I need help to calculate the SNR at output of ICA

[ednan]

SNR of ICA

Hello..


I need help to calculate the SNR at output of ICA ..please tell me if any one done this ...so that i can tell my problem....

waiting for your reply..


thanx

Regards
Adnan

 

[mmaah]

Re: SNR of ICA

Hello mate..

I've done a part of my master project related to ICA and the primary evaluation factor was the SNR. So, please state your problem to see if i can help...

Good Luck for you.

 

[ednan]

Re: SNR of ICA

Thanx for ur response ...i working on speech signal ..i have seperate the mixed components,input is Speech signal and Gaussian Noise..n mixed it and then seperate the compenents..1 component is Enahanced speech signal and other is Noise

I am calculating SNR with this formula

SNR(i/p)=10log10(power of input signal/Power of input noise)


SNR(o/p)=10log10(power of input signal/Power of (Enhanced signal component-input signal))..

Is it Correct Formula..???

and my results using this formula is
SNR(i/p)=-21.47 dB
SNR(o/p)=19.01 dB
is this results are correct???

please tell me ...waiting for your response ...
Thanx

regards
Adnan

 

[mmaah]

Re: SNR of ICA

Hi mate..

My dear mate, please try to help me to understand you..

You have separated components (pure speech (PS) + Gaussian noise(GN)) and you mixed them together and applied to ICA and separated the independent components that you have, namely (speech and noise), is that right?

Now, watch this:

The general equation for SNR is given by:

SNR = 10log(S/N), (in other words, the level of the desired signal compared to the noise)

where

S- is the desired signal power
N- is the total noise power (might be more than one)

In case of the input, we have two main signals (PS + GN). Thus,

SNR = 10log(PS/NG)... And your formula for the input is fine.

But in the output, we also have two main signals (enhanced speech (ES) and other separated noises (N)). Thus, we have to maintain our relationship:

SNR = 10log(ES/N)

Now,

1 - I would wanna ask you, why did you use the power of the input signal again in your equation? I don't think that's right!

2- If you're considering the output noise that you calculated after ICA as difference between the enhanced signal and the input signal, then you have to use this formula (input - enhanced) because the enhanced is a part of the input not the opposite.

3- When you deal with the ICA, just calculate the SNR individually, that is, the input is not really related to the output, because we have (an enhanced signal and it's not mixed with any other signals). Therefore, if we want to calculate the SNR here, you have to assume something, for example, the level of the noise that you have at the output equals to the one that you measured at the input or you simply use the measured one after the ICA. Hence, you can apply this equation:

SNR (output) = 10log(ES/GN) or 10log(ES/N)

Then you can compare the results with the acceptable range.

Note that, all the signals in here must be in the power format. Besides, if your signals are digital signals, then use the summation to calculate the total power, that is (total power = Sum_from 0 to end_ (signal))..

I hope that's helpful

If you want to discuss in private, then this is my email address (mmaah82atgmail.com)


Good Luck for you.

 

[ednan]

Re: SNR of ICA

THANX ALOT FOR YOUR REPLY ... I ADDED U IN GOOGLE TALK ..PLEASE TELL ME UR WHEN U CAME ONLINE SO I CAN DISSCUSS IT ...AND THE INFO U GIVEN ITS RELLLY HELP FUL FOR ME ..THANX

aDNAN

 

[ednan]

Re: SNR of ICA

Hello...if i am using White Gaussian Noise ...then can we calculate Its power ,to Calculate its SNR?????