If I have a average normalized power spectrum of noise is like this $W(f)=N_0$ while frequency is $-B$ to $B$.

Why the average normalized power is $1/4 N_0 (2B)$ insteat of $N_0 (2B)$?

  • $\begingroup$ kept staring at it and thought.. that makes no sense! Remark that it is $(1/4)N_0(2B)$. not sure if that is right though. gotta work out the math. $\endgroup$
    – thang
    Jan 25, 2013 at 9:35
  • $\begingroup$ Yes, I have completely no idea where is the 1/4 come form. $\endgroup$
    – Samuel
    Jan 25, 2013 at 10:50
  • $\begingroup$ Well I think that if you work out the formula, it comes out to 1/4. It will take me awhile to do it, and I am kinda lazy now (it's 3am here... :(). To get an intuition, just look at sin(t). $\endgroup$
    – thang
    Jan 25, 2013 at 10:52
  • $\begingroup$ Where did you get these expressions? $N_0$ is typically used to denote a flat power spectral density, in which case its units are power per Hertz of bandwidth. In that case, the total power across a bandwidth $[-B,B]$ should be $2N_0 B$. $\endgroup$
    – Jason R
    Jan 25, 2013 at 15:25
  • 1
    $\begingroup$ @JasonR, as you can see from the formula, you don't need to know $f$ at every point in time to get the average power. If you do, all the better. In fact, consider for example $g=f+h$ where $h$ is nonzero only on a set of measure 0, then average power of $g$ is the same as that of f. There is a formula that relates the power spectral density and average power. He's trying to get a derivation of this formula. I suggested, for simplicity to save time, just do $sin(t)$ instead of generally. I have been meaning to derive it, but got caught up with a stupid thing called the job :p $\endgroup$
    – thang
    Jan 25, 2013 at 16:40

1 Answer 1


If a filter with transfer function $H(f)$ has a noise process as input, then the output noise power $P_{\scriptstyle{\text{out}}}$ is given by $$P_{\scriptstyle{\text{out}}} = \int_{-\infty}^\infty |H(f)|^2 S_x(f)\,\mathrm df $$ where $S_x(f)$ is the power spectral density of the input noise process. If the input process is modeled as a white noise process, then $S_x(f)$ is a constant, call it $K$ for now, and so we have that $$P_{\scriptstyle{\text{out}}} = K \int_{-\infty}^\infty |H(f)|^2\,\mathrm df $$ For an ideal low-pass filter with cutoff frequency $B$ Hz, $$ H_{\scriptstyle{\text{lowpass}}}(f) = \begin{cases} 1, &-B \leq f \leq B,\\ 0, & |f| > B,\end{cases}$$ we have $P_{\scriptstyle{\text{out}}} = 2KB$. More generally, for an arbitrary linear time-invariant system with maximum passband gain equal to $1$, the noise-equivalent bandwidth of the system is defined as $$B_{\scriptstyle{\text{ne}}} = \frac{1}{2}\int_{-\infty}^\infty |H(f)|^2\,\mathrm df$$ and so we get $P_{\scriptstyle{\text{out}}} = 2KB_{\scriptstyle{\text{ne}}}$. Put another way, the noise-equivalent bandwidth of an ideal LPF with cutoff $B$ Hz is just $B$ Hz.

It is conventional in the engineering literature to write $N_0/2$ instead of $K$ because then the output noise power expression becomes $N_0B_{\scriptstyle{\text{ne}}}$. The constant of proportionality $N_0$, measured in watts/Hz as Jason R pointed out, is convenient for use in ordinary language when conversing with other engineers. If we have a filter with noise equivalent bandwidth $B_{\scriptstyle{\text{ne}}}$, and the thermal noise power at the input of the filter is $N_0$ watts/Hz, then the output noise power is $N_0B_{\scriptstyle{\text{ne}}}$. That's the way people talk during discussions. Nobody will say that an ideal lowpass filter with cutoff $B$ Hz has "bandwidth" $2B$ even though the transfer function has value $1$ for $f \in [-B,B]$ and thus the support of $H_{\scriptstyle{\text{lowpass}}}(f)$ has length $2B$.

Turning to your question, it is not a good idea to write $W(f) = N_0$ for $f \in [-B,B]$, but if the paper you are reading does insist on this notation, then the output power is $N_0(2B)$. If the paper says that it is $\frac{1}{4}N_0(2B)$, then it is mistaken. Another possibility is that there are other conditions that are mentioned in the paper that make $\frac{1}{4}N_0(2B)$ the correct answer, but you have not included these conditions in your question. For example, if the passband gain of the ideal LPF is $\frac{1}{2}$, then the expression $\frac{1}{4}N_0(2B)$ for the output power is perfectly correct.

  • 1
    $\begingroup$ Thanks Dilip for very clearly articulating what I was trying to say above. $\endgroup$
    – Jason R
    Jan 26, 2013 at 0:41

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.