>It seems to me that a simple way to eliminate the feedback issue is to separate the microphone from the speaker.
That won't necessarily fix the problem. Feedback comes from bad electronic/DSP design. Physical mic/speaker positioning makes certain solutions harder, but in practical designs it's not (usually) the limiting factor.
Phones of all kinds, Skype, etc include adaptive echo/feedback cancellation already. It's a well-understood technology - adaptive cancellation has been used since the 1960s - although the fact that it exists is maybe not as well known as it could be.
(Protip - if Skype starts feeding back, you can often reset the adaptive filter by clapping once loudly.)
As for hearing aids - the first papers mentioning multiband compression date from the 1980s, so there's nothing new here, except maybe a lack of research.
My mother's hearing is pretty bad now, and I had to talk a professional audiologist through tuning her aids for maximum intelligibility. I couldn't fault his personal skills, but the consonant/vowel heuristic he'd learned in training was oversimplified and not giving good results.
He'd basically set up a phone curve filter, but in fact you need some low-mid for good intelligibility, especially on male voices. Once he dialed that back in everyone was happy.
Thing is, we needed three one hour sessions to get it in the ballpark. The real problem with aids isn't the technology, it's the fact that setting up a good prescription is really difficult and time-consuming - even more so for elderly people who may have problems describing what they're hearing.
I see that you mention an adaptive filter, and that might be a good solution, however, you are limited by CPU power of the system and other constraints (like having to work continuously)
Bonus points if you tell me what are the poles and zeros of that system to avoid feedback.
> other constraints (like having to work continuously)
I'd vote that next to no delay is a stronger constraint. With skype, you're already delayed, so adding a little processing isn't an issue. With a hearing aid, if there is too strong a delay it because off putting and, at times, dangerous.
But yeah, CPU and Power constraints in something about the size of the first joint of my pinky (including battery) are tight and very limiting.
That won't necessarily fix the problem. Feedback comes from bad electronic/DSP design. Physical mic/speaker positioning makes certain solutions harder, but in practical designs it's not (usually) the limiting factor.
Phones of all kinds, Skype, etc include adaptive echo/feedback cancellation already. It's a well-understood technology - adaptive cancellation has been used since the 1960s - although the fact that it exists is maybe not as well known as it could be.
(Protip - if Skype starts feeding back, you can often reset the adaptive filter by clapping once loudly.)
As for hearing aids - the first papers mentioning multiband compression date from the 1980s, so there's nothing new here, except maybe a lack of research.
My mother's hearing is pretty bad now, and I had to talk a professional audiologist through tuning her aids for maximum intelligibility. I couldn't fault his personal skills, but the consonant/vowel heuristic he'd learned in training was oversimplified and not giving good results.
He'd basically set up a phone curve filter, but in fact you need some low-mid for good intelligibility, especially on male voices. Once he dialed that back in everyone was happy.
Thing is, we needed three one hour sessions to get it in the ballpark. The real problem with aids isn't the technology, it's the fact that setting up a good prescription is really difficult and time-consuming - even more so for elderly people who may have problems describing what they're hearing.