How We Tested V.90 Modems
To test V.90 connections, we simulated a variety of network conditions found on the PSTN (public switched telephone network); not all connections from the end user to the remote-access server experience the same network conditions, and they vary from location to location. To simulate these PSTN and local loop conditions, we used a Telecom Analysis Systems (TAS) Series II+, Model 240 Loop Emulator and Gemini controller. To test PC Card modems, we used a Toshiba Tecra 510 laptop. Where appropriate, we modified AT commands to increase the likelihood of making V.90 connections.

For V.90 to work, the remote-access server must be connected to the PSTN via a fourýwire trunk using either T1 or ISDN. To simulate an ISDN remote-access server, we used a 3Com Corp. Courier I-Modem, which terminates ISDN BRI, V.90, x2, and V.34. For the T1 connections, we used an Ascend Communications Corp. Ascend MAX 4048, with the Rockwell chipset and software revision 6.1.3.

The TAS network-simulation equipment tests several aspects of modem connections, including performance and connection reliability, over a variety of network conditions. We focused on performance with V.90 conditions over ISDN and T1 (robbed-bit signaling) digital trunks. It wasn't feasible for us to test all possible network conditions; this would have required an unwieldy number of tests. For example covering 99 percent of the PSTN would necessitate more than 560 individual tests. Most of the individual network conditions represent less than 1 percent of the PSTN, while no individual test represented more than 9 percent. The proliferation of network equipment and the impairment requirements of V.90 have resulted in a flatter and fatter network model than described in TSB (Telecommunications System Bulletin) 37A and TSB 37B modeling for V.34 testing.

To determine our battery of tests, we choose a relatively small, though common, subset of the total number of network conditions. We derived this subset by selecting the network conditions that individually had a greater than 2 percent chance of occurrence. For ISDN testing, this amounted to 15 conditions that introduced moderate impairment to the connection; for T1 testing, this presented 21 conditions with moderate to severe impairments.

We tracked connection speed (by protocol) and throughput for each network model. The data represents the percent of V.90, V.34 and negotiation failures when connecting to the I-Modem and Ascend MAX digital back ends. A successful negotiation occurred when the client modem returned a "CONNECT" string. A failure occurred when the "NO CONNECT" string was returned. We also measured the average connection speeds (as denoted by the upper number in each bar) for each connection protocol. Note that the higher the average, the more aggressive the negotiation. We then measured the average throughput for each connection protocol; once again, the higher the number, the better the throughput.

To evaluate overall modem performance, we considered both V.90 negotiation and raw performance. For example, Ositech's Five of Hearts Trumpcard exhibited a relatively high connect speed with the I-Modem, but its overall performance was low because of throughput issues.

Because we didn't expect the modems to make V.90 connections in every test, those that successfully connected 100 percent of the time fared better than modems with higher V.90 connection rates that experienced failures. This capability is especially important for PC Card modems, which rarely obtain V.90 connections on the road.