I am developing a mobile application which sends some encrypted data to a Bluetooth device and this Bluetooth device sends the data to server. My question is that in this case how can I prevent replay attacks. Someone might use a fake Bluetooth device to get the signals and send them to the server.
The mobile application works in offline mode and has no connection to server. So using a synchronized nonce or counter doesn't help.
I can not also use time-stamp to narrow the attack window because mobile phone's time might not be correct (synchronized with a time server).
Communication between my mobile application and Bluetooth device is one-way and my mobile application can only send data to the device.
One way to do this would be to use a counter, but allow it to skip a large number of steps. For example if the last counter value you've seen from phone A is 123 and you get something with a counter value of 156 then you accept it, but anything outside the range of [124, 1000123] you discard (1000000 being completely arbitrary and dependent on your use case).
This would prevent a replay attack, but you do have to take care that the transmissions aren't malleable or it would be trivial to forge counter numbers. This could be accomplished by having a secret per device MAC key (which would only be possible if the server and phone communicate beforehand).
It's worth stating that it would be good for the transactions to be authenticated (only phone A has the capability to generate a message saying it's from phone A) or an attacker could move up the counter very quickly and do a denial of service on phone A. However, from the way you phrased the question it sounds like it's something you've already dealt with.