Network Computingwww.networkcomputing.com

The Vanguard 320 offers all the common voice-/data-tuning features, such as voice priority, data priority and packet segmentation. Unfortunately, none of these settings seemed to make any difference in voice quality. Despite our best efforts, we were not able to get the unit to perform voice signaling as well as the other units. To make matters worse, voice calls were disconnected during high-volume data transactions more than once during our tests.

Low Price, Few Features The Vanguard 320 was the least expensive unit tested, but in this instance, you get what you pay for. For example, the unit offers only one serial data port and one voice port. Other units, such as the NetPerformer and 1100 QIK Model 305, offer four serial data ports for multiple WAN connections. And all of the other units offer two voice ports. You can add a second voice port to the Vanguard 320, but you must sacrifice the optional CSU/DSU to do so. The unit provides a modest two plug-in slots for adding voice or data cards; all other units offer a much greater ability to scale.

The Vanguard 320 was fairly easy to manage once configured. Its text-based menus were easy to navigate, but configuring the entire unit was more difficult than the Marathon 3K, which had a similar interface. Additional steps were required to set up voice services, routing, and tuning the connection. Unfortunately, this extra effort j ust did not pay off in the area of voice quality.

Jeff Newman can be reached at jnewman@nwc.com.

How We Tested Voice & Data FRADs
All FRADs in this roundup were tested at MCI Developers Lab, our test partner, facilities in Richardson, Texas. We used a 56-Kbps frame relay circuit that spanned from Richardson to Omaha, Neb., to Wichita, Kan., and back to a separate network within MCI's lab. To eliminate variable switching delays, the network was hosted by MCI's Hyperstream OC-3 network, which provided ample bandwidth for our tests. Two network segments were used: One segment represented a remote office site; the other, the central office site. One FRAD was placed on each segment, and a 56-Kbps frame relay circuit was established with a 16-Kbps CIR and b ursting up to 56 Kbps. Verilink Corp.'s AS2000 external CSU/DSUs with V.35 interfaces connected FRADs to the local circuit with a clock speed of 56 Kbps.

Once a data connection was established between the units, Wandel and Goltermann DA-30 LAN/WAN analyzers and Ganymede Software's Chariot benchmark software generated traffic and monitored throughput and response times. They also were used to create data overhead during voice conversations

to measure each FRAD's ability to handle voice while routing IP traffic. FRADs were subjected to a series of voice tests while under the following data loads: 0-Kbps, 8-Kbps, 28-Kbps and 56-Kbps data streams. Four different voice samples were recorded at MCI's recording studio with sentences chosen to generate a range of vowels and consonants that might expose weaknesses in a voice codec, but did not exceed the vocal range or patterns found in a typical business telephone conversation.

In all, four voice recordings were used: a clean male voice, a clean female voice , a male voice with Ð20 dB of background chatter and a female voice with Ð20 dB of background chatter. Each series of recordings was transmitted and recorded across the FRADs using as close to an 8-Kbps voice-compression algorithm as possible.

CMOS (comparative mean opinion score) was used to quantify the voice quality of each FRAD. The sampled voice recordings were given an index code and randomly presented to 32 people. Our ratio of female to males was 40 to 60. Subject were asked to compare the recorded samples to the same studio recorded samples transmitted over a long-distance POTS network, spanning approximately the same distance, and rate them according to the following scale: 3= much better, 2=better, 1=slightly better, 0=same, -1=slightly worse, -2=worse or -3=much worse.

A positive score means that the subject found the quality of the compressed voice transmitted by the FRAD to be better than that of the POTS network transmission. A negative score means that the subject found the quality of th e compressed voice to be worse than the POTS network transmission.