I was lucky (or unlucky) enough to have to rebuild my company’s Asterisk server to prepare to have a backup. I took a slightly less powerful machine, installed Debian Etch on it and threw Asterisk 1.2.13 on it. The goal was to mimic the Asterisk configuration on its sister machine which was a Gentoo 1.2.10 install (eventually to be upgraded to 1.4.4). The FXO cards in both machines are exactly the same. They are the TDM400P. Everything went smoothly except when I got the computer in place, the echo that came in was unbearable.
This brings me to the topic title. There is a lot of information that one needs to know to be good at this. And like usual, I didn’t have time to get to it all. Therefore I will show a summary of commands and tasks and provide you with a few links that helped me out. Long story short, read the docs so you don’t completely mess everything up.
It should also be noted that the config files that some of these options reside in may differ slightly depending on your configuration.
One of the first things that should be done is to run fxotune. Ensure Asterisk isn’t running when you run this and beware because it took approximately 20mins to run on my P4 2.8GHz w/ 256M RAM. Run it using the following command:
The eventual result came out to be below in my /etc/fxotune.conf. Just be sure that run
before starting Asterisk so your settings get used.
That did a good job, but it just wasn’t where it needed to be yet. The next thing I did should have been the first thing I did. Ensure all the telephone wires are as short as possible (while still being long enough to serve their purpose) and are away from all sources of power. This helped with the slight hum I would hear on some calls (this whole scenario is scary to me and thankfully it will be fixed shortly).
Next I moved on to the echo cancellation internals that Asterisk has. First, in the phone.conf, I changed the variable echocancel to high. Obviously, you should step through the possible values incrementally, but mine was already set on medium, so high was the next logical value.
Most of the work was done here in the zapata.conf file. The first value that I tinkered with here is the echocancel (same name as in the phone.conf file). It was initially set to yes which means that Asterisk automatically defaults to a value of 128 taps. Knowing that this variable has to be a power of 2, I decided to have some fun. I started at 2 and went straight up through 256. As it turned out, 64 ended up being the best. They say it is impossible to hear with the normal ear, but there was enough of a difference that I was able to discern which sounded better. Sometimes I couldn’t hear a difference at all (which is what I assume the docs were referencing), but as soon as I heard the difference between 64 and 128, I left it at 64. The last variable I toyed with was echotraining. echotraining was off initially. I tried calls with it on and off and there was a significant difference in initial call quality when echotraining was on. If these didn’t work, I would have messed with the value of the jitterbuffers. However, it is sufficient at 10 because of the small amount of memory that this machine has.
Eventually, once I have more time (and all sysadmins know this is a rarity), I want to move on to adjusting the rxgain/txgain. More information on this can be found here. I didn’t have a need for it now since I have reached what is believed to be tolerable. But ultimately I don’t want to have to deal with this again and I want to finish the job.
Hope this allows at least one person to have a one stop shop for information on echo cancellation and can save them a long night or headache.