Universal LoRa(WAN) gateway limitations, because physics?

Just some notes:

1. FHSS makes LBT very difficult. In fact, I can't think of a great way to do LBT using FHSS while also meeting the hopping requirements. CSS or DSSS, or just some method of FDMA or even "trunking," however, can work well with LBT.

2. To optimize LBT, you'll also want a method for low-power group synchronization. DASH7 has the best scheme for this, but there are others.

3. Semtech's sensitivity figures are a bit rigged. They factor-in FEC, which is OK, but there's basically no description of the implementation of LoRa's FEC model. I've reverse-engineered it, however. I'm pretty sure it's a base-32 Reed Solomon scheme performed on 40 bit payload chunks. You can simulate that or look up book values of the processing gain, here, but it's about 4dB. On top of that, they utilize a coherent receiver at the gateway, which adds 3dB over the endpoint receiver (this is the somewhat disingenuous part). Semtech also tends to use low SF in their sensitivity, which will reduce some of the SNR loss due to de-spreading. At the end of the day, Semtech's sensitivity figures have more to do with the quality of the front end AS WELL AS the over-engineered reference design (it has TX/RX switches ahead of the antenna, unlike most of its competitors... again, disingenuous, because any compact endpoint can't accommodate the size of this reference design) than it does with LoRa modulation. A good DSSS QPSK scheme with a more advanced error-correction model (e.g. convolutional code concatenated with base-256 Reed Solomon) will outperform LoRa any day of the week, even in multipath environments. TI's CC13xx devices can do this, although you need to do the RS in firmware.

4. TDMA is perfectly achievable with low overhead, low-duty, and roaming as long as you have a good clock source on the endpoint as well as some protocol mechanisms for self-synchronization and re-synchronization. There's emerging work on this topic, although the first part of it is little more than adding a good RTC to your device. Incidentally, this is a hell of a lot cheaper and smaller than implementing the LoRa reference design (big IC with lots of extra components).