Thanks all, for filter info. For reasons that will become evident when I
describe the LF/DC situation, I plan to use an all-passive LC LPF. I
assume I'll be needing a fairly high-order (like 9 or so) Butterworth
type response for good flatness, and enough stop-band rejection for the
higher frequency junk.

The filter that I'm using for now is an old (1970s - 80s?) K&L brand,
marked 4L52-20-0/0-100. I've never found specific info on it over all
the years, but I did just find some data on current products that look
similar:

https://klmicrowave.nyc3.digitaloceanspaces.com/products/attachments/_plk147_1_LLSeries.pdf

The descriptions look familiar enough to get some idea of what it should
do, except the passband ripple is nowhere near these specs (0.05 dB).
The modern part numbering scheme is different, but the "4" and the "20"
seem to jive, for 4-sections, and 20 MHz fc. It looks like K&L describes
the number of sections as the number of choke elements, and that's
what's in this filter. It has 4 chokes and 5 capacitors, so 9th order,
as I understand.

I picked this up in some junk long ago, and it was in bad shape -
someone had opened it up, and the cover was left hanging by a thread -
literally - a single 0-80 screw managed to keep the lid associated with
the rest. I thought the other corners were drilled out, so I just taped
it shut, and since it seemed to work, I started using it for experiments
over the years.

A couple weeks back, I began looking at it closely. I was going to
replace the original SMAs with SMBs for this project, so figured I'd
pull the guts out so I could try to ID the part values. The caps were
easy, just regular mica types with markings. The chokes are small
ferrite toroids, apparently identical cores. I counted the turns, and
found 20 on the outer pair, and 22 on the inner. I didn't want to risk
damage by removing any parts (the chokes are silicone-gooped to the
shielding), but I did manage to get ballpark in-circuit values using the
HP4276A LCZ meter at 20 kHz, so the error from the caps isn't too bad.

I also had a bit of good luck in finding that a screw was loose - one
that anchors the assembly to the floor of the machined Al box, and is
critical for grounding the circuit board. I had assumed this was not all
that great of a filter, barely keeping the stop-band 40 dB down, or that
maybe it would get better if the lid was properly attached. I also found
that there was enough intact thread left in the corner holes, that
digging up some 0-80 screws of just the right lengths fixed the lid
mounting.

So, I got my values, got my SMBs, and the filter is rebuilt almost like
new. It still leaks some at the higher frequencies, but it's now better
than 70 dB down, which is on par with the modern spec, which only shows
to that level. It is a fairly sharp cutoff filter, dropping about 80 dB
at 50 MHz

Here's the parts info:

Caps (pF labeled, unknown tolerance) 68, 150, 150, 150, 68
Chokes (uH +/- 20% possible measurement error) 2.25, 2.43, 2.5, 2.28
Choke ratio inner/outer 1.21 according to turns count

So, anyway, I know it's symmetric, supposedly 50 ohms, and 20 MHz fc.
Since then, I've been looking at filter design tools, trying to match
what's in there to any kind of "standard" filter response, and tweaking
fc and impedance too. So far, I've found nothing that's close.

The actual amplitude response looks very much like the Chebyshev example
that Gerhard posted, and the datasheet says that's what this product
line is, so it's probably in there somewhere. It's just that the part
values don't make sense.

Ed

I managed to build a filter, using the values for a 9th order
Butterworth, 50 ohms, 25 MHz fc. The caps were fairly straightforward to
get nearly right on in values, with one or two (paralleled) selected
micas for each spot. The chokes were tricky. I decided to use IF-can
style adjustable ones, since I managed to scrounge up a few that were
close enough. The whole thing is built on special double ground plane
board (with 0.15" via grid shorting the sides together) stock, which
took a lot of hand crafting to mount the cans and lay everything out right.

I checked it with the TG, and it looks like a filter, kind of as
expected. After much tweaking of the chokes, I got it to look fairly
good, but it's all open-loop, part-wise - the chokes are set for
appearance of the response, not necessarily right values. So, it's some
kind of LPF, but that's about all I can say. The chokes are the weak
link, since they're hard to measure accurately.

I put the filter into the noise project, and the result looks pretty
good. Measuring the actual noise output on the SA, and zooming in, I
found it was flat to less than half a dB p-p, looking at 1 dB/div. Not
bad considering my eyeball-controlled adjustment using 10 dB/div and the
TG beforehand. This flatness is the net effect of the noise output
itself, and the filter, and a little bit the SA, so pretty decent. The
fc is around 22 MHz at the "best appearance" setting, and the Z-match
seems OK. There's no pad at the filter input, and about 3 dB at the
output, then that same 20 feet of cable to the SA. The high frequency
rejection looks pretty good too, with the 70 MHz and 140 MHz (the worst
offenders) below -85 dBm. This can be improved with more grounding
enhancement, and possibly adding shielding - it's kind of open
construction now, just on the board. The chokes are fairly well
contained and shielded in the cans, but the caps are exposed.

Anyway, for this purpose, it's way better than the original filter,
which can now be returned to its other project. I'm fairly happy with it
so far, but expect it to be one of those never ending projects - always
room for improvement.

Ed

Thanks Mike, for info on LCR alternatives. It's good to know of others
out there, if needed. I have an HP4276A and HP4271A. The 4276A is the
main workhorse for all part checking, since it has a wide range of LCZ,
although limited frequency coverage (100 Hz - 20 kHz). The 4271A is 1
MHz only, and good for smaller and RF parts, but very limited upper LCR
ranges. I think it works, so I can use it if needed, but would have to
check it out and build an official lead set for it. I recall working on
it a few years ago to fix some flakiness in the controls, so not 100%
sure of its present condition.

The main difficulty I've found in measuring small chokes is more of
probing/connection problem rather than instrument limitation. For most
things, I use a ground reference converter that I built for the 4276A
many years ago. It allows ground-referenced measurements, so the DUT
doesn't have to float inside the measuring bridge. The four-wire
arrangement is extended (in modified form) all the way to a small
alligator clip ground, and a probe tip, for DUT connection, so there is
some residual L in the clip and the probe tip, which causes some
variable error, especially in attaching to very small parts and leads.
When you add in the variable contact resistance too, it gets worse.
Imagine holding a small RF can (about a 1/2 inch cube) between your
fingers, with a little clip sort of hanging from one lead, and pressing
the end of the probe tip against the other lead. All the while, there's
the variable contact forces, and effects from the relative positions of
all the pieces and fingers, and the stray C from the coil to the can to
the fingers. I have pretty good dexterity, and have managed to make
these measurements holding all this stuff in one hand, while tweaking
the tuning slug with the other.

I had planned on making other accessories like another clip lead to go
in place of the probe tip, but not yet built. I also have the official
Kelvin-style lead set that came with the unit, so that's an option that
would provide much better accuracy and consistency, but the clips are
fairly large and hard to fit in tight situations, and the DUT must
float. Anyway, I can make all sorts of improvements in holding parts and
hookup, but usually I just clip and poke and try to get close enough -
especially when I have to check a lot of parts, quickly.

The other problem is that the 4276A is near its limit for getting
measurements below 1 uH, with only two digits left for nH. The 4271A
would be much better for this, with 1 nH vs 10 nH resolution.

If I get in a situation where I need to do a lot of this (if I should
get filter madness, for instance), then I'll have to improve the tools
and methods, but I'm OK for now, having slogged through it this time.

Ed

Now that I have the "official" filter in place, I can wrap up the LF/DC
issues. This is the other extreme, so no SA here, just time domain view
with a Tek 7A22 vertical, which gets down to 10 uV/div, and has settable
BW steps from 100 Hz to 1 MHz. For very low f and DC, I use a HP3456A.
There are some limits, especially in the 7A22, which is a little flaky,
but mostly puts on a good show. In either instrument, there may be
errors caused by the large HF part of the noise up to the 25 MHz or so,
way beyond what they're trying to see.

One thing that immediately showed up is the mixer DC offset (about -1.2
mV) due mostly to imperfections in the mixer, distortion in the LO, and
LO leakage into where it doesn't belong. I built a photo-voltaic
circuit to generate a current to cancel it out, but had to wait until
other issues were settled before final design adjustments.

Why a PV generator? This relates to the fundamental design plan. You may
recall that in the earlier talk on the mixer, I wanted to be able to
have galvanic isolation of the IF port, in order to eliminate or reduce
ground loop interference. Indeed, I found out right away that this was
the way to go. On the 7A22, I could see several mV of line-related junk,
and figured it was time to lift the IF off ground. For the RF
experimenting, I had the IF chassis-grounded, but had all the provisions
in place to float the whole works, from the IF port all the way to the
front panel BNC. I chose to overdo the capacitance from the IF common to
earth, with two 100 nF caps. The common-mode chassis noise disappeared,
as expected.

But, all this forces various compromises between the requirements.
First, there's not much point in making a thing that can go essentially
all the way down to DC, and possibly at very tiny signal levels
(depending on BW and noise power level), if you can't convey the signal
to an external piece of gear or experiment without ground loop
interference. So, this isolation is necessary - it raises the
common-mode impedance of the source so that the (hopefully) small
inter-chassis voltages can't push much current between equipment.

But, this is all frequency dependent too. If the ground loop
interference has higher frequency content (like in something with a SMPS
that's not very clean), the caps isolating the floating section present
much lower CM impedance, allowing more current. For this, you'd want
minimal CM capacitance.

But, minimal CM capacitance is minimally effective in shorting out the
LO and RF at the mixer - whatever leaks through due to the limited
isolation of the mixer becomes CM and additional IF signal at the IF
port. For this, you'd want as high a CM capacitance as possible, or
solid ground (which is the non-isolated form).

So, it all boils down to making appropriate trade-offs in that CM
capacitance. As mentioned earlier, I started with 200 nF, which was
sufficient for line/harmonic interference rejection, and was a good RF
short at the mixer. Next, I tried a lower extreme of 2 nF total, which
would have been great for medium frequency rejection, but alas, not a
good enough short for the LO and RF, indicated by increasing power at
the output, and increasing DC offset - it nearly doubled it.

The present compromise is about 9 nF total (the previous 2 nF plus three
2200 pF tacked on). This seems to be pretty good, with reasonably small
(maybe -90 dBm) LO showing, and only slightly higher offset compared to
the 200 nF version. I think when all's said and done, I'll end up with
about a 10-20 nF compromise value.

There's also some CM choking involved. The most important one isolates
the LO and RF CM right at the IF port, formed with three loops (about 10
uH) through a ferrite toroid of the SMB pigtail cable the goes to the
filter. A second one will be included on the output cable to the front
panel, to help at the medium to high frequencies.

I edited the box's board ground plane to form the isolated section that
carries the filter, padding, associated interconnects, and PV generator
circuits. Since this all floats, the PV method is used, and no power
supplies or chassis ground returns (which would spoil it) are needed.
The generator is two paralleled 4N37 opto-isolators operating in PV
mode, with variable LED drive for setting the offset current.

The concept of "floating" is somewhat arbitrary. In reality, the whole
output could float to any applied voltage until something breaks down,
but I decided it was safest to just hard-clamp it to chassis ground with
Si rectifiers (1N5401). Unfortunately, their zero-bias capacitance adds
to the total CM capacitance, while they can't help with any RF shorting
at the mixer - they're too big to fit near there, and are too far
removed from the action by distance and the CM choke.

Next up will be more details. It's getting close to the end. I can tell
that it's near time to wrap up or quit this project, because the
connectors are starting to wear out from all the puts and takes of the
box into the instrument - I'd say it's well over a hundred times already.

Ed

I've been working on final design cleanup, mainly in the RF. I found
quite a bit of spurious LO harmonic content up to almost 2 GHz, with
some quite strong (-75 dBm). It was time to clean up the experimental
wiring layout, so I simplified the cabling and consolidated the RF stuff
onto the LPF board. This improved things a bit, but some spurs were
still pretty big. I presumed most of it was going right through or
around the LPF, and some due to common-mode and cavity resonances inside
the box, which can have many modes.

I added a small LPF about 300 MHz (10 pF/50 nH/10 pF), inside its own
tiny shield box, forming the last bastion of filtering, right at the
inlet of the pigtail cable that goes to the isolated SMA bulkhead
fitting, and including another CM choke (only 1 pass of cable). This
filter is high enough up (over ten times the fc of the main LPF) that
they shouldn't interact very much - they are isolated only by the 3 dB
pad in between.

All along, I've wondered what to do about the reflected power from the
main LPF, that mostly has to go back to the mixer. They are separated by
maybe 300 pSec of cable, which could be in the range for resonances at
the upper end. But, various experiments during development, including
padding the LPF input, and even making a diplexer with a 50 MHz HPF to
take the HF content into a terminator, showed no difference in the noise
output flatness, although the spurious levels likely would have changed
a little - some up, some down. So, I decided to keep it simple and just
let 'er rip, with nothing extra at the LPF input.

Things are now at levels where the fine (and subtle) details show,
mostly cable dress, and grounding. I'll probably be adding bits of
shielding here and there, and maybe fooling with some RF absorbing foam
to see if any box resonances are a problem.

Speaking of subtle effects, here's something interesting. The little
shield box for the 300 MHz LPF is a type with a fold-down lid, on a
hinge formed by thinning the sheet steel. It's only good for a few open
and close operations before the hinge breaks apart, so I kept it open
while building and testing the filter. It looked great, and the time
came to close everything up and look at the spurs again. I closed the
lid, and bent the retainer tangs a little, for good closure. Virtually
all the higher frequency spurs got a few dB worse. So, was it that the
lid isn't really grounded thoroughly, and acting as an antenna to bypass
the filter, or did it affect the choke Q or part values enough, or is it
that I also changed the cable dress a bit while putting it all back
together? I'll have to figure it out.

Anyway, it's looking pretty good right now. With everything closed up,
including the box lids, as it would be when completed, all the spurs
show around -90 dBm or less. There were maybe two dozen noticeable spurs
identified earlier. Some are now in the noise floor (around -105 dBm,
but some remain, sticking out. I think most will disappear if I figure
out that 300 MHz filter box lid, which would leave the 70 MHz as the
main offender. This isn't surprising, since it's the biggest signal of
all, and it's not filtered all that much - it's too close to the main
LPF fc, and below the 300 MHz LPF. I should be able to knock it down
enough with detail work mentioned above, and I'm also pondering ways to
make a 70 MHz trap, if it won't go away. I have a couple of 70 MHz
crystals, so I could try this kind fairly easily. Does anyone have any
handy design info for crystal notch filters in this frequency range? For
an LC trap, it looks like a single L and C would be enough to get the
job done, without interacting too much with the other filters.

Ed