The evolution of PON

The Australian Network Operator's Group (AusNOG) meetings have always managed to assemble a program that represents a fascinating window into the diversity of activities in the Australian communications environment. This year the program included the use of femtocells raised to a useful height by tethered drones to provide fast response to areas affected by natural disasters, the construction of radio links in remote locations, and the fascinating use of fibre optic systems as acoustic microphones. Here are my thoughts in response to a presentation by Ciena's Pradap Rajagopal on the evolution of Passive Optical Network (PON) access networks.

The evolution of wired access networks for suburban reticulation has been driven by a special set of economic and technical circumstances. In general, consumers are resistant to paying a very high fee to have their residence connected to a high-speed access data network.

To illustrate these issues, a summary of the costs in the US five years ago noted that it was costing the network constructor some USD 1,000 (on average) to pass each urban or suburban house on average, and if the house elected to connect to the service, then another USD 800 would be spent on the tail line to complete the connection. All of this was on par with the experience of many other FTTH deployments. While the government-funded National Broadband Network (NBN) in Australia used the mandatory conversion of every suburban home to this NBN service, in most other environments it's an optional decision by the consumer.

In the American context, the take up of high-speed broadband is a consumer choice, and less than 40% of households were taking up the option. This means that each connected service has to cover not only the USD 800 tie line plus the USD 1,000 per house network costs of 'passing' the house with FTTH, but, assuming that this service is deployed in a revenue-neutral or even positive manner, the total revenue from all households that opt to take the service needs to cover the total network deployment costs, effectively including those costs of passing the 'missed' houses that do not take up the service.

This implies that the 'miss rate' has a big impact on service costs. For example, with an average per passed-house cost of USD 1,000, and a 'miss' rate of 40%, instead of costing USD 1,800 per household that would result from a 100% service take-up rate, each connected service costs some USD 3,300 per household. That's a premium of 83% to cover the costs that result from partial take-up of the service. In many local markets, the alternatives to FTTH, namely DOCSIS, xDSL, and even 4G/5G services, look a whole lot cheaper, to both the network operator and the consumer.

The outcome of this situation is that it may take many years to achieve a positive return on the capital invested in building the wired access network. It's no surprise that the copper pair telephone access network had such a long service life. There was just no economic motivation to upgrade this network, and, on the whole, it was cheaper and easier for the network operator to deploy equipment that improved the digital signal processing capabilities at both ends of the access network than it was to perform any broad upgrade of the wired infrastructure.

The shift to DSL technology and the simultaneous deployment of high-speed mobile services also meant that the margins of revenue opportunity for a new generation of access technology, namely optical fibre cable, were small enough to present some significant challenges to both public authorities and network operators.

While each copper pair subscriber was provisioned with their own discrete physical copper pair from their location to the network's access points (or central office), this approach was not feasible for a fibre-based access network. The commonly adopted solution was to use a form of shared infrastructure that shares a common fibre run from the Optical Line Terminal (OLT) located in the Central Office (CO) to an optical splitter, that then sends the same optical signal to each of the attached Optical Network-termination Units (ONUs). This system is entirely passive in that the signal splitting (and combining in the upstream direction) is performed at the optical level and there are no active electronics within the network.

OLTs in PON networks typically support between splitting the single optical feed to between 16 and 32 ONUs, and in certain cases, up to 256 ONUs. The point-to-multipoint topology of PON requires different modes for downstream (OLT to ONU) and upstream (ONU to OLT) transmission. In the case of downstream transmission, typically all ONUs receive the same optical signal from the OLT, allowing the OLT to broadcast in continuous mode and requiring the equipment attached to each ONU to filter out all data frames that were not directed to that particular ONU.

For upstream transmission, the ONUs cannot all broadcast in continuous mode simultaneously. Instead, the ONUs transmit upstream signals in burst mode and share the upstream channel using time-division multiplexing through a dynamic bandwidth allocation process. For a Time-Division Multiplexing (TDM) PON, each ONU is assigned specific timeslots to transmit a burst of upstream data, and at any given time, the OLT is only receiving signals from a designated ONU. Because of the potential varying distance between the OLT and the ONUs and the random phase of the burst mode packets received by the OLT, the phase and amplitude variation in the PON network needs to be compensated by adopting clock and data recovery.

The first generation of PON technology, Broadband Passive Optical Network (BPON), was specified in ITU-T G.983 and featured a shared downstream data rate of 622Mbps and an upstream rate of 155Mbps. The signal was encoded using on/off keying and used 1530nm (C-band) for the downstream signal and 1410nm (L-band) for the upstream signal (the nomenclature for these optical bands is shown in Figure 2).

PON networks have seen a steady stream of improvements, though the optical components - such as fibre cables and passive optical splitters - remain unchanged. The advancements have primarily involved changes in optical wavelengths and the optical carrier signal.

In 2003 GPON (ITU-T G.984.x) was released, which defined a 2.5Gbps downstream capacity at 1490nm and 1.25Gbps upstream using 1310nm lasers. The framing of the signal used Asynchronous Transfer Mode (ATM), extending the conventional framing format used in broadband DSL systems to these PON systems. Soon after, in 2004 the IEEE published a specification for EPON, IEEE 802.3ah, which was a 1.25Gbps symmetric system using Ethernet framing. EPON has weaker levels of error correction and has lower split ratios, but uses 1310nm optics, and this represented a significant cost efficiency over the 1530nm optics used by GPON at the time.

The higher speeds in these two systems relied on advances in Digital Signal Processing (DSP) capabilities, due to the inexorable work of Moore's Law in providing higher processing capability at a lower cost per gate. These advances in signal processing brought on by continual refinement of the chip production process were the underpinning factor in the general observation at the time that capacity improvements in access networks increased by a factor of 100 across 20-year intervals.

Some four years later, around 2008, the IEEE and ITU-T both defined their 10Gbit/s PON solutions, namely IEEE 802.3av, 10GE-EPON, and ITU-T XG-PON, G.987.x, respectively. 10G-EPON provides symmetric (10Gbit/s downlink and uplink) and asymmetric (10Gbit/s downlink and 1Gbit/s uplink) connections. The optical properties of these 10G systems were the same as the earlier 1G systems, namely a 28db maximum loss in the 1310nm and 1500nm bands. The fundamental difference between 10G-EPON and XG-PON is that XG-PON is a transport technology for Ethernet as well as TDM and ATM.

At this juncture, it may be useful to pause for a second and consider what a 10G PON system actually means. The total capacity in the cluster of ONUs served by a single OLT is 10G. If each of the 32 ONUs that are fed by this OLT are receiving at 'full speed', then while each ONU would be receiving at a 10Gbps total rate, the data feed that is directed to each individual ONU is some 312Mbit/s. If a single ONU is the only active ONU in this network, then the dynamic allocation behaviour of the OLT would assign the entire capacity to this active ONU, and the received data feed would be 10Gbit/s. There are differences between the shared data clocking rate, the average clocking rate under 'normal' load, and the minimum data clocking rate when all PON clients are equally active at the same time, and they relate to the splitter ratio used in the PON and the mix of client behaviours that are served by the PON.

These speeds of up to 10Gbit/s have been achieved over fibre systems using simple On / Off Keying (OOK) signal modulation. Some years ago, the point-to-point (P2P) fibre world turned to using coherent optics to obtain higher speeds. Coherent Optics uses polarization and phase and amplitude modulation to increase the number of bits that are represented by each transmitted symbol. Using a basic optical system that these days can support the transmission and reception of 25G symbols per second and using a Quadrature Phase Shift Keying (QPSK) system to encode two bits per symbol, and also by using two polarizations, the system can support a data rate of some 100Gbit/s (4x25Gbit/s). The same coherent optical techniques can be used in the context of PON systems, as the passive splitter does not alter either the phase or the polarization of the light signal. With shorter lengths (up to 20km in general) and lower signal distortion levels, PON solutions that utilize coherent optics for access networks have a better signal-to-noise ratio, which allows for higher modulation orders than other technologies, and thereby achieve high capacity in the PON network.

In 2023, CableLabs issued the architectural specification of Coherent PON networks using 100G single wavelength capacity (CPON). CPON supports an extended optical power budget and higher spectral efficiency, which for new PON builds implies a capability for greater splitter ratios or extending the reach of the PON network from around 25km to up to 80km. It's not just 100G PON systems, as a rather enthusiastic CableLabs blog post reports: "As consumer appetite for bandwidth continues to grow, CPON will prove a scalable and extensible technology for fibre networks for the next 30 years." Considering the extent of the changes over the last 30 years that's a mighty claim!

Where does it go from here? The 400Gbit/s Point-to-Point (P2P) optical systems have been around for some years now, and it looks like these systems will be heading into the Point-to-Multipoint (P2M) PON environments in the very near future. New PON rollouts can take advantage of coherent optics to support greater numbers of clients per OLT through the use of higher splitter levels and also longer maximum reach from the OLT to the ONUs.

However, it's not just new PON deployments. One of the key requirements is to be able to upgrade existing PON networks by changing the optical drivers and receivers at the OLTs and ONUs, leaving the entire passive optical infrastructure unchanged. In such scenarios of upgrading existing infrastructure, the result is an increase in the capability of these systems per client, and not necessarily an increase in the scalability of the PON network to service more clients over the same infrastructure.

With Coherent PON and digital signal processing capability that supports between 32 and 64 discrete encoding points in the phase, amplitude and polarization space it is not unreasonable to contemplate 400G, 800G and even terabit-per-second PON networks. Technically, that's a stunning achievement! The evolutionary path of PON capacity over the past few decades is shown in Figure 3.

There are several tempering thoughts about future PON evolution, particularly in relation to the unbounded future 30-year vision espoused by CableLabs. One such thought is that it is unlikely that a residential user of an access PON network would pay a significant price premium over a 100G PON service for such a high-speed service, at least in the near-term future. While terabit-per-second PON access networks certainly look to be feasible from a technology perspective, working out if such very high-speed systems have an associated use profile and value proposition that can support an economically feasible deployment model for network operators currently falls in the realm of unfinished work.

Secondly, there are very real issues with the continued functioning of Moore's Law in silicon chips. So far, we've been able to reduce the width of the planar tracks in silicon dyes and maintain the same production cost and yield per chip. The cost per gate falls and the unit cost of processing falls. The implication is that for a constant cost (or even a declining cost) the evolution of digital signal processors that support ever-higher sensitivity and increasingly crowded discrete encoding points in phase-amplitude signal modulation can be sustained. With track widths heading to sub 3nm, it's not clear how we can keep Moore's Law functioning with planar integrated circuits, and it is a distinct possibility that we are getting to the end of these 60-year-long adventures with silicon chips and Moore's Law.

I don't know about you, but between these issues with technology and economics, I'm certainly not holding my breath for a terabit optical service delivered directly to my suburban residence. I just don't think that it's going to happen in my wired Internet access service for many years to come, if at all!

The views expressed by the authors of this blog are their own and do not necessarily reflect the views of APNIC. Please note a Code of Conduct applies to this blog.