
BM-TX-version1 was set at 2.5 dBm with an extinction M
dynamic range reaches 23 dB ( = 6) at BER 10 < .

ratio of 10 dB. Fig. 12 shows the measured BERs for different

avalanche gains versus the average input optical power. bursts up to 125 s long, spaced with a guard time of 25.6 ns,
The receiver sensitivity threshold was found at an optical and compliant with the GPON standards.
input power of 31.6 30.8 30.2, and 28.9 dBm, for an
avalanche gain of 8, 7, 6, and 5, respectively. In addition, the
VII. CONCLUSION
maximum receiver overload was 11.0 10.6 10.1, and
9.2 dBm at 8, 7, 6, and 5, respectively. This paper presented the development of a 1244-Mb/s GPON
Furthermore, the burst-mode BER was measured for a burst-mode chip set and its integration into upstream PMD op-
back-to-back 1.25 Gb/s uplink containing one US-TX-version2 tical transmitter and receiver prototypes. Based on the P2MP
(the second optimized chip BM-LDD), an APD-TIA module, uplink system study and modeling, innovative design concepts
and a recently developed fully functional US-RX-version2 (the were proposed, implemented, and proven. Valuable contribu-
second optimized prototype), showing even better performance. tions were provided to FSAN/ITU-T for GPON PMD standard-
The improved performance is mainly due to the fact that this ization.
receiver is fully integrated, avoiding any capacitive loading In the GPON burst-mode receiver, an APD gain of 5 to 6
on sensitive nodes that leads to eye closure. The measured proved to yield a good compromise between dynamic range and
burst-mode BER at the BM-RX-version2 output is shown in sensitivity. For the first time, to the authors’ knowledge, test
Fig. 13. The two different slopes in the curve can be explained results are demonstrated and confirmed by uplink simulations,
by the fact that, for optical powers beneath 34.0 dBm, the which comply with the new ITU-T Recommendation G.984.2
differential signal at the output of the TIA is too small to be supporting the overall PLM. The GPON PMD and system pro-
correctly recovered by the BM-RX. The optimized receiver totypes with high performance will be demonstrated within the
sensitivity is 32.8 dBm 6 , and its dynamic range framework of the IWT SYMPATHI and IST GIANT project,
reaches 23 dB 6 at BER . The experiment respectively, which are expected to be one of the first proof-of-
carried out so far shows that the US-RX is capable of handling concept demonstrations of the advanced GPON technology.
QIU et al.: DEVELOPMENT OF GPON UPSTREAM PMD PROTOTYPES 2507

ACKNOWLEDGMENT [20] Y. Ota and R. G. Swartz, “DC-1 Gb/s burst-mode compatible receiver for
optical bus applications,” J. Lightwave Technol., vol. 10, pp. 244–249,
The authors would like to thank all SYMPATHI and GIANT Feb. 1992.
project partners for their cooperation, especially STMicroelec- [21] S. Han and M.-S. Lee, “AC-coupled burst-mode optical receiver
employing 8B/10B coding,” IEE Electron. Lett., vol. 39, no. 21, pp.
tronics and Alcatel Bell, which were coordinators for the SYM- 1527–1528, Oct. 2003.
PATHI and GIANT projects, respectively. [22] , “Burst-mode penalty of AC-coupled optical receivers optimized
for 8B/10B line code,” IEEE Photon. Technol. Lett., pp. 1724–1726, July
2004.
REFERENCES [23] P. Ossieur, Y. C. Yi, J. Bauwelinck, X. Z. Qiu, J. Vandewege, and E.
[1] X. Z. Qiu, J. Vandewege, F. Fredricx, and P. Vetter, “Burst mode trans- Gilon, “DC-coupled 1.25 Gbit/s burst-mode receiver with automatic
mission in PON access systems,” in Proc. 7th Eur. Conf. Networks Op- offset compensation,” IEE Electron. Lett., vol. 40, no. 7, pp. 447–448,
tical Communication, 2002, pp. 127–132. Apr. 2004.
[2] P. Vetter et al., “Study and demonstration of extensions to the standard [24] P. Ossieur, X.-Z. Qiu, J. Bauwelinck, and J. Vandewege, “Sensitivity
FSAN BPON,” in Proc. Int. Symp. Services Local Access, 2002, pp. penalty calculation for burst-mode receivers using avalanche photodi-
119–128. odes,” J. Lightwave Technol., vol. 21, pp. 2565–2575, Nov. 2003.
[3] B. De Vos, F. Duthilleul, F. Fredricx, P. Vetter, P. Ossieur, J. Bauwelinck, [25] B. Meerschman, Y. C. Yi, P. Ossieur, D. Verhulst, J. Bauwelinck, X.
X.-Z. Qiu, J. Vandewege, and J. Watté, “Demonstration of extended Z. Qiu, and J. Vandewege, “Burst bit-error rate calculation for GPON
split APON,” in Proc. Conf. Optical Fiber Communication (OFC 2002), systems,” in Proc. IEEE/LEOS Benelux Chapter, 2003, pp. 165–168.
2002, pp. 437–439. [26] Z. Lou, S. Verschuere, Y. Yi, D. Verhulst, X. Z. Qiu, and J. Vandewege,
[4] I. Van de Voorde, C. Martin, J. Vandewege, and X. Z. Qiu, “The Su- “Lab. test bed development for evaluation of the GigaPON uplink per-
perPON demonstrator: An exploration of possible evolution paths for formance,” in Proc. IEEE/LEOS Benelux Chapter, 2003, pp. 137–140.
optical access networks,” IEEE Commun. Mag., vol. 38, pp. 74–82, Feb. [27] X. Z. Qiu, P. Ossieur, J. Bauwelinck, Y. C. Yi, D. Verhulst, S. Verschuere,
2000. Z. Lou, W. Chen, Y. Martens, X. Yin, J. Vandewege, B. De Vos, and E.
[5] F. Fredricx, J. Angelopoulos, P. Solina, X. Z. Qiu, S. Fisher, and E. Laes, Gilon, “FSAN GPON upstream burst-mode transmission experiments,”
“The IST-GIANT project (GIgaPON access NeTwork),” in Proc. 7th presented at the 30th Eur. Conf. Optical Communication (ECOC’04),
Eur. Conf. Networks Optical Communication, 2002, pp. 119–126. Stockholm, Sweden, Sept. 5–9, 2004.
[6] K. W. Cobb and R. Mercinelli, “QoS and customer base for an FSAN
compliant GPON,” in Proc. Int. Symp. Subscriber Loops Services, Ed-
inburgh, U.K., Mar. 2004.
[7] J. D. Angelopoulos, H.-C. Leligou, T. Argyriou, S. Zontos, E. Ringoot,
and T. Van Caenegem, “Efficient transport of packets with QoS in an
FSAN-aligned GPON,” IEEE Commun. Mag., pp. 92–98, Feb. 2004. Xing-Zhi Qiu (M’98) received the Ph.D. degree in
[8] Gigabit-Capable Passive Optical Networks (GPON): Physical Media electronics engineering from Ghent University, Gent,
Dependent (PMD) Layer Specification, ITU-T Recommendation Belgium, in 1993.
G.984.2, 2003. She joined the INTEC design laboratory of Ghent
[9] Transmission Convergence Layer for Gigabit Passive Optical Networks, University in 1986. She has been active in the
ITU-T Recommendation G.984.3, 2004. development of optoelectronic systems, especially of
[10] X.-Z. Qiu, J. Bauwelinck, Y. Martens, D. Verhulst, J. Vandewege, B. burst-mode transmitter and receiver front ends. She
Stubbe, F. Fredricx, and P. Solina, “Design of ONT transmitters for gi- has managed the development of the gigabit-capable
gabit PON’s,” in Proc. Int. Integrated Circuits-China Conf., 2003, pp. passive optical network (GPON) burst-mode chip
127–132. set design and GPON upstream PMD subsystem
[11] T. Van Caenegem, E. Gilon, X.-Z. Qiu, P. Solina, and K. Cobb, “Pro- development within the INTEC design laboratory.
totyping ITU-T GPON, the new efficient and flexible FTTP PON solu- She is author/coauthor of 70 international publications in the field of advanced
tion,” in Proc. Int. Symp. Subscriber Loops Services, Edinburgh, U.K., telecommunication systems, optical access network demonstrations, and
Mar. 21–24 2004. mixed-mode analog/digital chip design.
[12] T. Van Caenegem, B. De Vos, and E. Ringoot, “A comparison of
emerging packet based PON standards,” in Proc. 29th Eur. Conf.
Optical Communications, 2003, pp. 862–863.
[13] X.-Z. Qiu, J. Vandewege, Y. Martens, J. Bauwelinck, P. Ossieur, E.
Gilon, and B. Stubbe, “A burst-mode laser transmitter with fast digital
power control for a 155 Mb/s upstream PON,” IEICE Trans. Commun. Peter Ossieur (S’03) received the Eng. degree in ap-
(Japan), vol. E86-B, pp. 1567–1574, May 2003. plied electronics from Ghent University, Gent, Bel-
[14] J. Bauwelinck, D. Verhulst, P. Ossieur, X. Z. Qiu, J. Vandewege, and B. gium, in 2000, where he is currently working toward
De Vos, “DC-coupled burst-mode transmitter for a 1.25 Gbit/s upstream the Ph.D. degree.
PON,” IEE Electron. Lett., vol. 40, no. 8, pp. 501–502, Apr. 2004. He has been a Research Assistant in the INTEC
[15] J. Bauwelinck, D. Verhulst, P. Ossieur, X. Z. Qiu, J. Vandewege, and design laboratory of Ghent University since 2000.
B. De Vos, “Current mode circuits for fast and accurate optical level His research focuses on analog integrated circuits
monitoring with wide dynamic range,” IEICE Trans. Commun. (Japan), for burst-mode laser drivers and receivers in passive
vol. E87-B, no. 9, Sept. 2004. optical network telecommunication systems and in
[16] E. Säckinger, Y. Ota, T. J. Gabara, and W. C. Fischer, “A 15-mW, 155- the modeling of burst-mode communication.
Mb/s CMOS burst-mode laser driver with automatic power control and
end-of-life detection,” IEEE J. Solid-State Circuits, vol. 35, pp. 269–275,
Feb. 2000.
[17] M. Doci, C. Fiocchi, U. Gatti, A. Profumo, and G. Promitzer, “A dig-
itally programmable burst-mode 155 Mb/s transmitter for PON,” pre-
sented at the 26th Eur. Solid-State Circuits Conf., Stockholm, , Sweden,
Sept. 19–21, 2000. Johan Bauwelinck (S’03) received the Eng. degree
[18] Y. Martens, X. Z. Qiu, J. Bauwelinck, K. Noldus, P. Ossieur, J. Van- in applied electronics from Ghent University, Gent,
dewege, E. Gilon, F. Duthilleu, and G. Lorusso, “Design and implemen- Belgium, in 2000, where he is currently working to-
tation of a dual mode digital APC algorithm for an APON burst-mode ward the Ph.D. degree.
laser driver,” in Proc. Int. Conf. Electronics Hardware Systems, 2002, He has been a Research Assistant in the INTEC
pp. 1401–1406. design laboratory of Ghent University since 2000.
[19] J. Bauwelinck, Y. Martens, P. Ossieur, K. Noldus, X. Z. Qiu, J. Van- His research focuses on analog integrated circuits for
dewege, E. Gilon, and A. Ingrassia, “Generic and intelligent CMOS 155 burst-mode laser drivers in passive optical network
Mb/s Burst mode laser driver chip design and performances,” in Proc. telecommunication systems.
Eur. Solid-State Circuits Conf., 2002, pp. 495–498.
2508 JOURNAL OF LIGHTWAVE TECHNOLOGY, VOL. 22, NO. 11, NOVEMBER 2004

Yanchun Yi received the Bachelor’s degree in elec- Benoit De Vos received the M.Sc. degree in elec-
tronics engineering from Tsinghua University, Bei- trical engineering from the Faculté Polytechnique de
jing, China, in 1998 and the Master’s degree from the Mons, Mons, Belgium, in 2000.
University of York, York, U.K., in 2002. He is cur- He has been working in the Research and Innova-
rently working toward the Ph.D. degree at the Infor- tion Department of Alcatel Bell, Antwerp, Belgium,
mation Technology Department of Ghent University, since September 2000. His research focuses on PMD-
Gent, Belgium. layer-related system studies for passive optical net-
His research interest focuses on the simulation of work (PON) systems. Currently, he is leading, within
burst-mode optical communication systems. Alcatel, the SYMPATHI project, which investigates
a gigabit-speed PON network.

Dieter Verhulst (S’03) received the engineering

degree in applied electronics from Ghent University,
Gent, Belgium, in 2001, where he is currently
working toward the Ph.D. degree.
He has been working at the INTEC design labora-
tory of Ghent University in a passive optical network
research project with a focus on the high-speed digital
functions in burst-mode laser drivers and clock-phase
alignment circuits.

Paolo Solina received the electronics degree from the

Technical Institute G. Peano of Turin, Italy, in 1974.
Jan Vandewege (M’96) was born in Gent, Belgium, He joined Telecom Italia Lab (TILab, formerly
in 1949. He received the Electronic Engineering and CSELT), Turin, Italy, in 1974. He contributed to
Ph.D. degrees from Ghent University, Gent, Belgium, the development of high-bit-rate optical communi-
in 1972 and 1978), respectively. cations systems within several projects: Esprit 169
He founded the INTEC design laboratory of LION, Esprit 2512 IACIS, RACE 2024 BAF, and
Ghent University, in 1985, which provides training ACTS 050 PLANET. He was also responsible for
to Ph.D.-level electronic engineers in the design the service trials based on passive optical network
of telecom and radio-frequency hardware and fast platforms within the EURESCOM projects P917
embedded software. He is currently a Full Professor BOBAN and P1015 FREEHANDS.
at Ghent University. He has authored or coauthored Mr. Solina has been a Member of the Full Services Access Network Optical
152 international publications and ten international Access Network (FSAN OAN) group since 1996. He has been the Editor of the
patents in the field of telecommunication systems, optical access networks, and new ITU-T Recommendation G.984.2, which specifies the physical layer of the
mixed-mode analog/digital chip design. gigabit-capable passive optical network systems.

View publication stats