[jrockway]

This doesn't sound that unbelievable to me. You don't need a lot of power to send a receivable radio signal. (GPS is a good example, the satellites are 20,000km away and transmit at 25W, but GPS still works just fine!)

The basic equation is:

  channel capacity in bits/second = bandwidth in Hz * log2(1 + signal power / noise power).  

So let's say they are using 1MHz of spectrum (WiFi uses 20-80MHz), are transmitting at 1mW EIRP, and have an omnidirectional receiving antenna (there is no such thing, but "assume a spherical cow", it's the worst-case anyway). At the claimed range of 4 miles, the signal will lose 100dB because of "path loss" (actually spreading out), giving you ~ 1e-10mW (-10dBm) of signal at the antenna. Let's set the noise floor at -60dBm which I have not really measured, but sounds good. That's 1e-6mW. Plug this into our formula and you get a bitrate of about 1 bit every 7ms. That's ~20 bytes per second!

So let's say that we want to collect a sample once a second from 256 devices. We get 20 bytes per second, so can sample every 256 seconds (let's say that's 5 minutes). There are 256 devices, so each one gets a 1 byte ID. Then you have 19 bytes of payload, which is fine for things like thermometers or your FitBit or whatever. Fill the rest with an error-correction code.

Now you have 256 devices, each using 0.01W of power for 1/256 of the time. With a 5Wh cell-phone battery, that's enough for 14 years of transmissions.

Now, to get thousands of devices, you can use more spectrum; there is plenty more in the ISM band. You can transmit with more power, 0.01W is what a Raspberry Pi IO pin can transmit connected to a long wire. You can get a directional antenna, since you probably aren't listening for signals from the sky or underground. You can also send less data.

Anyway, I ran the numbers and I don't think this company is claiming to violate any laws of physics! It all sounds quite possible, actually, with the right engineering work. I'm looking quite forward to purchasing an eval board!

[amscanne]

Isn't your math off?

Due to path loss, rx signal is 1e-10mW => -100dBm, not -10dBm.

Plug into the formula and we have a maximum theoretical bound of ~ 0.14 bits/second, or one byte every 42 seconds.

[jrockway]

I switched from W to dBm at the last minute, so there might be mistakes. Anyway, I think it ends up being feasible, because you have some play with transmit power, the noise floor, reflections, and a better antenna, which will all change your number by a few 10s of dB in whatever direction you choose.

Definitely not the worst startup I've seen on HN. I think they can make something useful, since their goal is quite modest.

[bryanlarsen]

Even 1 byte every 42 seconds would be very useful for many types of sensors.

[amscanne]

That's with the transmitter on 100% of the time.

[gonzo]

Assuming the path loss exponent as '2' is a mistake.

[pharkmillups]

jrockway, would love for you to test with us.