My cellular data plan with T-mobile UK had been working perfectly with the N900 since I got it around Christmas time. For the last month or so though, it will show a 3 or 3.5 g connection in the status area and work brilliantly for about a minute, then any new connection attempts start timing out - browser, FeedingIt, gPodder, whatever, although there is still a connected symbol and strong signal strength in the status area. But if I disconnect and reconnect the data service, suddenly everything's working again. Only for another minute though, then same behaviour.

Recently I've installed the 2g/3g connection switcher applet. When the connection is in this frozen state, switching back and forth between Dual and 3G wakes it up, the same as disconnecting and reconnecting but without breaking the connection. Interestingly, if I choose 2g only the connection works fine all day with no problems at all... but sloooooooow.

When the 3g connection is working it's great - good signal strength, speeds often in 100s of k/sec. But it's irritating to have to go into the status area and 'wake' the connection countless times a day.

Does anyone have any ideas why my cellular data seems to go to sleep on 3G? Any way to get the N900 to wake up the connection itself the way the 2g/3g switcher seems to? It's driving me nuts... Things I've tried so far:

• A new SIM for the N900.
• Trying my old handset (Windows mobile Touch Pro) with the same SIM. It works fine. Tried it with a 2nd SIM at the same time, only the N900 has the problem. Also other T-mobile data users in the same location as me aren't having this problem.
• Re-sending network settings to device from T-mobile.
• Manually entering correct T-mobile settings.
• Creating a new data connection with fAPN and repeating the previous step.
• Using OpenDNS or Google DNS servers in previous 2 steps.

I think it could be caused by network congestion - the problem seemed less evident outside London. But then why can't the N900 deal with it the way my crappy old Win Mo phone does!??! Could this be an effect of power saving implemented in PR1.1?

Other similar reports found here:

http://talk.maemo.org/showthread.php?t=52944