Broadband wireless access (BWA) networks, such as LTE and WiMAX, are inherently lossy due to wireless medium unreliability. Although the Hybrid Automatic Repeat reQuest (HARQ) error-control method recovers from packet loss, it has low transmission efficiency and is unsuitable for delay-sensitive applications. Alternatively, network coding techniques improve the throughput of wireless networks, but incur significant overhead and ignore network constraints such as Medium Access Control (MAC) layer transmission opportunities and physical (PHY) layer channel conditions. The present study provides analysis of Random Network Coding (RNC) and Systematic Network Coding (SNC) decoding probabilities. Based on the analytical results, SNC is selected for developing an adaptive network coding scheme designated as Frame-by-frame Adaptive Systematic Network Coding (FASNC). According to network constraints per frame, FASNC dynamically utilizes either Modified Systematic Network Coding (M-SNC) or Mixed Generation Coding (MGC). An analytical model is developed for evaluating the mean decoding delay and mean goodput of the proposed FASNC scheme. The results derived using this model agree with those obtained from computer simulations. Simulations show that FASNC results in both lower decoding delay and reduced buffer requirements compared to MRNC and {\cal N}-in-1 ReTX, while also yielding higher goodput than HARQ, MRNC, and {\cal N}-in-1 ReTX.