INTRODUCTION

As Internet telephony gateways evolve from a fringe technology to a mainstream tool, their developers face the challenges of integration, reliability, audio quality, and gatekeeper functions. Covering the trend, TMC Labs is shifting its gateway testing methods from qualitative and stand-alone testing to quantitative and comparison testing. We will test T1 and fax options of gateways in the future, but Part I of our shootout is dedicated to the analog voice functionality of gateways from Linkon, Inter-Tel, and Nuera, based on a Sun platform, a PC, and a black box, respectively. We also chose these gateways based on their installation process, documentation, support, competitive pricing, and mainstream availability. Equipment we used included the Hammer IT 2.1.3 software and the Hammer VoIP 2.0.2 test suite, plus a TLS-4 analog line simulator from the Teltone Corporation. You can visit the Web sites of these companies at www.ham.com and www.teltone.com, respectively.

TOOLS & METHODOLOGY

The most important aspect of this equipment was Hammer's VoIP suite. Earlier versions of the Hammer IT VoIP suite included Perceptual Speech Quality Measurements, but version 2.0x is the first version to include testing for speech latency. Midway through our tests, Hammer provided us with an even newer revision because Hammer's disconnect tone conflicted with Linkon's handshake tone. This conflict is explained below in the Operational Testing section.

An important part of the prompt test is measuring latency. This works by listening for DTMF tones that have been played and time-stamped at the originating Hammer channel. These tones travel to the local gateway, where they are compressed, packetized, and sent across the network. At the remote gateway, the packets are reassembled, decompressed, and sent to the terminating Hammer channel, where they are time-stamped again. The difference between time stamps equals the latency. As shown in Figure 1, this path consists of an analog line connection (1) stemming from one channel on the Hammer to the first gateway, a compression phase (2) at the gateway, a network connection (3) to another gateway where the speech undergoes a decompression phase (4), and finally, another analog connection (5) from the second gateway to a terminating channel (final destination phone number) on the Hammer unit. This provided us with five distinct phases that all speech would travel across: Two analog line connections, a compression and decompression phase, and a network path connecting the two gateways.

Our goal was to examine only the latency introduced by the gateways because of compression and decompression algorithms, so we had to minimize the latency effect of the analog and network connections (phases 1, 3, and 5). Better gateways don't just compress and decompress the voice, they do it as quickly as possible to minimize any latency introduced.

The IP network between the two gateways was easy to manage because we used a "clean" network with a 10-BaseT hub connected only to a PC running Gatekeeper and network management software. Meanwhile, the analog circuit between each gateway and the Hammer channels depended on whether the gateway being tested used FXS ports, which provide their own dial tone, or FXO ports, which need dial tone provided to them. Only the Nuera gateway used FXS ports, so we connected it directly to the Hammer system. The Inter-Tel and Linkon gateways used FXO ports, so we used the Teltone simulator between the Hammer and the gateways to generate dial tone. For these two gateways, we also subtracted the Teltone-induced latency (12 milliseconds for each half of the Hammer-to-gateway circuit) from the final figures.

Another important component of the prompt test is the PSQM score, which determines the quality of speech according to the benchmark assessment of a group of human judges endorsed by the International Telecommunications Union (the ITU-T P.800 and P.861 standards). Visit the ITU's Web site at www.itu.org. The algorithm plays sound clips chosen for the range of sounds that they produce, not for the clips' actual meaning, which tend to have an eerie and nonsensical nature. (Examples include "I worship wooden idols ... I want a minute with the inspector;" "You will have to be very quiet ... There is nothing more to be seen;" and a nasal man's voice drawling over the word "hello.") Each of the sentences is read in a male, female, and child's voice, encompassing a range of tones. Like the DTMF tones in the latency test, the sound clips are compressed, packetized, sent across the network, and decompressed by the remote gateway, which plays the clip back to the Hammer. The Hammer then takes this signal and passes it to the PSQM algorithm along with the original prompt from the voice library. The PSQM algorithm compares the received prompt to the original one and reports a PSQM score. 

To complete our testing procedure, we configured the prompt test to produce 100 test calls with sufficient time between calls for each gateway to disconnect and reset itself for the next call. We found that 50 calls per hour provided an adequate time-out between calls (Figure 2). We performed the prompt test four times on each set of gateways: once for latency testing and once for PSQM scores, using two compression schemes. Finally, we also ran the PSQM test on the Teltone unit, but the results of this test are for comparison purposes only. As Hammer engineers explained to us, the PSQM scores are non-linear and are attained using complex algorithms. Trying to "subtract" the Teltone's PSQM scores from a gateway's actual scores would not give an accurate figure. 

Reprinted by Reprint Services with permission from TMC. For reorders call Reprint Services at 651-582-3800. For subscription information call TMC at 203-852-6800 or subscribe FREE online at www.internettelephony.com. 986773