ONU power saving modes in next generation

optical access networks: progress, efficiency and

challenges
Abhishek Dixit,1,*Bart Lannoo,1Didier Colle,1Mario Pickavet,1and Piet Demeester1
1
Department of Information Technology, Ghent University – IBBT, Gent, Belgium
*[email protected]

Abstract: The optical network unit (ONU), installed at a customer’s

premises, accounts for about 60% of power in current fiber-to-the-home
(FTTH) networks. We propose a power consumption model for the ONU
and evaluate the ONU power consumption in various next generation
optical access (NGOA) architectures. Further, we study the impact of the
power savings of the ONU in various low power modes such as power
shedding, doze and sleep.
©2012 Optical Society of America
OCIS codes:(060.4250) Networks; (060.4252) Networks, broadcast.

References and links

1. M. Pickavet, W. Vereecken, S. Demeyer, P. Audenaert, B. Vermeulen, C. Develder, D. Colle, B. Dhoedt, and P.
Demeester, “Worldwide energy needs for ICT: the rise of power-aware networking,” in Proc. of IEEE Conf. on
Advanced Networks and Telecommunications Systems, New Delhi, 2008.
2. K. Hinton, J. Baliga, M. Feng, R. Ayre, and R. S. Tucker, “Power consumption and energy efficiency in the
Internet,” IEEE Netw. 25(2), 6–12 (2011).
3. White paper: 802.3av Power Saving Adhoc Report, Seoul, Sep. 2008. Available at:
http://www.ieee802.org/3/av/public/2008_09/3av_0809_otaka_1.pdf
4. L. Shi, B. Mukherjee, and S. S. Lee, “Efficient PON with sleep-mode ONU: progress, challenges, and
solutions,” IEEE Netw. 26(2), 36–41 (2012).
5. A. Dixit, B. Lannoo, S. Lambert, D. Colle, M. Pickavet, and P. Demeester, “Evaluation of ONU power saving
modes in next generation optical access networks,” in Proc. of European Conf. on Optical Communication,
Amsterdam, paper Mo.2.B.5 (2012).
6. B. Skubic, E. In de Betou, T. Ayhan, and S. Dahlfort, “Energy-efficient next-generation optical access
networks,” IEEE Commun. Mag. 50(1), 122–127 (2012).
7. I. T. U.-T. Rec, G.Sup45, GPON Power Conservation, 2009.
8. GreenTouch, http://www.greentouch.org.
9. Code of Conduct on Energy Consumption of Broadband Equipment, European Commission, v. 4, Feb. 10, 2011.
10. 249025-ICT OASE Project, D4.4: Implementation and integration into new system concepts.
11. SFP Multisource level agreement, Available at: ftp://ftp.seagate.com/sff/INF-8074.PDF
12. E. Igawa, M. Nogami, and J. Nakagawa, “Symmetric 10G-EPON ONU burst-mode transceiver employing
dynamic power save control circuit,” in Proc. of Optical Fiber Communication Conf. and National Fiber Optic
Engineers Conf., Los Angeles, CA, paper NTuD5.pdf (2011).
13. E. Wong, C. A. Chan, P. I. Dias, M. Mueller, and M. C. Amann, “Vertical cavity surface emitting laser
transmitters for energy efficient broadband access networks,” in Proc. of Int. Conf. on Communications, Ottawa,
paper ONS02 (2012).
14. S. W. Wong, L. Valcarenghi, Y. She-Hwa, D. R. Campelo, S. Yamashita, and L. Kazovsky, “Sleep mode for
energy saving PONs: advantages and drawbacks,” in Proc. of IEEE Globecom, Honolulu, HI (2009).
15. L. Bolzani, A. Calimera, A. Macii, E. Macii, and M. Poncino, “Enabling concurrent clock and power gating inan
industrial design flow,” in Proc. Design, Automation & Test in Europe Conference & Exhibition, 2009.
16. K. Christensen, P. Reviriego, B. Nordman, M. Bennett, M. Mostowfi, and J. A. Maestro, “IEEE 802.3az:The
Road to Energy Efficient Ethernet,” IEEE Commun. Mag. 48(11), 50–56 (2010).
17. G. Kramer, B. Mukherjee, and G. Pesavento, “IPACT a dynamic protocol for an Ethernet PON (EPON),” IEEE
Commun. Mag. 40(2), 74–80 (2002).

#177139 - $15.00 USD Received 2 Oct 2012; revised 27 Oct 2012; accepted 7 Nov 2012; published 28 Nov 2012
(C) 2012 OSA 10 December 2012 / Vol. 20, No. 26 / OPTICS EXPRESS B52
1. Introduction
Expected global energy demand is growing faster than 2% per annum [1], and will most
likely become unattainable in the years to come. Among others, one of the factors leading to
the upsurge in the energy needs of society is the continuous colossal growth in the
information and communication technology (ICT) sector, particularly the Internet. Today, the
ICT and the Internet are the important constituent factors of power consumption (PC); and
they account for approximately 5 and 1% respectively of the total electrical PC in developed
economies [2]. Out of the total Internet PC, access networks consume about 60 to 80% of
power and an optical network unit (ONU), installed at a customer's premises, accounts for
about 60% of the energy consumed in current fiber-to-the-home (FTTH) technologies [3].
Thus, significant energy savings can be attained by low energy consuming ONU
architectures, and the ONU PC has remained as a key parameter in the conception of next-
generation optical access (NGOA) networks.
While the design of low power consuming ONU architectures should be considered in the
first place, the PC at ONUs can be further reduced by operating ONUs in low power modes.
Low power modes have been discussed in ITU-T Rec. G.Sup45 and have potential to impart
significant energy savings [4]. While all NGOA networks will likely benefit from low power
modes, the amount of savings that can be procured vary significantly [5]. A holistic view on
the PC of various NGOA architectures can only be derived by counting in the possible power
savings at the ONUs. In this paper, we evaluate the PC of various NGOA architectures in
active state and in low power modes. The results show that the architectures that have higher
active state PC may achieve lower PC in low power modes.
The remainder of this paper is organized as follows. Section 2 discusses the considered
NGOA networks. Section 3 proposes the ONU PC model. Section 4 discusses power saving
modes, progress and challenges. Section 5 presents the evaluation methodology, and section 6
presents the results. Finally, section 7 presents the conclusions.
2. Optical access technologies
As the NGOA candidate, different system concepts are actively considered such as: high data
rate time division multiple access (TDMA) passive optical network (PON) such as 40G-
TDMA-PON, wavelength division multiplexing (WDM) PON, hybrid TDMA/WDM PON
(TWDM-PON), point-to-point (PtP) and active optical network (AON) [6]. We will also
compare these NGOA solutions with present state-of-art solutions like Ethernet PON (EPON)
and 10G-EPON (10G-EPON).
The basic differences among various system aspects are the ways in which a user (or an
ONU) connects to the optical line terminal (OLT) at the central office, and accesses network
resources. While the network architecture of these solutions is quite different; in addition,
these solutions require a different set of functionalities at the ONU. This different set of
functionalities drives the PC of the ONU. First, the upstream and the downstream line rate of
the technologies impact the PC as they influence the transceiver design and processing
requirements; second, the need of tunability at the transmitter influences the PC, as either the
laser needs to be thermo-electrically controlled or is a high power consuming uncooled
tunable laser; third, if the ONU receives more than one wavelength, there is the need of a
tunable filter, which increases the PC; and lastly, every TDMA based solution requires burst
mode electronics, which increase the PC. We distinguish the important functionalities that
influence the PC in the ONU in Table 1, and discuss them in the perspective of each
architecture:
- TDMA-PONs: In TDMA-PON, the OLT accesses ONUs using a TDMA protocol over a
power splitter. The architectural configuration of 40G-TDMA-PON is same as EPON and
10G-EPON but with the support of a much higher upstream and downstream rate using a non-
return to zero (NRZ) on-off keying. It, however, suffers from reach limitations posed by

#177139 - $15.00 USD Received 2 Oct 2012; revised 27 Oct 2012; accepted 7 Nov 2012; published 28 Nov 2012
(C) 2012 OSA 10 December 2012 / Vol. 20, No. 26 / OPTICS EXPRESS B53
serious dispersion issues with high data rate (e.g., 40G NRZ) transmission. Hence, special
functionalities like electronic dispersion compensation (EDC) and optical amplification (OA)
are required along with burst mode electronics.
- WDM-PON: WDM-PON offers the most straightforward way of capacity increase
compared with TDMA-PON, where each user is given a separate wavelength. Since, users are
on a separate wavelength, WDM-PON does not require the complexities of TDMA. Either the
WDM-PON ONU is equipped with a tunable laser (TL) to tune to a separate wavelength or it
may use the downstream signal wavelength to transmit on a separate upstream wavelength.
For the second case, the ONU requires reflective semiconductor optical amplifiers (RSOA)
and frequency shift keying (FSK) modulators for the simultaneous transmission of the
upstream and the downstream signals. We consider these two variants of WDM-PON: with
tunable lasers (WDM-TL) and with RSOA (WDM-RSOA).
- TWDM-PON: TWDM-PON combines the flexibility of TDMA-PON with the increased
overall capacity of WDM technology. They use the TDMA functionality and tunable optics at
both the transmitter and the receiver.
- PtP: In PtP, each ONU is connected directly via a fiber to the central office. It has the
simplest ONU architecture.
- AON: AONs use an active remote node in the field, which requires powering and
maintenance. Since, ONUs are accessed over the active switch, they do not require TDMA
functionality and tunable optics.
Table 1. NGOA system concepts and key required functionality
System Data Rate Tunable Tunable Burst Special
Upstream Downstream TX Filter mode functionality
EPON 1G 1G / /  NA
10G-EPON 1G 10G / /  NA
40G-TDMA 10 G 40 G / /  OA, EDC
WDM-RSOA 1G 1G / / / FSK modulators
for upstream
WDM-TL 1G 1G  / / NA
TWDM 2.5 G 10 G    NA
PtP 1G 1G / / / NA
AON 1G 1G / / / NA
/ = does not require,  = requires

3. ONU power consumption (PC) model

In Fig. 1, we present the PC model of the ONU. The model takes data from [5–10] and the
large survey of component datasheets. To compare the technologies fairly, the power
consumption data of components are scaled relative to the power consumption value of 1.2 W
of the GPON TRX as adopted in [7]. For example, the value of uncooled tunable TRX of
WDM-PON and the GPON TRX are adopted as 0.75 W and 0.45 W respectively by [8]; these
values are scaled up to 2 W and 1.2 W. Note that, the value of 1.2 W adopted in [7] was
according to the maximum power consumption value found in the survey of component data
sheets. For the analysis, we divide the ONU PC model in three main parts: user network
interfaces (UNIs), core functional blocks (CFBs) and specific functional blocks (SFBs).
- UNIs: The interfaces towards client sides are referred as UNIs. Depending on the feature
set of the ONU, the ONU can host different UNIs such as voice interfaces, data ports,
multimedia over coax alliance (MoCA) interface, or video overlay interfaces. Voice
interfaces are assumed to provide plain old telephone service (POTS). Dual subscriber line
interface circuit (SLIC) modules are used to interface with the analog telephone line. The
gigabit Ethernet (GbE) interface is considered for the support of data traffic, and RF video
and MoCA for video and multimedia data.
- CFBs: The core functional block represents the power consumed by an optical
transceiver (TRX), digital processing and memory. The optical TRX consists of a transmitter

#177139 - $15.00 USD Received 2 Oct 2012; revised 27 Oct 2012; accepted 7 Nov 2012; published 28 Nov 2012
(C) 2012 OSA 10 December 2012 / Vol. 20, No. 26 / OPTICS EXPRESS B54
(TX) and a receiver (RX) block. The TX includes the PC for a laser diode (LD) and a laser
diode driver (LDD). The RX consists of an avalanche photodiode (APD) or a PIN
photodiode, transimpedance amplifiers (TIA) and limiting amplifiers (LA). Digital processing
functions are considered to be implemented in a system on chip (SoC). The SoC includes the
PC for a medium access control (MAC), a serialiser, a deserialiser, forward error correction
(FEC), etc. The memory requirement in different concepts may vary (cf. section 6.3), and we
have assumed the PC of 30 mW per MB of memory.
- SFBs: For a specific NGOA system concept, we add the PC for special functionalities
like TDMA, EDC, and OA. For example, we add TDMA PC for EPON, 10G-EPON, and
TWDM-PON, and TDMA, EDC and OA functionality for 40G-TDMA-PON.
Miscellaneous and power conversion losses are assumed at 5% and 20%, respectively.
Note that, the power conversion efficiency consists of AC to DC and DC to DC conversion,
which are both assumed to be 90%, resulting in the overall power conversion efficiency of
80%. In the model, we also show PC values used for the components in active (A), power
shedding (PS), doze (D), fast sleep (FS) and deep sleep (DS) state. These states are discussed
in the following section.
Generic ONU architecture Memory
(30 mW/MB) A 0.25 W
PS 0.15 W
SLIC
(2 W)

A 0.7 W
OA

Transceiver System On Chip

PS 0.14 W
Color GbE
TDMA MAC (0.2 W)
Code A 1.8 W
PS 0 W EDC (2W) A 2.2 W
SFB
PS 0 W
CFB
UNI RF optics MoCA RF

ONU power Sub-systems power

consumption model dissipation

SFBs CFBs UNIs Component Power consumption (W)

Transceiver A PS D FS DS
Voice EPON TRX [ 1.25 G Tx, 2.5 G Rx, APD] 1.2 1.2 0.3 0 0
Transceiver (Dual SLIC) 10G-EPON TRX [ 2.5 G Tx, 10 G Rx, APD] 1.4 1.4 0.5 0 0
Time Division
Multiplexing WDM TRX (TL) [Uncooled Directly Modulated
System on Data (DM) Tunable Laser 1.25 G, 1.25 G Rx, APD] 2 2 0.25 0 0
Chip (GbE) WDM TRX (RSOA) [DM RSOA, 1.25 G Rx,
Dispersion APD] 1.9 1.9 0.25 0 0
Compensation Memory Multimedia PtP [1.25 G Tx, 1.25 G Rx, APD] 1 1 0.25 0 0
= 30 mW/MB (MoCA) AON [1.25 G Tx, 1.25 G Rx, APD] 1 1 0.25 0 0
Optical TWDM [2.5 G Tx, 10 G Rx, APD, Tunable filter] 3.5 3.5 0.7 0 0
Amplification 40G-TDMA [10 G Tx, 40 G Rx, APD ] 4.5 4.5 1 0 0
RF overlay
SoC
WDM /AON/PtP 1 0.6 0.4 0.2 0.1
Miscellaneous losses = 5 % EPON 1.2 0.7 0.5 0.3 0.1
10G-EPON 2 1.2 0.8 0.4 0.1
Power Conversion efficiency = 80% TWDM 2 1.2 0.8 0.4 0.1
40G-TDMA 3 1.8 1.2 0.6 0.1

Fig. 1. ONU power consumption model.

4. Power saving modes: definitions, progress and challenges

In this section, we first define the low power modes in section 4.1, and then discuss progress,
and challenges associated with powering off the parts of the ONU in section 4.2.
4.1 Definitions
ITU-T G.sup 45 [7] proposes a number of power saving states, namely power shedding, doze
and sleep. These approaches differ based on the parts of the ONU that are switched off.
- Power shedding: The power shedding approach shut down the unused UNIs. Note that
there can be many classes of power shedding based on the interfaces that can be shut down.
For the analysis, we assume power shedding with MoCA and RF interfaces completely
powered down.

#177139 - $15.00 USD Received 2 Oct 2012; revised 27 Oct 2012; accepted 7 Nov 2012; published 28 Nov 2012
(C) 2012 OSA 10 December 2012 / Vol. 20, No. 26 / OPTICS EXPRESS B55
 - Doze state: In doze state, non-essential functions are powered off with an additional
powering off of the ONU transmitter while the receiver remains on.
- Sleep state: In sleep state, non-essential functional blocks and both the ONU transmitter
and the receiver are turned off.
We digress here to mention the difference between the term ‘state’ and ‘mode’ as used in
the paper. Note that a mode represents the combination of states according to traffic load. For
example, doze mode is referred to as the cyclic transitions between power shedding and doze
state, and sleep mode as the cyclic transitions between power shedding, doze, or sleep state.
Sleep mode can be classified further as deep and fast sleep mode based on the periods of
sleep.
4.2 Progress and Challenges
In this section, we review progress and challenges for the implementation of power saving
modes.
- Transmitters: The switch on/off time of TXs is important, as a large switch on/off time
will increase sleep overheads and guard bands, leading to the decrease in energy efficiency
and throughput. The switch on/off time of the TX depends on whether it is a burst mode (BM)
TX or a continuous mode (CM) TX. The BM-TX has inherited better ability to switch on/off
and can be switched on/off within microseconds. On the other hand, the CM-TX takes about
1 ms to switch on/off [11] and thus cannot support very short sleep cycles. TDMA based
technologies like EPON and TWDM-PON use the BM-TX, and WDM-PON and AON use
the CM-TX. In addition, switch on/off time depends upon transmission rates; high bit rate
TXs have a longer switch on/off time than low bit rate TXs. To minimize further the
switching time and the PC of the TX, there have been some recent proposals in [12] and [13].
-Receivers: The ONU receiver consists of the APD (or PIN), TIA, and LA. When the RX
is powered down, the ONU loses the downstream signal, which is required to recover clock
by the clock and data recovery (CDR) circuit, and is used to maintain synchronization
between the OLT and the ONU. Paper [14] has shown that CDR can take as long as 5 ms to
re-acquire clock. To minimize clock recovery time, there are the proposals of the burst mode
CDR that uses a local oscillator to keep the ONU in accord with the OLT. The BM-CDR can
reduce recovery time to as low as 10 ns for the 1 G receiver and even less for 10 G, as the
higher bit rate CDRs can scan a higher number of bits in a shorter time. Note that, for NGOA
architectures, ONUs can employ BM-CDRs, which will lead to negligible overheads
associated with switching off a RX, and thus doze mode will not give any additional
dividends compared with sleep mode. Doze mode makes sense only for ONUs that are
already deployed in the field and are equipped with CM-CDR, wherein overheads associated
with sleep mode are high compared with doze mode.
-Digital Processing circuits: Digital processing circuits can be either clock-gated, power
gated or can be completely powered off [15]. Clock-gating disables (i.e., gating) the clock
signal of the register that feeds a portion of the combination logic that is not performing
useful functions. Clocks consume power because they continuously toggle registers and
removing clocks from the parts that are not useful saves PC. Power gating uses low-leakage
transistors to shut off the power supplies of the parts of a design that are not used. The
advantage of clock gating and power gating is that the components can be turned on/off in
microseconds. On the other hand, powering off a whole digital circuit reduces PC to zero, but
it takes a longer switching time (depending on the functionality) of up to tens of milliseconds.
For the analysis, we assume a power saving between 60 to 90% by using clock gating, or
power gating, or both.
-UNIs: UNIs also support power savings. SLIC modules available today support no on-
hook transmission. Low power idle (LPI) mode has been recently proposed for the GbE
interface [16]. In LPI mode, the transmitter sends out a periodic pulse to keep the receiver in
sync instead of the normally required continuous transmission between the transmitter and the

#177139 - $15.00 USD Received 2 Oct 2012; revised 27 Oct 2012; accepted 7 Nov 2012; published 28 Nov 2012
(C) 2012 OSA 10 December 2012 / Vol. 20, No. 26 / OPTICS EXPRESS B56
receiver. LPI mode can lead to power savings of about 50%. For MoCA interfaces, also
power saving modes like “wake-on-MoCA” have been proposed.
Table 2 gives the summary of switching time and power saving tradeoffs in considered
mechanisms.
Table 2. Switching time and power saving tradeoffs in considered mechanisms
Mechanisms Component Switching Power
ON/OFF time Savings
Doze mode (TRX) 1.25 G CM 1 ms 100%
1.25 G BM 10 ns 100%
2.5 G BM 200 ns 100%
10 G BM 500 ns 100%
Sleep Mode (RX) CM-CDR 1-5 ms 100%
BM-CDR 10 ns 100%
Clock Gating Digital Processing Blocks 5 µs 30-60%
Power switching Digital Processing Blocks 10 ms 100%
Low Power Idle GbE 5 µs 0-50%
Wake on MoCA MoCA MAC 5 ms 0-40%

5. Evaluation methodology
For the evaluation of power saving modes, we first discuss the proposed algorithm, sleep
mode aware (SMA). Then in the next section, we discuss the assumptions for the analysis.
5.1 Sleep Mode Aware (SMA) Algorithm
To evaluate sleep mode efficiency, a well-suited dynamic bandwidth allocation (DBA)
algorithm is required, which can optimize the cyclic transitions between the states according
to the traffic requirement. For this purpose, we apply an Ethernet PON (EPON) based
protocol. EPON is chosen as an example; the approaches proposed, however, are generic and
can apply to any system concept. Interleaved polling with adaptive cycle time (IPACT) [17]
is considered as an important example of an EPON DBA algorithm for scheduling upstream
transmission. IPACT, however, does not support sleep mode and thus we propose a new DBA
algorithm, which we refer as the SMA algorithm. In traditional approaches, the downstream
traffic is broadcasted to all ONUs and each ONU has to continuously hear the broadcasted
traffic. This leaves no opportunity for the ONU to sleep and it wastes energy, as the ONU has
to remain awake at all the time and process packets that are not destined for it. In the SMA
algorithm, the OLT buffers the downstream traffic for each ONU and only transmits it during
a pre-determined activity slot of the ONU. The ONU transmits upstream traffic and receives
downstream traffic only during this activity slot. This removes the requirement of the ONU to
be awake at all times and gives an opportunity for the ONU to sleep. It, however, necessitates
buffering even in the downstream direction and increases packet delay. SMA, like IPACT,
polls ONUs in a round-robin manner and issues GATE messages to every ONU in each cycle.
Now, the OLT computes the transmission slot (TS) as the function of the buffer backlog of
the downstream and the upstream traffic of the ONU according to:

TS  Min{
T cycle
, Max( B u , B d )} (1)
N u R Ru d

where Min/Max represents the minimum and the maximum value of the function, Tcycle is the
cycle time in which ONUs are polled, N u is the number of users, Bu and Bd are backlogged
upstream and downstream bytes for the ONU, Ru and Rd are the upstream and the
downstream data rate, respectively. SMA also safeguards against very frequent polling of the
ONU by assuring every ONU a minimum sleeping time (MST). If the TS (calculated by Eq.
(1)) is smaller than MST / Nu , then the TS equivalent to MST / Nu is chosen. This insures that

#177139 - $15.00 USD Received 2 Oct 2012; revised 27 Oct 2012; accepted 7 Nov 2012; published 28 Nov 2012
(C) 2012 OSA 10 December 2012 / Vol. 20, No. 26 / OPTICS EXPRESS B57
even at a very low load, the ONU polling time is not very short and the ONU does not wake
up frequently. Note that, the ONU may be allocated a longer upstream TS than it requested, in
which newly arrived packets between the time of a previous request and present grant are
transmitted. If the ONU, however, has no additional packet arrivals, it goes to doze mode.
The ONU sleeps for a period of Tcycle  TS . The proposed algorithm is adopted to study sleep
mode in considered NGOA architectures. For example, for WDM PON, PtP, and AON, N u
is chosen as 1.
In addition, the OLT communicates the next wake up time to the ONU. The next wake up
time of the ONU is the time of issuing the next GATE message of the ONU. The time of
issuing the next GATE message of the ONU may not be known at the time of issuing the
present GATE message; and thus the OLT predicts the time epoch of the issue of next GATE
message. Figure 2 (a) explains it more clearly. We have assumed two ONUs for clarity. Let
us assume that at time T, the OLT knows the buffer statistics of both ONUs and their round-
trip time (RTT). Thus, at the transmission time of the first GATE G1 to ONU1, the OLT can
easily calculate the grant time of the next GATE message for ONU1. At the time of issuing
the second GATE message for ONU1, however, the REPORT message from ONU2 has still
not arrived and thus the OLT cannot calculate the time epoch of the next GATE message for
ONU1. In the SMA algorithm, the OLT assumes a minimum TS for the ONUs of which the
REPORT messages have not arrived at the time of decision. When the REPORT messages
from ONUs arrive, we determine the grant time of the next (in cyclic order) ONU. Using the
latest determined grant time of the ONU k, we can calculate the minimum time epoch
 MTp [i  1] at which the (i + 1)th GATE message to the pth ONU is transmitted and is
formulated by:
MTp [i  1]  GTk [i  1]  rtt[k ]  rtt[ p]   (2)
where r tt[p] is the round trip time of the p ONU and  is the minimum TS of the remaining
th

ONUs (ONUs for which the REPORT has not arrived) as shown in Fig. 2 (b). Note that, the
actual TS of the remaining ONUs is (Δ + ). From Fig. 2 (b), we can see that the sleep
percentage (SP) of an ONU is TS / TG . Logical point-to-point systems like WDM-PON, PtP,
and AON will not suffer from the impairments due to wake up time prediction. To the best of
our knowledge, for logical point-to-point systems, no other algorithm has been proposed to
exploit the sleep mode functionality. Nevertheless, even for the logical point-to-point
systems, the SMA algorithm achieves high power efficiency without adding a high
complexity. However, for future research, the DBA algorithms focused on logical point-to-
point systems can be used for the analysis of their PC.

Fig. 2. (a) A TDMA-PON with an OLT and two ONUs showing REPORT (R1 and R2 for
ONU1 and 2) and GATE (G1 and G2 for ONU1 and 2) messages transmission. (b) GATE
prediction in the SMA algorithm.

5.2 Assumptions
We have simulated considered NGOA architectures using the OPNET simulation
environment and employing the DBA algorithm as explained in section 5.1. Service quality is

#177139 - $15.00 USD Received 2 Oct 2012; revised 27 Oct 2012; accepted 7 Nov 2012; published 28 Nov 2012
(C) 2012 OSA 10 December 2012 / Vol. 20, No. 26 / OPTICS EXPRESS B58
considered as an important requirement with sleep mode scheduling [7] and the SMA
scheduling fulfills this requirement. For the study, we have assumed a symmetric upstream
and downstream data rate ( Rd ) between 0 to 100 Mbps, in line with [9]. To meet this end,
1:8, 1:16, and 1:64 split are chosen for EPON, 10G-EPON and TWDM-PON, and 40G-
TDMA-PON respectively. We have assumed the reach of 60 km, the buffer size of 5 MB, and
three durations of cycle lengths: short (5ms), moderate (20 ms) and long (100 ms). Different
cycle lengths are adopted to study the effect of QoS requirements in energy efficiency. The
shorter cycle lengths reflect the stringent QoS requirements. For long cycle lengths, we have
assumed that digital components can be completely powered off; but at short and moderate
cycle lengths, we have assumed the clock gating and the power gating approach. Overheads
(To ) in sleep and doze modes are assumed to vary with a  20% deviation from the switching
ON/OFF times given in Table 2. During transition period (i.e. active to sleep state), PC is
assumed as half in active state. The synthetic user traffic is self-similar with a Hurst
Parameter of 0.8 [16] and with a packet size varying exponentially in the form of Ethernet
frames (64 to 1518 bytes). All ONUs are assumed to be symmetrically loaded. Note that,
though we have employed EPON based multi-point control protocol (MPCP) even for PtP,
WDM-PON and AON, the overheads due to the use of MPCP are negligible, as these
technologies have considerably higher line rates (1 Gbps) compared to the considered
maximum data rate per user, and the overheads in scheduling only become significant at a
high network load. Furthermore, the transmission of a MPCP control message requires only
0.512 µs, which is negligible compared to the sleep cycle lengths and the overheads of 1 ms
due to continuous mode transmission.
6. Simulation results
In this section, we evaluate the PC of considered NGOA system concepts. We first discuss
the PC in low power states and then in low power modes. Finally, we discuss the effects of
low power modes on quality of service (QoS) performance and memory requirements.
6.1 Low power states
Figure 3 gives the PC of NGOA networks in low power states, viz. active, shedding, doze and
sleep. The PC is split into the following parts: TRX, SoC, memory, SFBs, UNIs, power
conversion inefficiency and a variable part. The variable part represents the extra PC that
results from worst assumptions of power saving possible. For example, it represents the extra
PC due to the worst estimates of the benefits of LPI [16], and the power savings that can be
achieved by clock gating or power gating or power switching a digital processing block. For
sleep state, the variable component is further divided into two parts to show the influence of
fast and deep sleep. In fast sleep, not all the functionalities of a SoC can be turned off, and the
SFBs cannot be completely powered off. Whereas in deep sleep, the SoC functionality can be
essentially reduced to maintain an internal timer to wake up at its expiry or to respond to local
stimuli like the off-hook condition, and all SFBs can be completely powered off. In addition,
the contribution of the variable part is highest in 40G-TDMA-PON, as it has a significant
power contribution from digital processing blocks like EDC and SoC. Due to variable power
savings possible for digital processing blocks, the resulting power saving varies significantly.
For example, for 40G-TDMA-PON, the PC of digital processing blocks can be reduced to
zero if they are completely powered off at long cycle lengths or can be as high as 60% at
short cycle lengths, where the large portion of a digital processing block has to remain awake
for a faster switch on. The variable part for the best-case scenario will be zero. The system
concepts are arranged according to ascending order of the active state PC. 40G-TDMA-PON
has the highest PC in active state because of the use of OA, EDC, and the high downstream
and upstream bit rates. The power shedding state minimizes the PC due to UNIs. Note that
the PC of WDM-PON concepts in doze mode is same as PtP and AON, as it has similar

#177139 - $15.00 USD Received 2 Oct 2012; revised 27 Oct 2012; accepted 7 Nov 2012; published 28 Nov 2012
(C) 2012 OSA 10 December 2012 / Vol. 20, No. 26 / OPTICS EXPRESS B59
receiver and other requirements. In sleep state, the PC due to the TRX becomes zero, and the
considered technologies achieve a similar power consumption for deep sleep or at long idle
conditions.

Fig. 3. Power Consumption of NGOA architectures in low power modes

6.2 Low power modes

We investigate the PC of considered NGOA contenders in doze and sleep mode. Note that,
the analysis corresponds to busy hour traffic conditions. Due to the continual increase in
online behavior of users and the requirement that the lifeline telephone services should
always be available [7], it is essential to evaluate the PC in busy hour traffic conditions. In the
long idle periods, the PC of various contenders is same as in deep sleep state shown in Fig.
3(d).
We show the PC as the split of four parts: 1) primary part, which includes the PC of TRX,
SoC, memory, SFBs, and power conversion inefficiency; 2) data rate, which shows the
variation in PC due to the variation in data rate; 3) overheads, which include the PC because
of the variation in overheads assumptions; and 4) a variable part, which includes the PC
because of the variation in the PC of UNIs and digital processing blocks. Figure 4 shows the
PC of various NGOA system concepts in doze mode. The PC performance is shown for short,
moderate and long cycle lengths. The PC contribution due to overheads in NGOA system
concepts employing the CM-TX is only significant at short cycle lengths. No significant PC
variation is observed for the variation in cycle lengths. The power savings in doze mode are
limited between 50 and 75%. It is easy to see that the PC in doze mode  PDM  is a function of
the PC in shedding state  PPS  , doze state  PDS  and doze period TD  as:

Rd T
TD  Tcycle .(1   o ) (3)
Ru Tcycle

#177139 - $15.00 USD Received 2 Oct 2012; revised 27 Oct 2012; accepted 7 Nov 2012; published 28 Nov 2012
(C) 2012 OSA 10 December 2012 / Vol. 20, No. 26 / OPTICS EXPRESS B60
 Rd T R T
PDM  PPS .(  o )  PDS .(1  d  o ) (4)
Ru Tcycle Ru Tcycle
Figure 5 shows the PC and the power savings in sleep mode. The power savings in sleep
mode are found to vary between 72 and 92%. The PC in sleep mode is largely impacted by
the cycle lengths. The PC for all technologies will drop down with the increase in the cycle
length. All technologies, however, will benefit differently. The technologies with BM
transmission and reception are found to benefit more. At short cycle lengths, CM-TX
technologies will have an increased PC because of the effects of overheads. As the cycle
length increases, the impact of overheads reduces. 40G-TDMA-PON has the highest PC at
short and moderate cycle lengths because of the use of EDC and a high power consuming
SoC. At long cycle lengths, the PC in sleep state is reduced as the large number of functional
blocks can be powered off and the SoC functionality reduces significantly. Note that, at short
or moderate cycle lengths, the functionalities, like phase locked loop (PLL), are still
maintained because of the requirement of shorter wake up times. Furthermore, because of the
high downstream and upstream rate, 40G-TDMA can transmit and receive packets in the
minimum time and can, thus sleep for the maximum period. Because of the combination of
the maximum sleep time and the low PC in sleep state for long cycle lengths, 40G-TDMA-
PON has a minimum PC at the long cycle lengths.
12 100
Power Consumption

95
10

Power Savings
90
85
8
(W, bars)

(%, line)
80
6 75
70
4 65
60
2
55
0 50
PtP

PtP

PtP
XGPON

XGPON

XGPON
WDM-RSOA

WDM-RSOA

WDM-RSOA
AON

HB-TDMA

AON

HB-TDMA

AON

HB-TDMA
GPON

GPON

GPON
WDM-TL

WDM-TL

WDM-TL
TWDM-PON

TWDM-PON

TWDM-PON
Short Moderate Long
Cycle Length
Variable Overheads Datarate Primary Power Savings

Fig. 4. Power Consumption (W) of various NGOA systems in doze mode during busy hours
with three values of the cycle length: short (5 ms), moderate (20 ms) and long (100 ms). The
bars representing variable, overheads, and data rate components will be zero for the best case
scenario.

#177139 - $15.00 USD Received 2 Oct 2012; revised 27 Oct 2012; accepted 7 Nov 2012; published 28 Nov 2012
(C) 2012 OSA 10 December 2012 / Vol. 20, No. 26 / OPTICS EXPRESS B61
 3.5 100

Power Consumption

Power Savings
3 95

(%, line)
2.5 90

(W, bars)
2
85
1.5
80
1
0.5 75

0 70

PtP

PtP

PtP
XGPON

XGPON

XGPON
WDM-RSOA

WDM-RSOA

WDM-RSOA
AON

HB-TDMA

AON

HB-TDMA

AON

HB-TDMA
GPON

GPON

GPON
WDM-TL

WDM-TL

WDM-TL
TWDM-PON

TWDM-PON

TWDM-PON
Short Moderate Long
Cycle Length
Variable Overheads Datarate Primary Power Savings

Fig. 5. Power Consumption (W) of various NGOA systems in sleep mode during busy hours
with three values of the cycle length: short (5 ms), moderate (20 ms) and long (100 ms). The
bars representing variable, overheads, and data rate components will be zero for the best case
scenario.

6.3 Effect on QoS parameters and Memory requirements

We have seen that large power savings can be obtained by increasing the cycle length but the
increase in cycle length also impacts the quality of service (QoS) performance. The queuing
delay is an important QoS parameter, which increases with the cycle length. The increase in
the cycle length also enlarges the queue size, which scales the memory requirements. From
the simulations, we found out that the delay and the queue size increase linearly with the
cycle length, and all NGOA systems exhibited the same increase in the delay and the queue
size. Figure 6 shows the upstream queuing delay and the queue size (for upstream traffic at
the ONU) variation with the cycle length and the data rate (H = high (100 Mbps), L = low
(1.5 Mbps)).The queuing delay and queue size requirement for downstream traffic at the OLT
was also found to be the same. There is another difference among the technologies. 40G-
TDMA, TWDM, and 10G-EPON have a higher downstream data rate (> 1 Gbps), but have
UNIs with only 1 Gbps capability. Hence, either UNIs need to be scaled to higher bit rates or
the downstream traffic has to be buffered at the ONU. The UNIs with higher bit rates have
significantly higher power dissipation, and hence buffering is a preferred solution. For 40G-
TDMA, where the effect of buffering is most severe, there is the additional queuing delay of 2
ms and the increase in buffer size of 0.05 MB, which will have insignificant additional PC.

Fig. 6. (a) Queuing delay vs. cycle length (b) Queue size vs. cycle length. H (high = 100 Mbps)
and L (Low = 1.5 Mbps) data rates are evaluated.

7. Conclusions
Sleep and doze mode, which are proposed as a promising mechanism to reduce the power
consumption of the customer premises equipment, are investigated for next generation optical
access (NGOA) technologies like 40G time division multiple access (40G-TDMA) passive

#177139 - $15.00 USD Received 2 Oct 2012; revised 27 Oct 2012; accepted 7 Nov 2012; published 28 Nov 2012
(C) 2012 OSA 10 December 2012 / Vol. 20, No. 26 / OPTICS EXPRESS B62
optical network (PON), wavelength division multiplexing (WDM) PON, hybrid
TDMA/WDM PON (TWDM-PON), point-to-point (PtP) and active optical network (AON).
In active state, the power consumption profile of the considered technologies varies
significantly and is shown in Fig. 3(a). By application of low power modes, the power
consumption of the technologies can be reduced significantly. For long idle periods, the
considered technologies achieve similar power consumption. However, during busy hours,
the power consumption of the technologies depends upon the cycle length. During busy
hours, doze mode can reduce energy consumption between 50 and 75%, whereas sleep mode
can reduce energy consumption between 72 and 92%, and technologies with burst mode
transmission and reception, like 40G-TDMA-PONs, TWDM-PONs, 10G-EPON, and EPON
achieve a better power saving compared to continuous mode transmission and reception
technologies like WDM-PON, AON and PtP.
Acknowledgments
The research leading to these results has received funding from the European Community's
Seventh Framework Programme (FP7/2007-2013) under grant agreement n° 249025 (ICT-
OASE) and grant agreement n° 257740 (Network of Excellence TREND). We would also like
to thank the GreenTouch consortium for its valuable feedback and suggestions.

#177139 - $15.00 USD Received 2 Oct 2012; revised 27 Oct 2012; accepted 7 Nov 2012; published 28 Nov 2012
(C) 2012 OSA 10 December 2012 / Vol. 20, No. 26 / OPTICS EXPRESS B63