Echo cancelation? (VoIP)

Mata Hari · May 4, 2003, 12:50 AM

Question for the DSP gods here...
Would an echo eliminater be possible? (forinctance to use Voice over IP)

this meaning, the echo could come as far as 200ms later, and has gone transcoding through 4 codecs (in local, out remote, back to mic in at remote, going to codec in remote, and codec out local)

I doubt that it's possible, or, if it is, then it is probably very hard to do? (not digital exact filtering inside the dsp, keeping track of unknown latencys...?)

eyagos · May 4, 2003, 10:14 PM

Sometime ago I wrote a pluging to reduce the typical echo produced with a microphone. It gives poor quality sound, but is enougth for IP voice. I called it "FeedBack Destroyer", and maybe the one you are looking for...
But it doesn't takes any delays into account. Just try it:

Feedback Destroyer.rar

Mata Hari · May 5, 2003, 01:35 AM

mmm, broken link?

eyagos · May 5, 2003, 12:25 PM

Yep, wait until tomorrow.

eyagos · May 5, 2003, 10:04 PM

Well, it's ready. This version works, but it is the simples't solution. I'm making a more elaborated one, wich eliminates more echoes with more sound quality...

Feedback Destroyer V1.0

Mata Hari · May 6, 2003, 01:30 AM

Although this also is very handy at certain points, and very thanks for this and helping me! But it's not exactly what I ment.

As I see it (and tested it) This kills feedback in the local sound system, but the thing I was thinking of should kill echo in a local/remote system...

let me try and explain it... and I'll try to explain, wich parts I don't think are possible in the DSP (which of I hope I'm wrong :-)

The normal system would be Mic -> encoder -> internet -> decoder -> output remote
And on the other side, other way around...

Now the echo is when the mic remote picksup the output from output remote, then encodes that, and sends it back.
This echo will arive at the local system at a latency of anywhere between 20ms to maybe even 500ms (for VoIP, 200ms roundtrip with echo cancalation is doable)

The normal way of doing this I think(!) is to cache the recorded sound for around 1 sec. and analyse the incoming sound to see if the original recorded data is there at any offset, and if it is, within a threshold, soften and filter that...

Now, the problem I see is the DSP caching audio data for that long, and if it CAN do that, filtering it without a known latency...

I hope this makes it a bit more clear :-)
And maybe there is another way of this kind of echo cancalation I'm overlooking?

eyagos · May 6, 2003, 04:22 PM

Well, I understand. This problem seems to be more complicated. And I think there not exists any simple and convincing algorithm to eliminate this echoes.

Just one question: Can you send and receive signals in stereo (two independent channels)?. If this could be posible, using various speakers and the same number of microphones (in example 2 and 2), you could succeed, in theory, and with a very simple system, to hear the voice of the opponent , but not to record it, so the echoes can't be propagated.

Mata Hari · May 6, 2003, 11:17 PM

No, not really, the signal itself is coded by the GSM codec, or even lower bitrate codecs, to keep the bandwith as low as possible
making it dual channel would mean increasing the signal by 2, wich is not accaptable in this form of application, and the goal here would be, low bandwith, accaptble sound, low latency, no echo, with standard live and mic's...

Maybe it is wiser to let the echo cancelation be done by some other type of programming, and not the DSP it self (wich there are solutions for, intel made lots of progress in that field)

Grasso · Jun 9, 2003, 03:15 AM

Echo cancellation is pretty standard nowadays. These algorithms are self-adjusting, that means that after the audio connection is up it takes a second or two until the echo cancellation fully works. They must be self-adjusting because every connection, speaker, microphone and room is different. You can hear that real-time self-adjustment when you make a connection to a portable telephone inside a car or a similar device with strong echo cancellation.

Uli

eyagos · Jun 9, 2003, 03:56 AM

Grasso, what kind of echoes are you talking about?

Grasso · Jun 9, 2003, 04:40 AM

I speak of all that noise that one gets when the speaker crosstalks into the microphone, or when the substraction circuits don´t fully substract the signal. More or less delayed feedback, ranging from capacitive crosstalk at 1KHz, i.e. less than a ms, to satellite transmission, i.e. several hundreds of ms, occurs.

May 4, 2003, 12:50 AM	#1
Mata Hari HardwareHeaven Lover Join Date: Feb 2003 Posts: 213 Rep Power: 0	Echo cancelation? (VoIP) Question for the DSP gods here... Would an echo eliminater be possible? (forinctance to use Voice over IP) this meaning, the echo could come as far as 200ms later, and has gone transcoding through 4 codecs (in local, out remote, back to mic in at remote, going to codec in remote, and codec out local) I doubt that it's possible, or, if it is, then it is probably very hard to do? (not digital exact filtering inside the dsp, keeping track of unknown latencys...?)

May 4, 2003, 10:14 PM	#2
eyagos DriverHeaven Addict Join Date: Dec 2002 Posts: 259 Rep Power: 0	Sometime ago I wrote a pluging to reduce the typical echo produced with a microphone. It gives poor quality sound, but is enougth for IP voice. I called it "FeedBack Destroyer", and maybe the one you are looking for... But it doesn't takes any delays into account. Just try it: Feedback Destroyer.rar

May 5, 2003, 01:35 AM	Thread Starter #3
Mata Hari HardwareHeaven Lover Join Date: Feb 2003 Posts: 213 Rep Power: 0	mmm, broken link?

May 5, 2003, 12:25 PM	#4
eyagos DriverHeaven Addict Join Date: Dec 2002 Posts: 259 Rep Power: 0	Yep, wait until tomorrow.

May 5, 2003, 10:04 PM	#5
eyagos DriverHeaven Addict Join Date: Dec 2002 Posts: 259 Rep Power: 0	Well, it's ready. This version works, but it is the simples't solution. I'm making a more elaborated one, wich eliminates more echoes with more sound quality... Feedback Destroyer V1.0

May 6, 2003, 01:30 AM	Thread Starter #6
Mata Hari HardwareHeaven Lover Join Date: Feb 2003 Posts: 213 Rep Power: 0	Although this also is very handy at certain points, and very thanks for this and helping me! But it's not exactly what I ment. As I see it (and tested it) This kills feedback in the local sound system, but the thing I was thinking of should kill echo in a local/remote system... let me try and explain it... and I'll try to explain, wich parts I don't think are possible in the DSP (which of I hope I'm wrong :-) The normal system would be Mic -> encoder -> internet -> decoder -> output remote And on the other side, other way around... Now the echo is when the mic remote picksup the output from output remote, then encodes that, and sends it back. This echo will arive at the local system at a latency of anywhere between 20ms to maybe even 500ms (for VoIP, 200ms roundtrip with echo cancalation is doable) The normal way of doing this I think(!) is to cache the recorded sound for around 1 sec. and analyse the incoming sound to see if the original recorded data is there at any offset, and if it is, within a threshold, soften and filter that... Now, the problem I see is the DSP caching audio data for that long, and if it CAN do that, filtering it without a known latency... I hope this makes it a bit more clear :-) And maybe there is another way of this kind of echo cancalation I'm overlooking?

May 6, 2003, 04:22 PM	#7
eyagos DriverHeaven Addict Join Date: Dec 2002 Posts: 259 Rep Power: 0	Well, I understand. This problem seems to be more complicated. And I think there not exists any simple and convincing algorithm to eliminate this echoes. Just one question: Can you send and receive signals in stereo (two independent channels)?. If this could be posible, using various speakers and the same number of microphones (in example 2 and 2), you could succeed, in theory, and with a very simple system, to hear the voice of the opponent , but not to record it, so the echoes can't be propagated.

May 6, 2003, 11:17 PM	Thread Starter #8
Mata Hari HardwareHeaven Lover Join Date: Feb 2003 Posts: 213 Rep Power: 0	No, not really, the signal itself is coded by the GSM codec, or even lower bitrate codecs, to keep the bandwith as low as possible making it dual channel would mean increasing the signal by 2, wich is not accaptable in this form of application, and the goal here would be, low bandwith, accaptble sound, low latency, no echo, with standard live and mic's... Maybe it is wiser to let the echo cancelation be done by some other type of programming, and not the DSP it self (wich there are solutions for, intel made lots of progress in that field)

Jun 9, 2003, 03:15 AM	#9
Grasso DriverHeaven Newbie Join Date: Jun 2003 Location: Wuppertal, germany Posts: 18 Rep Power: 0	Echo cancellation is pretty standard nowadays. These algorithms are self-adjusting, that means that after the audio connection is up it takes a second or two until the echo cancellation fully works. They must be self-adjusting because every connection, speaker, microphone and room is different. You can hear that real-time self-adjustment when you make a connection to a portable telephone inside a car or a similar device with strong echo cancellation. Uli

Jun 9, 2003, 03:56 AM	#10
eyagos DriverHeaven Addict Join Date: Dec 2002 Posts: 259 Rep Power: 0	Grasso, what kind of echoes are you talking about?

Jun 9, 2003, 04:40 AM	#11
Grasso DriverHeaven Newbie Join Date: Jun 2003 Location: Wuppertal, germany Posts: 18 Rep Power: 0	I speak of all that noise that one gets when the speaker crosstalks into the microphone, or when the substraction circuits don´t fully substract the signal. More or less delayed feedback, ranging from capacitive crosstalk at 1KHz, i.e. less than a ms, to satellite transmission, i.e. several hundreds of ms, occurs.