The past decade has witnessed a strong trend of the convergence towards a packet-switched all-IP infrastructure in wireless networks. The domination of IP packets in cellular data communications has made header compression a vital process in wireless networks because of its important role in substantially improving spectrum efficiency by increasing packet payload.  As competition intensifies for the limited bandwidth resource among a growing number of wireless applications, services, and users, it no longer suffices to only focus on PHY/MAC layers for spectral efficiency improvement.  In this work, we present an innovative, ``trans-layer'' approach to integratively improve overall network efficiency. We propose an innovative approach by modeling header control as a partially observable Markov decision process (POMDP) in order to maximize the success rate and hence the efficiency of ROHC decompression.  Unlike the ``cross-layer'' concept, our POMDP framework for ROHC relies on ``trans-layer'' informations observed through multiple interactive network layers, for optimized  header compression design and control decisions.  We develop novel methodologies and architectures to jointly optimize header compression and lower layer decisions in wireless networks to achieve significant improvement of transmission efficiency and robustness.