strange out of band fractional spur?

[eejli]

pfd nonlinearity sdm

For a fractional PLL:
Fref=1.728 MHz
divider N=1023+501/(2^10)=1023.48925
loop bandwidth=20kHz
3rd order loop filter and sigma delta modulator

In PLL output phase noise spectrum,a fixed fractional spur is just outstanding there at 37.14KHz offset. I tried around that how to relate this 37.14KHz to Fref and fraction part of the divider N, but cannot see any relationship among them.

This spur is bigger but locate in the same frequency if I increase the deadzone of the PFD.

Can you take a look and give me a hint?

Thanks.

 

[madengr]

Did you shut down (stop clocking) all other logic on the PCB? It may not be from the PLL.

 

[eejli]

no it is a simulation.

 

[RFDave]

If you change the divide ratio, does the spur move?
Since you are in the simulaiton domain, take a look at the spectrum at the output of the SDM. Is there a spur there?

 

[eejli]

Thanks RFDave,

You are right. The spur just moves or is jut gone if I chang the divider fractional part. It seems if the divider fractional part is much higher or much lower than the present one, the spur disappear. But for small offset from the present fraction, it moves around.

The SDM output is very similar to the textbook,i.e. the quantization noise is dithered and high passed and there is no obviours spur in the SDM output.

According to RFSystem's comment the spur comes from the folding of the SDM Qnoise. But I cannot have any quatitative idea about where will the spur locate.

 

[RFDave]

Based on that, I suspect that your SDM implementation isn't quite right. Take a look at the output spectrum with the same division ratio and see what it look's like. Your SDM output sequence may be not quite correct.

Dave

 

[eejli]

You are right RFDave. In my simulation only for one fraction division ratio, the SDM output looks like a good shape. Other fraction will end up with much bigger Qnoise.

But the question still there, what is the relationship between the fraction and the spur frequency.

If I only change the deadzone value the spur does not change in frequency but only change in amplitude.

 

[rfsystem]

It is the nonlinearity which disturb the noise shaping property of the SDM. You can double check if you apply simply a nonlinearity to the sigma delta number output. Then take a FFT and you get folded spurious similar to the closed loop simulation.

There is no other way than to improve the linearity or decrease the fractional phase. It is the major drawback of the fractional approach.

Could you post how you phase detector and charge-pump nonlinearity model is looking like?

 

[eejli]

Thanks RFsystem,

For some reason I cannot attch a figure. I will try later.

The simulation setup is in ADS. It has an PhaseFreqDetCP block which has nonlinearity parameter. They are:

Ihigh=up current, Ilow=down current, Deadtime=deadzone, Jitter=PFD input refered jitter: such as crystal jitter and divider PFD refered jitter.

If I assume the ADS example's setup is right I found that Qnoise fold-in spur is changing its location or just gone depending on the fraction value. And in-band noise shaping is also changing.

From a design point of view I should have some simple quantitative analysis on those spurs and in-band noise folding. Is there any good reference on this?

Thanks.

 

[rfsystem]

I will check how realistic the ADS model is. I use only my own models derived from circuit insight. If did made posting regarding the nonlinearity model and the conclusion is that the gain is reduced at the locking point. It is not simply having zero gain for some small phase range.

The analysis if the spur amplitude is difficult because the fractional modulation make a limit cycle in the phase domain. This limit cycle could have a very complex phase over time response. The main property is that the mean value over greater number of periods approaching zero. If the nonlinearity is second order you will have a DC phase component. A third order, typical for phase freqeuncy detector plus charge pump model, generate low frequency spurs.

My own expirence end here. I have read somewhere, possible under freshpatents, that a dithergenerator modifies the loop gain coeffcients instead of adding low amplitude noise. I did check that for a second order to dither the last bit of the modulator but the spurious under nonlinearity remain.

 

[eejli]

Thanks RFsystem,

For that example I tried some other analog block parameter and find the spur is just gone whatever the fraction is. ADS is very sensitive to setup.

Anyway a solid conclusin I can draw is that nonlinearity give big rise to noise floor increasement both for in-band and out-band PLL output.

I have one thing not understood. From past posts in this board people claim they have measured deadzone of the PFD-CP via cadence. But my simulation shows there is no deadzone at all since there is delay chain in the PFD reset path. Am I right?

 

[rfsystem]

Then you have proper relation of the current source switching time and the minimum pulse time coming from the phase/frequency detector.

If the ratio approaches 2-3 you have to do accurate simulations because the nonlinearity is small but there.

 

[eejli]

The problem is that how to do the simulation to find the deadzone. I already tried to limit the sweep size(phase difference) to be 1ps while the reference frequency is 27MHz. But still can not see any deadzone.

 

[RFDave]

I haven't yet found a good analysis of the spurious output of a SDM. I spent lot's of time doing SDM simulaiton's in Simulink, and as long as I had the SDM topology correct, the output of the sdm was spur free as long as I stayed away from rational division ration (1/2, 1/4, etc). You are doing this in ADS? Are you using the SDM in the Fractional-N example they provide?

From a PLL point of view, spurious components at the output of the SDM will show up as a noise source that is within the loop filter BW. In other words, the spurious output will be low pass filtered by the Loop Response. Inside the loop response, it should directly modulate the VCO output.

As far as dead zone in a PFD in a Fractional N Loop, my opinion is that it really doesn't matter. Because the division ratio is continiously changing, the PLL never truly acheive "Lock" where there is zero phase difference between the two inputs to the PFD. With that the case, I don't think that a dead zone will be significant.

Dave

 

[eejli]

Hi RFDave,

I am using ADS's SD Fraction-N PLL example. You can find it RF board examples.

As for the deadzone I see the noise floor if Dzone is added when runing ADS. It can only explained by SDM Qnoise folding just as RFsystem explained. Since there is no quantitaive analysis about this I am not sure whether it is actual the reason for noise floor jump.

 

[rfasic]

It seems like it is the second harmonic of the 'half-spur'.
Usually a Fractional pll has a large spur halfways between two integer divisions; In your case you should have a spur at; (1.728e6*1023 + 1.728e6/2 =) 1768.608 MHz. So when you get close to this absolute frequency it should appear.

With your setting your frequency must be (1023*1.728e6 + 1.728e6*(501/1024) =
1768589437.5 Hz. Which is only 18562.5 Hz from the spur. So I would expect that you had a spur at 18562.5 Hz offset (can you see this spur?), but apparently you can only see the second harmonic of this spur i.e. (2*18562.5 Hz) = 37125 Hz.

 

[davidwong]

For FN implementation with digital accumulator,
I would expect a fractional spur at fractional part
e.g. fraction = 1/1024, clk freq = 1.728e6 -->
fractional spur = 1.728e6/1024 =1687.5Hz.

I still do not unerstand why a sigma delta Fractional pll has a large spur halfways between two integer divisions?
Please tell me.
thx

Added after 1 minutes:

For FN implementation with digital accumulator,
I would expect a fractional spur at fractional part
e.g. fraction = 1/1024, clk freq = 1.728e6 -->
fractional spur = 1.728e6/1024 =1687.5Hz.

I still do not unerstand why a sigma delta Fractional pll has a large spur halfways between two integer divisions?
Please tell me.
thx

 

[eejli]

RFAsic,

Yes I only see this sub-harmonics of the 1.856kHz. Do not see a spur at 1.856kHz offset. Thanks for help.

 

[rfasic]

RFAsic,

Yes I only see this sub-harmonics of the 1.856kHz. Do not see a spur at 1.856kHz offset. Thanks for help.

eejli, I guess your reply should contain 18.56 kHz somewhere and not 1.856kHz which appears both as what you see and what you don't see?.