In an air combat game, a naive application of Kalman filtering does not work, because direct information about the enemy's inputs is not available. In this paper, we present two different approaches in estimating the states of the friendly as well as enemy forces based on the output observation and the friendly control inputs, when the enemy inputs are not available. The two methods are our extension of the Kalman filter due to Darouach et al. and the unknown input-decoupling observer due to Chen and Patton. We perform their stochastic simulation in the context of a game-theoretic feedback controller and compare their performance under noise.