[asterisk-users] Questions on X100P/X101P cards

[Wilton Helm]

>BTW, can someone explain to a libart major like me (;-)) where echo
comes on in a telephone conversation? I seem to recall it's due to the
length of the line between the CO and the local party, but I'm not
sure.

Yes, I'll tackle that.  It takes a finite amount of time for the electrical
signal originating in one phone to arrive at another phone over whatever
path it is taking.  If the path is copper, that time can be fairly small.
If the path is satellite, the time will have to exceed the 1/4 second round
trip to the bird.  If there are SIP packets involved, the time must be
larger than twice the packet size because of the time taken to collect the
data in the packet and then to serialize it at the other and after it
arrives.  If the path involves the internet, there is the path delay there
to be added in (ping will give you an idea of what that is, but it can often
be 50 - 200 ms).

All of this constitutes a delay.  It can be a bit annoying in its own right
because one person asks a question, and twice the delay time elapses before
they start hearing the answer.  However, if there are POTS analog circuits
involved anywhere, a second factor comes into play.  A POTS analog circuit
is two wires, which carry an electrical representation of sound.  Both sides
of the conversation are carried over the same wire. (its called a 2 wire
circuit.  There are also four wire circuits where each direction travels on
a separate pair of wires.  They don't have echo problems.  Digital circuits
also have separate paths for each direction, so are immune to echo)

The problem with a two wire circuit is how to separate the sound going in
both directions.  That is done by something called a 2 wire to 4 wire
converter, also commonly known as a hybrid.  It basically works by
subtracting out what it knows is being sent at the near end from what it
sees on the wire.  If that subtraction is perfect, only what came from the
other end is left and that is presented to the listener.  In the real world,
this isn't perfectly possible, but it can be done fairly well.

However, there is a side effect that comes with the transition from two wire
to four wire.  Some of the signal originating at one end of the wire gets to
the other end and is reflected back.  For an analogy, tie the end of a long
rope to a pipe, stretch it out and snap the other end.  You will see a wave
travel to the pipe and then come back.  If you were able to attach the rope
to the pipe with a suitable dashpot or something that would fully absorb the
wave, nothing would come back.  This reflection from the other end is the
cause of echo.  If the path is terminated in exactly the correct impedance,
there would be no echo.  However, for real circuits over the range of
frequencies that make up sound, that impedance is a complex quantity, and
cannot be exactly matched.

The bottom line is that any circuit with one or more 2 wire analog portions
is going to have some echo.  Since most of the circuits provided by a phone
company are POTS, they are two wire analog from the subscriber to the CO.
If the subscriber equipment is Asterisk, then a 2 wire to 4 wire conversion
and digitization takes place there.  Likewise virtually all telco links are
digital and a conversion takes place in the switch in the CO.  Then at the
other end the process is repeated.  That makes a total of 4 interfaces where
echo can originate in a typical phone call.  If part of the call is SIP, or
internet or satellite, the delay is large enough to guarantee it will be
noticeable.  Since there are several interfaces there can be several echoes.

Another example that illustrates the concept is a speaker phone.  If the
person on the other end is using a speakerphone, then some of what you say
comes out of the speakerphone, bounces off the walls of the room, gets
picked up in the mic and comes back to you.  Again, if the delay is very
large, it will be an echo by the time it gets back to you.  Speakerphones
(if they are full duplex--i.e. allow both parties to talk at once) have to
have echo cancellers to prevent this from happening.


>Is there a way to keep track of this issue, and overtime, to configure
it to answer a call by expecting such and such echo, and thus, avoid
starting sampling from scratch every time?

Unfortunately not.  If you've followed the discussion to this point, you
understand that the magnitude (loudness) of the echo depends on the
impedance mismatch which is unique to the circuitry at each end (for a
typical call) of the call.  The delay time is unique to the call path, which
is likely different for each call, and in the case of internet calls, can
vary within the call.  The echo canceller must constantly do pattern
matching to recognize changes and adjust for them.  Its job is to
subtracting out a signal of exactly the same amplitude as the echo, but of
the opposite polarity and delayed by exactly the path delay the echo is
travelling through.  Since there can easily be four or more echos, the math
gets even messier.

Wilton