Folks,
Here is a summary of my analysis of Shaw’s presentation on 25G ONU options for increasing network capacity in 100G EPON:
- There are 9 use cases based on capacity shared by 32 ONUs per PON:
- - Use case 1 = 25G ONU with 0.8 Gbps Sustained Speeds per ONU
- - Use case 2 = 25G ONU allowing 8.3 Gbps Flagship Speed on the PON
- - Use case 3 = 25G ONU allowing 12.5 Gbps Billboard Speed on the PON
- - Use case 4 = 50G ONU with 1.6 Gbps Sustained Speed per ONU
- - Use case 5 = 50G ONU allowing 16.6 Gbps Flagship Speed on the PON
- - Use case 6 = 50G ONU allowing 25 Gbps Billboard Speed on the PON
- - Use case 7 = 100G ONU with 3.1 Gbps Sustained Speeds per ONU
- - Use case 8 = 100G ONU allowing 33.3 Gbps Flagship Speed on PON
- - Use case 9 = 100G ONU allowing 50 Gbps Billboard Speed on PON
- There are four kinds of λ pairs: λ0, λ1, λ2, λ3
- An O-band fixed channel for λ0 makes the most sense to achieve the lowest costs (less burst mode dispersion compensation to deal with at the OLT) & Laser costs at par with 10G EPON.
- λ1, λ2 & λ3 should be DWDM in C-band and DWDM (to be in the contiguous tuning range of a Tunable DBR laser (limited to 10 nm))
- There are 7 kinds of ONUs Types A,B,C,D,E,F,G
- - ONU Type A = 1X relative cost = 25G/λ0 (1 O-Band ch.)
- - ONU Type B = 1.5X relative cost = 25G/λ1 (1 C-Band DWDM ch.)
- - ONU Type C = 1.5X relative cost = 25G/λ2 (1 C-Band DWDM ch.)
- - ONU Type D = 1.5X relative cost = 25G/λ3 (1 C-Band DWDM ch.)
- - ONU Type E = 2.5X relative costs = 50G fixed on λ1..2 (Two C-Band DWDM ch.)
- - ONU Type F = 3.5X relative costs = 25G Tunable in Tx & Rx across λ1..3 (1 Tuneable C-Band DWDM ch.)
- - ONU Type G = 5.5X relative costs = 100G fixed on λ0..3 (1 O-Band + 3 C-Band DWDM ch.)
- There are 3 options for network expansion all the way to 100 Gbps short of use cases 8 & 9: 1, 2A & 2B
- - Option 1 uses ONU Types A, E ,G
- - Option 2A uses ONU Types A, B, C, D
- - Option 2B uses ONU types A & F
- The relative costs for use cases 1 to 7 for each option are calculated as follows:
- - Option 1 = 8*1X + 8*2.5X + 16*5.5X = 116X
- - Option 2 = 8*1X + 8X 1.5X + 8*1.5X + 8*1.5X =44X
- - Option 3 = 8*1X + 24 * 3.5X = 92X
- The relative costs for use cases 8 & 9 are not calculated as depending on the number of ONUs servicing 33.3 Gbps Flagship or 50 Gbps Billboard Speeds on the PON
From draw I propose to draw the following conclusions:
- Use cases 4, 5, 6 & 7 associated with 50G & 100G ONUs can be accommodated by Type F ONUs
- Type F ONUs avoid use of CWDM/DWDM Mux Demux inside ONU
- Type F ONU can replace use of Type B,C,D ONUs
- Type F ONUs can satisfy Use Cases 1 to 7
Shaw’s presentation supports my postulate. Should the relative costs of the Type F ONU be reduced from 3.5X to become that of a Type B, C or D ONU, i.e. 1.5X, then the relative costs of Option
2B becomes 8*1X + 25*1.5X = 44X, which is same as 2A. Another option would be 2B’ based on 32 * 1.5X = 48X with the number of ONU types being reduced to 1, should λ0 is be the same
tuning range as λ1..3. Since Option 2A with a relative cost of 44X requires four kinds of ONUs, its deployment
OpEx will be much greater than option 2B or 2B’.
Therefore I offer the following proposal which may accommodate the best of both worlds. Having a low cost λ0 at
part with 10G EPON, along with that of accommodating the use case of allowing Type F ONUs to be used on λ0, thus allowing all users with Type F ONUs to make use of 100% of the capacity
of the 100G E-PON. Further, a low cost Type E ONU which may be able to be manufactured with two instances of a Type F ONUs and a Type G ONU may be able to be manufactured with four
instances of a Type F ONU. Further, should the number of channels be greater than 4, the ability to re-use multiple instances of Type F ONUs for channel bonding will make sense. Furthermore, should there be more than one instance of 100G EPON on two sets
of 4 channels on the same PON, ONUs based on Type F components tunable across all instances of 100G EPON will allow the same ONU to roam across all instances of 100G EPON.
The proposal is to accomodate at least a a fifth channel λ4 on top of λ0..3,
only applicable in the direction of the ONU to the OLT. This 5th channel would not need to exist in the direction of the OLT to the ONU, as the OLT will not need tunable transceiver like in NG-PON2. Further, unlike NG-PON2, in NG-EPON, it is not yet anticipated
that an external wavelength multiplexer would be utilized and rather, it seems to be expected that any increase in speed on the PON will require swapping out the OLT transceiver with a single 25 channel pair with one with two or four instances of 25G channel
pairs and that the OLT transceiver will include an integrated Mux/Demux. NG-PON2 allows for an external mux/demux with individual 10G channels coming from different OLT ports.
The reason why the 5th channel will not be required in the direction of the OLT to the ONU, is that I expect the 1.5X relative cost of Type F ONU to remain exactly the same for an ONU capable
of Tunable Rx across λ0..4 versus that of λ1..3, given the use of discrete tunable filters in front of a
single instance of a wideband avalanche photodetector. This is unlike the case of a III-V gain chip, which is limited in its gain range with the laws of physics defying the possibility of covering O-band and C-band with a single III-V chip instance.
The use case I fear is that of having to implement λ0 inside a Type G ONU requiring to include an O-Band as well as a
C-Band transmitter in the same ONU, which would not make sense from the perspective of a photonics integrated circuit design, as well as would nullifying the benefits of allowing NG-EPON to evolve its tuning range beyond that of four channels.
CONCLUSION: Thanks to Shawn for putting together this nice relative cost framework fulfilling IEEE’ policy for openly discussing about the costs of implementing standards. I have used it to
show how the relative costs for Option 2B (44X) could become at par with that of Option 2A (44X). I have also show how the
marginally higher relative cost of Option 2B’ (48X) (if λ0 is in the same tuning range as λ1..3)
would evidently have a lower OpEx for operators than Option 2A (as only 1 kind of 25G ONU would be needed in Option 2B’ versus that of four kinds of ONUs in Option 2A). I have used the
framework to demonstrates that a single kind of Type F ONU could cover 7 out of the 9 use cases of 100G EPON.
Since the deployment of 100G EPON is years away, users of 100G EPON will be disadvantage once Type F ONUs become available at a 1.5X relative cost over that of Type A ONUs, should 100G EPON
not accommodate use of Tunable ONUs.
-=Francois=-
--
Francois Menard
CTO & Co-Founder
AEPONYX inc.
Cell: +1 (819) 609-1394
E-Mail: francois.menard@xxxxxxxxxxx
I put together the attached presentation to try to move the 25G ONU philosophy forward. I will not be at Macau but maybe this will help the rest of the group. From
the Preface slide:
w
Limiting wavelengths and/or optics capability (i.e. wavelength-tunability) for 25G ONU makes writing the standard simpler, may reduce R&D expenses for
vendors, and may provide lower product costs due to increased manufacturing volumes.
w
More wavelengths and optics capability (i.e. wavelength-tunability) for 25G ONU may reduce CapEx and/or OpEx for operators to deploy additional wavelengthsλ1,
λ2,
λ3.
It may not.
§
Lowest cost for optics at
λ0must
remain #1 priority no matter what
w
Excluding capabilities in the standard at this early stage risks:
§
PON solution that is too expensive to deploy for λ1, λ2, λ3 -> bad for everyone
§
re-writing the standard in the future
§
Standard could have provisions for these capabilities and allow time & market forces to determine if they make sense to develop and deploy
w
Either way, a decision needs to be made quickly so we can move forward on the standard
At the end, I propose a straw vote question on this topic.
Any feedback is appreciated.
If anyone likes the graphics in the presentation, you are welcome to re-use them.
Francois,
I’m saying that ideally 25G ONU and 50G ONU can be fixed or tunable if not disallowed by wavelength selection (and technical feasibility)
On point 2 I agree with Marek
-Barry
Francois,
Just one note: we are already using 1260 – 1280nm band for 10G-EPON. XGS-PON is nothing more than 10G-EPON in disguise (yes, they did change a few things at logical
layer, so what). I am not sure though what point you’re trying to make – reuse of 1260-1280 band for NG-EPON would block the coexistence / backward compatibility with 10G-EPON, which is one of our goals for the project.
If we forgot backward compatibility with 1G-EPON, we have plenty of real estate in O band to work with, without the need to kill off compatibility with 10G-EPON.
Marek
Barry wrote > I agree with Ed that the second, third fourth channel is where we can start adding the cost and
this is where a tunable may make sense. Of course the cost of a tunable must not exceed 2x the cost of a non-tunable.
Barry, are you making the point that until such time the OLT has a Gen2 or Gen3 transceiver on the PON, it makes no sense for the ONU to be tunable ?
My view is that even if the OLT has a Gen2 or Gen3 transceiver, it would still make sense for a Gen1 ONU to be tunable across all two channels of a Gen 2 OLT transceiver or across all four
channels of a Gen 3 OLT transceiver. To the same end, a Gen2 ONU could be tunable on both of its channels, so it could roam on any of the four channels of a Gen3 OLT transceiver. Only Gen3 ONUs would not take advantage of tunability, for having a quad transmission
/ receiver array. Of course, if the number of channels is greater than four ( NG-PON2 already has eight), such as the use case I’ve pointed to earlier where two NG-EPON instances on two sets of four channels could be there on the same PON and the 100 Gbps
ONU could roam any of its four channels onto any of the channels of the two instances of NG-EPON on the same PON.
Barry wrote > If the goal is to just
provide a lot of inflexible cheap incremental bandwidth (not a bad goal) than we can declare that all ONU will use 4 CWDM channels in the O band and be done with the discussion.
I’m not sure CWDM is an adequate solution at this bitrate. I’ve been meaning to run some simulations on the effect of four wave mixing in the O-band at high bitrates. I like the idea that
gain chips with quantum dots are potentially easier to make in the O band (and less expensive) than in the C band, allowing better resiliency to back reflections and possibly sparing the costs of having to package an isolator in the ONU. Finally, I’m not
sure CWDM technology lasers have the required linewith to combat chromatic dispersion at such a high bitrate, even if its 25 Gbps over distances of 20 km. I think there is merit in being able to consider the 1260-1280 spectrum proposed for XGS-PON for more
than a single channel.
Absolutely. A gen-1 25 Gb/s single-lane EPON should provide 2.5x more bandwidth than 10G EPON at less than 2.5x the cost. That won’t happen if you toss in a tunable laser (and
tunable receiver filter) in the ONU, and put wavelengths on a DWDM grid. Let’s defer that cost and complexity to the 2nd lane.
Ed
Somehow, just because the end target is 100G-EPON, people think that we need to utilize the 100G EPON capacity from day one. And if we don’t have 100G ONUs at day
one, then we have to fill this capacity with 25G ONUs.
This is not what we set to do. The generation 1 is a
single-lane EPON. Yes, starting with tunable optics and utilizing 4 lanes will provide 4x of sustained throughput per ONU. But that would be at more than 4x the cost.
According to my latest market data, the cost of tunable ONU optics today is ~9x that of 10G/10G-EPON ONU optics. Projections out to 2020 show it to drop to 3.5x the
cost of 10/10 ONU optics, which is still very high.
Is this not why after completing NG-PON2 standard, SG15 shifted focus to XGS-PON that uses a fixed single wavelength 10/10 optics?
If the same cost ratio between tunable and fixed optics remains for the 25G tunable and 25G fixed, then the cost of 50G ONUs will likely be lower than tunable 25G
ONUs. A 50G ONU obviously can burst at 50Gb/s peak rate, but additionally can operate as a 25G ONU on either of the channels, or even can operate as two independent 25G ONUs. So, we may spend time and efforts developing the first generation based on tunable
optics, but then why wouldn’t operators just skip gen 1 and go directly to 50G ONUs with 2 fixed channels?
Glen
Here is what I understand so far:
Per what Glen has presented: The OLT starts with a Gen 1 transceiver, which is stuck at 25 Gbps until it is replaced
with a Gen 2 at 50 Gbps. Only the OLT transceiver is replaced with a Gen 3 transceiver, would it then become possible to add 100 Gbps ONUs on the PON. With a Gen 1 OLT transceiver on the PON, 100 Gbps ONUs would be limited to 25 Gbps.
However, in NG-PON2, the use of an external WM allows for different OLT ports (or different OLT’s) to be the source
of the additional instances of 10 Gbps channel (up to 8 from 8 different line cards or OLT shelves is allowed). Therefore this allows pay as you grow, in service, with no downtime without requirement of retiring out OLT transceivers. Is this a benefit or
a pain in the rear end for operators ? Benefits allow for greater reliability, pay as you grow from cheaper 10 Gbps fixed XFPs/SFP+ with burst mode receivers. Pain in the butt means dealing with the WM and increased footprint.
With regards to the benefits of being able to get a 25 Gbps Tunable Tx / Tunable Rx ONU to roam across channels, here
are the benefits:
-
Serviceability of the PON port. If an ONU can move to another channel while the one that is down is being serviced, then everybody is happy.
-
Enhanced average throughput on the PON at the expense of peak speed. For instances, with 32 ONUs and 8 lambdas @ 25 Gbps, average throughput would be 25 Gbps / 4 = 6.25 Gbps rather than 25 Gbps / 32 = 0.78 Gbps,
which is 8 times greater. Perhaps this is not required for all applications, but as soon as you have mobile fronthaul, or business services on the PON, this capability becomes important.
-
As soon as the tuning range at the ONU exceeds the number of users on the PON (and you have an equally sized multi-wavelength Comb laser + Rx Array in the OLT transceiver like what is being done in the OpenOptics
MSA), then it becomes possible to turn off burst mode operation both for transmission on the ONU as well as for reception in the OLT. You’re now doing WDM-PON on power splitters. Each user gets a dedicated channel. This is where we want to go in the end
with all of this.
-
Achieving greater 4 times the throughput of the maximum line rate of a single channel, is all about packaging the cost structure of 4 tunable ONUs in one ONU and to have the MAC and ASIC to deal with the four
transceivers. It is perfectly possible to imagine the OLT transceiver would support 32 channels and that an ONU would only be able to bond 4 channels. Then the question becomes which 4 channels to bond in a pool of 32. If the ONU is not tunable, then
you are not able to take advantage of assigning different bonded channels to different ONUs.
Francois Menard
CTO & Co-Founder
|