better 4046 PLL

The isolation resistor in series with the varicap is a problem in wideband loops. You can use a push-pull oscillator with two varicaps in series across the tank. The junction of the varicaps is theoretically at ground, so you don't need the resistor.

In a synthesizer, you may be multiplying a reference to a higher frequency. Say you have 10MHz and you want 100MHz. The PFD runs at 10 MHZ, and the spurious sidebands will appear at 90MHz and 110MHz. This is a problem in spectrum analyzers and other applications where you need to multiply a reference frequency and you need good phase noise.

You have to use a huge ripple filter in front of the op amp, otherwise you will drive it into saturation. That will make the main loop filter much more difficult to design.

The XOR is sensitive to the duty cycle of the input signals. The phase between the oscillators will drift unless you use the Extended Range option in Howard's paper:

Now you are back to 1F ripple and still have to keep it from the input of the op amp.

Why would you want to use an XOR when a PFD doesn't have these problems? JK

With an XOR phase detector, lots of times you don't need an opamp.

Because it's easy and has no deadband.

--

John Larkin                  Highland Technology Inc 
www.highlandtechnology.com   jlarkin at highlandtechnology dot com    

Precision electronic instrumentation 
Picosecond-resolution Digital Delay and Pulse generators 
Custom timing and laser controllers 
Photonics and fiberoptic TTL data links 
VME  analog, thermocouple, LVDT, synchro, tachometer 
Multichannel arbitrary waveform generators

Depends on the filter you put between the phase detector and the VCO.

Actually it doesn't. A phase-frequnecy detector can get it into lock a lot faster than a multiplying detector can, but - as Floyd Gardner point out - the multiplying detector can eventually do the job.

Not to nil, but presumably to an acceptably low value, if the filter was rationally designed.

But not impossible, or even all that difficult. If the worst comes to the worst you may have to add an all-pass phase-shift adjusting section to keep the loop stable.

A lot depends on how your input signals are being generated - putting a pair of divide-by-two bistables in series with the inputs XOR may make sense in some situations. In others it's not going to help at all.

That's what filters are for.

It has different problems.

--
Bill Sloman, Sydney

You lose the integrator. That won't give a type II loop, so the phase error changes with frequency:

"

"

Easy can come back and bite you later when you have to time to solve the problem and no space on the pcb for a solution. Better to do it right in the first place and not leave any openings for problems downstream. Quick and dirty have no place in a professional approach.

An XOR is useful only for the crudest, least critical applications. You get a sloppy lock and have to use a huge ripple filter. I still maintain you will have trouble with spurs.

The PFD has no deadband if you do it as we have discussed here. The ripple is zero when you are on time. The cost is negligible. An XOR with a huge ripple filter will take up a lot of space on the pcb. A PFD might actually fit in less space.

Unless you are locking a crystal oscillator to some reference, you will need a PFD. So why bother switching over to an XOR when you have lock?

I don't see any use for an XOR in any kind of instrumentation application. YMMV

JK

Sadly, the PDF still needs the ripple filter. It produces narrow correction spikes whenever the VCO drifts off frequency, or moves off frequency in du e to Johnson noise generated in the dissipative components of the oscillato r.

These spikes have harmonic frequency components all the way up from the osc illator frequency to an upper limit determined by the bandwidth of the comp onents used to build the PDF. You don't want to feed them directly into the frequency control input of your VCO, so you have to filter them.

The ripple filter may not need as much LF attenuation as it would with an X OR, so it might fit into less space, but not a lot less.

--
Bill Sloman, Sydney

Then you lose the integrator and are no longer a type II.

Easy can come back and bite you later when you have no time to find a solution and no space on the pcb to implement it. It is far better to do the job right in the first place and not leave any holes that can cause problems later.

The XOR requires a huge filter to get rid of the ripple. This takes up space on the pcb.

Unless you are locking a crystal to a stable reference, you will need a PFD to achieve lock. You will also need to know when you are out of lock and switch back to the PFD. This is similar to the AD9901 which is also an XOR.

The PFD cost is negligible. It has no deadband if you do it as we have discussed here. It requires much less filtering so it won't take up much room. It is edge-triggered so you know what the phase relation is at all times.

As long as you need a PFD to achieve lock, why bother going to an XOR?

About the only place I see a PFD can give problems is if you can lose some of the incoming reference pulses. For example, in handling data streams. Even then, you can disable the PFD when there is no incoming data pulse. You can do this by adding a delay in front of the PFD to detect an incoming pulse and enable the PFD for that sample. Then you need a constant stream of pulses to lock up to before switching to the data pattern. An example of this is data recovery in hard disk drives.

But an XOR would be totally useless in this application since zero phase error depends on the incoming pulse widths. If they drift, the data will no longer be centered in the data separation windows. Along with the huge ripple problem, the XOR is a non-starter.

So I don't see any real reason for considering the XOR.

JK

Sorry for the dupes. I thought I screwed up and exited the news client instead of sending the post. It didn't show up, so I was convinced I blew it. Of course, I didn't make a copy and had to go on memory to repost.

As you can see, that doesn't always work very well:)

JK

I agree entirely that RC oscillators are crap, and that wrapping a PLL around them will produce (at best) brightly polished crap. Last time I used a 4046 in the datasheet-approved fashion was to add a phase adjustment to the sync output of an optical chopper controller--i.e. something that was already slow and jittery, so that the crummy oscillator wouldn't make it noticeably worse.

I expect that the better behaviour of LC oscillators has a lot to do with the impedance level, and the noise bandwidth. Your average parallel LC tank has a pretty low impedance even at resonance, and its Q limits the noise bandwidth severely compared with an RC.

The 4046, being CMOS, doesn't have a saturation problem. The resistor doesn't affect the transient behaviour significantly, it just ensures that in its quiescent state the output pulse is at least several nanoseconds wide. For larger phase errors, it behaves exactly as normal.

Sure. I think that's what JL already does. Of course if the amp has a low open-loop output impedance, it's less of a problem since the feedback cap takes the charge anyway, even if it's happening too fast for the op amp to react. (I usually use a series RC plus a smaller parallel C as the feedback network.)

Cheers

Phil Hobbs

--
Dr Philip C D Hobbs 
Principal Consultant 
ElectroOptical Innovations LLC 
Optics, Electro-optics, Photonics, Analog Electronics 

160 North State Road #203 
Briarcliff Manor NY 10510 

hobbs at electrooptical dot net 
http://electrooptical.net

Sometimes phase error doesn't matter.

Well, get it right the first time.

Better to do it right in

There's nothing dirty about an XOR phase detector. In fact, its phase error vs voltage curve is beautifully linear.

Sometimes an XOR is an excellent choice, sometimes it's not.

--

John Larkin                  Highland Technology Inc 
www.highlandtechnology.com   jlarkin at highlandtechnology dot com    

Precision electronic instrumentation 
Picosecond-resolution Digital Delay and Pulse generators 
Custom timing and laser controllers 
Photonics and fiberoptic TTL data links 
VME  analog, thermocouple, LVDT, synchro, tachometer 
Multichannel arbitrary waveform generators

On a sunny day (Wed, 03 Jul 2013 09:51:02 -0400) it happened Phil Hobbs wrote in :

It all depends, many years ago I used a 4046 in a floppy drive data separator design, and it never ever produced one error, a lot better than some other designs I came across:

Floppies experience extreme speed variations... Of course I designed my own phase comparator, did not use the 4046 build in ones. It also had to maintain frequency when input was not present, so dropouts. mm 1984, 30 years ago...

Indeed. But for any digital use it seems foolish to use an XOR, when you can make a very fine PFD from a dual-D and a quad-NAND with no dead-band.

XOR's, or better yet Gilbert cells (the analog equivalent), are best when the incoming signal is analog with noisy transitions, or with extraneous spikes. But the price you pay is needing a filter with a low frequency corner to knock down the ripple, thus slow loop response.

I've used analog PD's (Gilbert cell variation) to extract TACAN signals that were buried so deep in the noise you couldn't see the signal on an oscilloscope.

Also XOR/multiplier PD's will harmonic lock if the tuning range of the VCO is not restricted. ...Jim Thompson

--
| James E.Thompson                                 |    mens     | 
| Analog Innovations                               |     et      | 
| Analog/Mixed-Signal ASIC's and Discrete Systems  |    manus    | 
| San Tan Valley, AZ 85142   Skype: Contacts Only  |             | 
| Voice:(480)460-2350  Fax: Available upon request |  Brass Rat  | 
| E-mail Icon at http://www.analog-innovations.com |    1962     | 
              
I love to cook with wine.     Sometimes I even put it in the food.

Phil Hobbs wrote:

[...]

[...]

You have not been following my LTspice analysis of this problem. I have previously shown the LC tank is immune to voltage noise on the ground plane. Injecting noise into the ground end of the capacitor has no effect on the phase of oscillation.

If the inductor and capacitor are grounded at the same spot, then noise on the ground plane merely bounces the entire tank up and down. No energy gets into the tank to affect the phase.

The only way to change the phase is to inject current into the tank. The optimum point is when th tank voltage passes through zero, as Hajimiri discusses in his paper. I show this effect in the LTspice file appended to this post.

However, it is very difficult to inject current into the tank in a common-collector Colpitts. You have to do it when the voltage waveform crosses zero. The transistor is off at that instant, so there is no mechanism to access the tank.

The Q of the tank has no effect on the resulting shift in phase. I tried various values of Q from 40 to 50,000. For the same current pulse, the time shift is the same. I should also mention the amount of shift is much less than the result in a multivibrator. The mvb is easy to shift, the LC tank is hard.

Solving one mystery seems to lead to another one. In the analysis, all the current goes into the bottom capacitor, C3. None appears to go into C1. This raises the questions of how does the transistor sustain oscillations, and how does the oscillator start up.

The Phillips and NXP 74HCT9046A claim there is a deadband problem in the

74HCT4046A, and that they solve it. Probably by increasing the delay in the phase detector feedback path. See Fig. 11a and Fig. 11b on page 12 of

"

12526886.pdf"

Of course, to paraphrase Neumann, anyone who puts a multivibrator vco in the same package as a PFD is in a state of sin.

JL posted his op amp circuit. There is no input filter:

"

"

My concern is what happens inside the op amp when you feed it very fast pulses. I don't think you can assume the feedback capacitance will take all the charge. You have to believe the output transistors have a low impedance to the fast pulse. This may not be true in all cases. Also, if some portion of the internal circuitry saturates, then control is lost. This could lead to the same effect as deadband.

It would depend on the type of op amp and its characteristics. Some may be vulnerable and others may not.

I posted several LTspice analysis files showing the effect of RC and LC filters using Bessel and Gaussian filters. I like the LC filters much better, but have to be careful about noise pickup in the inductor.

Besides helping to protect the op amp against fast pulses, the input filtering will help reduce the spurious sidebands at the pll clock frequency. This is important in frequency multipliers such as spectrum analyzers and synthesizers.

JK

Here is the LC tank noise analysis file with the PLT file at the end. As usual, watch for line wrap.

Version 4 SHEET 1 1260 800 WIRE -144 -80 -208 -80 WIRE -64 -80 -144 -80 WIRE 304 -80 240 -80 WIRE 384 -80 304 -80 WIRE -64 -64 -64 -80 WIRE 384 -64 384 -80 WIRE -208 16 -208 0 WIRE -64 16 -64 0 WIRE -16 16 -64 16 WIRE 0 16 -16 16 WIRE 240 16 240 0 WIRE 384 16 384 0 WIRE -208 32 -208 16 WIRE 0 32 0 16 WIRE 240 32 240 16 WIRE -64 48 -64 16 WIRE 384 48 384 16 WIRE -208 128 -208 112 WIRE -64 128 -64 112 WIRE 0 128 0 112 WIRE 128 128 128 112 WIRE 240 128 240 112 WIRE 384 128 384 112 FLAG -208 128 0 FLAG -208 16 L1R1 FLAG -144 -80 L1C1 FLAG 384 128 0 FLAG 240 128 0 FLAG 240 16 L2R2 FLAG 304 -80 L2C2 FLAG 128 128 0 FLAG 128 32 0V FLAG -64 128 0 FLAG 0 128 0 FLAG -16 16 C1C3 FLAG 384 16 C2C4 SYMBOL ind -192 -96 M0 SYMATTR InstName L1 SYMATTR Value 22µH SYMATTR SpiceLine Rser=1u SYMBOL res -224 16 R0 SYMATTR InstName R1 SYMATTR Value 1m SYMBOL cap -48 -64 M0 SYMATTR InstName C1 SYMATTR Value 17.68nF SYMATTR SpiceLine Rser=1u SYMBOL ind 256 -96 M0 SYMATTR InstName L2 SYMATTR Value 22µH SYMATTR SpiceLine Rser=1u SYMBOL res 224 16 R0 SYMATTR InstName R2 SYMATTR Value 1m SYMBOL res 112 16 R0 SYMATTR InstName R3 SYMATTR Value 1k SYMBOL current 0 32 R0 WINDOW 3 -118 131 Left 2 WINDOW 123 0 0 Left 2 WINDOW 39 0 0 Left 2 SYMATTR Value PULSE(0 200ma 7.6u 1n 1n 100n 0 1) SYMATTR InstName I1 SYMBOL cap -48 48 M0 SYMATTR InstName C3 SYMATTR Value 17.68nF SYMATTR SpiceLine Rser=1u SYMBOL cap 400 -64 M0 SYMATTR InstName C2 SYMATTR Value 17.68nF SYMATTR SpiceLine Rser=1u SYMBOL cap 400 48 M0 SYMATTR InstName C4 SYMATTR Value 17.68nF SYMATTR SpiceLine Rser=1u TEXT 16 -248 Left 2 ;'LCR Step Change In Phase, Current Pulse Split C TEXT 24 -224 Left 2 !.tran 20u TEXT -80 -176 Left 2 ;Change I1 Delay to see changes in zero crossings TEXT 336 -224 Left 2 !.ic V(L1R1) = 10 TEXT 336 -208 Left 2 !.ic V(L2R2) = 10 TEXT -80 -152 Left 2 ;XL = 50 ohms. Change resistors to change the Q TEXT -80 -128 Left 2 ;Q = 50 / R1

[Transient Analysis] { Npanes: 3 { traces: 1 {34603011,0,"I(I1)"} X: ('µ',0,0,2e-006,2e-005) Y[0]: ('m',0,0,0.02,0.22) Y[1]: ('K',0,1e+308,1000,-1e+308) Amps: ('m',0,0,0,0,0.02,0.22) Log: 0 0 0 GridStyle: 1 }, { traces: 4 {524290,0,"V(l1c1)"} {524292,0,"V(l2c2)"} {524293,0,"V (0v)"} {34668550,1,"I(C3)"} X: ('µ',0,0,2e-006,2e-005) Y[0]: (' ',0,-10,2,14) Y[1]: ('m',0,-0.24,0.04,0.24) Volts: (' ',0,0,0,-10,2,14) Amps: ('m',0,0,0,-0.24,0.04,0.24) Log: 0 0 0 GridStyle: 1 }, { traces: 2 {34603016,0,"I(C1)"} {34603015,0,"I(C2)"} X: ('µ',0,0,2e-006,2e-005) Y[0]: ('m',0,-0.24,0.04,0.24) Y[1]: ('_',0,1e+308,0,-1e+308) Amps: ('m',0,0,0,-0.24,0.04,0.24) Log: 0 0 0 GridStyle: 1 } }

The resistor has no memory, but the capacitor does, so this is nonsense.

If the other end of the capacitor is floating freely ...

Which is only possible if you are relying on the self-capacitance of the inductor for your capacitance ...

Correct theology, but not a very useful obsservation.

Because if you inject charge when the tank charge goes through zero, you only affect the amplitude, not the phase? That optimum "point" is actually a rather broad region of near maximal sensitivity.

So the common-collector Colpitts oscillator can't actually oscillate. Good thinking there.

For equal capacitances?

Snort.

Nice to see that you have finally got your brain engaged. It has taken a while.

And for the phase-frequency detectors in both parts. The other phase detector would be unaffected.

That's not the problem at all.

Actually, a delay in one of paths through the phase-frequency detector, so that both U and D outputs are "on" at zero phase shift. Because the outputs are both current sources, this doesn't waste much current

But the wages of sin are sales - RCA and other semi-conductor manufacturers have sold a lot of CD4046 parts and it's successors.

Even the NXP 74HCT9046 is still in production, which suggests that there are heretics out there buying them.

More heresy. JL doesn't think as hard about the stuff he designs as he might.

It doesn't. It leads to DC offsets

Bipolar op amps are more vulnerable than FET input op-amps. It takes a roughtly 50mV excursion on the input to push a bipolar op amp out of it's linear operating region, more like a volt to shift a FET/MOSFET input out of its linear range.

What a surprising conclusion.

--
Bill Sloman, Sydney

It depends. I had this integrator for ns pulses using an ordinary old LF356 opamp. At each input pulse it would go totally crazy for a while, to finally settle to about 0.1% or so. The trick is that no charge goes astray with FET inputs. It takes a while, but it all goes into the feedback elements, eventually.

Jeroen Belleman

fast

take

if

lost.

I'm still not convinced. The op amp conserves charge iff it is linear. It is no longer linear when it goes crazy.

It seems clear the charge would go into the integrating cap if the other end tied to the op amp output pin was held at a fixed potential. But I don't think the impedance looking into the output pin is constant or very low.

I have been able to find combinations of ap amp gain, bandwidth and slew rate in LTspice where the output signal is in phase with a pulse at the inverting input. This is using the LTspice 1 pole op amp model, which I assume cannot saturate internally. I'm sure there must be a description of the model somewhere buut I haven't had the time to search for it.

If the op amp output is in phase with the inverting input, something drastic is going on inside the model.

It seems reasonable that some op amps would handle this problem very well, and others not so well.

How did you measure the linearity? JK

It takes all the charge that goes through the input resistor, true. But if the virtual ground moves around, that charge is no longer proportional to the integral of the input voltage, since the voltage across the input resistor is now variable. If the input is inherently a charge anyway (or current) then it is OK.

--

John Devereux

The LF356 is a FET-input part. The "virtual earth" can move by up to a volt before it's excursions start feeding a non-linear component into the output stage.

--
Bill Sloman, Sydney

With resistive inputs, it's varying with the virtual ground bouncing around. ...Jim Thompson

--
| James E.Thompson                                 |    mens     | 
| Analog Innovations                               |     et      | 
| Analog/Mixed-Signal ASIC's and Discrete Systems  |    manus    | 
| San Tan Valley, AZ 85142   Skype: Contacts Only  |             | 
| Voice:(480)460-2350  Fax: Available upon request |  Brass Rat  | 
| E-mail Icon at http://www.analog-innovations.com |    1962     | 
              
I love to cook with wine.     Sometimes I even put it in the food.

John K, I have to ask, what exactly are you trying to accomplish?

Are you working on a Baseband or RF project?

Freq Synth, Data recovery???

Is this a real hardware project or a "spice" jerk off?

Just trying gain some context for my own edification.

No disrespect implied here.