Daniel sent us this one — he's at his in-laws' place in Storrs, Connecticut for a month or two a year, and he wants to set up a basic smart home sensor network. Flood monitoring in the basement is the big one, plus temperature and security. The house has two floors above a basement that's maybe ten to fifteen meters below grade, and the question is how to get connectivity through all that concrete. His instinct is a Raspberry Pi on the top floor with one ZigBee repeater on the ground floor — but he's wondering if that actually works, or if LoRa makes more sense for the flood sensor specifically. So we're designing a sensor network for a seasonal New England home where the basement wants to flood and the radio signals want to die.
I love this question because it's not a hypothetical architecture debate — it's a real house with real concrete and a real January thaw coming. Storrs gets cold. Average January low is minus eight Celsius, eighteen Fahrenheit. When the snow melts midwinter and the ground is still frozen solid, that water has exactly one place to go — straight into the basement. Daniel's father-in-law is not going to be there to hear the sump pump fail. So the system has to work, not just theoretically work.
The house is empty ten months a year. If the flood sensor can't get a signal through two concrete slabs, you find out about the flood when you open the door in June and smell it.
Exactly the stakes. Let's start with what Daniel proposed — Raspberry Pi on the top floor, ZigBee repeater on the ground floor, sensors in the basement. The question is whether that mesh actually meshes. And the answer is probably not, for reasons that are pure physics, not protocol design.
Walk me through the physics. What actually happens to a ZigBee signal when it hits a concrete floor?
ZigBee operates at two point four gigahertz, same as Wi-Fi. Concrete at two point four gigahertz attenuates the signal by about twenty-five decibels per ten centimeters of thickness. A typical residential concrete slab is fifteen to twenty centimeters. So a single floor slab eats twenty to thirty decibels of your signal. That translates to a sixty to seventy percent range reduction per floor.
One floor cuts your range by two-thirds, roughly.
Now Daniel's basement is ten to fifteen meters below the top floor, with two concrete slabs in between. Total attenuation is forty to sixty decibels. In linear terms, that's a factor of ten thousand to a million in signal power loss. The ZigBee repeater on the ground floor helps — but it only helps if the basement sensor can actually hear the repeater. And the basement sensor is still going through one slab to reach that ground-floor repeater. That's twenty to thirty decibels right there.
The basement node might have zero reachable neighbors. The mesh collapses not because ZigBee is bad at meshing, but because there's no one to mesh with.
There's real-world test data on this. Through one concrete floor, ZigBee loses sixty to seventy percent of its range. Through two floors — basement to top floor — packet loss exceeds fifty percent even with repeaters on intermediate floors. The mesh simply cannot maintain reliable routes. And for flood detection, fifty percent packet loss means half your flood alerts never arrive. That's not a sensor network, that's a coin flip.
Daniel's not going to be there to notice the sensor went silent. A flood sensor that works fifty percent of the time is worse than no sensor, because it gives you false confidence. It's like having a smoke detector that only goes off for every other fire. You install it, you feel safe, you stop worrying — and then the one time you actually need it, you're rolling dice.
There's a case study from a parking garage deployment that makes this painfully clear. They put ZigBee sensors three floors underground. Forty percent packet loss. The mesh couldn't maintain routes through multiple concrete slabs. Same team swapped in LoRaWAN sensors at eight hundred sixty-eight megahertz — sub-gigahertz — and got ninety-nine point eight percent packet delivery over six months. Same building, same sensor placement, different frequency.
That's the kind of reliability you need when you're five thousand miles away and the basement is filling with snowmelt. So what makes LoRa different? Why does eight hundred sixty-eight megahertz laugh at concrete while two point four gigahertz weeps?
Lower frequencies penetrate matter better. It's not a subtle effect. At eight hundred sixty-eight megahertz, concrete attenuation is about eight decibels per ten centimeters. At two point four gigahertz, it's twenty-five decibels per ten centimeters. That's a factor of three to five times better penetration. A single LoRa gateway on the top floor can reach a basement sensor fifteen meters below grade with ten to fifteen decibels of link margin remaining. That's not a marginal link — that's a comfortable, reliable connection with room to spare.
LoRa gives you a working link where ZigBee gives you a prayer. But LoRa is star topology, not mesh. Every sensor talks directly to the gateway. No repeaters, no daisy-chaining. Doesn't that make it less flexible if you want to expand later?
Which for this deployment is actually ideal. Daniel's talking about maybe three to five sensors — a couple of flood sensors, a temperature sensor or two, maybe a motion sensor for security. In a star topology, each of those talks directly to the gateway. No intermediate nodes to fail, no routing tables to maintain, no mesh to collapse. For fifty-plus sensors, ZigBee mesh scales better because you can daisy-chain. But Daniel doesn't have fifty sensors. He has a small, concrete-hostile deployment. Star topology with sub-gigahertz penetration is exactly the right tool.
The gateway doesn't need to be complicated. What are we actually recommending?
A RAK seven two four nine LoRaWAN gateway, about a hundred and fifty dollars. It sits on the top floor — attic if possible, height helps — and connects to the Raspberry Pi via USB or Ethernet. The Pi runs ChirpStack, which is an open-source LoRaWAN network server. The gateway receives sensor data and the Pi processes it locally. No cloud required.
That local processing part matters a lot for a seasonal home. If the house is empty and the internet goes down — which it will, at some point, over ten months — the system still needs to log data and trigger alerts.
The Pi and gateway need continuous power, so plug them into a UPS. Winter power outages in New England are common. A basic UPS gives you a few hours of runtime, enough to ride out most outages. For sensors, the flood sensors are Dragino L D S zero two — about forty dollars each, runs on two double-A batteries with a five to ten year battery life at hourly transmissions. That's set-and-forget for a seasonal home. Temperature and humidity, the Dragino L H T sixty-five, about thirty-five dollars. Same battery story. No repeaters needed, no power cables to run through concrete floors, no batteries to change every six months.
The bill of materials — one gateway at a hundred fifty, two flood sensors at forty each, one temperature sensor at thirty-five, a Raspberry Pi four at fifty-five. That's about three hundred twenty dollars total. Daniel's original ZigBee plan — Pi plus coordinator, two repeaters, three sensors — comes in around a hundred eighty. So we're asking him to spend an extra hundred forty dollars.
For reliable basement connectivity and ten-year battery life. That hundred forty dollars is the difference between knowing your basement is flooding in three minutes and finding out when you open the door in June. I'd call that cheap insurance.
There's a real deployment that proves this out. A vacation home in Vermont — same climate as Storrs, same freeze-thaw cycle, same concrete basement — used a RAK seven two four nine gateway and three Dragino L D S zero two flood sensors. After two winters, zero false alarms and one real flood event detected within three minutes of water contact. The homeowner was two thousand miles away, got the alert, called a neighbor to shut off the water. Without that sensor, they'd have had a finished basement full of water for weeks.
Three minutes from water contact to alert. That's the difference between a mop and a renovation. And the Vermont deployment had zero false alarms over two winters — which matters, because if your flood sensor cries wolf while you're in Jerusalem, you stop trusting it. Then when the real flood comes, you ignore it.
False alarms are a bigger problem than most people realize. Every time a sensor triggers incorrectly, you have to decide — do I call the neighbor again? Do I ask them to drive over in the snow at eleven PM to check a basement that's probably dry? After two or three false alarms, you stop calling. The sensor becomes background noise. So zero false alarms over two winters isn't just a nice stat — it's the difference between a system you trust and a system you unplug.
And LoRaWAN helps here in a way that's not obvious. Because the link is so solid, you're not getting spurious readings from corrupted packets. ZigBee at the edge of its range can produce garbled sensor data that looks like a flood alert. LoRaWAN's error correction at these signal levels is extremely robust — you either get a clean packet or nothing. No ambiguous maybe-it's-wet-maybe-it's-noise readings.
The architecture is clear — LoRaWAN gateway on the top floor, three or four sensors, local processing on a Pi. But Daniel also mentioned security sensors and temperature monitoring. Let's talk about those, because they have different requirements.
Temperature is straightforward. The Dragino L H T sixty-five does temperature and humidity over LoRaWAN. Place one in the basement near any plumbing, one on the ground floor. Set a threshold alert — if the basement temperature drops below two degrees Celsius, you've got a freeze risk and pipes could burst. But here's the thing about a seasonal home with intermittent internet. A notification on your phone is great, but what if the internet is down when the freeze hits? You need a local failsafe.
A freeze alarm with an actual siren.
Something that makes noise in the house. If the temperature drops below two degrees, a local alarm sounds. Will anyone hear it? Maybe not — the house is empty. But it's a belt-and-suspenders approach. The LoRa sensor logs the event and queues the alert. When internet comes back, you get the notification. But the siren means that if a neighbor is checking on the house, or if you arrive for a winter visit, you know immediately something's wrong.
You can wire that siren directly to the Pi, right? The Pi sees the temperature threshold crossed in ChirpStack and triggers a GPIO pin that drives a loud buzzer. Simple, no extra cloud dependency.
A five-dollar piezo buzzer and a transistor to drive it. The Pi's logic is: if basement temperature drops below two degrees, sound the alarm and don't stop until someone physically resets it. That way if the freeze happens during an internet outage, the alarm still triggers locally, and it keeps sounding so that when someone eventually checks the house, they know something went wrong even if the pipes already burst. At least you know to inspect before turning the water back on.
That's a great detail — the alarm latches. It doesn't just beep once and assume you heard it. And for security, LoRa has passive infrared motion sensors — the Dragino L D S zero one, for example. But there's a latency trade-off. LoRa PIR sensors have a two to three second delay between motion detection and alert delivery. For intrusion detection in a seasonal home, that's fine — you're not trying to catch a burglar in real time, you're trying to know if someone was in the house. But it's not instant.
If Daniel wants instant security alerts with video, a cellular-connected camera like the Reolink Go makes more sense. That adds a monthly data plan — maybe ten to fifteen dollars — but it gives you real-time motion alerts with footage. Mixing LoRa for environmental sensors and cellular for cameras is totally fine. They operate on completely different frequencies and don't interfere. LoRa at eight hundred sixty-eight megahertz, cellular at various bands — no conflict.
The full system is LoRaWAN for flood and temperature, cellular for security cameras if you want real-time video, and everything feeds back to the Pi for local logging. What about the backhaul? How does the Pi actually get data out when the house has internet?
The simplest approach — the Pi connects to the house Wi-Fi when it's available, which is when someone's there. It runs ChirpStack locally, stores all sensor data in a local database, and when internet is present it can push alerts via email or a messaging service. For the ten months the house is empty, you could add a cellular modem to the Pi — a basic four G hat is about forty dollars — with a low-data IoT SIM that costs a few dollars a month. That gives you always-on alerting regardless of whether the house internet is working.
If Daniel doesn't want the monthly cost, he can skip the cellular modem and just check the logs when he arrives. The sensors keep logging locally either way. The flood alert won't reach him in Jerusalem, but the data is there when he lands in Connecticut.
The system degrades gracefully. Data is logged. Internet comes back? Alerts go out. UPS keeps the gateway and Pi running for a few hours. Sensors are battery-powered and keep transmitting regardless. It's resilient by design, not by accident.
Let's talk installation, because placement matters. Where do you actually put the gateway and the sensors?
Gateway goes as high as possible. Attic is ideal — height improves line-of-sight and reduces the amount of material the signal has to punch through. If there's no attic access, the top floor ceiling is fine. Mount it on a wall, not inside a metal enclosure — metal is a Faraday cage and will kill your signal. For the flood sensors, place them at the lowest point in the basement — water flows downhill — and near any plumbing, the water heater, the sump pump. If the basement has a history of flooding in a specific corner, put a sensor there too. They're forty dollars each, put in two or three.
Test before you permanently mount anything. LoRaWAN has a link check command — it reports received signal strength and signal-to-noise ratio. If you're getting minus one hundred ten dBm or better at the basement sensor, you're golden. If it's worse than that, try moving the gateway higher or the sensor to a different spot in the basement.
The link budget math backs this up. A LoRaWAN gateway at typical indoor power levels, through two concrete slabs and fifteen meters of vertical distance, should deliver around minus one hundred to minus one hundred five dBm at the basement sensor. LoRa's sensitivity goes down to about minus one hundred thirty seven dBm at the lowest data rate. So you've got thirty-plus decibels of margin. That's not marginal — that's a rock-solid link.
Which brings us back to the core misconception Daniel was working with — that adding a ZigBee repeater would solve the concrete problem. The repeater only helps if the basement sensor can hear it. And through one slab of concrete, it probably can't, or at least not reliably. Mesh networks are not magic. They route around obstacles, but only if there's a route to find.
The second misconception — that LoRaWAN is only for agricultural sensors spread across kilometers of farmland. It's excellent for in-building sensor networks where concrete is the barrier. The parking garage case study proves this — ninety-nine point eight percent packet delivery three floors underground. That's better than most people's Wi-Fi in a single-floor apartment.
The third misconception worth busting — that all smart home protocols work equally well through walls. They absolutely don't. Two point four gigahertz protocols — ZigBee, Thread, Wi-Fi — all suffer similarly from concrete attenuation, about twenty-five decibels per ten centimeters. Sub-gigahertz protocols — LoRa, Z-Wave — perform three to five times better. Physics doesn't care about your protocol preference.
This is going to become a bigger deal as Matter and Thread roll out as the smart home standard. Both are two point four gigahertz. Thread has a better mesh than ZigBee — it's IPv6-based, more robust routing — but it's still two point four gigahertz. Twenty-five decibels per ten centimeters of concrete doesn't change because your routing algorithm is nicer.
The question becomes — will we see a sub-gigahertz version of Matter? Or will LoRaWAN remain the go-to for concrete-heavy buildings indefinitely?
I suspect LoRaWAN sticks around for exactly this use case. Matter is designed for consumer smart home — light bulbs, door locks, thermostats — where most devices are on the same floor or one floor apart in wood-frame construction. It's not designed for basement flood sensors in concrete buildings. LoRaWAN fills that gap, and I don't see that changing soon. The protocol is boring, it's reliable, it's cheap, and it works through concrete. Sometimes boring is exactly what you want.
Especially when boring saves your basement from a January thaw. Let's summarize what Daniel should actually buy. One RAK seven two four nine LoRaWAN gateway, a hundred fifty dollars. Two Dragino L D S zero two flood sensors, forty dollars each. One Dragino L H T sixty-five temperature and humidity sensor, thirty-five dollars. A Raspberry Pi four, fifty-five dollars. Total, about three hundred twenty dollars. No repeaters, no mesh headaches, ten-year battery life on the sensors. Optional — a four G hat for the Pi at forty dollars and a low-data IoT SIM for always-on alerting, and a UPS for the gateway and Pi to survive power outages.
Installation — gateway in the attic or top floor ceiling, flood sensors at the lowest basement point and near plumbing, temperature sensor near pipes. Test with link check before mounting permanently. Set a freeze alert at two degrees Celsius. Add a local siren if you want belt-and-suspenders freeze protection. For security, either LoRa PIR sensors with a few seconds of latency, or a cellular camera if you want real-time video.
The Vermont vacation home case study tells us this works. Two winters, zero false alarms, one real flood caught in three minutes. That's the standard Daniel should aim for.
For anyone listening with a similar problem — detached garage, garden shed, boat house, any structure with a concrete foundation or metal roof — the same logic applies. Two point four gigahertz will struggle or fail. Go sub-gigahertz first. It's not flashy, but it works when you're not there to fix it.
The best smart home system is the one that works when you're not there. For seasonal homes, reliability trumps features every time.
Now: Hilbert's daily fun fact.
Hilbert: In the late Victorian period, a French naturalist in Chad documented a freshwater jellyfish species, Craspedacusta sowerbii, whose polyps attach themselves to the shells of live Nile crocodiles — the jellyfish gets transport to new feeding grounds, and the crocodile gets a free exfoliation service that it appears to actively seek out by basking in jellyfish-rich shallows.
...I have questions about the crocodile's skincare routine that I'm not sure I want answered.
Nature is just out there inventing spa treatments we haven't monetized yet. I'm picturing a crocodile sliding into a jellyfish pool like it's a hot tub, just utterly pleased with itself.
One forward-looking thought before we go. As Matter and Thread become the smart home standard, both running at two point four gigahertz, the concrete penetration problem isn't going away. Twenty-five decibels per ten centimeters is a law of physics, not a protocol limitation. The question is whether the industry acknowledges this and builds a sub-gigahertz path into the standard, or whether LoRaWAN remains the quiet workhorse for anyone with a basement and a concrete floor. My money's on LoRaWAN sticking around for a long time.
Boring, reliable, perfect for the job. Just like a good flood sensor. And I think there's something almost philosophical here — the smart home industry is obsessed with the new thing, Matter this, Thread that, everything's got AI now. But Daniel's problem is a concrete floor in Connecticut. It doesn't need AI. It needs a radio wave that can punch through rock. Sometimes the right answer is the one we already had ten years ago.
Thanks to our producer Hilbert Flumingtop. This has been My Weird Prompts. If you've got a deployment problem you want us to solve, email the show at show at my weird prompts dot com.
Until next time.
