Caveat: This worked for me, but might not for you. If you take any of my advice and something goes horribly wrong, you are entitled to a full refund of goose eggs.
TL;DR: For a 2-floor home, with a mix of 5+ year-old wifi devices, to reasonably new iOS, Macs, & set top boxes, and 22+ competing network signals, the best mix I found was:
Central 2nd floor wifi router, 5Ghz band set to a fixed (not auto) channel, 802.11a/n, with 40Mhz width produced optimal speed and reliability, delivering 78-145 Mbps down and ~25 Mbps up on a ~200Mbps connection. 2.4Ghz band set to 802.11b/g/n with 20Mhz width for old devices or extended range. The specific channels used should be entirely based on signal maps using the apps mentioned below. Sit where you and your friends/family sit (in bed, on the sofa, at the table, on the floor, wherever, and take several samples to find the least congested channels).
The 5Ghz setup was the best setting for every nearly device (Nest, Roku 3, iPad 3rd gen, iPod 5th gen, and iPhone 6) on both floors. At that setting, the signal was even strong enough to pull 38 Mbps down outside the house sitting in the car. To push things to a really crazy level, I took some measurements in the basement as well, and though I could only sustain 1 bar, was still able to pull 12 Mbps down from two floors above.
For two really old devices (iPad original and ancient set top box), the second radio was set to 2.4 Ghz fixed (not auto) 802.11b/g/n with 20 Mhz width, and produced 25-35 Mbps down. Surprisingly iPad original did see and could work with the 5Ghz a/n signal, but seemed to drop off at half the distance as other devices (suspect it's using a suboptimal n protocol radio, but too lazy to verify).
Basically, it all comes down to the fact you can't change physics, so a central location is critical, as is using objective signal data, taken at different times during the day (especially in the evenings when congestion is likely highest).
If you don't have cable/fiber coming into a central room, consider the awesome Actiontec Ethernet-over-Coax gadget to use existing cable outlets to do point-to-point networking (https://www.amazon.com/Actiontec-MoCA-Ethernet-Adapter-ECB6000K02/dp/B013J7L6BW). This is fabulous, and can easily sustain 200 Mbps down.
Note #1: Highly recommend WiFi Explorer, and inSSider apps (OSX) to gather signal and congestion data throughout the entire network area. Particularly useful for identifying rude/noisy neighbors.
Note #2: I'm not generally a fan of UPS (battery backups), but it might be worth considering using one for the ISP's router, your router, and optionally, the Actiontec extender. Even a brief hiccup in the power line can cause havoc and weird problems. You don't want to wake up to a 50º home because the Nest thermostat lost a signal.
Note #3: TIL that if a Roku can't immediately find its wifi connection, it goes into broadcast mode and saturates both the 2.4Ghz and 5Ghz bands at the same channel as it last connected to, blasting out what I'm not sure, but enough noise to kill everything on the 2.4Ghz network from channel 1-6 for a range of 60 feet. Seriously, at several points, the Roku was the strongest of 22 signals I measured. It's basically a self-DOS, by design.
Note #4: Highly recommended to reset all network settings on devices (not just "Forget this network") between configuration tests:
- Nest thermostats: Reset/Network/Confirm
- Roku: Settings/Advanced Settings/Network/Reset
- iPhone/iPad: Settings/General/Reset (at bottom)/Reset Network settings
Note #5: For whatever reason, my Macbook Pro (circa 2014) could not seem to reliably detect or offer to connect to 5 Ghz bands above channel 44. This was true across multiple routers. Both WiFi Explorer and InSSIDer apps could see other networks using 5 Ghz channels 100-165, but never my own. Strangely, the iPhone 6 could see 100, 132, etc.
Very helpful network advice for Apple users: https://support.apple.com/en-us/HT202068 (though in my case, I had less interference using 40Mhz width signals (vs. 80 Mhz), and better reliability using fixed channels vs. auto). In observing WiFi Explorer maps over many hours, it's now clear that nearly every competing wifi signal contributing to any congestion auto-hopped, and so while there might be a 5-10 second drop of one bar on my devices, they never lost the connection, and showed no weird latency lag.
If you use Chrome and/or have click-to-Flash (which you really should) and don't want to futz with Flash, an HTML5 speed test is here: http://beta.speedtest.net/, and for mobile, just google "speed test" and you'll get a nifty widget from Google using Measurement Lab's network diagnostic tool (direct link here, though slighly less user friendly as Google's embed: https://www.measurementlab.net/tools/ndt/). Lastly, a good general-purpose desktop and mobile tool is Netflix's fast.com. Hat Tip to @a0viedo for the reminder.
In general the 5Ghz bands were far less congested, but as designed, the signal strength does drop considerably after a 150 feet or so radius from the router.
Results were nearly identical using Omnia Turris, TP-Link, and Netgear.
Hope that's helpful. Happy to take constructive criticism.
Nice tips! How was your signal loss through walls? I have almost no signal at 50 feet with a 5Ghz band but with 2 walls in between.
A good alternative to do a speed test that doesn't require Flash is fast.com.