Federated　learning,　as　a　novel　machine　learning　paradigm,　aims　to　collaboratively　train　a　global　model　while　keeping　the　training　data　on　local　devices,　which　protects　data　privacy　and　security　of　distributed　devices.　However,　the　model　cannot　generalize　to　new　devices　because　of　domain　shift　caused　by　the　statistical　difference　between　the　labeled　data　and　unlabeled　data　collected　by　different　devices　in　heterogeneous　internet　of　things　networks.　In　this　paper,　we　propose　a　method　named　Unsupervised　Federated　Adversarial　Domain　Adaptation　with　Controller　Modules　(UFADACM),　which　aims　to　reduce　the　distribution　difference　between　source　nodes　with　labeled　data　and　target　nodes　with　unlabeled　data,　and　reduce　the　parameter　cost　and　communication　overhead　while　achieving　a　comparable　performance.　We　also　conduct　extensive　experiments　to　demonstrate　the　effectiveness　of　the　proposed　method.　©　2021　IEEE.