Experiments on QoS adaptation for improving end user speech perception over multi-hop wireless networks

Ad-hoc wireless networks cannot easily support multimedia applications because of the media high probability, burstiness and persistence of errors. Real-time constraints and multicast make the problem even more difficult. Therefore, in order to improve their performance over the existing best-effort networks, multimedia applications must adapt their operation to constantly changing network QoS. In this paper we propose a programming model that allows audio applications to adapt to changes in network QoS. In our scheme QoS information is continuously fed back from audio clients to the audio server, which uses this information to adapt the characteristics of an audio stream to fit the current network conditions. We have implemented an audio-on-demand application for the Windows NT platform that uses this model. We present experiments that confirm the usefulness of our adaptation mechanism for improving the packet loss and delay jitter characteristics of an audio channel in networks with unpredictable QoS behavior. In introducing an ultimate speech layer, we incorporate techniques such as captioning, speech recognition and speech synthesizing. When the QoS notification indicates, this minimal layer takes over to maintain an acceptable level of meaningful communication. Our experiments, both in a simulated and in a real multihop wireless testbed, show that our QoS mechanism improves the characteristics of the audio channel. End user perception can be greatly enhanced, and meaningful communication can be sustained even at most adverse network conditions by using our speech transcription scheme.