HP 8591E spectrum analyzer non-linearity

[MarkL]

I was looking at some linear (non-log) sweeps using a HP 8591E spectrum analyzer and I noticed an odd dip in amplitude at around 59MHz.  Double checking on a 8595E and two different oscilloscopes, I concluded the dip was not in the source.

Here's what the 8595E shows (scale 1dB per div):



And here's the 8591E:



The approx +/-1dB variation is not terrible, but that's the max spec for the 8591E.  It looks unusual enough that I thought I'd poke at it a bit.  I've included the block diagram PDF from the service manual below if you want to follow along.

To get your first thoughts out of the way, the dip happens in log or linear mode, and it also happens at any attenuation setting.

Here's what I've discovered using the 8595E to look at various points:

  - The source is flat coming out of the attenuator and into the A4 first converter.  So the problem is not in the attenuator or anything previous to A4.

  - The A4 J2 LO input is flat during 8591E sweeps.  LO is ok.

  - The A4 J4 counter output is flat during 8591E sweeps.  LO passing through A4 is ok.

  - The A4 J5 tracking gen output is flat during 8591E sweeps.  (This unit does not have a tracking gen, so there was just a terminator on that port.)  More verification that LO is ok.

  - Using the 8595E in zero-span mode @ 2.1214GHz, I can see the dip looking at the J4 band-pass align port on the A5 second converter.  So, is it coming out of the A4 first converter?  Let's see...

  - Disconnecting the A5 second converter and looking at the signal right at that connector coming from A4 & FL1, again using zero-span, I do NOT see the dip.  Is it something with the A5 second converter?

  - Inspected the innards of the A5 second converter.  There's not much to it.  Re-assembled it and re-aligned the band-pass filter.  Dip unchanged.

  - Inserted the 8595E with a "T" at the A5 second converter input.   I could see the dip, although it's now at a different frequency.  I had to replace the hard line with a 12" jumper to get the "T" in place.  It's interesting the dip moved.

  - Removed the "T" and left the jumper.  The dip is still there but at a different frequency.  So, now the only difference is the connection between the A4 output filter FL1 and the A5 input is a different length coax, and it's RG174 instead of hard line.

  - Inserted an SMA elbow on the 12" jumper.  The dip moved.  Here's the difference with and without the elbow (yellow is without the elbow, green is with):



So this says there's sensitivity to the length of the signal path between the first and second converter.  The elbow adds about an inch to the path.

My theory is that reflections are occurring on the input of the A5 second converter and they're making it back to the A4 first converter and causing interference.  It could be happening as far back as the A4 mixer, but I'm not sure.

Any other ideas?

Or should I not worry about it and and just fix it in the calibration constants?  Oddly, there's already a noticeable delta from the factory at 41MHz:

  Freq(MHz)   Error
    4        -0.90
   41        -1.40
   78        -0.63
  115        -0.73
  152        -0.49
  (...etc...)

Maybe it's always been like this since the day it was made?

I haven't opened the A4 first converter (yet), but I suspect it more than the A5 second converter at this point.

Anyone have a 8591x that can compare their lower frequency response?

[Richard Head]

That's very interesting. Perhaps what you can try is to temporarily insert a 50R attenuator (say 6-10dB) between the A4 output filter and the A5 input. The return loss of the filter will be twice the insertion loss so should significantly reduce the reflected energy. The attenuator will introduce a gain reduction but that can be ignored.
If you have a VNA then you can sweep the A4 and A5 modules independantly to confirm that S22 of the A4 module and S11 of the A5 module are indeed 50R.

[MarkL]

The attenuator sounds like a great idea but I only have BNC ones at the moment which I don't trust at these frequencies.  I did try one but it changed the shape so completely nothing could be concluded.  I've ordered a nice set of SMA ones from Minicricuits which should be here in a day or two.

A VNA would make it pretty easy but unfortunately I don't do enough RF work to justify the investment.

The 8595E has a tracking generator to 2.9GHz, so I put a directional coupler in the Minicircuits order.  Hopefully a scalar view of the reflections (if that is indeed the issue) will help.

Thanks for the ideas!  I'll post back when I have some more results.

[Richard Head]

The TG and SA with a Minicircuits directional coupler should allow you to check the return loss. I'm not sure but I would think if the match had a RL of 20dB or better it should be fine.
I would have been inclined to make a temporary attenuator using 0805 resistors and SMA connectors but some Mini-circuits ones wil be a good investment anyway.
The frequency is only 59Mhz so you can get away with murder.
Good luck.

[KJDS]

What messes up mixers isn't the match at the wanted frequency but the match at the image frequency

ETA, there's only a need to get about 10dB match on a mixer to get decent performance. The part that is so often missed is the image frequency also needs a reasonable match. So if you have an RF signal at 800MHz and you mix it with an LO of 1000MHz to get a 200MHz IF, it's easy to remember to make sure that the match at 200MHz is good, but there is also the image frequency at 1800MHz which gets the same level of signal as the wanted 200MHz. This is why a mixer is usually followed by an attenuator or a diplexer to provide a reasonable match at both the wanted IF and the image frequency.

I wouldn't trust a Mini-Circuits directional coupler to accurately measure return loss, but that's probably for another post about RF measurement accuracy.

[MarkL]

I would have been inclined to make a temporary attenuator using 0805 resistors and SMA connectors but some Mini-circuits ones wil be a good investment anyway.
The frequency is only 59Mhz so you can get away with murder.
Good luck.

The frequency between A4 and A5 is 2.1214GHz plus mixing products, so perhaps not so trivial.  And since the signal path length makes a difference for this problem, I wanted two attenuators that were identical except for dB to try your suggestion.

I've been meaning to buy a decent set of attenuators anyway.

[G0HZU]

I'm not familiar with the HP8591 analysers but I did find and look at the block diagram you are referring to. I think I can offer a guess as the the design procedure for this part of the analyser if that helps? I would expect that the first mixer will be a fairly basic singly balanced mixer using two diodes. I would therefore expect that the conversion efficiency will be at its best if there is a very low impedance at the IF port at the LO (and IF) frequency. Conversely the conversion efficiency would be poor if the impedance at the IF port was high at the LO (and IF) frequency. Usually there will be a compromise made here (against signal loss) with an attenuator placed after the mixer to try and provide some impedance control here.

But this won't be enough on its own. The 5GHz LPF sections that follow this are designed to pass both the wanted IF signal at 2.1GHz as well as the LO and the image term. In an ideal world the input to the second converter will look like a flat 50R load across all these frequencies. But in reality there is a narrow first IF BPF at 2.1GHz here.

So an ordinary BPF would cause problems here because it will be highly reflective in its rolloff region and also the stopbands. I can only guess here but I think there will be some cleverness here in this IF1 filter design to make it look resistive (ideally close to 50R) in the stopbands. This means low reflections and the low Z at the IF port of the first mixer will be maintained across the wanted 2.1GHz IF1 passband and ALSO the far out stopband frequencies across several GHz.

But there is a problem... at frequencies very close to the IF1 BPF passband (i.e. a few tens of MHz away from 2.12GHz) the impedance looking into the second converter will not be close to 50R. i.e. the region where the IF1 BPF begins to roll off will have a wide variation in impedance. So this means that reflections will occur back to the first mixer in this frequency range. Maybe this is what you are seeing.

So my guess is that HP will have designed this whole circuit such that the IF1 BPF in the second converter and the length of the transmission line connecting all this will be optimised to present a LOW (or at least a controlled) impedance at the IF port of the first mixer in this critical range of frequencies. eg at frequencies just above the IF1 passband. If you change the length or tweak around in the IF1 BPF then the impedance presented to the IF port of the first mixer can rotate around the smith chart to the undesirable 'high impedance' area of the chart and the conversion loss will get higher.

So maybe the adjustment of the IF1 BPF and the length of the transmission line is critical to prevent poor flatness in the mixer conversion efficiency at frequencies that lie near the 'shoulders' of the IF1 BPF in the second converter. So you can expect the analysers frequency response to be worse here compared to slightly higher frequencies where the LO and image terms are probably going to see a resistive impedance looking into the stopbands of the IF1 BPF because I think it will be designed to look resistive in the stopbands.

Hope this makes sense. I think you may need to use a VNA here to see what part of the smith chart is being presented to the IF port of the mixer by the rest of the analyser. If there is a lot of impedance variation here at 2.15 to 2.25GHz then you can expect to see a fair bit of ripple in the analyser response in the low VHF region of operation.

In reality, the optimum adjustment point might be a halfway house between a low impedance and a high impedance because (maybe) this will give less overall ripple. Ideally you want an isolator here and maybe that is why some models like the HP8591C have one here? Maybe the isolator trades sensitivity against far better flatness in this critical region for the VHF TV bands.

Note that all of the above is just guesswork looking at a very basic block diagram so some or all of it could be wrong

[MarkL]

I received the attenuators and tried Richard's test which was to try them between the A4 first converter and A5 second converter.

To start, I made sure the insertion point before or after the FL1 4.9GHz LPF makes no difference.  That's confirmed, so there's nothing weird happening in FL1.

To keep the signal path length approximately the same, I used an SMA elbow as a stand-in for a 0dB attenuator.

Here's what I get, 0dB to 3dB:



There's quite a noticeable difference, confirming a reduction in reflected energy from the bandpass filter in A5.  I also tried 3dB/6dB and 6dB/10dB.  There was a noticeable reduction of artifacts in both of those steps too, but much less pronounced.


KJDS: You're quite right on the Mini-Circuits directional coupler to do any kind of accurate reflection measurements.  It's no substitute for a proper test set.  But I thought I'd at least give it a try to show which frequencies were reflecting without investing a couple of magnitudes more.  It might be an experiment doomed to fail.


G0HZU:  Thanks for the detailed interpretation of the block diagram.  (I did, BTW, include it in the first post.  Sorry it wasn't obvious and you had to rummage for it.)  Your description makes a lot of sense and matches what I've seen so far.

You're right that the first mixer is just two diodes.  I managed to find a schematic (pulled from a post by "Mike M" on Agilent's forum):



And here's the second converter (same source, PDF of this below if image is too fuzzy):



Unfortunately there's no VNA coming to the rescue to measure the actual impedance.  I'm going to try the coupler, and I'll also try tweaking the A5 BPF as you suggest to see if I can find a happy medium.  Or see if it even affects the dips and peaks at all.

But if not successful, it may be just be the way the 8591E is.  Your theory on the isolator makes sense that they knew about this issue and wanted to reduce the artifacts for the 8591C.

[KJDS]

There's four diodes in a balanced ring for the first mixer

[MarkL]

There's four diodes in a balanced ring for the first mixer

Oops - right.  Sorry.

[G0HZU]

The mixer circuit isn't quite what I was expecting to see but I think it is still just a singly balanced mixer despite the 2 extra diodes.

Also I think this circuit topology will be less sensitive to the IF termination impedance than I first thought. So the IF1 BPF in the second converter may well just be some form of cavity filter.

Some of the older HP analysers arranged the mixer in a different way. Also, They had a kind of directional IF1 filter at the IF port that maintained  a 50R termination across all the frequencies that exit from the IF port. eg the IF, the image and the LO frequency. This improves the flatness. So I thought they would do the same here.

[G0HZU]

I did a quick design and Sonnet EM simulation of a very basic directional filter for a post mixer filter with an IF1 of 2.12GHz and you can see the plot below.

This filter is a printed type with just a few strips of copper on a PCB and some termination resistors that provide a BPF response at 2.12GHz but it also gives a decent return loss in the filter stopbands. The filter is about 1.5" x 1.2" in size.

Obviously this filter on its own isn't adequate as an IF1 BPF as it doesn't have very good selectivity or stopband performance but it shows the technique. This filter would give a reasonably broadband termination at the mixer IF port as well as providing the initial selectivity at IF1. Some (higher cost?) HP analysers use this type of printed filter immediately after the mixer (instead of an attenuator?) with very similar construction techniques as mine.

I've designed more complicated versions in the past with higher performance for other bands but these needed multilayer technology.

[MarkL]

Well I tried adjusting the BPF in A5.  It's a cavity filter with 3 tuning studs.  There's no happy medium that I could find between peaking it and minimizing the dip at 59MHz.

Maybe I should just pick up a used isolator from an 8591C?  I don't know the specs of that specific unit, but a quick survey of similar ones says that I shouldn't lose more than a dB.

[MarkL]

I decided to try an isolator from an 8591C (part #0960-0084) at the input of the A5 the second converter.  This is where it goes in an actual 8591C.

The parts list says it's a 2 to 4 GHz isolator.  Testing it on a 8595E shows it's providing at least 21dB of isolation from 2.0GHz to 2.9GHz (that's as far as the TG will go).

The trace from the 8591E below shows without the isolator in yellow, and with in green.  At attempt was made to keep the with/without signal paths the same length.

It's certainly an improvement, but not as dramatic as I thought it would be.  Looks like it has an insertion loss of around 0.3dB, which isn't too bad.

Oh well.  Probably enough time spent on this unless there's any other interesting ideas.

[KJDS]

The problem with the isolator is one that I mentioned earlier in the thread. It's got good performance at the desired IF frequency, but the image products, at LO+RF will all be reflected back to the mixer. It needs a diplexer in there.

[G0HZU]

At the area where the ripple occurs , all three terms (IF, LO and image) will be very close together in frequency. However, unless something is placed after the mixer and before the IF1 BPF then the IF1 BPF will present a wide variation in impedance across this relatively narrow frequency range.

On higher cost/performance analysers from the 1980s they fitted a directional filter after the mixer to help with this issue. This is what is fitted to both my HP8566B and Advantest TR4172 analysers. The other thing to be wary of is the level of LO leakage from your first mixer. If this leaks at a high level  (when the analyser is sweeping at low RF frequencies) it can get through the first IF filter and may be high enough to cause issues with linearity and flatness in the second mixer.

Many years aso I made a flat noise source that is flat to about 175MHz and I sometimes use it to look at the flatness of my analysers as it is quick and easy to do this test.

The plots below show my 8566B and TR4172 up to 180MHz. Above 180MHz you can see the noise roll off because the noise source has a 180MHz LPF on its output.

But they both look quite flat. Maybe the best thing to do is see if you can find someone else with the same analyser as yours and make a comparison. The performance with the isolator does look better but maybe this is now masking the original problem.

[G0HZU]

Here's an old image of the IF1 directional filter in one of my TR4172 analysers. It's a bit fuzzy because I've cropped it from a much bigger image.

But you can see how simple this filter is. However, to make it accurate in frequency (in production volumes) and low loss at 2.05GHz requires an expensive PCB material.

It's probably about 30mm in width/length but I didn't measure it. You can see the 50R terminations in the filter. The idea is that only 2.05GHz can get across the 'square' filter. All other frequencies get dissipated in the 50R termination resistors depending on which direction they arrive at the filter.

I'm not suggesting you make/fit one to your HP8591 analyser but I thought some people may find it interesting to see this old technology in the flesh

 

[ConKbot]

Simple thing, but a potential gotcha? Whats the correction thats applied?  (CORR indicator) I had a coworker confused for a while about why a low pass filter had tons of gain, and a massivly sharp cutoff after I normalized a the spec-an with a high pass filter at a different freq. on it.   

[MarkL]

The problem with the isolator is one that I mentioned earlier in the thread. It's got good performance at the desired IF frequency, but the image products, at LO+RF will all be reflected back to the mixer. It needs a diplexer in there.

What diplexer would you recommend given the broad range of mixing products that are present?  The input to the A4 mixer is the input to the analyzer and can range from 9kHz to 1.8GHz.  That gets mixed with 2.1214GHz and ideally we'd want to absorb everything outside of 2.1214GHz before sending it to the A5 second converter, or if everything gets sent to A5 then absorb the reflections.

G0HZU's description of the directional filter in the TR4172 makes sense for this situation but I'm curious about your diplexer approach.

I guess I'm not seeing how it would work unless you can get one that has an extremely sharp cutoff to differentiate between 2.1214GHz and 2.1214GHz-9kHz and less.

[MarkL]

Simple thing, but a potential gotcha? Whats the correction thats applied?  (CORR indicator) I had a coworker confused for a while about why a low pass filter had tons of gain, and a massivly sharp cutoff after I normalized a the spec-an with a high pass filter at a different freq. on it.   

Interesting prank, but that's a good observation on the "CORR".

A couple of quick measurements.  Here's the sweep with the isolator still in line.  Yellow is with correction, and green is uncorrected.  The uncorrected trace has its reference 10dB higher, but I'm looking at the shape.



The sharp point in the corrected trace is exactly at 41MHz, which is one of the amplitude calibration points (mentioned in the first post).  Ah ha!

The uncorrected trace still has some variation and it definitely has a slope, but I'm thinking that recomputing the calibration points would flatten everything out sufficiently that it would be well within spec.

And to make sure there's not something else strange going on with correction, here's a shot without the isolator (back the original config), again yellow with correction and green without.  And again the uncorrected trace is 10dB higher.  The original problem is back, but you can see the slight difference in the corrected trace (yellow) at 41MHz.



So, what I'm concluding from all this is that the reflection problem always existed in this analyzer.  The factory correction value at 41MHz was a little bit out of line with the others.  And in the calibration procedure you don't sweep a range.  You calibrate specific frequencies and you would not have seen or been able to compensate for the shape of the reflection artifact.  But chances are high it would have impacted one or more calibration points that fell into the vicinity of the reflection.

At this point I think my options are to live with it (it's only +/-1dB after all), or leave the isolator in and re-do the amplitude calibration points.  It's a long procedure with 50 points.  We'll see.


I'm still curious though... Does anyone else with a 8591E see this at the lower frequency range?

[G0HZU]

G0HZU's description of the directional filter in the TR4172 makes sense for this situation but I'm curious about your diplexer approach.

In my experience with respect to mixers the term 'diplexer' is loosely used to cover a broad range of IF port networks that all seek to achieve the same goal. So I think the directional filter would be classed by many as just another form of mixer diplexer. I guess it could be debated if diplexer is the wrong term to use for the directional filter but I'd be quite happy to see it referred to as a mixer diplexer.

[MarkL]

G0HZU's description of the directional filter in the TR4172 makes sense for this situation but I'm curious about your diplexer approach.

In my experience with respect to mixers the term 'diplexer' is loosely used to cover a broad range of IF port networks that all seek to achieve the same goal. So I think the directional filter would be classed by many as just another form of mixer diplexer. I guess it could be debated if diplexer is the wrong term to use for the directional filter but I'd be quite happy to see it referred to as a mixer diplexer.

Ok, thanks for the broader definition.  Most of my RF experience is from broadband cable where diplexers are three port devices used to separate (or combine) downstream and upstream frequencies.

[KJDS]

G0HZU's description of the directional filter in the TR4172 makes sense for this situation but I'm curious about your diplexer approach.

In my experience with respect to mixers the term 'diplexer' is loosely used to cover a broad range of IF port networks that all seek to achieve the same goal. So I think the directional filter would be classed by many as just another form of mixer diplexer. I guess it could be debated if diplexer is the wrong term to use for the directional filter but I'd be quite happy to see it referred to as a mixer diplexer.

I'd forgotten that this was a broadband input, therefore it isn't possible to use a classic diplexer here.

I do have some 8594E here, I'll put one on the bench when there is some spare space on it and see how that performs.

[Eyecue]

Did you get the whole schematic or just the clip for that section?  I need the schematic for the A11 and A13 RBW filters.