Forums

Magnitude response of linear predictor (interference cancellation)

Started by Eric Huay January 12, 2014
I'm trying to implement an interference cancellation using linear prediction.
A block diagramtherefore is shown in [1]. The adaptive algorithm as a 
LMS (least mean square)-algorithm. This works quite well!

The overall system is something like notch-filter. How can I obtain 
this magnitude response of the overall system?

Thank you and best regards
Eric

[1] http://goo.gl/UtAkgM

On Sunday, January 12, 2014 10:58:42 PM UTC+13, Eric Huay wrote:
> I'm trying to implement an interference cancellation using linear prediction. > > A block diagramtherefore is shown in [1]. The adaptive algorithm as a > > LMS (least mean square)-algorithm. This works quite well! > > > > The overall system is something like notch-filter. How can I obtain > > this magnitude response of the overall system? > > > > Thank you and best regards > > Eric > > > > [1] http://goo.gl/UtAkgM
Unless the interference changes with frequency then all you need is a notch filter.
On 1/12/14 3:10 PM, gyansorova@gmail.com wrote:
> On Sunday, January 12, 2014 10:58:42 PM UTC+13, Eric Huay wrote: >> I'm trying to implement an interference cancellation using linear prediction. >> >> A block diagramtherefore is shown in [1]. The adaptive algorithm as a >> >> LMS (least mean square)-algorithm. This works quite well! >> >> >> >> The overall system is something like notch-filter. How can I obtain >> >> this magnitude response of the overall system? >> >> >> >> Thank you and best regards >> >> Eric >> >> >> >> [1] http://goo.gl/UtAkgM > > Unless the interference changes with frequency then all you need is a notch filter.
somebuddy's gotta tune it. -- r b-j rbj@audioimagination.com "Imagination is more important than knowledge."
On Monday, January 13, 2014 9:29:05 AM UTC+13, robert bristow-johnson wrote:
> On 1/12/14 3:10 PM, gyansorova@gmail.com wrote: > > > On Sunday, January 12, 2014 10:58:42 PM UTC+13, Eric Huay wrote: > > >> I'm trying to implement an interference cancellation using linear prediction. > > >> > > >> A block diagramtherefore is shown in [1]. The adaptive algorithm as a > > >> > > >> LMS (least mean square)-algorithm. This works quite well! > > >> > > >> > > >> > > >> The overall system is something like notch-filter. How can I obtain > > >> > > >> this magnitude response of the overall system? > > >> > > >> > > >> > > >> Thank you and best regards > > >> > > >> Eric > > >> > > >> > > >> > > >> [1] http://goo.gl/UtAkgM > > > > > > Unless the interference changes with frequency then all you need is a notch filter. > > > > somebuddy's gotta tune it. > > > > -- > > > > r b-j rbj@audioimagination.com > > > > "Imagination is more important than knowledge."
But if the freq is fixed so what. Trim it in. I suppose if you already have a DSP in place and plenty grunt then maybe an LMS.
Eric Huay <e.huay@nospam.com> writes:

> I'm trying to implement an interference cancellation using linear prediction. > A block diagramtherefore is shown in [1]. The adaptive algorithm as a > LMS (least mean square)-algorithm. This works quite well! > > The overall system is something like notch-filter. How can I obtain > this magnitude response of the overall system? > > Thank you and best regards > Eric
In the frequency domain, E(z) = D(z) - F(z) * z^(-delta) * D(z) = (1 - H(z) * z^(-delta)) * D(z) and so H(z) = E(z) / D(z) = 1 - H(z) * z^(-delta) where F(z) is the adaptive filter response. -- Randy Yates Digital Signal Labs http://www.digitalsignallabs.com
Randy Yates <yates@digitalsignallabs.com> writes:

> Eric Huay <e.huay@nospam.com> writes: > >> I'm trying to implement an interference cancellation using linear prediction. >> A block diagramtherefore is shown in [1]. The adaptive algorithm as a >> LMS (least mean square)-algorithm. This works quite well! >> >> The overall system is something like notch-filter. How can I obtain >> this magnitude response of the overall system? >> >> Thank you and best regards >> Eric > > In the frequency domain, > > E(z) = D(z) - F(z) * z^(-delta) * D(z) > = (1 - H(z) * z^(-delta)) * D(z) > > and so > > H(z) = E(z) / D(z) > = 1 - H(z) * z^(-delta) > > where F(z) is the adaptive filter response.
Correction: In the frequency domain, E(z) = D(z) - F(z) * z^(-delta) * D(z) = (1 - F(z) * z^(-delta)) * D(z) and so H(z) = E(z) / D(z) = 1 - F(z) * z^(-delta) where F(z) is the adaptive filter response. -- Randy Yates Digital Signal Labs http://www.digitalsignallabs.com
On 1/12/14 6:03 PM, gyansorova@gmail.com wrote:
> On Monday, January 13, 2014 9:29:05 AM UTC+13, robert bristow-johnson wrote: >> On 1/12/14 3:10 PM, gyansorova@gmail.com wrote: >> >>> Unless the interference changes with frequency then all you need is a notch filter. >> >> somebuddy's gotta tune it. > > But if the freq is fixed so what. Trim it in. I suppose if you already have a DSP in place and plenty grunt then maybe an LMS.
or your DSP can be a notch and some other alg to sense it's outa tune and to slowly adjust the resonant frequency and maybe the Q and the depth. like a parametric EQ. i thought that's how that Sabine Feedback Exterminator worked. not an LMS, which i don't think worked so well on the live stage. -- r b-j rbj@audioimagination.com "Imagination is more important than knowledge."