What is limited this FXLMS performance

Question

This is simulation data flow:

/*
                    Full wideband FX-lms schmatic
.....................................................................................
    disturbance                      -----          noise
----------------------------------->|  P  |----------------------
               |                     -----                       |
             -----                                               |
     ->     |  A  |                                              |
Duct in      -----                                               |     Duct outlet->
     ->     x  |                                                 |
               v   echo      -----     out      -----   anoise   v
              |+|<----------|  E  |<---------->|  S  |--------->|+|
         mic1  |             -----   \__|__/    -----            | mic2
...............|........................|........................|...................
    x_echoic + v  echo_h     -----      |                        |
              |+|<----------| EC  |<----|                        |
               | -           -----      | out                    |
               |                        |                        |
               | x_hat       --\--      |                  error |              DSP
               |----------->|  W\ |-----                         |
             ----            ----\                               |
            |  S |                \                              |
             ----  ----           -------                        |
      fx_hat   L--| SD |-------->|  LMS  |<----------------------

And here is the setup with about 4 meter long tube between error andprimary mic. with 110cm diameter:

Mic to Mic impulse response:

In my simulation some spot of frequency response have not been cancelled. Though total cancellation of about 10dB has been achieved. You can see the effect of cancellation on this, Also the secoundary path and echo path filter response is depicted which are FIR 1500 tap. here 2048 point FFT has been taken. ANC ON signal is after 7 miniute after turning ANC on.

Is there anyway to improve this response?

The amount of cancellation that you can reasonable expect depends A LOT on the details of your specific setup. Your current performance could be great it could be terrible. We can't tell without a lot more detail — Hilmar, Jun 22 '23 at 11:43
@Hilmar Dataflow has been added. I think it shoud be enough. Is there something else you needed? — mohammadsdtmnd, Jun 22 '23 at 11:46
Your current data flow (I would call it block diagram) is quite hard to follow. Could you possibly create a block diagram on some software other than a text editor to depict the system in an easier-to-read form? — ZaellixA, Jun 22 '23 at 13:41
@mohammadsdtmnd: It would be help to describe your physical setup. Your performance is already great if you trying to do ANC in a room where speaker and microphone are 1 meter apart. It's abysmal for a an ANC headset. Context matters here. — Hilmar, Jun 22 '23 at 15:23
@Hilmar : The setup is on the tube and I thibk performance is poor. I've updated by this information. Now infoes must be enouph doesn't it? — mohammadsdtmnd, Jun 24 '23 at 07:31
@mohammadsdtmnd: That's a long tube. How does the transfer function from the noise to the error mic look like? Unless the tube is well terminated you are going to see some severe resonances and reflections in there — Hilmar, Jun 24 '23 at 20:47
@Hilmar : It's impulse reponse added. And seems it's freq reponse is something like when ANC is off. — mohammadsdtmnd, Jun 25 '23 at 06:22

score 1 · Answer 1 · answered Jul 06 '23 at 06:57

1

To equalize the microphone I recommend using the Wiener Hopf equations to determine the least squares FIR solution for channel compensation, assuming a known sounding sequence can be used that spectrally fills the channel (such as a chirp or pseudo-random sequence with a bandwidth over the full audio band used). I detail the application of this here:

Compensating Loudspeaker frequency response in an audio signal

answered Jul 06 '23 at 06:57

Dan Boschen

50,942
2
57
135

Is there too many difference between lms and ls? – mohammadsdtmnd Jul 06 '23 at 07:00
Yes the key difference is this will be an FIR solution so numerically stable and easy to implement. Also if you review the code, the process is elegantly simple. – Dan Boschen Jul 06 '23 at 07:03
The compliment was for which algorithm? Ls or lms? – mohammadsdtmnd Jul 06 '23 at 07:05
Also here I have estimated echo and secoundary path with lms filter. Are you meaning to estimate them using ls? And that will enhance the performance? – mohammadsdtmnd Jul 06 '23 at 07:06
Please read through the referenced post and we can catch up later – Dan Boschen Jul 06 '23 at 07:50

What is limited this FXLMS performance

1 Answers1