Fixed Wireless Capacity Planning

Fixed Wireless Capacity Planning: Making Money from Thin Air

Fixed wireless can seem like a kind of magic. You’re using the open air and investing in reasonable costs for radios to transmit over that air, transforming bits into dollars.

Fixed wireless capacity planning is a must for ISP product and network managers. Understanding your system capacity is crucial for optimizing what speed plans you sell. It’s worth revisiting at least yearly. If it’s been a while since you last checked the match between your subscribers’ usage and your network capacity, let’s dive in!

In this blog post, we’ll talk about what wireless (radio frequency, or RF) capacity means and how that translates into business opportunities. In future posts, we’ll look at the RF environment itself. That is, how vendor specification sheets turn into a real-life working network. We’ll also provide a comprehensive guide to improving fixed wireless performance.

Operating Toolkit for Fixed Wireless Capacity Planning

Let’s get one thing straight first. Every WISP’s market area has unique features. If you operate a large WISP, you likely serve a variety of regions with different subscriber needs and traits. As a result, I’m not going to tell you what you should do with regards to fixed wireless capacity planning. However, I will give you some examples from our research at Preseem to help you build an operational toolkit.

A Series of Tubes

Selling internet access is a bit like buying water rights to a river. You purchase or lease a big pipe, pump water to a regional distribution point, and then start branching the pipe into a smaller and smaller series of tubes—the narrowest serving an individual subscriber. It’s not a perfect analogy, but it highlights the problem of Peak Busy Hour Load. In the UK, power distribution systems need to accommodate a large percentage of the population making tea at once!

Similarly, the Internet’s delivery system must be built to accommodate the busiest hours, and then some. If everyone turns on their tap at once, demand rises. Eventually, the big pipes reach capacity and pressure drops at the faucet. Those Zoom calls to Grandma stutter, and streaming video quality drops with insufficient supply.

Illustration: a week of aggregated backhaul download showing a healthy, pointy peak busy hour in the evening.

When we look at an individual subscriber’s daily activity, data consumed will be chaotic and ‘bursty.’ That is, there will be some streaming, some downloading, some browsing. It’s impossible to predict what an individual subscriber will do at any precise moment. But as we aggregate many subscribers together and look at these patterns over time, we can make firm predictions and projections about subscriber data usage as a whole.

Fixed Wireless Capacity Planning for Growth

In late 2020, active fixed wireless subscribers used an average of 4.8 Mbps of download capacity at peak busy hour. See the Preseem Fixed Wireless Network Report for further details on median and 95th percentile consumption. This number differs from traditional peak busy hour load which is total system usage divided by total users, both active and inactive.

You’ll also need to account for individual peak usage per plan. Preseem’s customer data shows good news: peak usage drops off rapidly in plans faster than 25 Mbps. Some (but not all) of the variance is likely due to operators selling faster plans than they can consistently deliver. Data from Preseem’s subscriber database 2020/05.

Illustration: bubble size indicates the relative sample size of users per plan.

The bad news is that unlike water and electricity, internet usage per subscriber has always grown. Over the life cycle of your last-mile technology, you need to account for:

• a whopping 17% peak busy hour load annual growth rate per subscriber. Historically, Cisco has tracked this for terrestrial and mobile internet through their Visual Network Index. This is also confirmed by our Fixed Wireless Network Report.
• 25% annual growth of total data transferred
• Any expected growth from increased market penetration

Ongoing Projections

At the individual level of an AP, you can project your capacity based on a few fundamental values. Measuring real-world performance becomes more accurate as you gain customers. (Preseem leverages the global data set to give you even more insights.) Each AP will need at least enough capacity to handle the fastest service plan you intend to sell on that type of hardware and in that market. You’ll also need additional peak busy hour capacity per the actual number of subscribers that AP will serve, reduced by some percentage of oversubscription. See Preseem’s Fixed Wireless Report for common oversubscription values.

You’ll then be able to project the previously-mentioned growth over the life of the system in years, not counting future upgrades. There are more precise formulas for calculating demand in a system using shared capacity. Let us know what method you’ve used in the past; we’d love to hear from you.

More Fun Facts About the Internet Busy Hour

• Rush hour on the internet is typically between 7 and 9 p.m. local time, with some variation for the day of the week.
• Internet consumption per subscriber is measured in two ways:

• Daily peak busy hour subscriber load (internet backhaul used, divided by the number of subscriber accounts) in megabits per second typically grows at 17% annually. This number can be difficult to determine in complicated network topologies. For ISPs that serve residential subscribers and business or wholesale customers, it’s helpful to ignore or exclude business traffic (largely off-peak), and include transit, non-transit, and any CDN traffic the ISP delivers.
• Monthly per-subscriber data transferred (primarily in the download direction) in gigabytes typically grows at 25% annually.

• A phenomenon known as ‘flat topping’ shows when Internet demand exceeds supply. When your daily peaks look more like daily plateaus, you’re seeing visual evidence of congestion. The bottleneck pictured is likely on an uplink and possibly on backhaul (middle mile) and access networks (last mile).
• As an ISP, a certain amount of oversubscription usefully keeps your backhaul costs down. But too much bandwidth compression and you’ll see subscriber satisfaction drop. For ISPs who also have performance requirements (for example, due to government subsidy), oversubscription should be carefully monitored. Chances are that many of the subscribers pictured below are not experiencing exactly ‘what they’re paying for.’ Bandwidth management systems like Preseem help keep oversubscription fair and equitable, where every subscriber is ‘squeezed’ proportionately.

Illustration: backhaul showing ‘flat topping,’ or demand exceeding supply at peak busy hour.

Airtime, Spare Time

Fixed wireless operators provision spectrum toward a subscriber using an AP with a theoretical capacity. What gets consumed (under load) is airtime. Subscriber demand creates transfers of data at the rates achievable by that subscriber at that moment. Thus, airtime is a finite resource.

In a perfect world, we’d have infinite bandwidth to sell appropriately-sized plans as needs rise over time. We don’t live in that world, however, and using more spectrum than necessary to reach subscribers raises costs. In addition to the readily apparent costs associated with acquiring a subscriber—spectrum licenses and radio hardware, among others—there are some hidden costs.

Unlicensed spectrum isn’t truly free. It comes at an added risk cost that competitors might start to use it. They have just as much right to it as you do. Over time the noise floor tends to rise, lowering the capacity available.

The Cost of Low Signal

Modulation vs. Airtime

Remember that peak busy hour usage per subscriber? Let’s look at a theoretical radio, a single AP transmitting 100 Mbps at full efficiency. We’ve modeled some examples using 12 subscribers per access point, showing four scenarios of a fictional peak busy hour load. You can see that airtime consumed rises as individual subscriber link quality (reported by average modulation rates) decreases. As of 2020, the average number of subscribers per AP is around 10. There are significant outliers, however This data table is not prescriptive of any vendor’s performance or your own network practices.

Equally striking is that even a ‘good’ access point—where the average or median subscriber has good signal and a strong connection—can be dramatically impaired by even a single subscriber running in a low modulation. Radio performance, as measurable in throughput (megabits per second, for example), typically fluctuates in a stair-step fashion through a number of achievable levels.

These levels define the specific modulation scheme used at the moment of transmission. Modulating is the process of converting bits into a radio signal that can be transmitted over the air and decoded at the receiver. Higher modulating rates, or modulations, require a higher signal to noise ratio while consuming airtime.

A subscriber with a low signal (circled in red above) could consume more than half the local airtime at peak busy hour. Such subscribers will be operating at a lower modulation, reducing the AP’s total effective capacity. Less access point capacity means less bandwidth for all active subscribers to share. Less capacity also means lower total revenue due to being able to serve fewer subscribers.

Distance vs. Signal

Illustration: Sample coverage of a low-gain radio in DFS frequencies illustrating a random distribution of users grouped by modulation.

While exact coverage parameters vary widely by deployed technology, there’s also the inverse correlation between link distance and quality to consider. Every radio theoretically can operate at full speed—until there’s not enough link margin and it can no longer do so.

The wedge pictured above exemplifies a typical AP’s sector coverage with a pseudo-random distribution of synthetic users. You can see that performance (measured in bits per Hertz) drops off rapidly.

In fixed wireless parlance, a sector is typically the combination of a radio, usually transmitting at least two simultaneous signals, with one or more antennas that shape the signal to cover a specific geographic area. Traditional directional sectors cover a swath about 6-7° vertically and 90-120° horizontally. Other types of antennas include horn antennas which have more symmetric coverage (for example 30° in both azimuth and elevation), and beamforming antennas which simulate wide coverage by electronically steering individual (narrower) beams of signal.

Frustratingly, the further from the AP, the more subscribers you can cover, but the less bandwidth you can deliver to them.

The greater distance you try to cover from an existing sector, the more disruptive impact an individual subscriber can have on your overall system efficiency. Engineering such a system requires tradeoffs.

Another way to think about this, harking a bit back to our water analogy, is to think of a sector’s capacity as a bucket.

Let’s imagine that bucket filled with happy little subscribers, each with their own bucket. When you launched, you set your minimum standards, trained your installers, and the first batch of installs worked well. Everyone had a good signal and operates at high modulation and efficiency:

But let’s imagine that before too long you hire some new installers who haven’t had as much experience. Or, word of mouth is that your service is the best one (not a bad problem to have!), or trees have grown up in the way, and you let your standards slip:

Now, a minority of subscribers have low signal and consume more than the average airtime. You’ve run out of water. The picture is a bit misleading; in a truly shared medium like wireless, the shortfall is borne by everyone together. Guess what happens now—all the subscribers are unhappy, not just the ones with low signal.

More about ensuring wireless performance in an upcoming blog post.

Tuning Your Service Plans

Now that you’ve considered peak load and airtime, you’ve likely also been considering your market and evaluating your competition. If you’re in a rural area where your biggest battle is with geosynchronous, capped satellite service, your pricing and plan structure will be very different from an ISP serving a high-density, competitive area.

You have also made projections:

• You looked at the Fixed Wireless Network Report for details on the average subscribers by vendors and AP models. You’ve also chosen one or more technologies that fit your intended markets by range, subscriber density, and overall capacity.
• You’ve modeled low, medium, and high-density areas where you have or will take subscribers. You’ve also performed radio network analysis to determine what over-the-air rates you’ll achieve. These rates set a cap on the fastest plans you’ll sell. You’ll need to achieve significantly higher per-subscriber over-the-air rates than the plans you sell for best satisfaction.
• You’ve looked at your local competition’s service plans and have chosen what your minimum plan speed will be.
• You’ve done real-world testing. There are many variables in a fixed wireless system. A vendor’s spec sheet helps, but nothing beats real-world testing for confirming performance.
• You’re now well on your way to create or modify your internet speed plan or range of plans.

0 to 60 in X Seconds

One temptation is to launch new technology with fast speeds, priced to steer subscribers toward the high end rather than the middle. This can work fine … at first. But soon, as your access network approaches middle age and counts of subscribers per AP rise, you find yourself unable to add more subscribers at the faster speeds they crave. Naturally, you want to squeeze maximum value out of your network before investing more into upgrading it. You have a few options:

• Keep adding subscribers. End-user satisfaction will drop, so you’ll want to pay close attention to their level of tolerance.
• Raise prices without raising speeds. You’ll also lose some subscribers, so you’ll need to model whether this will be a net improvement financially.
• Raise speeds and prices simultaneously. In general, subscriber demand will let you know when it’s appropriate to add faster plans. Make sure you have the headroom to deliver the new speeds. One common approach is to gradually phase out old plans while adding new, faster plans.
• Manage bandwidth for quality of experience. Hopefully, you’re already doing this to get the business value out of closely monitoring your network. Preseem performs the nifty trick of managing overall network consumption while simultaneously improving effective throughput and the end subscriber’s quality of experience.
• A healthy network is continually growing. You’ll eventually need to add capacity to your network by densifying (sector splits, getting closer to the subscriber, upgrading technology.)

TL;DR: Fixed Wireless Capacity Planning

With so many variables, trying to predict performance and implement fixed wireless capacity planning can feel like trying to nail pudding to a wall. Measuring the performance of an installed system once operational is possible, and critical to the ongoing health of your business. We at Preseem are incredibly passionate about metrics! Let’s review what we learned above.

• Everybody wants internet all at once (Peak Busy Hour)
• They likely can’t have it all at once (Oversubscription)
• They probably don’t use it all once (Active Bandwidth)
• Unfortunately, they want it more and more (Ongoing Projections)
• What they consume is airtime, not bandwidth (Airtime, Spare Time)
• One bad egg spoils the whole dozen (The Cost of Low Signal)
• Optimize your capacity to match demand (Tuning Your Service Plans)

If this article on fixed wireless capacity planning has been useful to you, now would be an excellent time to subscribe to our Blog Newsletter and our Interference Newsletter. In future posts, we’ll look at the RF environment itself and how spec sheets turn into a real-life working network. We’ll also create a comprehensive guide to improving fixed wireless performance.

Special thanks to Matt Jenkins and Dan Siemon for their feedback on drafts of this article.