In the $K$-user single-input single-output (SISO) frequency-selective fading interference channel, it is shown that the maximal achievable multiplexing gain is almost surely $K/2$ by using interference alignment (IA). However, when the signaling dimensions are limited, allocating all the resources to all users simultaneously is not optimal. So, a group based interference alignment (GIA) scheme is proposed, and it is formulated as an unbounded knapsack problem. Optimal and greedy search algorithms are proposed to obtain group patterns. Analysis and numerical results show that the GIA scheme can obtain a higher multiplexing gain when the resources are limited.