DOA estimation

i have implemented MUSIC and ESPRIT algorithm using Matlab.i want to
know what are the variations or modifications  that can be done to
these algorithms.i have studied effect of varying SNR,antenna
elements,angular separation on these algorithms.but something more is
required.

On 27 Nov, 14:29, preetigup...@gmail.com wrote:
> i have implemented MUSIC and ESPRIT algorithm using Matlab.i want to
> know what are the variations or modifications �that can be done to
> these algorithms.i have studied effect of varying SNR,antenna
> elements,angular separation on these algorithms.but something more is
> required.

With ESPRIT you can check the Least Means Square method
against the Total Least Squares method. With MUSIC you
can check e.g. Classical MUSIC agains Root MUSIC (I remember
having read somewhere about a Pop MUSIC variation too...)

The most interesting tests are with model mis-match scenarios.
Both these methods are designed with a non-damped unscaled
expunential signal in mind. Try them with either a damping term
or a scale term and see how well they perform. Then try and
increase the number of signals. At some point, near N/2 where
N is the number of elements in the array, you might notice
some changes in the behaviour of the methods.

Rune

>On 27 Nov, 14:29, preetigup...@gmail.com wrote:
>> i have implemented MUSIC and ESPRIT algorithm using Matlab.i want to
>> know what are the variations or modifications =A0that can be done to
>> these algorithms.i have studied effect of varying SNR,antenna
>> elements,angular separation on these algorithms.but something more is
>> required.
>
>With ESPRIT you can check the Least Means Square method
>against the Total Least Squares method. With MUSIC you
>can check e.g. Classical MUSIC agains Root MUSIC (I remember
>having read somewhere about a Pop MUSIC variation too...)
>
>The most interesting tests are with model mis-match scenarios.
>Both these methods are designed with a non-damped unscaled
>expunential signal in mind. Try them with either a damping term
>or a scale term and see how well they perform. Then try and
>increase the number of signals. At some point, near N/2 where
>N is the number of elements in the array, you might notice
>some changes in the behaviour of the methods.
>
>Rune
>

thanks for replying.i cant compare LS-ESPRIT with TLS.cant work with
existing methods.have to do some little modification in MUSIC and ESPRIT.i
am not getting your point that how to use damping term or scale term.can
you give an example that what exactly signal can i take.it will be of
help.

Preeti

On 28 Nov, 13:59, "preetigupta" <preeti_engin...@yahoo.com> wrote:
> >On 27 Nov, 14:29, preetigup...@gmail.com wrote:
> >> i have implemented MUSIC and ESPRIT algorithm using Matlab.i want to
> >> know what are the variations or modifications =A0that can be done to
> >> these algorithms.i have studied effect of varying SNR,antenna
> >> elements,angular separation on these algorithms.but something more is
> >> required.
>
> >With ESPRIT you can check the Least Means Square method
> >against the Total Least Squares method. With MUSIC you
> >can check e.g. Classical MUSIC agains Root MUSIC (I remember
> >having read somewhere about a Pop MUSIC variation too...)
>
> >The most interesting tests are with model mis-match scenarios.
> >Both these methods are designed with a non-damped unscaled
> >expunential signal in mind. Try them with either a damping term
> >or a scale term and see how well they perform. Then try and
> >increase the number of signals. At some point, near N/2 where
> >N is the number of elements in the array, you might notice
> >some changes in the behaviour of the methods.
>
> >Rune
>
> thanks for replying.i cant compare LS-ESPRIT with TLS.cant work with
> existing methods.have to do some little modification in MUSIC and ESPRIT.i
> am not getting your point that how to use damping term or scale term.can
> you give an example that what exactly signal can i take.it will be of
> help.

The basic signal model for MUSIC and ESPRIT is the complex
exponential

s(x) = exp(jkx)

where k is the wavenumber you use to derive the DoA,

k = cos(phi)/c.

In the real world there will be a damping term,

s(x) = exp(jkx + ax)

and with point sources there might be scaling terms
as well, if you measure the signal close to the source

s(x) = 1/sqrt(kx) exp(jkx).

So try and model the signals with daming terms or scaling
terms, and see how well the DoA estimators perform.

- How large daming terms can they cope with?
- How close to the source can you get and still
  get good results?

And so on.

Rune

>On 28 Nov, 13:59, "preetigupta" <preeti_engin...@yahoo.com> wrote:
>> >On 27 Nov, 14:29, preetigup...@gmail.com wrote:
>> >> i have implemented MUSIC and ESPRIT algorithm using Matlab.i want
to
>> >> know what are the variations or modifications =A0that can be done
to
>> >> these algorithms.i have studied effect of varying SNR,antenna
>> >> elements,angular separation on these algorithms.but something more
is
>> >> required.
>>
>> >With ESPRIT you can check the Least Means Square method
>> >against the Total Least Squares method. With MUSIC you
>> >can check e.g. Classical MUSIC agains Root MUSIC (I remember
>> >having read somewhere about a Pop MUSIC variation too...)
>>
>> >The most interesting tests are with model mis-match scenarios.
>> >Both these methods are designed with a non-damped unscaled
>> >expunential signal in mind. Try them with either a damping term
>> >or a scale term and see how well they perform. Then try and
>> >increase the number of signals. At some point, near N/2 where
>> >N is the number of elements in the array, you might notice
>> >some changes in the behaviour of the methods.
>>
>> >Rune
>>
>> thanks for replying.i cant compare LS-ESPRIT with TLS.cant work with
>> existing methods.have to do some little modification in MUSIC and
ESPRIT.i
>> am not getting your point that how to use damping term or scale
term.can
>> you give an example that what exactly signal can i take.it will be of
>> help.
>
>The basic signal model for MUSIC and ESPRIT is the complex
>exponential
>
>s(x) = exp(jkx)
>
>where k is the wavenumber you use to derive the DoA,
>
>k = cos(phi)/c.
>
>In the real world there will be a damping term,
>
>s(x) = exp(jkx + ax)
>
>and with point sources there might be scaling terms
>as well, if you measure the signal close to the source
>
>s(x) = 1/sqrt(kx) exp(jkx).
>
>So try and model the signals with daming terms or scaling
>terms, and see how well the DoA estimators perform.
>
>- How large daming terms can they cope with?
>- How close to the source can you get and still
>  get good results?
>
>And so on.
>
>Rune
>
>
Thanks for giving such valuable information.but i am facing problem.when i
add damping terms to my algorithm,then it is not giving accurate results.i
have tried both positive and negative term for damping.can u help.

Preeti

On 2 Des, 11:39, "preetigupta" <preeti_engin...@yahoo.com> wrote:
> >On 28 Nov, 13:59, "preetigupta" <preeti_engin...@yahoo.com> wrote:
> >> >On 27 Nov, 14:29, preetigup...@gmail.com wrote:
> >> >> i have implemented MUSIC and ESPRIT algorithm using Matlab.i want
> to
> >> >> know what are the variations or modifications =A0that can be done
> to
> >> >> these algorithms.i have studied effect of varying SNR,antenna
> >> >> elements,angular separation on these algorithms.but something more
> is
> >> >> required.
>
> >> >With ESPRIT you can check the Least Means Square method
> >> >against the Total Least Squares method. With MUSIC you
> >> >can check e.g. Classical MUSIC agains Root MUSIC (I remember
> >> >having read somewhere about a Pop MUSIC variation too...)
>
> >> >The most interesting tests are with model mis-match scenarios.
> >> >Both these methods are designed with a non-damped unscaled
> >> >expunential signal in mind. Try them with either a damping term
> >> >or a scale term and see how well they perform. Then try and
> >> >increase the number of signals. At some point, near N/2 where
> >> >N is the number of elements in the array, you might notice
> >> >some changes in the behaviour of the methods.
>
> >> >Rune
>
> >> thanks for replying.i cant compare LS-ESPRIT with TLS.cant work with
> >> existing methods.have to do some little modification in MUSIC and
> ESPRIT.i
> >> am not getting your point that how to use damping term or scale
> term.can
> >> you give an example that what exactly signal can i take.it will be of
> >> help.
>
> >The basic signal model for MUSIC and ESPRIT is the complex
> >exponential
>
> >s(x) = exp(jkx)
>
> >where k is the wavenumber you use to derive the DoA,
>
> >k = cos(phi)/c.
>
> >In the real world there will be a damping term,
>
> >s(x) = exp(jkx + ax)
>
> >and with point sources there might be scaling terms
> >as well, if you measure the signal close to the source
>
> >s(x) = 1/sqrt(kx) exp(jkx).
>
> >So try and model the signals with daming terms or scaling
> >terms, and see how well the DoA estimators perform.
>
> >- How large daming terms can they cope with?
> >- How close to the source can you get and still
> > &#4294967295;get good results?
>
> >And so on.
>
> >Rune
>
> Thanks for giving such valuable information.but i am facing problem.when i
> add damping terms to my algorithm,then it is not giving accurate results.i
> have tried both positive and negative term for damping.can u help.

That's what the test was intended to demonstrate. There is
nothing you can do about that, other than do tests and see
how sensitive the DoA estimate is with respect to all these
terms that are not included in the sum-of-sines signal.

Rune