ARRL Extra Class Q&A #3 - Phase-locked Loop Spectral Impurities

[Tom WA3KLR]

Here is another question & answer from the ARRL Extra class 2002 question pool that irks me, #3 in a series.  I am interested in your expert comments.

“E7E15  What are the major spectral impurity components of phase-locked loop synthesizers?

A.   Broadband noise.
B.   Digital conversion noise.
C.   Spurs at discrete frequencies.
D.   Nyquist limit noise.

A.   A phase-locked loop synthesizer is constantly “correcting” the output signal frequency.  This results in slight phase shifts from one cycle to the next.  This in turn results in phase noise, which is a broadband noise around the desired output frequency and which is the major spectral impurity component of a PLL synthesizer.”

As far as I am concerned, the major impurity component in the output of a phase-locked loop synthesizer is what is called the reference sidebands, which qualifies C. as the best answer of the 4 choices presented.

The ARRL chooses A. as the best answer.  What they are referring to is the “phase noise” or “single-sideband phase noise” (when referring to the slope on just one side of the carrier) on the synthesizer’s output spectrum.  I have never heard of the phase noise spectrum referred to as broadband noise.

The different regions of the slope on one side of a carrier, created by the various noise mechanisms have noise labels.  The part of the slope closest to carrier is the steepest and is called 1/f4  random walk FM.  The next region out, slope less steep is 1/f3 flicker FM.  The next region out, less steep, is 1/f2 random walk phase (white FM).  The next region out is called 1/f flicker phase noise.  The final region away from the carrier, where the slope is now becoming asymptotically horizontal is called f-0 white phase noise (per H.P. via U. Rohde).  Again, I have never heard the phase noise spectrum referred to as just broadband noise.

In the poor explanation of answer A. they talk about the corrections constantly being done to the output frequency.  This is done at the rate of the phase detector’s reference input frequency.  This is a very stable and known frequency that modulates the carrier and causes the sidebands at the (carrier – reference frequency) AND (carrier + reference frequency), hence two reference sidebands.  These are “spurs at discrete frequencies”.

The corrections and resulting modulation does not cause the phase noise.  In fact, you can DC steer the VCO of the PLL system (PLL not in closed-loop lock), and the phase noise is present there too and basically unchanged.

Now, in practice, it is usually much tougher to deal with the reduction of the phase noise than reduction of the reference sidebands.  It depends on the application how critical the phase noise is.  The use of the spectrum analyzer tool is referenced in the ARRL Extra class material.  On a spectrum analyzer, one is usually hard pressed to identify the phase noise unless it is very excessive.  The reference sidebands are easy to discern however.

A semantics note about answer D.  There is a technical term “Nyquist noise” and there is a technical term “Nyquist limit”.  They are two different things.  “Nyquist limit noise” is a bogus term as far as I am concerned.  I don’t know if this is deliberate or an inadvertent error.

“Nyquist noise” is also known as Johnson noise which is the noise generated in electronic circuits by the circuit material being above absolute zero and having a resistance.  This is the mechanism that defines the “0 dB. Noise Figure” level in amplifiers.

Nyquist noise does contribute to the phase noise spectrum of the VCO output signal.

The “Nyquist limit” refers to the input frequency in a sampled signal system above which the output signal will have aliasing components if the maximum input frequency exceeds ½ of the sampling rate.

This is my take on E7E15.  Any phase-lock loop synthesizer gurus out there with comments?

[Steve - WB3HUZ]

Yes, the phase noise are sidebands, but they are broadband sidebands, not discrete frequencies. The noise sideband will have a distribution equal to the distribution of the phase errors and/or perturbations. You will often see the phase noise specified at some carrier frequency offset. Depending on your requirement, the offset number may tell you what you need to know. But if it's not at the offset range you are interested in, it's nearly useless. It's always best to see the entire noise spectrum.

Also, I think these are FCC pool questions, not ARRL.

[WA1GFZ]

The problem with a pll is the phase isn't constantly corrected. There are periods where the next correction is being determined. Then there is the transient of the new correction. It is pretty easy to eliminate the reference side bands but phase noise is caused by a number of different inputs to the VCO. I find power supply ripple one of the hardest to deal with because the 120 hZ ripple frequency is well below the loop filter cutoff.
gfz

[Tom WA3KLR]

Thanks for the comments Steve and Frank.  I think that what is the worst type of impurity problem is a matter of opinion and the application.  Steve is talking about phase noise and Frank is talking about discrete frequencies.  The original questions asked about “impurity components”, note the plural.  So I think the question is o.k., just the choice of answers is the problem.


Current Test Pool: What are the major spectral impurity components of phase-locked loop synthesizers?
 
A    Broadband noise.
B    Digital conversion noise.
C.   Spurs at discrete frequencies.
D.   Nyquist limit noise.

Should be something like:
A    phase noise and spurs at discrete frequencies.
B.    Digital conversion noise and Nyquist noise.
C.   Spurs at discrete frequencies and digital conversion noise.
D.   Nyquist noise and Taft-Hartley oscillations.

A. is the correct answer.
- - - - - - - - - - - - - - - - - - - - - - - - - -

Frank, I would say by definition, that a PLL’s phase is constantly corrected.  You are asking for the loop to be corrected with infinite resolution it seems.

As far as the specifics of the 120 Hz. spurs:  being below the “loop filter cut-off” is a good thing and means that the 120 Hz. is within the loop bandwidth.  This means the loop should reduce the level of the 120 Hz. sidebands, if they are due to FM.  If the 120 Hz. sidebands are due to AM, the loop can’t do anything about it.  The loop can’t completely remove the 120 Hz. sideband if FM, only reduce them by XX dB, according to the open loop gain at that frequency.

Have you tried free-running the VCO with DC steer only from a 9 Volt battery and pot with ground lead right to the VCO ground area.  This can tell you if the 120 Hz. sidebands are due to coming in from the loop filter circuit or not.

Are your power supplies on the same p.c.b. as the PLL synthesizer?  It’s important to have the rectifier/capacitor current loops (txfmr secondary/diodes and filter cap. returns) on a common track that ties to a single point to ground at the voltage regulator i.c.

[WA1GFZ]

Tom,
I just disconnected the loop filter and let the control voltage float. The 120 hz side bands went away. In later 6830s Racal changed over to a linear power supply. This caused the 120 Hz sidebands because the unregulated voltage  for 5V from the supply has 1/2 volt of ripple on it. The switcher was a lot cleaner.  The local regulator on the synthesizer module has a couple mv of ripple coming through it. This voltage is used as op amp offset volltages in the loop so it is summed into the complicated loop filter. I know a guy who ran the radio off batteries and he measured the close in phase noise almost 20 dB cleaner. I think he was measuring phase noise in the -90s dBc at 100 hz.
Even stock once you get out a few KHz the 120 hz falls into the mud so I'm going crazy here.
Stock radio well below filter blow by levels but my hot rod has 3 filters in the IF so I can measure the noise so it needs to go.
There is room in the power supply to add a LM317 regulator to preregulate the voltage to 8 volts and eliminate a lot of the ripple. Should be easy to knock it down by 30 dB.
The RA6830 has a fractional n with a 1 MHz phase detector yet has 1 Hz resolution. The older 6790 ran at 100 KHz. I guess the lsi controller chip must have some cool guts.
yup I'm crazy....

[w3jn]

You can also get discrete spurs from the ref oscillator.   

There are all number of reasons why you can get 120hz spurs.  I saw it even when I ran a battery for the VCO control - the VCO input is so high impedance you can get capacitive or inductive coupling from other junk in the shack that's radiating 120 hz.

"Broadband noise" is a very poor and non-descriptive term for the various noise elements inherent in any oscillator - phase locked or not.

[Tom WA3KLR]

Frank,

Some of those "old" regulators have better ripple rejection and lower noise than the ubiquitous 78xx family.  Besides the LM317 regulator look at the LM723 also which was low noise output.

If you are going to feed voltage to a 7805, I would give it a little more than 8 Volts, the bare minimum at room temp.  If the 7805 is going to see colder, it needs a few more volts to be sure of good operation.  I've seen quite a few collapse at 32F with 8 Volts.   Note the 8 Volts spec. on the data sheet is with room temp.

But feeding the final regulator of a synthesizer with pure dc to begin with is a great idea.

[WA1GFZ]

John,
The later 6830s put a 5 volt regulator right under the reference module and I have verified it to be quite clean. Tom, I know about the 78XX regulators and used a lot of them. not the greatest for noise. I've also used the old 723 for years and it is hard to beat when it comes to ripple rejection.
The synthesizer module uses a 5 volt reference chip into an OP27 driving a darlington. It takes 1/2 volt of ripple ank knocks it down to a couple mv.
Friday morning I am going to build a little regulator up on a hunk of metal and mount it in a power supply to preregulate the only unregulated voltage coming out of the module.
The ultimate rejection of the IF filters in my hot rod is a bit over 100 dB. This means the CW filter has slightly better rejection then the synthesizer phase noise which keeps me up at night thinking about it.
I hope to fix this so I can get back to working on the transmitter....yup it's a sickness

[Tom WA3KLR]

Frank,

I'm wondering about if you made a crystal-derived 1st l.o. injection instead of the synsthesizer and tested the receiver with it at the resulting frequency and compared the performance to the PLL runing at the same frequency.  This would represent the ultimate l.o. spectrum and show if its really worthwhile getting the PLL's phase noise lower, regardless of whether you can achieve the PLL inprovement or not.

[WA1GFZ]

Tom,
It is sickness so if I look for more trouble I will find it. I found the synthesizer noise problem checking close in dynamic range. I would get a different number if my two test frequencies changed a hundred hz or so. If I landed on a 120 Hz sideband the dynamic range would be lower.
The synthesizer in a radio with a switcher supply works better close in. because it has a ripple signal at 45 KHz.
Actually I'm very close to the maximum performance of a crystal filter. The only way to improve on that is to run a pair of them in parallel driven by a quad hybrid each 90 degrees apart. I G3 guy did a qst article on the h mode mixer years ago and found that way to ram more performance through a filter. In time I will homebrew the mixer modules with these configurations.
I have a very rare RA6840 second mixer module that I modified to work in the 6830. It gives me a dynamic range of 108 dB wide band. My hot rod will do 100 dB at 10 KHz spacing on a good day.
Now that I'm done helping my son with his home work I can hit the shack and build a regulator.

[Steve - WB3HUZ]

Cavity tuned oscillators are the way to go for low noise.

[WA1GFZ]

Steve, ah to have the performance of the 8640B.....

Well the addition of a 7808 in the power supply killed the ripple into the synthesizer and the 120 hz sidebands are gone. The next thing to chase are the pulses from the phase detector. The phase detector does it's thing every 1 us for 200 ns so there is a douple pulse on the 5 volts every 1 us. the phase detector is a pair of 74AC74s and a 74AC00.

Now the close in phase noise falls way off once you get 3 khz. off center frequency and everything is down about 10 dB.

It's a sickness so maybe I'll get on 160 tonight.

[WA1GFZ]

Steve,
I put the 8640B on the spectrum analyzer after a lot of testing. The 8640B is quieter than the modified RA6830 but not by a lot. I suppose I'm at the limit of my HP141........If I could just get the guys at work to let me take a 8566 home..... I told them I would fix it for free.
Uncovered the PDM board for the 75 meter rig might be time to do something else.

[Steve - WB3HUZ]

The 8640 isn't bad, but it's noisy compared to some cavity tuned oscillators I've used.