- Hi All,
One last comment before I run off to the horse piddle and get my
pacemaker. When dealing with PLL op amps: Most PLL's are using tri-
state, current sourced charge pumps as the loop driver. If the loop
is in phase lock, the pll output will go into a "floating" condition.
It will remain inactive, and floating, if the vco has no noise, and
nothing in the loop circuit pulls or pushes the vco control voltage
line away from its inactive, "floating" voltage. However, in op amps,
the input bias current will pull or push that output line away from
its lock voltage. A perfect op amp would have 0 bias current (ain't
gonna happen). But an op amp with the least amout of input current is
best. Compare two "perfect" PLL's with "perfect" loop op amps, with
one exception, one op amp has some input bias current and the other
has no bias current. The PLL system, with op amp input bias current,
will have phase detector frequency sidebands on the vco output. The
other, with no bias current will have no PDF sidebands (or they will
be extremely low).
Loops that draw bias current will guarantee that its loop will never
be in full lock, thus, creating PDF spurs. Every cycle of PDF, the
PLL is adjusting the loop, trying to get the system into lock. This
cyclical "adjustment" is putting small pulses on the vco control
voltage line and creating the PDF spurs. The lower input bias op amps
will give the best PDF spur performance. These are spurs, not phase
noise. Phase noise creation from op amps is a different story.
--- In firstname.lastname@example.org, "swetterlin"
> The AD8671 is a great op amp, partly because of its 3na bias current.solder
> Playing with SimmPLL seems to show that significant phase noise
> results if you get above the 3na level. But remember that 3na at 12V
> corresponds to a resistance on the order of 4 giga-ohms--a huge
> resistance. This means any alternative path from an input to ground
> that is less than 4 Gohms will dominate in determining the effective
> bias current, so the input pins and the area around them need to be
> VERY CLEAN, and there should be no nearby traces unprotected by
> mask. Flux residue--especially if it absorbs humidity--probably has
> far too much conductivity. I even wonder if the input pins should be
> painted with conformal coating or something similar.
> Sam W
Thanks and good luck.
--- In email@example.com, "Scotty" <wsprowls@y...> wrote:
> Hi All,
> One last comment before I run off to the horse piddle and get my
- OK, the DigiKey fairy left some new opamps under my pillow last night, so here are some new results:This plot shows the composite noise at 10 kHz from the carrier measured with 3 different opamps in my hybrid synth board. This particular board still has the UMS-2150 VCO installed. Loop bandwidth is about 1950 Hz (R208=100 ohms, R209=100 ohms, C212=2 uF, C213=0.2 uF, C216=0.1 uF).Green = the LT1677, at -89.8 dBc/HzPurple = the AD8671A, at -90.0 dBc/HzBlue = the LT1128, at -91.3 dBc/HzThere wasn't much difference between the LT1677 and AD8671A. The big winner is the LT1128 (data sheet at http://snipurl.com/fzjh).This opamp is unity-gain stable, +/- 22V supplies, less than 1 nV/root-Hz input noise at 1 kHz. As the data sheet points out, this is less than the thermal noise from a 50-ohm resistor. Input bias current is good, if not record-setting, at 25 nA. Like the LT1677, it is considered a drop-in replacement for the OP-27/OP-37 family... but it is not a so-called rail-to-rail part like the LT1677 is.I am not sure I understand the "maximum output voltage swing" parameter on the data sheet. I'd assume it's just the total available output swing about the Vcc/Vee midpoint. But are they saying that with +/- 22V supplies, the output can't swing any closer than about 10V from the supply rails? I didn't see any problems with 0/24V supply rails, although I didn't try to drive the output any closer than about a volt to either rail.Next, I installed the LT1128 in my homebrew rx's first-LO synthesizer, which is (was) identical to the example in my QEX article. It uses an ROS-2150VW VCO in a 2500-Hz loop. The improvement was smaller but still worthwhile -- from -90.3 dBc/Hz to -91.2 dBc/Hz. The ROS-2150VW's tuning sensitivity is lower than the UMS part, and the loop bandwidth was wider than in my first experiment above, so it's not surprising that differences in opamp noise aren't as noticeable.I lowered the loop bandwidth to 1750 Hz, and the noise at 10 kHz went down to -94.9 dBc/Hz, which is about the best I've seen from my synthesizer. Lockup time from 1-2 GHz went from the 10 ms neighborhood to approximately 30 ms -- a bigger jump than I'd expected. I should probably hang a scope off the VCO line to see if anything ugly is going on with the larger capacitors I'm using now.All in all, the LT1128 looks like the best opamp for PLL filter applications. Its drawback is cost -- $8.63 each at DigiKey versus $2.79 for the AD8671A and $4.50 for the LT1677.-- john, KE5FX-----Original Message-----
From: John Miles [mailto:jmiles@...]
Sent: Wednesday, June 29, 2005 12:56 PM
Subject: RE: [spectrumanalyzer] AD8671Yep, that looks like a good part. I'd been meaning to order some LT1128s from DigiKey to try them, so I threw in a few AD8671s as well. They are sure a lot cheaper!Input bias current is important, but I'd note that people have been putting 1 M resistors on the output of charge-pump PFDs for years to avoid dead-zone problems. I would sure be surprised if 4 Gohms would make any difference in the real world! I'd be skeptical that this is really an important factor.-- john, K5FX-----Original Message-----The AD8671 is a great op amp, partly because of its 3na bias current.
From: firstname.lastname@example.org [mailto:email@example.com]On Behalf Of swetterlin
Sent: Wednesday, June 29, 2005 12:09 PM
Subject: [spectrumanalyzer] AD8671
Playing with SimmPLL seems to show that significant phase noise
results if you get above the 3na level. But remember that 3na at 12V
corresponds to a resistance on the order of 4 giga-ohms--a huge
resistance. This means any alternative path from an input to ground
that is less than 4 Gohms will dominate in determining the effective
bias current, so the input pins and the area around them need to be
VERY CLEAN, and there should be no nearby traces unprotected by solder
mask. Flux residue--especially if it absorbs humidity--probably has
far too much conductivity. I even wonder if the input pins should be
painted with conformal coating or something similar.
- It's fun to see experimental results to see how these things really
do. The "maximum voltage swing" you cited in the Electrical
Specifications is measured with +/- 15V supply (see fine print at the
top). By subtracting from 15 you get the margin from the rails, which
typically stays pretty constant as you change the rail voltage.
The data sheet also illustrates my point that op amps are the one
component where you do not need to worry about sub 10kHz noise getting
through the power supply. With PSRR of better than 80db from 0-10kHz,
noise of 10uV at the supply pins is the equivalent of noise of 1nV at
the input! At 1kHz it is equivalent oto 0.1nV at the input.
--- In firstname.lastname@example.org, "John Miles" <jmiles@p...> wrote:
> OK, the DigiKey fairy left some new opamps under my pillow last
> here are some new results: