Evaluating NMP Quality of Service : Experiment with JackTrip regarding Latency versus Packet Jitter/Dropouts with High Quality Audio via LAN and WAN

Abstract: This study has developed a method to create an, to a big extent, automated testing system for NMP (Networked Music Performance) communication over LAN and WAN to be able to benchmark the UDP streaming engine JackTrip using a client-server model. The method is not locked into using JackTrip only, it could be used to do experiments with other engines too. The study tried to answer the question if latency correlates to amount of correctly aligned audio, and to what extent the audio is correctly aligned in respect to tolerated latency (based on earlier research) when at least two musicians remote-conducting musical pieces together. There were 13 different buffer settings tested, which used no redundancy and redundancy of 2, and which were sent through 4 different LAN/WAN-scenarios. A big dataset was produced, with about 82 minutes’ worth of audio per test. To post-process the data a phase cancelling method was used to measure correctly aligned audio, while the latency was measured by counting the number of samples from the start of each audio file to the first sample that were not null or not under a certain threshold. The results showed clear correlations of buffer sizes impact of latency and amount of correctly audio sent over the network. If the buffer sizes are greater, it will produce higher latency and increase the amount of correctly aligned audio, and on the opposite side, if using less buffer, it will produce lower latencies and less correctly aligned amount of audio. The study also showed that there was very little impact of using higher redundancies, both regarding latency and amount of correctly audio. When analysing the amount of correct data when respecting the tolerated level of latency, the study showed a support for correctly aligned amount of streamed audio up to 65% when using JackTrip.