Hello. I am working on virtually similar topic too. can we brain storm together?

It's hard to answer because you did not provide any details. But generally you can try to find either something analytically or use experiments (simulation or implementation).

For simulations:

Let's say your attack tries to reduce network performance (like throughput). What people generally do is to use a simulator like ns2 and run simulations of networks. Run it first without the attacker so see what the performance is like in "normal" settings. Then run it on the same networks with the attacker (you have to implement the behavior of your attacker) and see how much the performance is reduced. From this you can see the impact of the attack.

Of course there are many attacks already in related work, and your attack should be better than any existing attack (the attacker always chooses the best). You can check this by implementing the other attacker and simulate it as well. If your attacker reduces the performance more than others across most networks (you have to simulate several different network layouts to be sure that your claim of being better is generally valid) then it is better and you can publish it.

There are many questions here on RG on what how to run simulations, what is the best simulator, etc, so you can search the questions too.

If your attack affects another factor, say privacy, you have to somehow measure this factor in the simulation and then compare to existing work like above.

Matthias, am using matlab to simulate my network at the moment and am working on prevention techniques on Primary user emulation Attacks. have you got any idea on that?

Yeh, you can combine different attacks to make a new attacking technique. Every target network may attacked through different idea and some attacks may be used on every network.

To create a new attack, first you have to find loopholes in the network and then design your own type to exploit its security.