No Connection Without this Packet

Modern communication is inconceivable without Internet-based networks (IP networks). And although this technology is incredibly well developed, unexpectedly long and hugely variable run times and packet losses aren’t uncommon. These problems result from brief network overloads. Classic Internet applications use transmission control protocol (TCP) which uses repeated transmissions to compensate for packet losses. As a result, users hardly notice anything. In real-time communications such as telephony (voice over IP/VoIP), user datagram protocol (UDP) generally comes into play. This doesn’t provide a replication of missing information. Yet even in this application area, things can be done to counteract packet losses. Various methods attempt to compensate for such losses in the terminal, either before, during or after transferring the user information. Working with the German federal association for telecommunications (VAF Bundesverband Telekommunikation e.V.) in Hilden, Germany, the Stuttgart-based Steinbeis Transfer Center for the Technology of Networks investigated just how well this approach actually works and how various devices on the market behave under disruptive conditions.

The various terminals behave quite differently, even when used in the same networks under identical conditions. While some of the devices still delivered very good speech quality (despite high packet losses), others delivered noticeably poorer performance. As a result, two users in the same network might rate the connection quality very differently if different terminals are used. What one user might experience as an excellent connection could be deemed unacceptable by another user on account of frequent connection drops and poor audio performance. Communication problems are often attributed to the transmission network, but according to the study this is not always the case.

In the first stage of testing, the Steinbeis experts worked with the VAF to develop a test device they called PacketRaptor. The device was then prototyped by the Flensburg-based company Nextragen. It allows for targeted simulation of specific network loads: a VoIP terminal is connected to an IP telecommunication system through an Ethernet interface. Then a generator is switched on to produce a continuous sinusoidal tone. IP packets traveling in both directions can be influenced with the PacketRaptor. A connected telephone does not necessarily receive all of the packets during real-time transmission. Speech samples can also be fed into the test system and the output can be recorded for acoustic comparisons.

The project team was able to identify two general categories of terminals: one group doesn’t react at all, meaning a packet with speech information (RTP packet) is simply missing; the other group tries to replace the missing packet. If terminals simply fail to react, the gaps caused by the missing packets are clearly audible. The signal curve shows the missing packet with a clear gap. In terminals that try to replicate a missing RTP packet, there is no obvious interruption in the signal curve. The correction curve closely resembles the original signal. Any deviation from the original tone, which is nonetheless visible on the graph, is inaudible in the sound sample. The third tested signal image shows frequency changes as well as additional peaks in the signal. Changes in the frequency of the original signal are particularly disruptive in communication and significantly impact audibility.

The tests carried out by the Steinbeis team and the VAF reveal surprising variations in how different VoIP terminals react to packet losses, runtime delays and runtime jitters. Some VoIP terminals can’t handle packet losses very well, others offer users very high speech quality, despite large packet losses in the transmission network. The same network with identical network performance and matching packet losses can seem completely different to two users depending on their end devices. Even poor networks with large data losses and long runtimes can offer good transmission quality if the right terminals are used. The researchers concluded that achieving the right quality with a VoIP solution should not simply be the design of the transmission network (LAN/WAN). The test of various VoIP telephones clearly shows that it should be underestimated how important it is to use a high-performance end device and that this is a factor that can positively influence the quality of the overall VoIP system.