Oscillators' noise nmeasurement with amateur equipment.
3rd south Italy homebuilders meeting (MAdS) presentation report held in
Acicastello (Catania province), Sicily, Italy, JM77NN, E15.1268/N37.56036, April 17th and 18th, 2010.
by I4SBX Eraldo Sbarbati

1) Thanks to the meeting organizers for inviting me to this meeting and thanks to the audience.

2) Continuing on the path taken by the previous meeting I will try to demonstrate how to make
measures with professional results, but with the means available to every radio amateur.
We will measure the oscillator's noise.

3) The signal generated by an ideal oscillator should contain a single spectral line
(be a monochromatic signal), unfortunately it doesn't. A crystal oscillator (high Q) will be spectrally
better than an LC oscillator. A PLL improves an oscillator near the f0.
At the base we will always find the thermal noise= -174dBm/Hz, added to the noise characteristic
of active circuits.

4) The effects of the phase noise are: in transmission it is easy to understand, in
reception it causes the reciprocal-mixing. In the graph (red line) it is shown the selectivity and
the dynamic range of the IC-706, due to the oscillator's phase noise we have a loss of dynamic 
of about 20 dB (blue line), the green line is obtained by a homebuilt PLL.

5) I like to describe the phase noise with this analogy: the water tap and a bucket at the side of the 
water flow. The bucket will get only a few droplets ( the phase noise) or some small secondary flow (spurious).
The ratio between the main flow and the water collected in the bucket will give us the measure. A drop
in a billion drops is a 90 dBc ratio, we are going to make measurements at the 150 dBc level. A large bucket
collects more water but increases the uncertainty of the distance from the source. We will put the bucket
under the main stream just to make calibrations. To unify the results and make them independent of the size
of the bucket will divide the amount of water for the surface of the bucket. The Y axis will calibrate the
carrier to 0 dB then, at increasing distances from f0, we will measure the power of the noise with a filter.
To unify the measure we will divide the amount of noise for the filter BW in dBc/Hz. In the graph the output
of an equipment, from Agilent, specific for the measurement of the phase noise.

6) In ham radio publications the same standard is used, to the left we see the noise from an IC-7800 published
in QST and to the right a series of measures for various receivers published by FunkAmateur. Note that the
values go down to -160dB from a few Hz up to one MHz, but the significant part, for us, stops before the 100 kHz.

7) This is an instrument used by a radio amateur, N1UL, unfortunately it is not affordable for most of us.

8) There are several ways to do this kind of measures, we will examine here the last 3, in particular the last.
With a good spectrum analyzer we can have a measure as in figure (on page 8). The red line shows a signal
generated by a crystal oscillator at 100 MHz. It is visible the form of the filter (10 Hz) where you can not make
noise measures (main flow inside the bucket), the rest is the analyzer background noise, and is also the
limit of the measurement. The blue line indicates the response of the Marconi 2019 generator. The difference
between the red and blue lines is the phase noise. for a standard measure we have to divide the value read on
the analyzer to the width of the filter. In this case BW= 10Hz equals -10dBc/Hz. The modern analyzers with the
option "Noise" automatically calculate the value referred to 1Hz.

9) On the left we see the same screen above, but with the axis to the left instead of the center.
To the right, the same signal, but with a filter of 1kHz and 10kHz span/div. It is clear that with
wide filters we have more precision on the Y axis, but with less definition on the X axis.
Limitations of the measures with this analyzer: X= 100Hz, Y= -100dBc/Hz; X= 10kHz, Y= -120dBc/Hz.

10) We try to understand how a spectrum analyzer is made. The input signal goes to a mixer, after appropriate
attenuators and there is a swept local oscillator. The product from the mixer is filtered, detected and
displayed on a screen. The mixer introduces attenuation, a bit of noise, but doesn't introduce phase noise.
The local oscillator's phase noise contributes to the analyzer total noise. This is the measurement limit
with this instrument.

11) A spectrum analyzer based on the Fourier transform is a completely numeric device and acts as a row of
filters each one next to the other.

12) There are many commercial analyzers based on the FFT, but they are not affordable for all amateurs.
Using a PC sound card with appropriate programs we can have an FFT analyzer for free. The DL4YHF's program
Spectrum Lab allows us to see a signal from 0Hz up to half the sampling rate with a dynamic to 160dBc/Hz and
resolutions of the fraction of Hz. [Oscilloscope at the bottom right, red trace -3 dB, figure on page 12].

13) Setup: Spectrum Only. FFT bin number, Hanning windowing, Noise Bandwidth, average 0/8. Y 0-160dB,
Copy from current spectrum.

14) Sensitivity and noise of the most common sound cards suitable for this use. The dynamic
noise was remeasured for our purposes in dBc/Hz, the official ones from the manufacturers, are for the
entire audio spectrum and have a "weighing" according to the audio standards.

15) With the Spectrum Lab we can work from almost 0Hz up to 24kHz with standard cards, and up to 48 or 96kHz
with professional cards. In order to make measurements on our frequencies you need a mixer. As in the classic
analyzers, mixer doesn't introduce phase noise. We need two oscillators at the same frequency or
nearby frequencies and both contribute to the phase noise.

16) This is a common double balanced mixer, just to make you notice that, contrary to the official pinout,
only pin 2 is physically connected to ground.

17) This is a possible scheme for mixer, diplexer and low noise amplifier. If you use
an amplified sound card as the E-MU 0202 or 0404 the amplifier is not needed. The diplexer,
instead is essential to present to the mixer a 50Ohm load from DC upto the RF input harmonics
and the audio signal has to be filtered so that the RF residues do not saturate the OpAmp.
The second OpAmp with gain 1 is used to invert the signal to drive a differential professional card, this is
equivalent to an additional gain of 6 dB.

18) This is another possible low noise amplifier. Here I used the fact that in the SRA-1
mixer pins 5 and 6 are not grounded, so I went out on a differential manner to directly enter, in this way,
on the OpAmp. Here the diplexer is simplified, but nothing prevents the use of the type just seen or vice versa.
The output of this OpAmp is differential, therefore suitable for professional cards.

19) I have made a table of minimal gains for Low-Noise OpAmp depending on the
mixer used. For maximum noise measure dynamic it is good to work at the point of mixer
compression, you should not exceed more than a few dB from the tabled values.

20) For low level oscillators a Gilbert cell mixer can be used, like the NA602-SA602-SA612.
With this device you do not need the OpAmp, input and output impedances are 1.5 Kohm, to take into account for
the diplexer.

21) Let's see the results. These are simply the lines of the noise of a
standard card ~ -120dBc/Hz, and a professional card E-MU 1212m in the order of -160dBc/Hz. These are the limits
of our system, similar or higher than professional ones.

22) Let's try to use the sound card supplied with the PC. The blue line shows the limit of the system, the
blue line indicates the noise of a Marconi 2019 generator (two equal generators) you can
see the noise near the carrier (span ±500Hz) it is shown the typical "rise" of the PLL.

23) Here, as in the next slides, I use professional cards E-MU. The yellow line is as
the previous one but with scan ±12kHz (note the central zero), the blue line indicates the phase
noise af the IC-706 local oscillator, see the differences.

24) This is what comes out of a DDS, the phase noise is very low, but you see all
the spurs. With a common analyzer would all be invisible, less than 100dBc/Hz.
Blue Line: system noise, E-MU 0202 sound card.

25) This slide is dedicated to Pippo I0FTG who, two years ago, at the 2nd MAdS, for a PLL,
advised us to use the 9046 instead of the 4046. These are the differences.

26) Back to the initial theme. This is the measurement made by N1UL with an R&S instrument. Note the X axis
useful from 10Hz to a maximum of 100KHz and the Y axis downto -150dBc/Hz.

27) This is a measure with the carrier fully to the left as in the professional instruments. There are
two ways to achieve this.
        a) mixing the two generators to get the product at the center of the screen,
           select the view from the center, adjusting the values with the offset on the
           X axis, half of the band is lost.
        b) Make a zero beat of the two generators, in this case the spurious that we have on the left
           are mirrored to the right, along with the others. Also, the noise sums up with an increase of
           3dB on the values. In this case either you compensate with an offset on the Y axis or more
           simply, calibrate the carrier 3dB below.
The blue line shows the noise of two crystal oscillators, there are still more than 10dB of
measurement margin, at least at frequencies close to the carrier. The blue line shows the noise
of two Marconi 2019 generators

28) The blue line is the same as previously shown. The yellow line shows the noise of a R&S SMFS test set
In this case, having only one test set, I mixed it with a crystal oscillator,
being the signal of the test set worse more than 40dB compared to that in the oscillator,
the contribution to the noise of the latter is infinitesimal, less than 1/10000.
In this case we don't need to removed the 3dB due to the second oscillator.

29) Thanks and have fun.