I suspect the most accurate measurements would be done using wideband SDR, iff I were able to acquire absolute position references for the signals I was seeing. Not likely, especially for something I might like to let others play with and/or generalize.
Everywhere I read about the subject the general consensus is that using accelerometer and gyro data from the average phone is a fool's errand. I have zero experience with the field so I wouldn't know if failure was incorrect signal processing or just had sensors.
With a particle filter and enough computation, that error can be eliminated after the fact. Ie. At the time you won't know where you are, but after you get out of the metro and the particle filter reconverged you'll know where you were with more accuracy at some point in the past.
They make external GPS units that plug into your car and connect to your phone wirelessly for example Garmin GLO that are supposed to improve accuracy although I haven't used one myself and thus can't vouch for it.
Only the former would match the car's odometer, of course.
I suspect you could go hours driving round a city with the GPS and WiFi location turned off before losing your position - simple wheel speed, gyro and compass is sufficient for most stuff.