# Summing of phase noise masks

Discussion in 'Electronic Design' started by koxe, Apr 18, 2007.

1. ### koxeGuest

Dear All!

I looked nearly one week to find this information, but I failed so I
want to ask this forum for help.

My problem is to calculate the phase noise of a satellite radio link.
The link consists of L-Band up-converter block up-converter, LNB and L-
band downconverter. All these components introduce phase noise, which
I have measured. My question is how can I calculate the resultant
phase noise if I concatinate all blocks.

First consideration:
All blocks are frequency converters based on mixing, hence a
multiplication in time domain will be represented by a convolution of
the frequency domain. So can I convolve all phase noise measuremenst
to obtain the eintire phase noise?

Second consideration:
Phase noise is per definition a phase modulation with noisy character.
represents the magnitude of a spectral line compared to the carrier
power. So if I simply convolve phase noise measurements I ignore the
character of the phase noise and would assume Gaussian behavior.

2. ### MarkGuest

I assume you have phase noise plots for each unit, not just a single
number...

Take the dBc/Hz value for each frequency, convert each value to a
numerical power value , P =10^(dBc/10), then ADD the powers form
enach unit for each frequncy, i.e add all the powers at 10kHz offset
for example, then convert each summed power back to dB dB =
10*log(P). You have to add the phase noise power at each frequency
for each unit...you can't add dB directly so you have to convert dB to
numberical power, ADD, then convert back to dB.

OK?

If you have just a single number at a single offset frequency , you
probably don't have enough meaningfull information

Mark

3. ### Rene TschaggelarGuest

By the way, noise is added as squares and then taken
the root of it, not just adding the amplitudes.

Rene

4. ### MarkGuest

I beg to differ,,, in the procedure above, we are adding noise POWER
numbers which can be added directly. If we were combining noise
VOLTAGE numbers, then yes squares/root are needed. But since the

Mark

5. ### koxeGuest

But I am not sure if your method is mathematically correct. A
frequency shifter like the L-band converter uses a mixer. Mixer are
mathematically represented by a multiplier. Hence two time signals are
multiplied, which corresponds to a convolution in the frequency
domain. So if I have two signals with the measured phase noise, my
approach would be the convolve these two signals to obtain the entire
phase noise. As you described the convolution has to be performed with
non dB values.

I tried both method in Matlab, and observed only slight differences in
the result. The convolution gave a more smoothed result, but in
average the results mainly the same.

Many thanks again for your ideas!

Koxe

6. ### MarkGuest

what you say above is true but the end result is that the mixer simply
takes whatever phase noise is present on the LO signal and "adds" it
onto the converted signal. Adds means using the method described.
Another way to think about it is....phase noise is unwanted FM
modulation with noise as the modulating singal.... Whatever FM is
present on the LO will also be imparted onto the converted signal. If
the LO shifts up by 10 Hz, the converted signal will also shift (up or
down) by 10 Hz.

Mark  