Fixed SNR with unitary noise variance

I've just one question :

How can I write a model like y = x + w,( with w a white gaussian noise) with a fixed SNR and a noise variance equal to 1. What coefficient may I have before x ?

Thanks !

• What is the amplitude/power of $x$? A standard model is that $y$ is a sample of the output of a (possibly matched) filter, $w$ is a standard Gaussian random variable, and $x$ is a sample of the signal output of the filter. If $|x| = A$, then the SNR is $A$. If your model for $x$ is a unit-amplitude output, then you want to write $y = Ax+w$. Warning: there are different meanings ascribed to SNR depending on the application. Some people like to write $P_e = Q(SNR)$, others $P_e = Q\left(\sqrt{SNR}\right)$, still others $P_e = Q\left(\sqrt{2\cdot SNR}\right)$. All get the same $P_e$ .... – Dilip Sarwate Aug 26 '14 at 14:17
• ... because they are using different definitions of SNR. – Dilip Sarwate Aug 26 '14 at 14:18

If you define the SNR as the ratio of the signal power and the noise power in dB, you have

$$SNR_{dB}=10\log \left(\frac{P_s}{P_w}\right)\tag{1}$$

where $P_s$ is the power of the desired signal and $P_w$ is the noise poiwer. If the noise $w$ has a mean of zero, then $P_w=\sigma^2_w=1$. From (1) (with $P_w=1)$ you get the desired value of $P_s$ for a given value of $SNR_{dB}$:

$$P_s=10^{SNR_{dB}/10}$$

In order to normalize the signal $x$ such that it has the desired power $P_s$ you first need to know its power $P_x$. Dividing $x$ by $\sqrt{P_x}$ will give you a unity power signal, which can then be multiplied by $\sqrt{P_s}$ in order to obtain the desired SNR:

$$s=\sqrt{P_s}\frac{x}{\sqrt{P_x}}=10^{SNR_{dB}/20}\frac{x}{\sqrt{P_x}}$$

• Hey! Just another question. How does it work when you work with complex variable ? A coefficient of 2 ? – Henry Aug 29 '14 at 8:50
• @Henry: complex variables will change nothing. You just compute the powers (which are real-valued) and you do the same as before. For computing the powers you need to use the squared magnitude instead of simple squares. – Matt L. Aug 29 '14 at 16:13

protected by jojek♦Jun 1 '15 at 22:54

Thank you for your interest in this question. Because it has attracted low-quality or spam answers that had to be removed, posting an answer now requires 10 reputation on this site (the association bonus does not count).

Would you like to answer one of these unanswered questions instead?