For the purpose of this project, we implemented a discrete event simulator using Java. In the simulator, we have events corresponding to packet and acknowledgements sent and received. In each such event further events are scheduled in the future. The program generates a trace of packets received at the wireless host, and notes the corresponding time at which they are received. The source generates packets at a constant rate, which is sent out for the wireless host. To simulate the performance of the extended TCP, we study the performance under all levels of network congestion. The network load (network congestion parameter) is made to vary between low load, and peak load in a quasi-periodic way, with a mean period of 1 second. It must be noted that the period of variation is not very relevant, as long as the period is not too less for the source estimate of the round-trip time to somewhat stabilize, since the basic aim is to study whether the extended TCP can distinguish between network congestion and the channel fade periods.
The smoothing parameter for the sRTT 
is taken to be 1/8
and that for variance is taken to be 1/4 (as is done in the implementation
of the BSD TCP stack).
The packet delay statistics between the source and the base station
is taken as a function of the current network congestion, which
more specifically is given by -
packet delay for a specific packet = base packet delay + A
network congestion parameter + p
where p represents the perturbation in the delay, and is given by a Gaussian distribution, and A is an arbitrary constant. The base packet delay is an estimate of the mean delay of packets to travel between the source and base station, and for our example plots have been chosen at 40 ms. The network congestion parameter results in the un-perturbed packet delay to vary between 30 ms and 50 ms, while the perturbtion adds some randomness in the delays. The network congestion parameter is a time dependent function that is used by the simulator to determine the actual packet delay for a packet being delivered at that time. To simulate peak load and congestion, we keep the time to live in the packets to be less than the maximum delay that is possible.
The propagation and transmission delay over the wireless link is assumed to be 2 ms, and the ratio between the good and the bad states is 100:1.
The source makes its sRTT and variance estimate using the round trip delay characteristics of each packet, that is acknowledged before the timer runs out.