|
|
|
|
go to Japanese page (latest information)
Adaptive howling canceller demo program for Windows PC and ASIO audio device is avairable. (Sorry, English documentation is not avairable)
Basic howling canceller cofiguration, only for elementary experiment, poor
sound quality, with long echoes and large latency.
But it works very well, howling is suppressed perfectly.
Howling canceller LSI is under development now.
Our research & development of howling canceller is not the research of the simulation of howling canceller.
Block Diagram canceller_blockdiagram.pdf (65KB)
Block Diagram (Expanded) canceller_blockdiagram_expand.pdf (743KB)
FIR filter section
LMS algorithm section
Block Diagram of Adaptive Howling Canceller / Block Diagram of Adaptive Acoustic Feedback Canceller (1 page abstract)
adaptive_howling_canceller_block_diagram.pdf
adaptive_howling_canceller_block_diagram.png
Improvement of Howling Canceller using Adaptive Filter with Delay Circuit
[in Japanese, with English abstract and figure captions]
IEICE technical report EA 108(491), 13-18, 2009-03-20
https://ci.nii.ac.jp/naid/110007328695/en/
IEICE = The Institute of Electronics, Information and Communication Engineers
EA means Electro-Acoustics
Basic adaptive howling canceller block diagram (without limiter)
Acoustic path is not feedback path.
Cut feedback path → There is not closed signal path. It means the cutted signal path is really feedback path.
Cut acoustic path → System is not always stable because still there is closed signal loop via adaptive filter. Acoustic path is not feedback path!
H(ω) : Acoustic system
W(ω) : Adaptive filter
A limiter circuit is added to ensure the stability of adaptive howling canceller. (see the figure of experiment below)
Limiter prevents saturation of D/A, power amp, speaker, microphone, mic amp and A/D.
Feedback Path Position of Adaptive Howling Canceller / Feedback Path Position of Adaptive Acoustic Feedback Canceller (1 page abstract)
feedback_path.pdf
feedback_path.png
Wrong model of adaptive howling canceller
Don't use wrong model (left).
Many research of the simulation of adaptive howling canceller use this wrong model.
(e.g. Toon van Waterschoot and Marc Moonen, "Fifty years of acoustic feedback control: state of the art and future challenges," Proc. IEEE, vol. 99, no. 2, Feb. 2011, pp. 288-327)
Acoustic path including power amp and mic amp has gain grater than 0dB. It means acoustic path is not feedback path. (Correct model, right )
Feedback path don't have gain grater than 0dB.
Wrong and Correct Model of Adaptive Howling Canceller / Wrong and Correct Model of Adaptive Acoustic Feedback Canceller (1 page abstract)
howling_canceller_model.pdf
howling_canceller_model.png
How to measure acoustic path gain of adaptive howling canceller
y[n]rms/x[n]rms is the gain of the acoustic path of adaptive howling canceller system.
How to measure acoustic path gain of adaptive howling canceller / How to measure acoustic path gain of adaptive acoustic feedback canceller (1 page abstract)
gain.pdf
gain.png
Link to the article about open loop gain measurement of P.A. system (without adaptive howling canceller) on ProSoundWeb
- Acoustic Feedback Part 2 – Open Loop Function Measurement (page 1/2) https://www.prosoundweb.com/acoustic-feedback-part-2-open-loop-function-measurement/
- Acoustic Feedback Part 2 – Open Loop Function Measurement (page 2/2) https://www.prosoundweb.com/acoustic-feedback-part-2-open-loop-function-measurement/2/
Open loop model of howling canceller
If the gain of H(w) increase, noise level in d[n] decrease relatively.
Theoretically higer gain public address system with howling canceller become more stable. (Incredible!)
Open loop model of adaptive howling canceller / Open loop model of adaptive acoustic feedback canceller (1 page abstract)
open_loop_model.pdf
open_loop_model.png
Change gain simulation program for Scilab
- gain=6dB sim6db.sce
- gain=60dB sim60db.sce
Higher gain system has better miss alignment characterlistics.
Simulation to test the effect of decorrelator
simulation program for Scilab decorrelator_effect.sce
left : simulation model without decorrelator
right : simulation model with decorrelator (delay)
input signal spectrum (spectrum of s)
colored signal
output signal spectrum of identification system without decorrelator (spectrum of e, x in left simulation model figure)
whitened signal
output signal spectrum of identification system with decorrelator/delay (spectrum of e, ze, x in right simulation model figure)
Experiment at Tokyo Metropolitan Industrial Technology Research Instuitute (TIRI)
Limiter characteristics
K is a threshold value.
Anechoic room
Experiment result
Please see and listen following video.
video.avi (61MB),
video.wmv (4MB)
Adaptive filter coefficient
Adaptive filter gain (grater than 10dB) = PA system gain
This PA system gain includes A/D gain, D/A gain, scaling gain and the gain
of fixed gain block (15dB)
Howling canceller simulation program for Scilab
Scilab simulation program (Acoustic feedback cancellation code) scilab_canceller.zip (710KB)
Simulation output signal (howling is suppressed) sim_out.wav (400KB)
Howling canceller demo system for echoic room using DSP (Digital Signal
Processor)
This system is used for demonstration at the meeting of AES (Audio Engineering Society) Japan Section.
AES Section Meeting Reports, Japan - December 15, 2016
Inside of DSP unit
We use C6713 DSK DSP board (TI/Spectrum Digital).
Example of coeffcients of the adaptive filter of howling canceller demo
system
and its amplitude response (= acoustic sysytem gain grater than 0dB)
This is experiment result in rather small room (Reverberation time = 300ms)
Example of the acoustic path impulse response of the room we use to test
our adaptive howling canceller (fs=8kHz). room_impres.txt
Please consider there is conversion delay of A/D, D/A and omitted leading
zero data etc.
The distance between speaker and microphone is 2.3m.
It looks slight strange response shape because it includes phase rotation
of measurement system etc.
Experiment at Tokyo University of the Arts, Senjyu Campus
The microphone suited for adaptive howling canceller is omnidirectional
! (or binomial array)
This is one of incredible characteristics of adaptive howling canceller.
Experiment at large stadio (reverberation time = 1second)
Experiment at calss room
Experiment with Windows PC and low price external USB audio interface (no DSP/FPGA board)
Hardware
- Desktop PC, Intel Core i3-2100T CPU @2.5GHz
- external USB audio interface, Zoom U-22 (fs=48kHz)
The latency of U-22 is rather longer than other ASIO audio interface but very stable without large latency drift.
link to ASIO loopback test program page (in Japanese)
Software
- LMS adaptive howling canceler including sample rate converter (48kHz/8kHz) by FreeBASIC
- with PortAudio ASIO interface library (open source software)
- with OpenBLAS fast vector math library (open source software)
FreeBASIC version of howling canceller software doesn't have any auxiliary processsing routine to improve perfomances and the frequency shift of decorrelator is large (5Hz).
So its sound quality is not good but easy to demonstrate howling canceller in large room with long reverberation time using notebook PC.
The length of adaptive filter is 2048tap/8000tap. (fs=8kHz)
Block Diagram
Experiment result
filter coefficient of adaptive filter and its frequency response (peak gain = 10dB)
2048 tap
8000 tap
Download adaptive howling canceller demo program for Windows PC and USB audio interface (ASIO)
audiokit_demo07.zip (14Mbyte)
souce code of adaptive howling canceller by FreeBASIC (only main routine) asio_canceller.bas
improved version binary asio_canceller2.exe
Basic howling canceller cofiguration, only for elementary experiment, poor sound quality and large latency.
English documentation is not avairable.
Tips : "higher D/A output and powr amp gain, lower A/D and mic amp gain" setting is better than "lower D/A gain, higher A/D gain" setting.
audio equipment example for
howling canceller experiment
Line array speaker for howling canceller experiment
Japanese page is here
|
|
