OK, I've attached a sketch of a version using TLO72 opamps for everything (except the power amp!)

By using separate pairs for the two audio paths, you can remove all the reactive elements from the 'null' point. This should allow better than 50dB attenuation of the internal mic/ speaker routing, and effectively remove the feedback problem.

Remember this is not a circuit to remove acoustic feedback. As Audioguru has already described, that is not possible with opamps.
This circuit simply allows you to control the audio routing, so that you can effectively 'null out' the gain path from the microphone to the internal loudspeaker, and simply allow only the audio from the other unit to be amplified.

Using Cat 5 cable, you could use the other two pairs for DC supply, and some form of calling sounder.

The system would need a 12V DC supply, at about 500mA.

 

The routing is fixed with local cancellation so that the mic does not activate the local speaker. But sounds can still go down to the distant end's speaker, be picked up by the distant end's mic and be delivered back to the local end so there is long distance feedback howling.

Turn down the gain or use an echo canceller circuit.

 

I see what you mean! The idea above works pretty well if one end is a handset -- as used in door entry systems - but where both ends are hands free, I can see the acoustic return problem, if the loop gains are too high.
It would seem that echo cancellation is the only way out of that.

Although I might have a look and see if there's any mileage in using relative mic gains, from send and echo return levels, to introduce effective attenuators into the 'loop'. If can be sure that your own mic is 'active' then you can attenuate the echo return signal.
That would introduce an element of 'semi duplex' into the system, but hopefully not too much ?

Finding reliable 'trigger' levels might be a problem though.....

 

I worked for a teleconferencing company.

Big bank head offices used a complicated and expensive echo cancelling circuit that worked perfectly most of the time but it occasionally got confused and made funny sounds for a couple of seconds then started to work perfectly again.

Small companies used half-duplex speaker-phones which almost always cut off words.

 

We used semi-duplex speaker phone circuits a lot where I worked. As you say, they tend to cut off words. We used to call it 'Neil Armstrong syndrome'. His most famous quote has a vox switch trigger right in the middle, which cuts off the beginning of 'giant leap'.
So, on the original recording at least, you get 'That's one small step for man, one 'iant leap for mankind'
But we know what he meant!

I think the use of really effective echo cancellation is never going to be a simple project, so perhaps the OP might be better concentrating on trying to control the gain loops, to get the best from a 'ducking' gain loop, rather than a complete switching one. Should be possible, if you are careful with controlling maximum loop gains available.

 

A company in the USA obtained used half-duplex speakerphones made by Northern Telcom and added an 8W power amplifier to drive speakers in the room.The mic modules were made from a hex housing with 3 electret mics and an RF transmitter. The RF receiver connected to the used speakerphone circuit. The company I worked for sold many of them to small and medium sized businesses. Background noises caused them to cut off the distant end. The first spoken syllable was always cut off. One end frequently cut off the other end.

I worked with very expensive boardroom tele-conferencing and video-conferencing systems that used echo cancellers. They worked pretty well.

I also worked with Polycom very expensive little speakerphones that had an echo canceller circuit. They cheated because they also used half-duplex to avoid acoustical feedback howling.

 

I have had some luck with the kind of biased gain control, highlighted on the attached sketch. By using the received audio from the remote unit, to sequentially reduce the local mic gain, and then open up the power amp gain, I find you can get quite lot of extra loop gain, without too much 'echo'. And you avoid the loop gain 'race' you can get, without the biasing.
Not as effective as real echo cancellation of course, but quite effective as a simpler option.
Using values in the sketch, you control the loop gain by about 12dB, so although there is some 'semi duplex' to it, it's not a complete 'cut off' situation.
Obviously dependant on system gains, speaker levels, mic sensitivity etc, but it does seem to work quite well, if you don't need to make huge gain changes.

And of course you still retain the local 'same end' mic/speaker cancellation options.

 

About 27 years ago I made a speakerphone with a Motorola MC34018 IC that used voice switching but it worked a lot better than speakerphones that were sold. Then they made a better MC34118 speakerphone IC that I never tried.

Motorola ICs have had the manufacturer's name changed to ON Semi and Freescale Semi and those ICs are not made by them anymore but maybe are made by Unixxxx.

Today you can buy a Cirrus Logic CS6422 full duplex speakerphone IC at Digikey. It has an acoustic echo-canceller and a network echo canceller.

 

I've used MC34118. Same old story. Quite a clever analogue analysis, in real time, of the prevailing requirement for audio 'direction' control, but dependent (as always!) on the state of the background noise.
In use with high background ambient, the +4dB 'speech' threshold can mean you really need to scream to overcome the bias state!
Not good!

I have no idea whether the Cirrus chip solves all the problems. From your earlier comments about the need to still 'cheat' slightly, with semi duplex features introduced into DSP echo cancellation, maybe not?
I certainly had to apply a guaranteed direction control override into a comms systems I designed, to solve the problem of forced air movement (extraction fans) from confusing analogue speaker phones about actual ambient levels, but it was rather a specialised case.

I think you can achieve pretty good results by 'tweaking' some of the ideas we have discussed in this thread.

Maybe the OP will want to go the full 'DSP' echo cancellation route. Perhaps he will share some of the results of the experiments with us here?

If there is a simple answer, I suspect he's sitting on a commercial fortune!

 

Tending to go a bit off topic here, I think. Sorry about that!

Question to the OP. You say your system is full duplex?
Are both units microphone and loudspeaker, or can one end be a handset?

If it can, then the circuit in post #21 should solve your feedback problem.
If both ends must both be mic and loudspeaker, then, as you will have now read, the problem is much more difficult.
True echo cancellation will not be 'simple', I'm sorry to say.

 

OK, one last schematic! Easier to read, I think, and includes the correct 'biased' gain controls, to prevent the possibility of feedback from too high a gain in the loop. Shown with -12dB remote loop gain attenuation values.

Not as comprehensive a solution as echo cancellation (see the linked Cirrus datasheet in Audioguru's last post), but a lot simpler, and quite effective