Understanding High-speed Modem Tones

by Toby Nixon

The sequence is somewhat different depending on the type of modems you're using, and get's more complicated when you combine different modem types into one box. I'll give you two examples: V.22bis, and V.32.

V.22bis

The answering modem detects ring, goes off hook, and remains silent for at least two seconds (required by phone company rules so that no data can pass before the network recognizes the call has been connected; this is known as the "billing delay").

Then, the answering modem transmits an "answer tone", described in CCITT Recommendation V.25. This is at 2100Hz, and lasts 3.3 +/- 0.7 seconds. The purpose is two-fold: so a manual-dial originator knows they've reached a modem and can put their calling modem in data mode, AND to inform the network that data is going to be transferred, so that echo supressors in the network can be disabled. If the echo supressors remained enabled, you couldn't transmit in both directions at the same time. (The originating modem remains silent throughout this period).

After the answer tone, the answering modem goes silent for 75 +/- 20 milliseconds, to separate the answer tone from the following signals. It then transmits "unscrambled binary 1 at 1200 bit/s" (USB1). This is the "static" sound you hear after the answer tone. It is slightly higher in pitch, because the signal's major components are at 2250 and 2550 Hz.

Now, the originating modem comes into play. It detects the USB1 signal in 155 +/- 10 ms, then remains silent for 456 +/- 10 msec. It then transmits "unscrambled double dibit 00 and 11 at 1200 bit/s" (S1) for 100 +/- 3 msec. Note that a Bell 212 or V.22 modem does not transmit this S1 signal, and it is by the presense or absence of this single 100 msec signal that V.22bis knows whether or not to fall back to 1200bps operation. After sending S1, the originating modem switches to sending "scrambled binary 1 at 1200 bit/s" (SB1). Note that "scrambling" has nothing to do with encryption or security, but is simply a method by which the signal is "whitened" to even out the power across the entire bandwidth.

When the answering modem (which is still transmitted USB1) detects the S1 signal from the originator, it also sends 100 msec of S1 (so that the originating modem knows that the answerer is capable of 2400bps operation), then itself switches to sending SB1. It does this for 500 msec, then switches to sending scrambled 1's at 2400bps. It does this for 200 msec, and then is ready to pass data.

600 msec after the originating modem hears SB1 from the answerer, it switches to sending scrambled 1's at 2400bps. It does this for 200msec, then is ready to pass data.

And you thought it was simple!!

V.32 (Automode)

The signals involved in V.32 handshaking are MUCH MORE COMPLICATED that V.22bis, because of the need to measure the total round-trip delay in the circuit so that the echo cancellers work. So, I won't go step-by-step through the entire handshake (unless you really want me to), but give a simplified description so you can understand what you hear when an Ultra 96 connects.

First, as in V.22bis, the answering modem detects ring, goes off-hook, and waits two seconds. It then transmits a V.25 answer tone, but it is different -- the phase of signal is reversed every 450 msec, which sounds like little "clicks" in the signal. These phase reversals inform the network that the modems themselves are going to do echo cancellation, and that any echo cancellers in the network itself should be disabled so as not to interfere with the modems.

The originating V.32 Automode modem DOES NOT wait for the end of the answer tone. After one second, it responds with an 1800Hz tone, which in V.32 is known as signal "AA". Sending this signal before the end of answer tone allows the answering modem to know, very early, that it is talking to another V.32 Automode modem.

When the answer tone is completes (3.3 +/- 0.7 seconds), if the answering modem heard signal AA, it proceeds to try to connect as V.32 immediately. If it didn't hear AA, it first tries, for three seconds, to connect as a V.22bis modem (sends signal USB1 and waits for a response). If it doesn't get a response to USB1, it goes back to trying to connect as a V.32 modem, because of the possibility that the calling V.32 modem didn't hear the answer tone, was manually dialed and switched to data mode late, or is an older V.32 model that doesn't know to respond to answer tone.

To connect in V.32, the answering modem sends signal "AC", which is 600 and 3000 Hz sent together. It sends AC for at least 64 "symbol intervals" (a symbol interval is 1/2400 of a second), then reverses the phase of the signal (making it into signal "CA"). When the originating modem detects this phase reversal, in 64 +/- 2 symbol intervals, it reverses the phase of it's own signal (making "AA" into "CC"). When the answer modem detects THIS phase reversal, in 64 +/- 2 symbol intervals it again reverses the phase of it's signal (makes "CA" back into "AC"). This exchange of phase reversals allows the modems to very accurately time the total propagation (round trip) delay of the circuit, so that the echo cancellers can be set to properly cancel far-end echoes.

The modems then do into a "half-duplex" exchange of training signals, to train the adaptive equalizers, test the quality of the phone line, and agree on the data rate to be used. The answering modem transmits first, for from 650 to 3525 milliseconds, then goes silent. The originating modem responds with a similar signal, but then leaves its signal on, while the answering modem responds one more time, establishing the final agreed-upon data rate. Both modem switch to sending scrambled binary 1 (marks) for at leat 128 symbol intervals, then are ready to pass data signals.

This is probably more than you wanted to know, but at least now you have an idea of what you're hearing through the modem speaker.

V42 AND V42BIS

The negotiation of V.42 and V.42bis doesn't start until after the modems have connected their carriers. Basically, the process is this:

The originating modem transmits a stream of XON characters with alternative parity ($11, $91), separated by 8 to 16 mark idle bits, for about 750 msec. At the same time, it listens for a response from the answerer. If no response is heard before the timer expires, the originator assumes that the other end is not a V.42 modem. It can then either attempt MNP (send a Link Request negotiation frame, wait for a response, possibly repeat), or fall back to non-error-control mode, or disconnect (if so configured).

The answering modem initially sends nothing. It waits for a timeout period (manufacturer-defined, but suggested default is 750 msec) to hear the XONs with alternating parity from the originator. If these are not detected, the modem may wait for an MNP Link Request frame from the originator, or fall back to non-error-control mode. If it does detected the XONs, it responds by sending the characters "EC" ten times (meaning, "error control"), then prepares to receive the first LAPM protocol frame (which is in synchronous, rather than async, mode).

The V.42 originator usually sends an HDLC "XID" (Exchange Identification) frame as its first LAPM transmission. The contents of the XID frame are encoded as "type, length, value" parameter strings, and can include a wide range of parameters including maximum frame size, maximum outstanding number of unacknowledge frames, LAPM optional features (selective reject, 32-bit frame check sequence, etc.), and also additional functions such as V.42bis negotiation and "manufacturer private parameters".

The answerer receives the XID frame, ignores the portions it doesn't understand (e.g., manufacturer options from a different company, or V.42bis options if it doesn't support V.42bis), selects from the values offered by the originator, and returns an XID response frame to establish the actual operating mode.

The originator then sends a SABME frame to enter data mode; the answerer replies with a UA frame, and data begins to be transferred.

-- Toby Nixon, Principal Engineer, Hayes Microcomputer Products, Inc.