[Top][All Lists]

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Discuss-gnuradio] Google Summer of Code -- Ideas List!

From: Kartik Patel
Subject: Re: [Discuss-gnuradio] Google Summer of Code -- Ideas List!
Date: Mon, 06 Feb 2017 19:55:50 +0000

Hi Marcus,

I was interested in implementing this myself. Sorry for not clarifying. It would be my first time contributing a whole new feature to GNU Radio. I believe, the mentoring should be from someone who is more frequent contributor? If someone is interested in being the mentor to the project, it would be great.

I can add to wiki, but I don't have account on redmine. It is waiting to be approved from Admin for a long time.

Kartik Patel

On Tue, Feb 7, 2017 1:19 AM, Marcus Müller address@hidden wrote:

Hi Kartik,

sorry, we've all been pretty busy over the Weekend – FOSDEM and stuff.

So, I personally think this is a pretty great idea that you should definitely put on the GNU Radio wiki page for GSoC ideas – if someone has a great idea how to improve what you're proposing, it's a wiki for a reason – so frankly, go for it. Notice that it'd be awesome if you putting this on the page also meant that you'd agree to at least partially mentor the student that picks that topic!


On 02/06/2017 08:26 PM, Kartik Patel wrote:
Hello all,

Any discussion over statistical toolbox?

Thank you.

Kartik Patel

On Wed, Feb 1, 2017 1:32 AM, Kartik Patel address@hidden wrote:
Hi Marcus,

Sorry for replying late. I was travelling.

My point is we can have a statistical module for GNU Radio. Although Scipy has extensive library available, we can have it's wrappers for GNU Radio. We can use those wrappers in GRC. Basically, all major statistical analysis can be done at GRC level instead of going to the python/c++ backend.

There are some fundamental statistical tools (can be extended with suggestions from community): 1. generation of RV, 2. various distributions and distribution fitting, 3. regressions 4. hypothesis testing (including non-parametric testing which basically check whether current samples matches a particular distribution or not) 5. parameter estimations. We will need various distributions/functions from Scipy.

So, consider a scenario where we have a block of "random variable generators" which will get input from a block called "distribution" which will specify the distribution as well as it's parameters.
There can be another block for "distribution fitting". Which will take two inputs: vector of samples and input from "distribution" block.
Consider a hypothesis testing scenario: Get a input vector: Provide a condition of testing (like energy of vector should be greater than some value).
Consider a testing mechanism where we test whether a sample vector is taken from a distribution or not (aka non-parametric goodness-of-fit based testing): It may take input from a "distribution block" and set of samples. and based on value of some "false alarm probability", it will give the decision.

We can try to make these testing completely generic. Like, you can write whole equation in textbox in GRC (may be. need to see how can we do it). It's similar to some blocks in Simulink (not sure exactly which one, but I remember those).

Note1: the "distribution" block will provide a distribution object. It may work internally, or externally. That's debatable.
Note2: This is a idea. We can discuss on various implementation approaches once the scope of project etc are discussed.

Kartik Patel

On Thu, Jan 26, 2017 11:51 PM, Marcus Müller address@hidden wrote:

Hi Kartik,

I heartily agree with you, you need a lot of random variables, but the question is: in which shape?

Do you need the noise source to produce more different types of amplitude distributions? Do you need those in the channel models?

"Blocks for hypothesis testing" sounds pretty interesting. Can you flesh out that idea a little more? In my head, I'm not sure what a hypothesis is here.

Best regards,


On 01/26/2017 05:24 PM, Kartik Patel wrote:
Hi Martin,

Till now, based on my experience in communication systems, I saw extensive need of probability and random variables.

So, now, if we are considering GNU Radio to be a full-fledged communication systems simulator, I think we can have wrappers of statistical analysis functions of Scipy. We can have GRC blocks for the same.

So, for an example, for spectrum sensing applications, instead of writing a code with Scipy library, we can have some blocks for direct hypothesis testing.

Kartik Patel

On Thu, Jan 26, 2017 4:07 PM, Martin Braun address@hidden wrote:

On 01/26/2017 12:07 AM, Kartik Patel wrote:

> Hi,


> I am not sure how relevant is this, but it's worth a consideration.


> Can we have a probability and statistical toolbox? It may include

> various probabilistic distributions, their random number generators,

> their PDFs and CDFs. These are very much useful in a communication

> system analysis. (Example: middleton noise etc. for simulations). Even

> adding various statistical functions like hypothesis testing,

> regressions, distribution fitting etc. can be added.

Sure, although scipy has pretty good ones already. Can you elaborate on

how this would be useful for GNU Radio specifically?

-- M


Discuss-gnuradio mailing list



_______________________________________________ Discuss-gnuradio mailing list address@hidden https://lists.gnu.org/mailman/listinfo/discuss-gnuradio 

reply via email to

[Prev in Thread] Current Thread [Next in Thread]