Throughout the long history of dial-up modems, higher connection speeds were always desired, but limitations in signal processing techniques and the phone lines themselves meant that achievable rates were always going to be limited. At various stages in time, the “absolute” limit of what could be put through a phone line have increased thanks to more complex modulations, error correction protocols, improved echo cancelling for full duplex operation and finally, by exploiting the digitally connected nature of later telephone networks.
While the ITU had done the market a big favour by standardising modem modulations, the need for speed meant that a number of manufacturers were willing to push the boundaries to develop their own modulations ahead of the ITU standard. Such non-standard modulations include Hayes Express96 (9.6kbit/s), USR HST (9.6-24kbit/s), Telebit (18.4kbit/s), V.32terbo (19.2kbit/s), V.FC (28.8kbit/s), K56flex and X2 (56kbit/s) to name a few.
Owing to the nature of the (now) practically all-digital carrier network, even in a perfect world, a phone call is only represented by 64kbit/s of data each way, thus representing the absolute limit of the dial-up technology. Real-life impacts of signal noise, amplitude and frequency response non-linearities, robbed-bit signalling, digital pads and codec mismatch mean that even getting 56kbit/s is rare. Where the network is analog at both ends, or the call is converted to analog and back again to digital, the realistic limit is 33.6kbit/s.
US Robotics X2 Technology
At the time when V.34 was just around offering 28.8kbit/s at its peak, US Robotics (USR, later 3com) sought to deliver twice as much bandwidth, calling their technology X2 (sometimes written as x2). USR saw success in the past with their proprietary HST modems and as a leading name in modems at the time holding 40% of the retail modem market, were well placed to disrupt the market.
The X2 technology worked (according to the whitepaper from USR) by exploiting the fact that the ISP’s modem banks are often connected to the switching network on a pure digital connection. If the connection remains digital through to the line-card that is serving the customer with no other analog-to-digital connections in the path, the signal to noise ratio on the downstream (from the ISP) link could be high enough to support greater rates than the V.34 standard allowed. Part of this boils down to a reduction in quantisation noise resulting in more voltage levels being able to be reconstructed. This advantage is not available in the upstream, as the customer modem and telephone line are analog, and thus the upstream remains V.34.
According to the USR Sportster Voice 56K modem manual, X2 technology is capable of customer download rates of 33333, 37333, 41333, 42666, 44000, 45333, 46666, 48000, 49333, 50666, 52000, 53333, 54666, 56000, 57333bps. However, rates above 53333bps are not permitted in the USA by the FCC due to power output limitations. Using X2, the customer upload rate can be 14400, 16800, 19200, 21600, 24000, 26400, 28800, 31200bps, although all V.34+ link rates could possibly be available.
USR had a healthy head start on the technology, especially as existing modems that they sold were DSP based and firmware upgradeable to X2. They started an ISP Xtreme Advantage Program to give ISPs some free exposure if they deployed X2 technology in their points of presence. By early 1997, they were shipping X2 modems to retailers and Stephen Hawking and Steve Wozniak were both chosen to be the face of X2 on a TV advertising campaign. The last capture of the news from the USR X2 site shows that even as late as February 1998, ISPs were considering deploying X2 technology.
Rather unfortunately, X2 was not the only “quasi-standard” in town. K56flex, championed by Rockwell (who had a commanding 80% share of the chipset market), Lucent and Motorola also made it to market in early 1997.
Both X2 and K56flex had some rocky beginnings and 56kbit/s modem adoption was (to some extent) dampened by the incompatibility of these standards. Specifically with regards to X2, it seems that there were gripes with the ISP-side modem equipment, as well as problems from users getting X2 connections which angered early adopters.
Rather luckily, change was on the horizon with the ITU V.90 standard being developed between March 1998 and February 1999. Rumours about this must have been around, as even as early as mid-1997, USR was promising customers free upgrades to a future ITU-standard to try and maintain market share. However, it seems that soon after the introduction of V.90, USR pulled support for X2 altogether. Some modems couldn’t get upgraded to V.90 in the end due to a lack of flash memory, and 3com settled a class action lawsuit over their 56k “claims” without admitting fault. Rather interestingly, this exchange from a 3Com engineer seems to point to the introduction of the DIL sequence in V.90 as being a big reliability improvement over X2.
Unfortunately, history will recall X2 as a market failure, if it even recalls it at all. It’s unclear how far X2 got outside of the USA – USR’s market dominance of modems was probably not as strong outside of the USA and as a result, I don’t recall there being an X2 capable ISP in Australia. The X2 system may not have been optimised for operation over A-law companding phone networks outside of the USA, which natively uses mu-law companding.
Nonetheless, it has always been a wish of mine to hear an X2 handshake “in the flesh” and (if possible) make it happen with my own equipment in real-time.
Getting an X2 Handshake
This whole blog post was inspired by an e-mail I received from a reader. The e-mail gave me some information which led to a modem bank in the USA which still supported X2. The reader did attempt a number of connections which were able to at least partially-negotiate X2 before falling back to V.34. As a result, I set my sights on the case.
Configuration Profile... Product type Australia External Product ID: 04562501 Options V32bis,V.80,V.34+,x2, V.90 Fax Options Class 1/Class 2.0 Line Options Caller ID Voice Options Speakerphone, TAD Clock Freq 92.0Mhz EPROM 256k RAM 32k FLASH date 11/12/98 FLASH rev 14.2.15 DSP date 5/7/97 DSP rev 10.3.79
I grabbed my USR Message Modem Pro PCB (as the case had long disintegrated into dust) and hooked it into my Grandstream HT702 to give it a whirl with my local VSP, placing a paid international call. I was able to achieve a V.34 connection but not stably. The audio was captured through Wireshark sniffing of a bridged interface where my ATA was connected, which showed amplitudes which were all over the place. I suspect the modem itself may be a little old and screwy.
Eliminating variables, I moved back to my venerable Linksys PAP2T ATA and instead terminated my calls through a US-based VSP. This method surprisingly bore more fruit, as an attempt at an X2 connection was made. Tweaking the input gain to obscene levels finally resulted in an X2 connection being successfully made.
An X2 connection sounds like this. The left channel is the digital modem at the ISP, the right channel is my modem. The call starts off similarly to most V.34 calls, but mid-way through the first ranging signal, the calling modem sends scrambled data upward. From there, it’s only a few blips in-between scrambled segments and a connection is made.
U.S. Robotics 56K Professional Message Link Diagnostics... Chars sent 3 Chars Received 0 Chars lost 0 Octets sent 0 Octets Received 0 Blocks sent 0 Blocks Received 0 Blocks resent 0 Retrains Requested 0 Retrains Granted 0 Line Reversals 0 Blers 0 Link Timeouts 0 Link Naks 0 Data Compression NONE Equalization Long Fallback Enabled Protocol NONE Speed 44000/4800 x2 Peak Speed 42666 Last Call 00:00:04 Disconnect Reason is Escape code U.S. Robotics 56K Professional Message Link Diagnostics... Modulation x2/V.34 Carrier Freq (Hz) None/1920 Symbol Rate 8000/3200 Trellis Code None/64S-4D Nonlinear Encoding None/ON Precoding None/ON Shaping ON/ON Preemphasis (-dB) 0/0 Recv/Xmit Level (-dBm) 18/16 Near Echo Loss (dB) 8 Far Echo Loss (dB) 0 Carrier Offset (Hz) NONE Round Trip Delay (msec) 559 Timing Offset (ppm) 7655 SNR (dB) 41.3 Speed Shifts Up/Down 0/0 Status : 0000,0001,0000,0000,0000,0000,0000,0000
Connection statistics from ati6 and ati11 are above. I disconnected the call pretty soon after connection with +++ followed by ath0 simply because I knew the call isn’t likely to hang on for long, the server at the other end never sends anything and I didn’t need any more seconds of white-scrambled-noise in the recording. The connection did complete at 42666bps, although the speed line says 44000/4800bps, so a speed change may have been on the cards. The three sent characters are the escape code.
In order to make an X2 connection, the command ats32=66 is needed to disable V.90, as most X2 capable servers have also V.90 capabilities and modems which are both V.90 and X2 capable will prefer V.90 mode.
A call without this S-register setting completes in V.90 as below.
The sound of the V.90 connection is the standard USR “bong” type which represents the DIL. Notice the V.34-like standard ranging signals complete fully, rather than being interrupted.
U.S. Robotics 56K Professional Message Link Diagnostics... Chars sent 3 Chars Received 0 Chars lost 0 Octets sent 0 Octets Received 0 Blocks sent 0 Blocks Received 0 Blocks resent 0 Retrains Requested 0 Retrains Granted 0 Line Reversals 0 Blers 0 Link Timeouts 0 Link Naks 0 Data Compression NONE Equalization Long Fallback Enabled Protocol NONE Speed 44000/16800 V.90 Peak Speed 44000 Last Call 00:00:03 Disconnect Reason is Escape code U.S. Robotics 56K Professional Message Link Diagnostics... Modulation V.90 Carrier Freq (Hz) None/1920 Symbol Rate 8000/3200 Trellis Code None/64S-4D Nonlinear Encoding None/ON Precoding None/ON Shaping ON/ON Preemphasis (-dB) 0/4 Recv/Xmit Level (-dBm) 20/16 Near Echo Loss (dB) 8 Far Echo Loss (dB) 0 Carrier Offset (Hz) NONE Round Trip Delay (msec) 680 Timing Offset (ppm) 7629 SNR (dB) 41.4 Speed Shifts Up/Down 0/0 Status : uu,5,12N,12.2,-0,1N,0,38.9,25.0
The connection was successful at 44000/16800bps. The last status line shows that the codec is properly matched – uu meaning mu-law on both sides (important as Australia is a-law by default). The higher rate of connection may reflect pure luck or more refinement on the V.90 standard as a result of being the combined efforts of a number of different 56k efforts. The upstream rate is still rather low, but trialling V.34 connections to the same server, it was possible to achieve either 31200/31200bps or 33600/26400bps connections, so it doesn’t seem to be one particularly “bad” direction but more-so that maybe the high energy of the digital downstream may be echoing into the upstream enough to reduce the SNR significantly.
Noting that all of these connections were achieved over a VoIP link carried over RTP/UDP on the public internet where delivery is not assured, across from Australia to the USA makes this especially significant as I have no landline anymore.
US Robotics was a leader in the modem industry and tried to bring 56k technology to consumers in their own way through their ill-fated proprietary X2 technology. While they were not the only ones to try, eventually, the ITU’s V.90 standard rendered both X2 and K56flex obsolete. Having only encountered one recording of an X2 connection online in the past on a website that no longer exists, it has been a longstanding desire of mine to actually witness a live X2 connection and record it. I can definitely say I have achieved that, which is amazing considering how long ago X2 was discontinued and the relative lack of land-line and dial-up users at present. To be able to do this in 2018 is very surprising.
Might the same be possible for K56flex as well? I don’t know … as almost everything I’ve tried is V.90 capable and won’t do K56flex due to being firmware upgraded. Likewise, I don’t have much choice in modem banks either … Maybe it’s time to crack out an older Rockwell or LTWinmodem, Windows 98SE, an old driver and make some more calls?