COMPGROUPS.NET | Search | Post Question | Groups | Stream | About | Register

### LMS convergence

• Email
• Follow

```I built an LMS filter, and I have better convergence when I use an
alternate signal as input, than a DC(all x(n)>0) one. Did someone have an
thanks

```
 0
Reply mmoctar (16) 11/28/2008 1:00:14 PM

See related articles to this posting

```On Nov 29, 2:00=A0am, "mmoctar" <mmoc...@gmail.com> wrote:
> I built an LMS filter, and I have better convergence when I use an
> alternate signal as input, than a DC(all x(n)>0) one. Did someone have an
> explanation about it?
> thanks

The signal needs to be "persistently exciting" so the nearer it is to
white noise the better. You could use a square wave and it would work
when it changes.

SS
```
 0
Reply sheepshaggerx (35) 11/29/2008 7:59:59 PM

```On Nov 28, 6:00=A0pm, "mmoctar" <mmoc...@gmail.com> wrote:
> I built an LMS filter, and I have better convergence when I use an
> alternate signal as input, than a DC(all x(n)>0) one. Did someone have an
> explanation about it?
> thanks

What you want from the training symbol? To experience the channel at
all frequencies or just at the DC?
A DC signal has a fourier transform of an impulse at the 0 frequency.
When this signal is passed through the channel, it can tell receiver
what channel did to the DC component only and not all the frequency.
So a better training symbol choice would be one that contains
components of all the frequencies. The only signal satisfying this
criteria is an impulse function (remember an impulse has fourier
transform of DC i.e. all the frequencies with the same amplitude).
This signal can probe the channel at all the frequencies. You can use
a truncated-in-time impulse function as training symbol.

Ubaid Abdullah
DSP & Communication Engineer
http://dspdotcomm.blogspot.com
```
 0
Reply ubaidabdullah (25) 12/1/2008 6:24:56 AM

2 Replies
52 Views

Similar Articles

12/7/2013 7:00:35 PM
[PageSpeed]

Similar Artilces:

region of convergence in matlab
I have equations of 7 circles.I need to plot the region of convergence of all these circles using matlab.How do i do that? arjun <arjunmvgr504@gmail.com> wrote in message <30390214.1232859778470.JavaMail.jakarta@nitrogen.mathforum.org>... > I have equations of 7 circles.I need to plot the region of convergence of all these circles using matlab.How do i do that? What is the 'region of convergence' of circles?

PROC MCMC + Convergence
Hi, I am trying to use PROC MCMC to run a linear Bayesian model and truncate the posteriors and am running into issues with convergence . I am unable to do the latter using PROC GENMOD or PROC PHREG. Could you please advice on what the problem could be. As an alternative if you could let me know if any other SAS procedures allow truncation of posteriors? Thanks Susan

bvp4c convergence problems
hi all, i'm solving 4th order boundary value problem with an unknown parameter for the eigenenergy.i'm using matlab's bvp4c and it seems the routine cannot limit itself to a specific eigenfunction/value. my solutions should be all ground state solutions, but my code keeps switching to higher states with unsuitable solutions and eigenenergies. now i know bvp4c was not made to find a specific eigenvalue, so my question is: is there any other BVP routine in matlab which can find a specific eigenvalue? or, alternatively - does anyone have any suggestions how i could force bvp4c to li

problem with fzero convergence
hi, I have a function which which should be strictly monotonic and has a zero but the problem is that the slope is extremely low close to the zero. I did fzero and it gave me: X=0.969143 FVAL = 1.2573 EXITFLAG =1 Clearly, it's not a zero... The output is: intervaliterations: 1 iterations: 48 funcCount: 50 algorithm: 'bisection, interpolation' message: 'Zero found in the interval [0.95984, 0.987778]' I also calculated the value of the function close to the "zero" and it gave me x f(x) 0.9

Momentum doesn't speed up the convergence
Hi: I use multilayer perceptron to do the function approximation, the training function I used was Traingdm, the problem is: no matter how I changed the value of momentum, the convergence speed was the same, was I missing anything? Here is the partial code(Matlab) I used: TF1='logsig'; TF2='purelin'; BTF='traingdm'; BLF='learngdm'; PF='mse'; net=newff([PX;PY],[s1 s2],{TF1 TF2},BTF,BLF,PF); net=init(net); net.trainParam.epochs=7000; net.trainParam.mc=0.9;%I changed the value of momentum here! net.trainParam.lr=0.2; net=train(net,p,t); Any help will be appreciated! Cutty

Hey guys Does anyone know the big-O of adaptive Gauss-Kronrod quadrature- quadgk? Any paper out there about the quadgk and also the way it treats singularity? Thanks Ron

PROC GLIMMIX vs %GLIMMIX
Hi everyone, I used the %GLIMMIX macro before but have switched to the PROC GLIMMIX recently to take advantage of the added features. The problem is that for the same model and dataset, the convergence criterion is met with the macro but not for the procedure...I get a "DID NOT CONVERGE" message. Anyone can tell what appears to be the difference between my two programs? Proc glimmix data=pairs4 ; class ID_pair year; model TARGETp (order=data)= REL1 DUMq MA_ACTIVITY SIZEp MBp LEVERAGEp MGNTp EPp / solution dist=binomial link=logit error=binomial; random intercept / subject=year; run; vs %glimmix(data=pairs4, procopt=method=reml, stmts=%str( class ID_pair year; model TARGETp = REL1 DUMq MA_ACTIVITY SIZEp MBp LEVERAGEp MGNTp EPp ; random INTERCEPT/ SUBJECT=year; ), error=binomial, link=logit, solution ); run; What can explain the no convergence? Thanks in advance, Claudia