Hi,
For an ambulance intercom project, we want to use the analog output from the siren controller as a reference for the AEC algorithm, to help reduce siren noise in the communication channel.
Has anyone tried this approach, or can you share recommendations on integrating an external analog noise reference for AEC on platforms like XMOS ?
Thanks in advance!
Using Analog Siren Signal as Reference for AEC
-
- Newbie
- Posts: 1
- Joined: Fri Jul 04, 2025 7:28 am
-
Verified
- Experienced Member
- Posts: 78
- Joined: Wed Feb 17, 2016 5:10 pm
Hello,
That sounds like a really cool project. In terms of design recommendations I would just keep the following in mind:
- the AEC reference should match the analog signal that gets played out as closely as possible, so any filtering, especially non-linear filters, need to be done before the reference gets fed to the AEC (regardless of which AEC you use)
- since the siren plays out incredibly loudly, there may still be some non-linear effects that you cannot control, this will slow convergence and limit how much of the siren can be removed, however the result may still be OK, in fact XVF3800 has specific tuning guidance for non-linear residual echos https://www.xmos.com/documentation/XM-0 ... v3.2.1.pdf (page 39 FYI, but if you do go this route please go through the guide in order before attempting to tune the nonlinear echo!)
- acoustic overload point, this applies to both your intercom mics and a mic used to gather the echo reference (if you go that route): you will need a mic with a high AOP, most MEMS mics will already meet this spec, but if you are sampling up close to a siren there may be other considerations! Ideally neither your reference or comms would ever clip, as this is an example of a nonlinear distortion and a loss of information that may cause instability. If you are new to MEMS mics, then this is a great reference: https://www.infineon.com/dgdl/Infineon- ... 62021&da=t
Cheers,
-Andrew
That sounds like a really cool project. In terms of design recommendations I would just keep the following in mind:
- the AEC reference should match the analog signal that gets played out as closely as possible, so any filtering, especially non-linear filters, need to be done before the reference gets fed to the AEC (regardless of which AEC you use)
- since the siren plays out incredibly loudly, there may still be some non-linear effects that you cannot control, this will slow convergence and limit how much of the siren can be removed, however the result may still be OK, in fact XVF3800 has specific tuning guidance for non-linear residual echos https://www.xmos.com/documentation/XM-0 ... v3.2.1.pdf (page 39 FYI, but if you do go this route please go through the guide in order before attempting to tune the nonlinear echo!)
- acoustic overload point, this applies to both your intercom mics and a mic used to gather the echo reference (if you go that route): you will need a mic with a high AOP, most MEMS mics will already meet this spec, but if you are sampling up close to a siren there may be other considerations! Ideally neither your reference or comms would ever clip, as this is an example of a nonlinear distortion and a loss of information that may cause instability. If you are new to MEMS mics, then this is a great reference: https://www.infineon.com/dgdl/Infineon- ... 62021&da=t
Cheers,
-Andrew