Because of the way people perceive speech, Voice over Internet Protocol (VoIP) is a transmission-sensitive application that requires certain conditions in the network. Finding and fixing network problems on Open Systems Interconnection (OSI) model Layer 2 or 3 is almost trivial compared with VoIP analysis. In Internet telephony, the sources of error migrate to higher OSI Layers 4 through 7. VoIP also requires the correct interaction of all layers and thus increases the complexity of troubleshooting. Of particular importance is the end-to-end assessment of network parameters, such as available bandwidth, packet loss, delay, and jitter. In local area networks (LANs), a VoIP administrator can use a special VoIP analyzer to control the available bandwidth and the quality of voice transmission by monitoring the relevant network parameters (e.g., load, packet loss, and delays).

Effective Bandwidth

Effective bandwidth (EFB) describes the bandwidth available over the entire network path (end-to-end) at a specific point in time for the application or data flow in question. In a network, EFB is continually changing and depends on the number of simultaneous data streams. In most cases, EFB is determined by a few overloaded network connections or coupling elements. EFB is measured in bits per second (bps) and must be sufficient to transport the data successfully from the transmitter to the receiver. A deficit of bandwidth can lead to jitter and packet loss. The required bandwidth also depends on the voice codec selected.

VoIP places high demands on the available bandwidth, so you must know the network and prevailing bandwidth conditions. A VoIP analyzer displays these changes so you can see them at a glance (Figure 1).

Figure 1: A detailed display of a VoIP session in the analyzer gives you an idea of the number of different reasons an error can occur. (See also the box entitled "VoIP Analyzer Features.")

Avoiding Packet Loss

Packet loss is the norm and occurs on every network. The packet loss rate reflects the percentage of data packets lost on a transmission path. Packet losses are typically the result of network congestion (high utilization level of the queues in routers or L3 switches) and can often be avoided by prioritization mechanisms on the network. Packet loss rates of up to five percent are hardly perceived given equidistant packet losses and non-compressing codecs. If the losses exceed 10 percent of the transmitted packets, voice quality deteriorates significantly.

The network protocols used today usually compensate for any packet losses that occur. For example, TCP retransmits a lost packet after a certain delay. However, the VoIP mechanisms and the underlying RTP/user datagram (UDP) protocols do not provide for retransmission. Packet losses must therefore be compensated (to some extent) in a different way for VoIP.

VoIP applications use packet loss concealment (PLC) to suppress the effect of packet loss. Short-term interruptions in the digital data stream can thus be bridged. The task of the PLC technology in the receiver is to generate the best possible estimate of the missing signal section and thus keep audible interference as low as possible. The achievable quality depends on several factors – in particular, the length of the lost segment, the stationarity of the speech signal at the time of loss, and the amount of information available from the surrounding speech frames. In the case of speech codecs with high compression, replacing lost speech frames is made even more difficult because the dependencies between successive frames cause errors to propagate beyond the lost frames.

The simplest PLC procedure replaces lost data with silence. More complex procedures hold the last transmitted sound or try to interpolate the sound. Older systems use waveform substitution, which fills the lost signals with artificially generated substitute signals. However, this procedure often leads to an unnatural robot voice with serious packet losses.

Newer algorithms interpolate the resulting signal gaps and achieve better sound quality. However, this is at the expense of the required computing capacities. In general, dropouts with a length of up to 30ms or a loss rate of up to 20 percent can be bridged without the receiver being aware of it. A VoIP analyzer displays the number of lost packets and visualizes them in real time. A precise root cause analysis based on the recorded VoIP packets enables the necessary measures to be taken to reduce packet losses on the network.

Special VoIP Latency

VoIP transmitted over an IP network experiences end-to-end signal delays. Delay is measured in milliseconds and is also referred to as "latency." The delay is the time interval between the occurrence of an event and the expected subsequent event. For VoIP, the delay time is the time between speaking and remote reception of the spoken message. In networks, the delay is often described with round-trip time (RTT), and round-trip delay describes the total delay (i.e., the outbound and return path between two IP endpoints). In VoIP applications, the one-way delay (from endpoint to endpoint in one direction) is important.

The delay is characterized by unwanted speech pauses or overlaps between transmitter and receiver during a conversation (echo effects). The end-to-end delay, according to International Telecommunication Union (ITU) recommendation G.114, should not last longer than 150ms. In VoIP applications, a delay that is too high results in a reduction of the quality of service (QoS).

A data stream delay cannot be determined by passive measurement. Because the packets are only recorded at one measuring point in the network, it is not possible to obtain measured values for the end-to-end delay, just for arrival time variations (jitter). A correct delay measurement always requires an active measurement on an end-to-end basis.