When computing power spectral density via Fourier transform and Parseval's theorem, one gets units of $V^2/Hz$ (e.g. Wiki). As a popular next step, the following unit conversion is widely used: $V^2/Hz \to dB/Hz$, and it's done by taking the famous $10.\log10(\cdot)$.

Question: while $\mathrm{dB}$ is a log of a ratio of two $\mathrm{V}^2$'s (second $\mathrm{V}^2$ is a fictive one) and I get where $\mathrm{dB}$ comes from, I'm struggling to understand why $1/\mathrm{Hz}$ remains unchanged after taking the logarithm?


P.S. Some obscure explanation is in the last paragraph here (and I'm afraid the word 'loosely' conceals all the real stuff here):

Since the argument of the logarithm is in units of $\mathrm{Hz}^{-1}$, this spectral measure can loosely be said to be in units of `dB/Hz'

  • $\begingroup$ i have trouble with the notion of dB/Hz in the first place. with V$^2$/Hz, we know that when we double a thin bandwidth, the power (or square of voltage) will also double. but the dB will not. $\endgroup$ – robert bristow-johnson Apr 22 '17 at 22:33

Say you have a measurement in unit $U$ (V$^2$) per unit $u$ (Hz). I do claim (although not in public) that the logarithm is a kind of unitless transformation, see for instance What is the logarithm of a kilometer? Is it a dimensionless number? In other words, it is somehow $0$-homogeneous, while a power is $2$-homogeneous: multiplying by a constant yields a constant to some power.

First, the logarithm is kind of homogeneous to a $0$-th power: it is the integral of $1/x$ (power $-1$). The derivative of an affine function is a constant. Derivatives decrease powers by one unit, integrals increase them of the same amount. The logarithm somehow stands with constants between $x^{-1}$ and $x^{1}$. Additionally, the coefficient of variation of $x^p$ when $p\to 0$ is about:

$$ \frac{x^p-x^0}{p-0}\approx \log x\,.$$

Second, a dB scale is often relative: to some signal measurement, to some "absolute" value. The $\log$ argument is better unitless: something in U divided by some unit $1$ U, called U$_0$. Thus, computing $\log(\mathrm{U}/\mathrm{U}_0)$ could be considered.

So converting U/Hz into dB$_\text{U}$/Hz, to me, only affects the numerator.

  • 1
    $\begingroup$ it is a meaningless unit of measure. even when referenced to a specific voltage. $\endgroup$ – robert bristow-johnson Apr 22 '17 at 22:34
  • 1
    $\begingroup$ I see your point! Thanks for your answer! $\endgroup$ – agronskiy Apr 23 '17 at 13:52

Simply because the intention of converting to a dB scale, here, is only to scale the power density, and not the bandwidth.

So, that's simply by definition.

Look at it this way: if you have a PSD plot with frequency on the x-axis and power density on y, you'd usually only want to scale the y-axis logarithmically.

You can, of course, also scale the frequency axis, but then, you get a different thing.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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