Thread Links Date Links
Thread Prev Thread Next Thread Index Date Prev Date Next Date Index

Re: [802.3_DIALOG] Considerations on technical feasibility of 1GE POF and being future proof



Peter and George:

Thank you for your feedback. 

I'd like to clarify some points concerning to worst case POF channel issue:
  • The 1mm SI-POF is categorized as A4a.2 in IEC 60793-2-40. I copied the basic transmission requirements specific for A4a.2 fibers in perezaranda_GEPOF_01_0914.pdf, page 33. I am not aware about the source to claim 0.4 dB/m attenuation.
  • The attenuation requirement of <= 18dB/100m is defined under specified measurement conditions, that is EMD (equilibrium mode distribution). The rational behind this is as follows:
    • The PMMA SI-POF bandwidth-length product as well as attenuation per length are parameters that depend on the mode distribution launching conditions, i.e. how the light power injected by the light source is distributed among all the excited modes of the optical fiber.
    • For step index fibers, the mode is very closely linked to the angle that light rays experience respect to fiber axis.
    • A very well known physical phenomena in PMMA SI-POF is the mode mixing. The energy of higher modes is transferred to lower modes and viceversa, as the light is propagated through the fiber, in such a way, the mode distribution is modified as the length of the fiber increase, until it get an equilibrium. This phenomena is produced by the rays scattering in the discontinuities and impurities of the core material and also affected by the characteristics of the core-cladding boundary. The equilibrium (EMD) is achieved when no changes in mode distribution are observed by means of e.g. far-end field measurement from a given length. This length is called equilibrium length.
      • The mode mixing, also called mode coupling, is well modeled by a diffusion partial derivative equation similar to heat transfer. This is the Gogle’s equation, to which many authors have proposed several improvements to match the equation solution with real lab measurements. 
    • Based on this it is easy to understand two extreme cases:
      • Light source with wide launching condition: the light coupled to fiber presents a mode distribution exciting all the modes with equal energy until acceptance angle defined by the proper numerical aperture of the fiber (NA 0.5 for A4a.2 POF). In this case, the bandwidth-length product will be lower for shorter distances than for longer distance, until the equilibrium is achieved, producing constant bandwidth-length product characteristic from equilibrium length. In the same way, the attenuation per length will be higher for shorter lengths than for longer lengths until EMD is achieved. One way to facilitate the mode mixing speeding up the EMD in shorter lengths is using a mode mixer as it is specified in annex I of IEC 60793-2-40, which do the work by mode conversion in the core-cladding boundary.
      • A light source with narrow launching condition, smaller than EMD NA of fiber. The bandwidth-length product in shorter fibers is higher and the attenuation per length smaller. This is a best case, that could not be considered, because any bending along the fiber will produce mode conversion, producing leaky modes and unpredictable bandwidth and attenuation.
    • The FOT manufacturers design light sources (e.g. LEDs plus coupling lenses) trying to get a launching condition similar to EMD, in such a way, the BLP and attenuation per length parameters keep constant. This criteria matches with the annex I of IEC 60793-2-40. When significant bending is produced in the fiber, the bandwidth increases, specially when it is located close to receiver, although attenuation will also be increased. When bending is located close to transmitter, a worst case consideration is only attenuation increase, because  mode recombination in the rest of fiber until reaching receiver eliminates the bandwidth increase. Because, full-duplex communication over POF is over 2 POF wires, does not make sense to consider bandwidth increases due to bending location in link budget analysis (the advantageous location for one wire can be only pure attenuation in the other wire). The worst case analysis should consider the bending as pure extra attenuation to be included in link budget. The extra attenuation produced by bending depends on radius. 
    • Manufacturing tolerances in producing FOTs (launching condition, lens focus adjustment, etc) as well as aging of PMMA core and maximum temperature and humidity conditions, is the reason to consider 0.2 dB/m attenuation in automotive environment and  0.19 dB/m in home applications. These numbers are coming from MOST and home applications experience during the last 15 years.
  • Summarizing:
    • Worst case channel conditions have been considered for GEPOF channel (perezaranda_GEPOF_1_0514.pdf):
      • POF attenuation, based on 15 years of experience (aging and FOT tolerances), under EMD launching specification.
      • Inline connectors attenuation considering tolerances and aging under EMD spec. for different markets.
      • Light source power variation, based on device characterization during the last 15 years of LED used in MOST (automotive)
      • Sensitivity requirements and link budget analysis
    • Shannon’s theory as well as real experiments under worst case conditions have demonstrated the technical feasibility
  • Parameters like: wavelength center,  wavelength width (max, min values), launching distribution masks, numerical apertures, max and min injected power, sensitivity, link budget, etc, in all the details that maybe are not defined in current international standards, have to be defined during the task force (like in any other optical PMD). 


Other comments that I would like to point out regarding to channel model and knowledge about non-linearity:
  • A set of presentations were presented during the Norfolk interim meeting, with objective of demonstrating the technical feasibility of gigabit over POF, under the link budget requirements also presented in the same meeting. In Norfolk were presented the characteristics of all the elements composing the communication channel.
  • The approach for demonstration by means of simulations the technical feasibility was to use well known concepts from Shannon’s information theory that are very easy to manage when the channel is considered as linear with additive noise. All the elements composing the channel excluding the LED are linear; of course, the DAC and ADC operations are not linear, but well known by everybody in this business (see  perezaranda_GEPOF_5_0514.pdf)
  • The LED non-linearity was analyzed in two ways in perezaranda_GEPOF_2_0514.pdf:
    • Real measurements of harmonic distortion of a MOST LED with several input tones and different temperatures. These measurements show the non-linear problem, but  show nothing about the morphology of it.
    • Non linear response based on Volterra’s truncated series, which is a very well know technique to model “soft” non linearities, that is the typical you can find in analog circuitry. I mean, Volterra models are not appropriate for hard discontinuities like saturations, dead-zones, non-linearities in a DAC nor ADC, etc. However, Volterra series result to be a good model for the LED device.
  • Why Volterra is a good model? - Because, under the same input signal, provide to us the same output of the real device. That is the point. And this is true for a LED operating from -40 up to 105 ºC. In perezaranda_GEPOF_2_0514.pdf are shown the Volterra responses for the different conditions.
  • How are calculated the Volterra models? - We use RLS adaptive filtering algorithm over a training sequence, because Volterra series can be seen as linear combination of non-linearities. 
  • The problem of equalization of non-linear channels is very old in the literature, because this topic has been investigated for read out channels of HDDs, and other devices. This is a complex topic and the optimum equalization technique always depend on the nonlinearity characteristics. The approach used in perezaranda_GEPOF_2_0514.pdf is linearization + equalization of a linear channel.
  • Linearization consists on compensating the Volterra kernels of order 2 and higher, but preserving the first order response of the channel, which could be optimally equalized by DFE or THP. DFE/THP non linear structure does not exist (it can be mathematically demonstrated), therefore, the linearizer operates like a linear equalizer, but for higher order, producing an small noise enhancement, this noise whitened by DFE or THP. 
  • This approach is implemented in KDPOF ICs, which are used in the experiments reported in the record of presentations. 
  • Finally, in perezaranda_GEPOF_2_0514.pdf, is shown how the linearizer + DFE operates for real Volterra channels obtained in the lab (LED + POF + receiver), and the capacity loss in terms of  SNR in detector is evaluated as a function of the effective SNR in the channel. We conclude that capacity loss is < 1 dB effective SNR < 30 dB. This cap loss is introduced as implementation loss in Shannon’s analysis for linear channel.
  • How is implemented the linearizer? - from perezaranda_GEPOF_2_0514.pdf page 33, it is clear that linearizer is a Volterra filter, like the channel model. Other topic, is how to calculate the coefficients, and I think it is receiver / implementor topic.  

Finally, some comments about using other wavelengths:
  • Attenuation of PMMA SI-POF as a function of light wavelength is well characterized by the POF manufacturers from decades ago. The IEC 60793-2-40 includes only specifications for 650 nm. The reason behind this is that optoelectronics for other wavelengths are not mature today or is not a reality in the market. This does not mean that for example a green light could not be used in the coming years to achieve 1Gbps over more than 100m SI-POF. The specification of all the parameters related to this wavelength operation could be defined in a future 802.3 annex.
    • In 802.3 there are a set of different PMDs that share a same PCS (e.g. clause 36), and not all these PMDs were published at the same date.
  • Feasibility at 450 - 580 nm range is a very well known technical topic. 
    • POF bandwidth: Chromatic dispersion significantly increases compared to 650 nm window. However, that is not the point, because modal dispersion is orders of magnitud greater than the chromatic dispersion for let say link lengths longer than 5 m, where in addition the bandwidth limitation of POF is negligible compared to bandwidth limitations of e.g. LED. The real experiments with green sources demonstrate that not only bandwidth, but also, impulsive response of PMMA SI-POF, is basically equal to that obtained with red light, if similar launching condition is used.
    • Optoelectronics: Si-PIN diodes optimized with antireflective coating for green range are available from decades, providing only 1.5 dB responsitivity loss respect to red range. Concerning to green LED,  the industry has not demanded high speed devices for communications. However, fast green LEDs with 350 MHz bandwidth are there and it could be considered in a future annex, after achieving mature knowledge in reliability, driving techniques, etc. The semiconductors are the same that are used today for lighting. 

_____________________________________
Rubén Pérez de Aranda Alonso
CTO
KDPOF - Knowledge Development for POF
http://www.kdpof.com
_____________________________________

El 08/10/2014, a las 19:41, Carlos Pardo <carlospardo@xxxxxxxxx> escribió:

FYI


-------- Mensaje original --------
Asunto:Re: [802.3_DIALOG] Considerations on technical feasibility of 1GE POF and being future proof
Fecha:Wed, 8 Oct 2014 16:42:52 +0000
De:George Zimmerman <george@xxxxxxxxxxxxxxxxxxxxxxxxxxxxx>
Responder a:George Zimmerman <george@xxxxxxxxxxxxxxxxxxxxxxxxxxxxx>
Para:STDS-802-3-DIALOG@xxxxxxxxxxxxxxxxx


Peter –
I thank you for your diligence.  I have been considering these issues, and find as some of these arguments overreaching.  Particularly the ‘future proof’ discussion for the home network market.  I would note first that the most prevalent home networking technologies are not even considered in the presentations from September (a hybrid of wired BASE-T Ethernet and advanced 802.11 wireless – and yes, I know that in some markets these have been argued as unavailable by construction and regulation, but the presentation environment was not limited by those considerations). 
What I see are market justifications that don’t necessarily get linked to objectives.  For example, if future proofing were to be considered as important to enabling this market, the objectives should make some provision for an autonegotiation protocol (at least an optional one) – to build a base for that future.
 
I will have to leave the optical testing issues to your analysis.  Coming from the copper world, when we have simulation models they are fairly detailed, and call out each of the residual impairments, using widely known and demonstrated techniques for cancellation. I don’t get that same feeling from the analysis I have seen in the record of presentations. I have been reviewing the detail of the presentations but get the impression that they are predominantly using a linearized model on a channel known for nonlinearities.  While some reference in the documents is made to signal processing for removing nonlinearities, the required accuracy and the performance limitations of those in the target environments is neither well known nor tracked in multi-vendor presentations.  I am not saying it cannot be done – just that technical feasibility and noise components for this channel need to be better shown before they live up to the ‘effects are well known and understood’ criteria for technical feasibility.   Hopefully these will be addressed in the tutorial in November.
 
George Zimmerman
Principal, CME Consulting
Experts in Advanced PHYsical Communications Technology
310-920-3860
 
(PLEASE NOTE NEW EMAIL ADDRESS.  THE OTHER WILL STILL WORK, BUT PLEASE USE THIS FOR CME BUSINESS)
 
From: Peter Stassar [mailto:Peter.Stassar@xxxxxxxxxx] 
Sent: Wednesday, October 08, 2014 8:09 AM
To: STDS-802-3-DIALOG@xxxxxxxxxxxxxxxxx
Subject: Re: [802.3_DIALOG] Considerations on technical feasibility of 1GE POF and being future proof
 
Hi Bob,
 
Thanks for your email response sent about 3 weeks ago.
 
I would like to start to re-emphasize that my main concern was in the end of the email which I sent on 22 July:
In conclusion I have not been convinced that 1Gb/s Ethernet over 50m of worst case POF  is technically feasible and even more that an installed POF link can be upgraded to higher speeds, which I believe is necessary to justify long lifetimes of home-installed POF networks.”
 
As you suggested I took a close look to the presentations to the meeting in Ottawa, 4 weeks ago.
Before commenting to those I will first address the 5 points from your email. I inserted comments in “red in the thread below.
 
So back to the September presentations.
In Eugene_GEPOF_01_0914 I see a strong claim that POF is future proof and that it is superior to existing technologies like twisted pair, coax and power line. This presentation also says “Tomorrow SMART HOME needs future proof networking medium”, which is a statement which I fully support and, I believe, have always clearly stated. This is however also the weak point of all presentations which I have seen so far. I continuously fail to find any evidence that POF and the 1GbE technology to work with it, is sufficiently future proof. In Tsukamoto_GEPOF_02_0914 the following is said  “Customers may use the network over 30 to 50 years. Is GbE enough in such a far future?”. This is exactly the point I tried to make during the CFI consensus building meeting in Beijing and during the San Diego meeting. It’s my strong belief that 1 Gb/s is NOT sufficiently high speed to justify deployment of POF in the home. Some references have been made to shorter wavelength windows of POF which could provide lower attenuation than at 650nm. Having looked at the Ottawa presentations I see some inconsistencies. While in Eugene_GEPOF_01_0914 I see a claim that in the range 450 to 580 nm the attenuation will be lower than at 650 nm, I conclude from Tsukamoto_GEPOF_01b_0914 that the POF standard is only specified for the 650nm window and that therefore operation is other wavelength windows is not specified and thus not guaranteed. A lower attenuation in the range 450 – 580nm would indeed be great, but what will be the bandwidth of POF in that wavelength range? It is not trivial that it will be the same or better or worse. I am under the impression that POF is a kind of multimode fiber and bandwidth is strongly dependent on the wavelength.
 
In conclusion, having looked in detail at the presentations to the Ottawa meeting and the clarifications in your email, I am still of the opinion that insufficient evidence has been provided that the proposed solutions are capable of multi-vendor interoperability (capability of developing a multi-vendor interoperable transceiver design from the specification under worst case conditions) and that the POF medium is sufficiently future-proof (or in other words upgradeable to speeds well in excess of 1Gb/s) to justify installations in the home for many decades to come.
 
Kind regards,
 
Peter Stassar, 施笪安
Technical Director, 术总监
Huawei Technologies Ltd, 华为技术有限公司
European Research Center, 欧洲研究所
Karspeldreef 4, 1101CJ Amsterdam
The Netherlands
Tel: +31 20 4300 832
Mob: +31 6 21146286
 
From: ROBERT GROW [mailto:bobgrow@xxxxxxx] 
Sent: Monday, September 15, 2014 6:57 PM
To: STDS-802-3-DIALOG@xxxxxxxxxxxxxxxxx
Subject: Re: [802.3_DIALOG] Considerations on technical feasibility of 1GE POF and being future proof
 

Peter:

Thank you for documenting your concerns following the July plenary meeting.  We reviewed your concerns during our interim Study Group meeting, and I offer some general response.  I believe these thoughts are consistent with the consensus of the GEPOF SG.  Detailed responses to your specific points will likely follow from me and others.  A few general points:

1.  We have certainly noted your acknowledgement that you are satisfied with POF applicability for two of the three markets our participants want to address (i.e., automotive and industrial).  Study Group members though did express some confusion because many of the presentations you choose to comment about were clearly only addressing requirements for the automotive market.  Please see September presentations for additional support for the application to home networking.

Peter: I just looked at the available documentation. I understood that many were addressing the 15m automotive market.

2.  In response to your repeated comment “15m (typical?)”, yes 15  meters is the requirement for automobiles.  That is reflected in one of our link length objectives for 15 meters with four in-line connections.  If you look, you will see similar 15 meter objectives for p802.3bp and p802.3bw. 

Peter: I do understand the 15m objective. I only put some questions marks on the testing done on 15m links. In the optics world a laboratory test on a TYPICAL fiber is completely different from a test on a worst case fiber under worst case operating conditions. Worst case conditions have a big impact on maximum loss and (I assume for POF) also on bandwidth. Therefore optics people are generally not convinced by a laboratory test on just 15 km of typical fiber. What kind of margins are available, etc.

3.  You also frequently commented on the lack of detail on modulation techniques.  We apologize for that, your question would have been answered if you had been able to attend SG meetings.  In defense of SG presenters, please understand that most of our participants helped develop a VDE specification for operation at up to and including 1 Gb/s with link lengths consistent with our automotive and home reach objectives.  Consequently, much is assumed to be a given with those folk, and it is perhaps excusable that presenters did not have an outside reader trying to understand everything from only the presentation in mind when preparing the presentation. 

Peter: I am of the opinion that details on how to actually “do it” needs to be provided by presentations and not via verbal clarifications. I understand that often things are a given for a limited group of people, but I also think you are very much aware that in 802.3 one needs to convince 75% of the plenary and not only the “insiders”.

There are implementations of the VDE specification from multiple vendors.  All tests that used a PHY used one of these implementations.  If you look at September presentations, you will see that additional presentations on testing include detailed one slide summaries of the characteristics of the VDE PHY.

Peter: Thanks for pointing this out. It would be great to see some evidence of those different implementations and interworking testing. I am under the impression that all test results shown at the September meeting (Lichtenegger_GEPOF_0914 and perezaranda_GEPOF_01_0914)  were from the same PHY vendor and not between PHYs from different vendors. So it would have been very interesting to see results from interoperability testing between devices from different vendors.

4.  You also repeatedly commented that simulation was not sufficient for justification of technical feasibility.  While you are entitled to that opinion, but give no rationale why simulation is not valid for demonstration of technical feasibility, I would point out that we have approved projects in the past based on simulation (in my personal experience as far back as 1000BASE-T).  Please also note that simulation is listed in IEEE 802 rules as an acceptable evidence of technical feasibility.

As is true for most engineers, the participants in our SG find it useful to look to both simulation and lab testing and when possible implementation tests expecting consistency.  As noted in my point 3, many of the test presentations you critiqued used an existence proof of technical feasibility – an implementation of the VDE specifications tested at 1000 Mb/s.  We do not only relay on simulation as some previous 802.3 projects have.

Peter: regarding to your statement “While you are entitled to that opinion, but give no rationale why simulation is not valid for demonstration of technical feasibility, I would point out that we have approved projects in the past based on simulation (in my personal experience as far back as 1000BASE-T)”, I would like to point out that for optical PHYs many years of (bad) experience have taught most optical engineers that simulations in the optical field in most cases need to be confirmed by experiments in order to judge the accuracy of the simulations. That’s why I (and probably not only myself) keep repeating that to demonstrate technical feasibility of Optical PHYs results from simulations are generally not sufficient.

5.  You also repeated in your comments a question about worst case channel.  Please note that the environmental requirements for automotive applications have higher temperature, greater EMC, and specific connector requirements.  Industrial applications similarly will also have higher temperature than home, and greater EMC challenges.  While vibration effects on optical power is important for the automotive channel, it is not considered relevant to the channel for nome networking, so our participants have looked at the effects of vibration on optical power, and consequently the automotive channel will include losses for vibration.  With our three major markets asking for a standard POF solution, and different objectives that address those markets, it is appropriate for a presentation to focus only on one particular market application, and unfailr for you to criticize those presentations for not also covering the home network application.  

Peter: on the topic of worst case channel I am NOT referring the range of environmental characteristics, despite the fact that they do play a significant role. I am referring to an optical link model with worst case characteristics appropriate for a distance of at least 50m. A test with a spool of 50m of typical POF is generally not regarded as testing a link under worst case conditions. I am under the impression that in many in-force optical specification in 802.3 a lot of work has been done on defining a worst case channel. This generally is defined such that it will support manufacturers in designing, manufacturing and testing transceivers. Performing a test on a spool of 50m POF is indeed proof of concept, but by far not enough to claim complete technical feasibility, or in other words sufficient maturity of the technology that it actually can be manufactured in high volume and with a high yield, work on worst case links of 50m POF in the home in a multi-vendor environment. I am afraid that this has not yet been demonstrated. The fiber shown on page 16 of perezaranda_GEPOF_01_0914 had an attenuation of 0.19 dB/m, but according to page 5 of Tsukamoto_GEPOF_01b_0914.pdf, IEC 60793-2-40 category A4a.2 has a maximum loss of 0.4 dB/m.  What was the bandwidth of the test fiber and how does that relate to the worst case specification?

Furthermore it was not my intent to criticize those presentations not covering the home market. If I came across to criticize them then I apologize for that. I was just putting remarks to those presentations what they were aimed at and what they were not aimed at.

Please see September and upcoming November presentations for additional support of feasibility for home networking.

—Bob