Eh, I did it quick-and-dirty tonight.
- Google TP-Link OnHub router acting as WAP (only 2.4GHz considered)
- Raspberry Pi 3 Model B (BCM43438 chipset) as acquired from CanaKit via Amazon in the official Pi Foundation red/white case acting as WAP
- Moto X Pure Edition 2015 phone (Qualcomm MSM8992 Snapdragon 808 chipset) with no case acting as client
- "Remote" test measured at 14.6 feet (4.5 meters)
- "Local" test measured by propping phone upright horizontally directly against the tested device's antenna housing (propped against the side of the OnHub and the narrow side of the RPi board where the ceramic antenna is)
- RPi running latest Raspbian in CLI-only mode with hostapd running via direct command-line as root, using the kernel/firmware from the "next" branch via rpi-update
- Phone was -not- connected to any AP for the testing, but merely measuring raw signal strength
- Graphs were filtered to focus only on the AP being tested at the time
- Each test was run for 2 minutes
- Standard hostapd on the RPi, configured as per https://frillip.com/using-your-raspberr ... h-hostapd/
- Wifi Analyzer, free from the Google Play Store
Myths I'd like to bust before I begin as found from my testing:
- Orientation of the RPi and phone with respect to one another (horizontal vs. vertical, 90 degree angles to one another, etc.) had -no effect- over a 2-minute window, locally or remotely
- Turning power management on or off on the BCM43438 had -no effect- over a 2-minute window, locally or remotely
- Removing the metal heatsinks from both chips on the board had -no effect- over a 2-minute window, locally or remotely
- Removing the RPi from the case and running it 'bare' had -no effect- over a 2-minute window, locally or remotely
Graphs demonstrating the above are deliberately excluded for brevity, so you're going to have to take my word for it, but I was -very- meticulous with the tests.
First test: Phone measuring OnHub router at remote distance
Very solid at -44dBm. From a reliability and performance standpoint, you're talking the equivalent of a fixed 100mbit Ethernet cable. Excellent latency and throughput.
Second test: Phone measuring OnHub at local distance
Ouch, that's a hot signal! (Actual measurement with same tool from a different screen is -2dBm, obviously cut off on the graph)
Third test: Phone measuring RPi at remote distance
Ugh...that's no good. We're averaging about -72dBm, which falls into the category of 'maybe it will work'. Usually, it doesn't.
Fourth test: Phone measuring RPi at local distance (again, allow me to remind netizens out there that this means the separation between the RPi antenna and the phone antenna is approximately _one centimeter_)
I'm going to be generous and say that visually this averages out to around -30dBm...which would be tremendously respectable if it was *consistent*. Why isn't it? (I'm not going to take an easy shot at Broadcom here, because their myriad of historical wireless issues aside, this should be a no-brainer. The client and WAP are *physically touching*!)
Last shot: All 2.4GHz WAPs visible from the location of the OnHub (again, roughly 15 feet from the RPi 3)
A little perspective: This means that the signal strength I'm seeing from the RPi, at a distance of 15 feet, is less than half the strength of service provider broadcasts nearly a quarter mile away and roughly the same strength I'm seeing from a WAP (pchang) in a neighbor's home *250 feet away through two sets of exterior walls*!
Thus, I stand by my original findings that the issue - at least with my device - lies nearly entirely with *transmit* signal strength...all measured reception strength is consistently at quality 62/70 or higher at 15 feet distance from the OnHub.
I admit that I may very well have a malfunctioning device, new out-of-the-box for only two days...but given the number of reports of poor wifi performance with the RPi 3, I rather doubt that is the case.
I used a test methodology and tools that should be *easily* replicated by anyone with access to a simple Android phone, of virtually any manufacture or version of the OS. Don't take my word for it...test for yourself!
EDIT: Added a final comparison graph that hopefully drives the point home more visually