Forecasting bandwidth demand is important. Under-predicting demand results in frequent network upgrades or an increased threat from competitors; over-predicting increases cost-per-user and creates a longer return on investment. 

GPON has been enabling 1 Gigabit services for years, but its bandwidth will be exhausted by the end of this decade. This is one reason why operators are rapidly stepping up deployment of XGS-PON, which supports multi-Gigabit services and will last through to the next decade. At the same time, commercial services are starting to be lit up on 25G PON, which can co-exist on the same fiber as GPON and XGS-PON, and can provide a true 10G service level, which XGS-PON cannot. 

In the second half of this decade operators will have an additional choice: 50G PON. It will offer twice the speed of 25G PON at a higher cost, higher power dissipation, and lower OLT port density. 

Evaluating these two technologies today, there are two key questions we need to answer. How long will 25G PON bandwidth be sufficient? When will operators need to upgrade to 50G PON?

Bandwidth demand forecasting

Nokia developed a tool to forecast bandwidth demand some years ago^[1]. This model accounts for both sustained and bursty downstream residential bandwidth demands. To all intents and purposes, sustained demand is video, for which the primary variables are:

1. How many simultaneous streams per subscriber in peak hour?
2. How many of those streams are VR? 
3. How many of those streams are cloud gaming?
4. What are the video resolutions of each of those streams?
5. For each resolution, what is the required bit rate?

The first four of these are actually probability distributions, and the last four vary greatly over time. All of these inputs are fed into a Monte Carlo simulation. If we assume aggressive availability of higher resolution video content (4k, 8k), aggressive uptake in cloud gaming and VR (and at high resolutions) through 2030 (the current model’s limit), and then extrapolate (albeit with reduced confidence) that CAGR to 2050, the forecasted demand for 64 subscribers on a PON is the orange curve in the chart below. The maximum capacity required for bursty traffic is approximately equivalent to the highest offered service level. In the chart we have subtracted 10 Gb/s of headroom, in order to support a 10G service level, from the 21 Gb/s capacity of 25G PON. The forecast shows that a 25G PON offering 10G service level will still have enough capacity to serve FTTH traffic demands in the year 2040. At this point, providers may want to start considering 50G PON but, given the aggressive assumptions we’ve made in our modeling, 25G PON will very likely last much longer than that.

Wi-Fi considerations

Another angle to check is the capacity of future Wi-Fi networks that will sit on the end of PON networks as there is some debate that future generations of Wi-Fi will outpace 25G PON in a few years. In most of North and South America, 1200 MHz of 6 GHz spectrum is allocated for unlicensed use, which is being used in the latest generation of Wi-Fi, Wi-Fi 6E, and will be further exploited by Wi-Fi 7. Wi-Fi 7 will have a downstream plus upstream headline capacity of 40 Gb/s. This is a theoretical number based on 16 spatial streams at close range in a line-of-sight-controlled environment. In the real world, practical Wi-Fi access points and client devices support a maximum of four spatial streams, and there are obstructions, reflections, and interference. A real world downstream plus upstream maximum in the Americas will be about 8 Gb/s. Other parts of the world have allocated only the lower part of the 6 GHz band for Wi-Fi while other countries still need to take action on it. These regions will experience lower Wi-Fi capacities.  Anyone citing potential speeds of 40 Gb/s with Wi-Fi 7, caveat emptor.

By the way, it's interesting that some companies mistakenly apply Nielsen's so-called “Law of Internet Bandwidth” to fiber access. First, Nielsen’s law does not say anything about bandwidth demand, and secondly has nothing to say about offered speeds on fiber networks. It merely tracks the availability of offered speeds on metallic networks to a single person, and finds a historic 50% growth rate. It doesn’t work for fiber because it forecasted 1 Gigabit services more than 10 years after they were offered. And it doesn’t work for future bandwidth demand because it predicts that XGS-PON will run out of bandwidth in two years, not to mention that 50G PON will run out in 2029, just a few years after its commercial availability. Drawing straight lines into the future can be dangerous. 

Service providers must do their own bandwidth forecasting to consider the capacity they need in their networks, and whether an early investment in 25G PON that helps them capture new opportunities will make a better business case than waiting for 50G PON. 

Learn more about our 25G PON solution. 

References:

^[1]Our original model was published in this article: E. Harstead and R. Sharpe, “Forecasting of Access Network Bandwidth Demands for Aggregate Subscribers Using Monte Carlo Methods”, IEEE Communications Magazine, March 2015.  It has been updated multiple times since.