Generate the "true" item parameters as shown in Table 1. Then generate the "true" latent traits for each person. Following the formula (15), you can calculate the response function. Getting the probability values in hand, compare every probability to an new uniform random number to generate an observable pairwise response. Repeat it for each probability in turn, you are going to finish the task. As for estimation, I am not familiar with Mplus. But I'll give it a try.