Optimizing federated learning in wireless networks: a knowledge distillation approach to mitigating data heterogeneity and resource scarcity