1GHz clock source

[phenol]

Thanks

I'm not sure how well it performs in terms of phase noise contribution. In terms of spurs, there's a prominent 2GHz one generated by the mmic itself.
If SRD works (and it should), I could get away with just one amplification stage after a sharp bandpass filter.
Minicircuits now have some very low noise (<1db) highly linear gain blocks.

[Berni]

That is built in to quite a neat small footprint there.

Phase noise is the most important part when running a DAC chip. The harmonics probably don't matter at all since its treated as a digital signal. You might be able to see the phase noise with your spectrum analyzer if it is high enough spec.

Would be interesting to see how your contraption compares to the internal PLL on that chip.

[phenol]

well, for really low phase noise I shouldn't use a SRD at all, as it seems to inject shot and recombination noise. NLTL have superior performance.

[daqq]

Maybe this one? :

http://uk.farnell.com/texas-instruments/lmk61e2-siat/oscillator-prog-1ghz-3-465v-smd/dp/2520002

[phenol]

with its tolerance of 50ppm it just doesn't cut the mustard in terms of stability. Plus, the datasheet has phase noise plots up to ~160MHz or so. 1GHz purity is likely going to be worse. Other than this and the potential spur injection in fract-N mode, I don't see a reason why it shouldn't work.

[phenol]

PLL (x40) and on-board 25MHz xtal vs ocxo+x10 multiplier (1GHz) in 50k, 1000k and 10000k spans using the FFT function of the scope. CF=145.45MHz. A lot of the spurs are internal to the scope, but the difference between PLL and external clock is obvious, i believe.

I did not spend any time tweaking the PLL loop filter components. Loop BW is around 50kHz with phase margin of 65 degrees according to the PLL loop filter excel spreadsheet developed by Analog.

the internal VCOs' phase noise is very sensitive to power supply noise, esp on dac and clock supply lines...open loop gain in the last band is 850MHz/V!

That is built in to quite a neat small footprint there.

Phase noise is the most important part when running a DAC chip. The harmonics probably don't matter at all since its treated as a digital signal. You might be able to see the phase noise with your spectrum analyzer if it is high enough spec.

Would be interesting to see how your contraption compares to the internal PLL on that chip.

[Berni]

Now that is a difference as clear as night and day. Great work!

[phenol]

I gave the srd mentioned earlier a go and while it is possible to get remarkably good conversion efficiency (17-18dbm 100MHz in->5-7dbm 1GHz out), i found it to be particularly fussy with output loading and harmonics reflections back onto itself. It would burst into what looks like parametric oscillations and generate all sorts of products like subharmonics (50M apart) and what not.
I then figured i should ask Macom how much their NLTL (MLPNC-7100S1-SMT580) cost and the nearly $3000 a pop made me pursue a different route-- schottky multipliers (x5+x2) with low 1/f noise parts (HSMS2815).
Would this be more efficient or cheaper than the current arrangement? Probably not, but im willing to give it a try

[phenol]

NLTL can be used as a very low phase noise comb generator, 10-15 db better than a SRD, probably close to the 20logN theoretical limit.
In terms of conversion and cost efficiency among passive circuits #1 would be srd, followed maybe by varactor multipliers, schottky and last is NLTL. The $2900 i was quoted really sounds absurd. Either it's a slow moving part and they have to produce it from a scratch or the small ceramic package is a host of a sizable diamond to justify the hefty price tag...

[KE5FX]

It's unlikely you'll be able to observe the difference between an SRD multiplier and an NLTL without going to a lot of trouble and/or expense.   The NLTLs have some nasty parametric instabilities of their own; if anything, I've found SRDs to be easier to work with.

Cheap MMIC amps make good comb generators if you overdrive them through small coupling capacitors. 

For a 1 GHz clock source it's almost crazy not to build a PLL with one of those Crystek SAW oscillators.  They're really nice parts.

[phenol]

I've seen those Crystek parts before, but still, complexity- and price-wise I believe that a passive multiplier with intermediate gain/filter stages is better. When i say complexity, i really mean the PLL+CPU burden around the SAW oscillator. Other than that, it would definitely yield a compact design.

As for SRD, what would be the optimum output loading/matching for stable operation? It only seems to run stable with some sort of resistive elements/attenuator pads, which does of course degrade conversion loss.

[KE5FX]

I've seen those Crystek parts before, but still, complexity- and price-wise I believe that a passive multiplier with intermediate gain/filter stages is better. When i say complexity, i really mean the PLL+CPU burden around the SAW oscillator. Other than that, it would definitely yield a compact design.

As for SRD, what would be the optimum output loading/matching for stable operation? It only seems to run stable with some sort of resistive elements/attenuator pads, which does of course degrade conversion loss.

You can make an SRD multiplier favor a particular harmonic with a reflective bandpass filter that bounces the unwanted comb lines back towards the diode.  But unless you're really feeling adventurous, I would use a 3 dB attenuator instead (which will provide at least 6 dB of return loss from the diode's point of view.)

[phenol]

Well, that's more or less what i ended up doing. The filter i used for experiments is a simple 2-stage helical unit with 50-ohm tap points. Connecting the low-z input of the filter directly to the SRD or even 3 SRD's in parallel to get the impedance down, was disastrous.
Some designs would use a lambda/4 line between the filter and the diode and lightly couple the other end of the line to the high-z end of a sharp cavity filter.

[phenol]

The following screenshot is the output of a x5 HSMS-2815 diode multiplier (~100MHz in/500MHz out) based on this topology: http://www.techlib.com/files/RFDesign2.pdf. Conversion loss is ~19.5dB. Even order harmonics are somewhat suppressed. x3 output is rather too strong, so further bandpass filtering is needed.

[phenol]

This is a sketch of the rev.2 x10 low noise 100MHz in->1GHz out multiplier. I don't have a clue how low a noise it actually is.
OCXO delivers ~20.5dBm into 50ohms. The input attenuator brings that down to 15dBm and improves input return loss. What follows is the Wenzel-style odd order x5 schottky multiplier based on low 1/f noise diodes. The 500MHz bandpass filter is a canned Minicircuits component. That and the coaxial ceramic resonator 1GHz bandpass filter (Minicircuits) are pretty darn expensive, yet easy to use, blocks. Cheaper helical filters should work, too.
The 500MHz->1000MHz doubler is again HSMS-2815 and a tiny Coilcraft transformer. I tried out the AMK-2-13 doubler, but, while it may have superior harmonic and fundamental rejection, its conversion loss was some 2dB worse than my version.
Interstage amplifiers are PGA-103.
The knife-carved pcb prototype puts out ~13dBm @ 1GHz with the second harmonic at -50dBc.

[KE5FX]

This is a sketch of the rev.2 x10 low noise 100MHz in->1GHz out multiplier. I don't have a clue how low a noise it actually is.
OCXO delivers ~20.5dBm into 50ohms. The input attenuator brings that down to 15dBm and improves input return loss. What follows is the Wenzel-style odd order x5 schottky multiplier based on low 1/f noise diodes. The 500MHz bandpass filter is a canned Minicircuits component. That and the coaxial ceramic resonator 1GHz bandpass filter (Minicircuits) are pretty darn expensive, yet easy to use, blocks. Cheaper helical filters should work, too.
The 500MHz->1000MHz doubler is again HSMS-2815 and a tiny Coilcraft transformer. I tried out the AMK-2-13 doubler, but, while it may have superior harmonic and fundamental rejection, its conversion loss was some 2dB worse than my version.
Interstage amplifiers are PGA-103.
The knife-carved pcb prototype puts out ~13dBm @ 1GHz with the second harmonic at -50dBc.

That should be pretty quiet.  Remember that anything better than about -155 dBc/Hz is wasted if you're driving a DAC, ADC, or DDS, due to its white noise floor.  You'll need a -175 dBc/Hz-class OCXO to get down there. 

[phenol]

As far as AD9910 goes, someone had noticed that its dac bandgap reference is not bypassed/filtered resulting in excessive AM noise.