Anyone encountered this problem when using NLMS? The h shifts up after running for sometime, I noticed this for both white noise and color voice signals. What caused the shift and how to avoid it? thanks Andrew
DC shift of the h(impulse response of the room) when using NLMS adaptive filtering
Started by ●September 16, 2003
Reply by ●September 16, 20032003-09-16
Andrew Xiang wrote:> > Anyone encountered this problem when using NLMS? The h shifts up after > running for sometime, I noticed this for both white noise and color > voice signals. What caused the shift and how to avoid it?Place a good HPF in front of NLMS. Vladimir Vassilevsky DSP and Mixed Signal Design Consultant http://www.abvolt.com
Reply by ●September 17, 20032003-09-17
DC is high cut off, however, there is some low freq stuff. The HPF is a second order IIR filter with slow roll off but about -80dB at DC. Vladimir Vassilevsky <vlv@abvolt.com> wrote in message news:<3F679238.95F1E113@abvolt.com>...> Andrew Xiang wrote: > > > > Anyone encountered this problem when using NLMS? The h shifts up after > > running for sometime, I noticed this for both white noise and color > > voice signals. What caused the shift and how to avoid it? > > Place a good HPF in front of NLMS. > > Vladimir Vassilevsky > > DSP and Mixed Signal Design Consultant > > http://www.abvolt.com
Reply by ●September 18, 20032003-09-18
LMS requires an infinite attenuation at DC. Even the smallest DC shift at the input will eventually cause an error buildup which I believe you are observing. Vladimir Vassilevsky DSP and Mixed Signal Design Consultant http://www.abvolt.com Andrew Xiang wrote:> > DC is high cut off, however, there is some low freq stuff. The HPF is > a second order IIR filter with slow roll off but about -80dB at DC. > > Vladimir Vassilevsky <vlv@abvolt.com> wrote in message news:<3F679238.95F1E113@abvolt.com>... > > Andrew Xiang wrote: > > > > > > Anyone encountered this problem when using NLMS? The h shifts up after > > > running for sometime, I noticed this for both white noise and color > > > voice signals. What caused the shift and how to avoid it? > > > > Place a good HPF in front of NLMS. > > > > Vladimir Vassilevsky > > > > DSP and Mixed Signal Design Consultant > > > > http://www.abvolt.com
Reply by ●September 18, 20032003-09-18
Hmmm, infinite attenuation? how to realize that? Is it possible? Vladimir Vassilevsky <vlv@abvolt.com> wrote in message news:<3F693329.F22BEB71@abvolt.com>...> LMS requires an infinite attenuation at DC. Even the smallest DC shift > at the input will eventually cause an error buildup which I believe you > are observing. > > > Vladimir Vassilevsky > > DSP and Mixed Signal Design Consultant > > http://www.abvolt.com > > Andrew Xiang wrote: > > > > DC is high cut off, however, there is some low freq stuff. The HPF is > > a second order IIR filter with slow roll off but about -80dB at DC. > > > > Vladimir Vassilevsky <vlv@abvolt.com> wrote in message news:<3F679238.95F1E113@abvolt.com>... > > > Andrew Xiang wrote: > > > > > > > > Anyone encountered this problem when using NLMS? The h shifts up after > > > > running for sometime, I noticed this for both white noise and color > > > > voice signals. What caused the shift and how to avoid it? > > > > > > Place a good HPF in front of NLMS. > > > > > > Vladimir Vassilevsky > > > > > > DSP and Mixed Signal Design Consultant > > > > > > http://www.abvolt.com
Reply by ●September 18, 20032003-09-18
There is no problem with achieving an infinite attenuation at DC or any particular frequency. Just create zero in the transfer function at that frequency. Blocking DC is very simple: the numerator should look like z - z[-1] for the first order filter or z - 2*z[-1] + z[-2] for the second order. That condition is satisfied for any Butterworth, Bessel or Chebyshev-1 filter. Vladimir Vassilevsky DSP and Mixed Signal Design Consultant http://www.abvolt.com Andrew Xiang wrote:> > Hmmm, infinite attenuation? how to realize that? Is it possible? > > Vladimir Vassilevsky <vlv@abvolt.com> wrote in message news:<3F693329.F22BEB71@abvolt.com>... > > LMS requires an infinite attenuation at DC. Even the smallest DC shift > > at the input will eventually cause an error buildup which I believe you > > are observing. > > > > > > Vladimir Vassilevsky > > > > DSP and Mixed Signal Design Consultant > > > > http://www.abvolt.com > > > > Andrew Xiang wrote: > > > > > > DC is high cut off, however, there is some low freq stuff. The HPF is > > > a second order IIR filter with slow roll off but about -80dB at DC. > > > > > > Vladimir Vassilevsky <vlv@abvolt.com> wrote in message news:<3F679238.95F1E113@abvolt.com>... > > > > Andrew Xiang wrote: > > > > > > > > > > Anyone encountered this problem when using NLMS? The h shifts up after > > > > > running for sometime, I noticed this for both white noise and color > > > > > voice signals. What caused the shift and how to avoid it? > > > > > > > > Place a good HPF in front of NLMS. > > > > > > > > Vladimir Vassilevsky > > > > > > > > DSP and Mixed Signal Design Consultant > > > > > > > > http://www.abvolt.com
Reply by ●September 19, 20032003-09-19
Please give me a complete example. 2nd order, 3dB <=180hz. Vladimir Vassilevsky <vlv@abvolt.com> wrote in message news:<3F69CD29.98C97F52@abvolt.com>...> There is no problem with achieving an infinite attenuation at DC or any > particular frequency. Just create zero in the transfer function at that > frequency. > > Blocking DC is very simple: the numerator should look like z - z[-1] for > the first order filter or z - 2*z[-1] + z[-2] for the second order. That > condition is satisfied for any Butterworth, Bessel or Chebyshev-1 > filter. > > Vladimir Vassilevsky > > DSP and Mixed Signal Design Consultant > > http://www.abvolt.com > > > > > Andrew Xiang wrote: > > > > Hmmm, infinite attenuation? how to realize that? Is it possible? > > > > Vladimir Vassilevsky <vlv@abvolt.com> wrote in message news:<3F693329.F22BEB71@abvolt.com>... > > > LMS requires an infinite attenuation at DC. Even the smallest DC shift > > > at the input will eventually cause an error buildup which I believe you > > > are observing. > > > > > > > > > Vladimir Vassilevsky > > > > > > DSP and Mixed Signal Design Consultant > > > > > > http://www.abvolt.com > > > > > > Andrew Xiang wrote: > > > > > > > > DC is high cut off, however, there is some low freq stuff. The HPF is > > > > a second order IIR filter with slow roll off but about -80dB at DC. > > > > > > > > Vladimir Vassilevsky <vlv@abvolt.com> wrote in message news:<3F679238.95F1E113@abvolt.com>... > > > > > Andrew Xiang wrote: > > > > > > > > > > > > Anyone encountered this problem when using NLMS? The h shifts up after > > > > > > running for sometime, I noticed this for both white noise and color > > > > > > voice signals. What caused the shift and how to avoid it? > > > > > > > > > > Place a good HPF in front of NLMS. > > > > > > > > > > Vladimir Vassilevsky > > > > > > > > > > DSP and Mixed Signal Design Consultant > > > > > > > > > > http://www.abvolt.com