We consider a bistatic radar network that consists of multiple separated radar transmitters (TXs) and receivers (RXs), aiming to detect a target on a set of points of interest (PoIs). In contrast to the disk-based sensing model in a traditional sensor network, the detection range of a bistatic radar depends on both locations of the TX and RX, and is characterized by the Cassini oval. First, we study the placement of radars to minimize the maximum distance product between each PoI and its closest TX-RX pair. Then, given the radar deployment, since the TXs use different frequencies to illuminate signals for interference avoidance, we study the problem of frequency selection for the RXs to form a bistatic radar network. In particular, for the case with an intelligent target which adaptively changes its location, we treat the dynamic interaction between the radar network and the target as a repeated game. Based on their respective histories, we propose a learning algorithm for each player. For the radar network, a model-based algorithm is proposed to maximize the expected utility for the next round based on the formed belief about the target's strategy. For the target, with only its obtained utility history available, a model-free algorithm is proposed, and we prove that on average the upper bound of the difference between the expected utility by using the globally best action and that by using the proposed algorithm is arbitrarily small when the time horizon is sufficiently large.