On Mon, 09 Jun 2008 09:10:57 -0500, Vladimir Vassilevsky
<antispam_bogus@hotmail.com> wrote:
>
>
>Steve Underwood wrote:
>> Vladimir Vassilevsky wrote:
>>
>>
>> I try to be a novelist, and do something that's at least a little bit
>> novel in everything I do. :-)
>
>
>Returning to the sorrowful and miserable discussion of QPSK, Costas and
>Gardner, the greatness of Eric and the reinventing of the wheels:
lol
Funny thread for it, tho...
>obviously there is a relation of BER vs carrier sync bandwidth vs symbol
>sync bandwidth vs Eb/No. It would be convenient to know that if the
>QPSK Eb/No is, say, 4dB and if we want the losses due to the imperfect
>sync to be below 1dB, then the carrier PLL bandwidth should be X and the
>symbol sync bandwidth should be Y, so the impact of X is equal to the
>impact of Y. The numbers are easy to derive at Eb/No > 10dB, when
>everything is linear, but it is not so simple at low SNR. It is not too
>difficult to get it by simulation however I am pretty sure somebody had
>already done that. Do you know of a book/article where the results are
>represented in this form?
I think I know what you mean and I've never seen a book/article
treatment on the subject. It's not a hard simulation task, as it is
just a loop BW vs BER experiment at various SNRs. I'm guessing it's
the sort of thing where a paper would be very appropriate but the
reviewers would say something like "this is not new research" and
therefore reject it from publication, regardless of how useful it
might be.
Eric Jacobsen
Minister of Algorithms
Abineau Communications
http://www.ericjacobsen.org
Blog: http://www.dsprelated.com/blogs-1/hf/Eric_Jacobsen.php
Reply by Vladimir Vassilevsky●June 9, 20082008-06-09
Steve Underwood wrote:
> Vladimir Vassilevsky wrote:
>
>
> I try to be a novelist, and do something that's at least a little bit
> novel in everything I do. :-)
Returning to the sorrowful and miserable discussion of QPSK, Costas and
Gardner, the greatness of Eric and the reinventing of the wheels:
obviously there is a relation of BER vs carrier sync bandwidth vs symbol
sync bandwidth vs Eb/No. It would be convenient to know that if the
QPSK Eb/No is, say, 4dB and if we want the losses due to the imperfect
sync to be below 1dB, then the carrier PLL bandwidth should be X and the
symbol sync bandwidth should be Y, so the impact of X is equal to the
impact of Y. The numbers are easy to derive at Eb/No > 10dB, when
everything is linear, but it is not so simple at low SNR. It is not too
difficult to get it by simulation however I am pretty sure somebody had
already done that. Do you know of a book/article where the results are
represented in this form?
Vladimir Vassilevsky
DSP and Mixed Signal Design Consultant
http://www.abvolt.com
Reply by Giulio Petrucci●June 9, 20082008-06-09
Hi Dave,
Dave ha scritto:
> You may also want to look at MATPlotlib - which provides matlab type
> plotting functions.
I know.
> One thing I found was that the libraries are kind of spread around.
> Most are in scipy, but I found spline 1d interpolation in MATPlotlib.
Yes, I noticed it too...
> There are a couple of gotchas which you have to watch when switching
> from Matlab to Python
[cut]
I know... moreover I know python and I'm quite a newbie to
MatLab\SciLab\GNU Octave...
> Indexing starts from 0 and something I like is that you can use
> negative indexes as well i.e. -1 gives the end of a vector.
> 2 - The default matrix multiply is not what you think. If I recall
> directly the multiply is an element wise multiplication. And the other
> multiply tends to give you Kronecker products - There is a way to
> change this, so it gives the standard matlab type matrix multiply - I
> believe that is done by importing a different interface than the
> normal scipy or numpy (It is already defined in one of those
> libraries).
When you say the default multiply is element wise, this is correct if
you use things like arange
or convert a list to an array. Numpy uses a more generic object
"array" and the base type.
If you have a "Matrix" type the Matrix multiplications will be the
same as Matlab.
I personally have found this very useful.
Reply by Dave●June 6, 20082008-06-06
On Jun 5, 1:06 pm, John Hadstate <jh113...@hotmail.com> wrote:
> On Jun 5, 11:40 am, Giulio Petrucci <sis...@nonono.boh> wrote:
>
> > John Hadstate ha scritto:
>
> > > I'm trying exactly the same combination and running into the same
> > > problems with regard to documentation and learning curve.
>
> > Add that AFAIK you should also run ipython... and it can be a problem
> > under WinXP (or may be not... anyway to install ipython you have to
> > install a lot of other stuff...).
>
> Are you sure? Maybe I'm missing something, but I thought that the
> latest distribution of NumPy eliminated the need to install all that
> stuff as individual pieces. My CD has Python 2.5.1, NumPy 1.0.4,
> SciPy 0.6, PyWin32 2.10, and a PDF file called scipy_tutorial.pdf by
> Travis Oliphant.
>
> You can find the tutorial on-line athttp://www.tau.ac.il/~kineret/amit/scipy_tutorial/
> I believe it's pretty close to the PDF and the dates are both in
> October, 2004.
>
> You can find more of the latest stuff athttp://www.scipy.org/Download
You may also want to look at MATPlotlib - which provides matlab type
plotting functions.
One thing I found was that the libraries are kind of spread around.
Most are in scipy, but I found spline 1d interpolation in MATPlotlib.
There are a couple of gotchas which you have to watch when switching
from Matlab to Python
1- Most of the time the matrices are created by passing along
references, rather than actually copying the data. You have to
explicitly tell it to create a copy. E.g. in matlab you could have
A=B(2:10), and A is a seperate copy, but in scipy A point to element
2:10 in B, so if you change A you change B.
2 - The default matrix multiply is not what you think. If I recall
directly the multiply is an element wise multiplication. And the other
multiply tends to give you Kronecker products - There is a way to
change this, so it gives the standard matlab type matrix multiply - I
believe that is done by importing a different interface than the
normal scipy or numpy (It is already defined in one of those
libraries).
Indexing starts from 0 and something I like is that you can use
negative indexes as well i.e. -1 gives the end of a vector.
I think the basic signal processing functions are there, but not
things like a polyphase interpolation i.e. upfirdn()
Hope that helps.
Cheers,
David
Reply by Steve Underwood●June 6, 20082008-06-06
Vladimir Vassilevsky wrote:
>
>
> Giulio Petrucci wrote:
>
>> I'm trying different alternatives in order to choose a software for
>> scientific calculation, expecially for the resulution of (simple) DSP
>> problems. Currently I'm using Scilab. Anyway I'd like to learn to use
>> SciPy too bt I wasn't able to find any useful "for-dummies"
>> documentation about DSP in SciPy.
>
> I want to be a novelist. Please tell me which colour of the pen is the
> most appropriate for writing the novels. Or should I use a pencil?
I try to be a novelist, and do something that's at least a little bit
novel in everything I do. :-)
Steve
Reply by RJS●June 5, 20082008-06-05
[posted and mailed]
Giulio Petrucci <sisisi@nonono.boh> wrote in news:4847f0b3$0$35960
$4fafbaef@reader2.news.tin.it:
> I'm trying different alternatives in order to choose a software for
> scientific calculation, expecially for the resulution of (simple) DSP
> problems. Currently I'm using Scilab. Anyway I'd like to learn to use
> SciPy too bt I wasn't able to find any useful "for-dummies"
> documentation about DSP in SciPy.
> Do you have any interesting link?
On Jun 5, 11:40�am, Giulio Petrucci <sis...@nonono.boh> wrote:
> John Hadstate ha scritto:
>
> > I'm trying exactly the same combination and running into the same
> > problems with regard to documentation and learning curve.
>
> Add that AFAIK you should also run ipython... and it can be a problem
> under WinXP (or may be not... anyway to install ipython you have to
> install a lot of other stuff...).
Are you sure? Maybe I'm missing something, but I thought that the
latest distribution of NumPy eliminated the need to install all that
stuff as individual pieces. My CD has Python 2.5.1, NumPy 1.0.4,
SciPy 0.6, PyWin32 2.10, and a PDF file called scipy_tutorial.pdf by
Travis Oliphant.
You can find the tutorial on-line at http://www.tau.ac.il/~kineret/amit/scipy_tutorial/
I believe it's pretty close to the PDF and the dates are both in
October, 2004.
You can find more of the latest stuff at http://www.scipy.org/Download
Reply by Vladimir Vassilevsky●June 5, 20082008-06-05
Giulio Petrucci wrote:
> I'm trying different alternatives in order to choose a software for
> scientific calculation, expecially for the resulution of (simple) DSP
> problems. Currently I'm using Scilab. Anyway I'd like to learn to use
> SciPy too bt I wasn't able to find any useful "for-dummies"
> documentation about DSP in SciPy.
I want to be a novelist. Please tell me which colour of the pen is the
most appropriate for writing the novels. Or should I use a pencil?
Vladimir Vassilevsky
DSP and Mixed Signal Design Consultant
http://www.abvolt.com