[asterisk-dev] Zaptel Echo Cancellation

Steve Underwood steveu at coppice.org
Fri Aug 4 09:41:45 MST 2006


John Lange wrote:

>On Fri, 2006-08-04 at 11:31 +0200, Koopmann, Jan-Peter wrote:
>  
>
>>On Friday, August 04, 2006 2:11 AM Steve Underwood wrote: 
>>
>>    
>>
>>>You are comparing a well designed echo canceller, with something very
>>>crude which addresses none of the key issues with relible
>>>cancellation. This has nothing to do with software versus hardware. 
>>>      
>>>
>
>Again I'd like to ask the question, where can one find out more about
>echo cancellation? If not online are there technical books on the
>subject?
>  
>
If you want an introduction, try 
http://www.spiritdsp.com/pdf/article_4.pdf, If you want something 
deeper, try Simon Haykin's book on adaptive filter theory. However, if 
you don't have a good grounding in DSP, the latter will be a waste of time.

>>Which is exactly the point I was trying to make, actually... :-) A
>>software canceller can probably perform as well as good hardware
>>cancellers if the algorithm used is as efficient and the CPU is
>>sufficiant.
>>    
>>
>
>In the vast majority of cases you should not have CPU issues since the
>server's CPU is probably many magnitudes more powerful then the onboard
>processor in a hardware echo canceler and Asterisk with a single span T1
>is probably not under much load.
>  
>
Wrong. EC devices have a lot of processing power.

>As an example, software raid can often outperform hardware raid in
>certain benchmarks when the machine is otherwise idle. But under load
>this advantage quickly disappears.
>
>This begs the question, what algorithm is the digium hardware echo
>canceler using? If its better then why hasn't that code been put into an
>asterisk software echo canceler?
>  
>
Those algorithms are private to the silicon vendor. Nobody with the 
requisite knowledge has tackled building a decent canceller from 
scratch. I started a couple of years ago, but haven't had time to work 
on it since.

People are actually not very interested in making EC work. They have 
little bursts of enthusiam for doing something, then disappear. :-\

Regards,
Steve




More information about the asterisk-dev mailing list