Messy Networks for the Internet of Things

Warning: this post gets a bit technical. It’s for the LPWA geeks.

You probably know I’m a big fan of new low-power wide-area networks for the Internet of Things. And most people know about two major benefits of these networks: years of battery life and miles of wireless range. (I talk about why that combination matters here.)

But there’s a third, equally important benefit of these new IoT networks. It’s discussed less frequently, maybe because it’s about the networks rather than the devices. Maybe because it’s harder to understand.

It’s this: New IoT networks are designed for unplanned network deployments. And messy networks are awesome for IoT.

Cell phone networks aren’t messy. They’re carefully planned, with minimal overlap, to maximize the efficiency of the network. Network engineers shoot for this type of tower plan:

For a phone network, the handoff is all-important. Handoffs happen when one cell tower transfers a phone’s connection to another tower. Handoffs have to work right every time, at 60 miles an hour, without interrupting your conversation with grandma.

Seamless handoffs take hard science. They made cell phone networks possible. And they require that each tower know to which other towers it’s adjacent, so it can coordinate handoffs with those towers.

New IoT networks don’t do handoffs. They don’t need to, because they don’t carry phone calls. They send messages.

Once you’ve eliminated handoffs, your cell towers (or “gateways” or “access points”) don’t need to be very intelligent. They don’t coordinate with adjacent towers, they just forward packets from devices to the cloud and back.

That means you can move the network logic to a “network server” in the cloud. That network server understands the network holistically and dynamically, and handle “messy” network architectures.

This is where it gets interesting.

With dumb towers and network management in the cloud, overlapping cells add no real complexity to the network. A cell phone only connects to one tower at a time (except during handoff), but on a new IoT network, a packet transmitted by a sensor might be received by several different towers. And that’s just fine. The network server just de-duplicates the redundant packets. It also decides which tower to use to send messages back to the device.

This flexible (but I’ll call it messy) architecture is really useful for network deployments.

First, it lets carriers build for coverage first, and add capacity later, with amacro/micro-tower approach. They start with one radio on a high tower to cover a city, then drop in more towers where they need more capacity, without any real planning up front. LTE networks do something similar with cell-splitting, but it’s much more complex and requires real network coordination. With IoT networks, you just drop in another tower and the network server figures it out.

Second, you can build reliability through redundancy. Instead of putting one cell tower in a location, you might put in four, so if one goes down, you have 3 backups for fail-over. While this might seem more expensive, it actually allows you to use cheaper tower sites, cheaper power supplies, and cheaper backhaul. If you save an order of magnitude in cost per tower site, a redundant network might be both more-reliable and cheaper to build.

Dumber towers and redundancy make the tower equipment far less expensive. Equipment on an LTE tower might cost $50,000. On a new IoT network it’s more like $1,000–5,000, and coming down quickly.

Finally, hybrid networks that combine user-deployed equipment and carrier-deployed equipment are easy to build. This is the messiest network of all, where carriers allow customers to place their own towers or gateways wherever they want them. This would quickly break a cell phone network, but a modern IoT network can sort it all out in the cloud.

Customer-deployed gateways solve for a major weakness of existing 3G and LTE networks: they don’t really have coverage everywhere. Say you need to connect a vending machine in the basement, or a moisture sensor on a rural farm — chances are, you won’t have cellular coverage there. But if your IoT network carrier lets you place your own low-cost micro-tower (imagine a wi-fi router on steroids), you can always get coverage wherever you need it.

This flexibility in approach makes building IoT networks far cheaper than traditional cellular networks. It empowers customers to get coverage where they need it, while saving carriers backhaul costs. And it allows for much-cheaper initial rollout of networks. All by eschewing handoffs and putting network logic in the cloud.

I think messy, hybrid, unplanned cellular networks for IoT have a bright future. What do you think?

Daniel is co-founder of Beep Networks, a maker of location-aware sensors and systems for low-power wide-area networks.

Crazy Wireless Range in San Francisco. (LoRa radio tests.)

At Beep, we’ve spent the past year working with a new long-range wireless technology called LoRa, designed specifically for the Internet of Things.

LoRa is different because, though it operates on unlicensed spectrum like Wi-Fi, it works at very long range. So instead of just covering one home, an access point can cover several miles. That means you can very easily deploy wireless sensors across an entire corporate campus, or track assets on a large site, with a single access point.

It also supports multiple access-points or towers, meaning cellular-like networks, for the Internet of Things.

You probably don’t believe me when I say miles of range. You shouldn’t yet. Every year a new wireless technology comes along claiming insane range that doesn’t pan out in the real world. Manufacturers tout ranges based on line-of-sight deployments with no interference. That’s all fine and good, but irrelevant for most real-world deployments.

(By the way, we know of a LoRa deployment achieving 100 miles line-of-sight, with a very tall tower in rural California. But really, who cares?)

We’ve been testing our LoRa system here in San Francisco, which is probably the toughest place to build a wireless network in the US, due to interference (lots of other radios) and topology (lots of hills). But our sensors still get several miles of range.

We’d like to share some more granular data on network quality, that’s coming later. To start, here’s a map tracing out a route I took on my scooter, testing out the range of our sensors and a single tower on the roof of our office. This certainly not a scientific, and it’s not representative of high-quality coverage. But we found it to be an interesting snapshot and thought we’d share. If you’re testing LoRa radios, we hope you’ll do the same. (We’ll even have hardware you can use for testing soon.)

First, here’s a quick look at the trip. It may not make a lot of sense until I talk it through, so please read on.

Okay, let‘s break this down.

Here’s the “tower.” It’s our LoRa gateway — the equivalent of a cell tower. It has a 6dB-gain antenna, and this one’s solar powered, just for fun. It’s on the roof of our 2-story office. Both of the buildings next to us are taller than ours, so it’s not a great location, but it’s not bad. Consider it a B+ site.

“Tower” on the top of Beep Networks HQ, 21st and Mission

Here’s where it is on the map.

Now, the test setup. I put this sensor in my backpack, with its big whip antenna (~3 dB gain) hanging out the side.

Inside the blue box is a LoRa radio operating at 27 dBm (500 mW). That’s quite a bit higher power than a standard off-the-shelf LoRa radio — we boost the signal with a power amplifier. (It’s still within FCC limits, just a bit more complex design. Maximizing range is part of what we do here at Beep.)

The sensor is set to send its location about every 10 seconds. It only transmits once, so if a packet is lost, it doesn’t show up on the map (in a standard LoRaWAN setup you would retransmit several times if you don’t receive acknowledgement, but we turned that off).

So every circle on the map is a point where the sensor got a GPS fix and transmitted its location back to our tower. The line that connects the dots shows the order in which they were received, but it doesn’t show the actual path traveled. If you see a lot of dots right next to each other, that suggests that most packets are getting through. If you see gaps with no dots, no packets got through in that area.

This is not a complete picture of network coverage. It’s a very forgiving test, really. For coverage testing we look at error rate (% of packets that were not received + packets that were received with errors / total number of packets transmitted). You’ll always lose some packets, which is okay, wi-fi works the same way. You just need a high enough percentage of them to get through to consider an area covered by your network.

It’s better to consider this a test of range, not of coverage. But I digress.

Okay, on to: Transportation. We used to walk, but that’s not fast enough. Biking those hills is exhausting. Enter, the perfect range-testing transport vehicle: the Buddy 125 scooter.

I rode this scooter West around Twin Peaks, then North to the Marina, and back. Then out around Giants Stadium in South of Market. Then South around Bernal Heights, to test all our local hills.

If you’re not familiar with San Francisco, we have a lot of hills. We’re in a relatively flat spot, the Mission neighborhood, but less than a mile in any direction there are hills.

Source: Mike Ernst.

The results were interesting. Here’s that map again, with the data points.

First, let’s look at the extreme end of the city. That’s the Marina Green, at the northern edge of San Francisco. I hung out there for a few minutes and the quality of the wireless connection was terrible, most packets transmitted by the tracker were lost. But a few of them got through.

What’s surprising here is that any packets got through at all. Seriously, that’s 3.7 miles away and there’s a 250-foot hill in the way. There’s no way that signal got through the earth, it must have bounced off… something.

Even more amazing, look at the point marked WTF on the map. Now, to be clear, I was cruising around Forest Hill for like 10 minutes, and only one packet got through. But it shouldn’t have. That’s on the other side of Twin Peaks, which is basically a small mountain in the middle of San Francisco, about 1,000 feet high.

Consider these links a fluke. You certainly couldn’t count on them. But — wow. It’s interesting to see the extremes.

On the other hand, at the location marked “Climbing Pacific Heights,” I was riding up the South slope of Pacific Heights. That’s not line-of-sight to our tower, but it’s not too far off. You can see there were lots of dots even though I was cruising pretty fast, so the link quality there wasn’t bad. Getting up on a hill makes a big difference.

Let’s look at one more interesting location: SoMa.

This is 2.7 miles away from our tower. It’s pretty flat along the route, but there are plenty of tall (5–10 story) buildings in the way.

As I was cruising down Bryant Street towards the water, the link was pretty good. Bryant is a large open road, and while it’s nothing like line-of-sight, there’s plenty of open space for radio signals to bounce around. As soon as I got to the water and turned the corner, I lost the connection (notice, no dots along Embarcadero, except where it intersects another large road).

The implication is that the signal doesn’t do a good job passing through buildings, but it does just fine bouncing around them, when there’s a nice broad street. So, on a grid-based city, you might have a few miles of range down the avenues but far less on the streets. Or something like that.

Once again, network coverage is a function of the topology and layout of the city as much as the radio’s spec sheet. This is a local game.

Interested in trying this for yourself? We’re making a developer’s version of the gateway we have on our roof, plus a GPS tracker kit to let you do similar tests. Will be ready in about a month.

We’d love to see how much range you can get in your city. Just drop us your email here and we’ll keep you posted:


Questions, comments? Let me know on Medium.

The Internet of Things is looking for its VisiCalc

There’s an interesting dualism in attitude toward the Internet of Things:

We’re all sure IoT is going to be huge. But no one needs it right now.

Microcomputers were kind of like this in the 1970s. Intel and others put the processing core of a computer onto a single chip, making the microcomputer possible. Which allowed Steve Wozniak to build a microcomputer, all on one circuit board.

But no one had any idea what to do with it. Steve Jobs thought people might store their recipes on a home computer. I don’t think anyone has ever done that.

So, for the first few years microcomputers were nothing more than a toy for early adopters. It wasn’t until 1979, three years later, that VisiCalc shipped. VisiCalc was the first viable spreadsheet for the Apple II, and it’s largely credited with making the microcomputer useful.

Suddenly, you could build financial projections, calculate budgets, make plans, all from your own thousand-dollar computer. This was the “killer app” that made the micro computer matter. It was quickly followed by desktop publishing, CAD, and a bunch of other killer applications.

It feels like IoT is in a similar phase. We’re just hitting the point where cloud-connectivity is cheap — new wireless chips make networks easy to deploy and dramatically reduce costs. Most people that don’t build connected devices probably don’t realize that this is a very recent phenomenon. We now have chips that enable low-cost IoT at scale. (More on that in this post.)

So now we can connect anything to the internet, but we’re not quite sure why we need to do that. To monitor temperature in the fridge? Improve physical security? Do smart-city-things? Track location of… everything? What will be the killer app for IoT that puts devices in every room of every home, every corner of every office, and maybe even in every piece of clothing we wear?

The truth is, we just don’t know. But I’m not worried.

Quite the contrary, I’m betting my career on a belief that we’ll see orders of magnitude more connected devices in the coming years, and that they’ll change the world in ways we can’t even imagine.

The IoT is coming. We’re just not quite sure what to do with it yet.


Daniel is co-founder of Beep Networks, an enabler of easy-to-deploy long-range IoT networks and systems.

Comment on this article on Medium. Or sign-up below for more.

Why Amazon Is Going to Build Its Own Cellular Network

Yes, I think Amazon is going to build its own cellular network.

But not for phones. For IoT devices that track its global logistics operation.

A few months ago I wrote about new long-range wireless technology that allows anyone to build low-cost cellular-like networks. I predicted then that we’d see new network operators pop up using this new tech. And we have.

But why should the network operators have all the fun? The point of this technology is that anyone can build a network. Covering the US with cell towers is not exactly trivial, but the cost would be a drop in the bucket for a company like Amazon.

And Amazon would benefit in a bunch of ways.

First, their new network would give them months or years of battery life, which would greatly simplify tracking pallets and packages. Most supply-chain tracking today is actually truck-tracking — because trucks have big batteries to power cellular radios. With long battery life, you can track containers or pallets or even individual parcels.

Next, Amazon would be able to put coverage exactly where they need it, starting with their transport hubs and growing from there. Say they park delivery vans in an underground parking garage — Verizon’s not likely to install a $100,000 cell tower to add coverage there, but Amazon could drop in a gateway for a few hundred bucks. Every nook and cranny of their rural distribution centers? Covered. Ports and airports? Done.

Also, they wouldn’t have to worry about the network going away. AT&T is retiring its 2G network this year, so every vending machine and point-of-sale device using that network must be thrown out or upgraded. If Amazon controlled the own network, they could upgrade on their own timeline.

Finally, they could avoid working with cell carriers, who are just kind of a pain in the ass. Certifying a device to work on a cellular network costs upwards of $50,000 and takes months. With your own network, you can just move a lot faster. Also, consider how long you wait on hold when you call Verizon’s billing department. Then scale that up to thousands of devices. You’d need an whole team of people just to manage the bills.

And while we’re listing the benefits, let’s not forget Amazon’s consumer devices.

Dash Buttons use Wi-Fi today, so they need set up. Wouldn’t it be cool if they worked straight out of the box? If Amazon owned their own network, they could. Same with Kindles. They could even put gateway radios into Echos to build out their network quickly — every home becomes an access point for their new network.

A cell tower?

Of course Amazon isn’t the only company that would benefit from owning their own IoT network.

Wouldn’t it be nice if your Nest Protect stayed connected when wi-fi failed? If Nest/Alphabet had their own IoT cellular network, it would. Setup would be seamless for all devices. I’d say the OnHub router would be an interesting place to put a gateway chip.

Security companies could cover a metro region, then drop in battery-powered security sensors with zero installation. You can’t do that with cellular because the batteries would die. Hospitals could track temperatures of medical samples as they travel to labs and back. Farmers could drop down a solar-powered gateway and light up their fields with soil sensors.

You can see why I think we’re heading into an interesting new world.

Drop us your email at the bottom of this page and we’ll keep you posted on this technology and how our company is working to develop it. Thanks!

How New Long-Range Radios Will Change the Internet of Things

This post is about a new radio technology.

No, hang on, keep reading — I promise it’s interesting. These radios are really significant for IoT, especially industrial sensors.

Despite that, most people don’t know about them. Maybe it’s because the radios are so new, or maybe it’s because everything written about them is so technical.

Here’s my attempt at an introduction for the rest of us.

Miles of Range, Years of Battery Life

Radios drain batteries.

By definition, in fact, radios radiate energy into space. The more energy they use to transmit, the further their signal will travel.

So device developers have traditionally faced a tradeoff between range and battery life. You can build a Bluetooth sensor that lasts for years on a battery, but you’ll only get 30 feet of range. Or you can build a cellular sensor with fantastic range, but you’ll need to recharge it every week.


A new category of radios has emerged which breaks this paradigm, delivering both miles of range and years of battery life. Several different standards are hitting the market now, with LoRa, Sigfox, and Ingenu in the lead. Collectively, these radios are known as Low-Power Wide Area Network radios, or LPWAN. (Which is a terrible name. Hopefully an opportunistic journalist will give us a better one soon.)

How do they work? Each standard uses a different technique to maximize range while minimizing transmission power. Sigfox uses a well-known modulation technique, but transmits slowly in a very narrow band of spectrum to maximize signal penetration. LoRa radios use a modulation technique that can find signal well below the noise floor. Ingenu uses their own novel form of spread spectrum modulation.

Put simply, these radios do some crazy math.

In fact, most of the underlying radio technology here isn’t new — the techniques for encoding signals were invented decades ago. What’s new is that the math has been committed to silicon, and is being produced in scale. So now you can buy a radio chip for a few bucks and add it to any device.

Low Throughput, High Capacity

What’s the downside? If the chart above had a z-axis, it would be throughput. Think kilobits-per-second, not megabits-per-second. So you can send lots of sensor data, but you won’t be streaming Netflix. LPWAN does not replace Wi-Fi or cellular.

On the flip side, unlike cellular, a single tower or basestation might support tens of thousands of devices. So these networks are built to scale on quantity of devices supported, not on bandwidth per device.

They’re built to support billions of battery-powered wireless sensors.

We Are All Carriers Now

There’s a second advantage to transmitting at low power: access to unlicensed spectrum.

You can’t blast at high power levels on unlicensed spectrum without screwing up cordless phones and Wi-Fi routers. But when you’re dribbling out a low-power signal, as LPWAN radios do, unlicensed frequencies become an option.

All of these radios can operate on unlicensed wireless spectrum. So the radios are cheap and the spectrum is free. When you consider that buying cellular licenses to cover the US would cost billions of dollars, that’s a pretty big deal.

It means that anyone can deploy an LPWAN network, no license required. With a couple of gateways, you can build your very own network to cover a corporate campus or basement or farm where cellular doesn’t reach, and run thousands of devices on it.

It also means that, with a relatively modest investment, new companies can become wireless carriers for IoT devices. I expect we’ll see a number of new carrier-style IoT networks built in the next few years, if only because the barriers are so low. Some cellular companies are even starting to build test networks, though mostly in Europe where competition is more intense.

Do They Really Work?

You should be a little skeptical here — I was. So I bought some radios.

Our team at Beep Networks is now working with LoRa radios here in San Francisco, and we’re getting signals through at over a mile of range. That’s in the city, through walls, with a tiny battery-powered sensor device — no towers or giant antennas involved. We know folks who are getting 10 miles in every direction when they put these radios on towers in rural areas, where there’s less interference.

It’s kind of amazing to see it in action.

Set It and Forget It

Even better, these radios have the potential to enable IoT devices that Just Work. Devices that are connected before you even pull them out of the box: no passwords, no hubs, no SIM cards. Devices which never need recharging, because their batteries effectively last forever.

For sensors, it means a truly “Set it and Forget it” experience, which is something you just can’t do with Wi-Fi, Cellular, Bluetooth or Zigbee.

Those sensors, in turn, enable cheap and easy instrumentation of real-world event data. All of which I believe will usher in a new “big-data” era for operational efficiency through industrial IoT.

But I’m getting ahead of myself. That’s another post for the future. Suffice it to say, these radios are really cool.

How do we learn more?

Obviously I’m pretty excited about these radios. I hope you are, too.

At Beep Networks we’re building a network to cover San Francisco, and we’re building a bunch of sensors that will run on it. Which means that over the next few months we’ll be learning exactly how these radios perform in an urban environment. We’re also building up the software infrastructure needed to support a real network. Check out our site to sign up for our beta program and get some sensors.

Comment on this post or follow me on Medium.