Signal generator - Now, new to me E4432B - now fixed

[Joel_l]

Hi,

I'm looking for a signal generator to replace my fading Wavetek 2410. I want something that goes to 3GHz ( why I just don't repair the 2410 ). Ii have been looking at R&S SM300 ( just a bit more than I want to spend ), SMIQ03,and SME03 type generators.

I recently looked at an Anritsu MG3700A. It looks interesting but seems highly biased towards digital communications. This interest me, but primarily I will be using AM and FM sweeps and some PWM. Looking at the instrument layout, it seems it should be able to do these things but not in an easy deliberate way. Looks like you would have to construct some digital wave files.

Anyone here have one of these? How messy is it to generate analog type AM, FM, PM, sweep type signals?

Open to other thoughts, want to stay < 1K ( lower = better ). I'm probably 90% < 100MHz with the mentioned modulations, but do want the digital capability at 2.4GHz.

Thanks

[Ice-Tea]

Let's see what I have... I have a SMIQ03B, around your 1k$ mark. The one I have doesn't have FM modulation, though.

A R&S CMU200 (includes RF gen and SA) would also do the trick, you'll probably land on the wrong side of 1k though.

A R&S SMH only goes up to 2GH but otherwise would fit you requirements rather well? Easily below 1k.

EDIT: you being in the US probably makes all of this kinda pointless.

[DaJMasta]

I haven't used one myself, but I think the MG3700A is a pretty capable unit... since it's based off ARB generators, I don't know if it's configured to mimic AM/FM modulation, but I'd expect it to be able to produce at least something like.  The ones I've seen have a lot of sample memory for it too.

In my mind, the SMIQ03 would be the go to in this category, it tends to be a bit cheaper than the ESG-D series (though an ESG-A would do the job if you're not needing digital modulation options), and from what others have said, it's a little more capable (better harmonic supression and faster settling time, if I remember right).  You can often find them below $1000 - partly functional SMIQ06s sometimes go below that now.  Could take some time to get the options you're looking for, though, since FM is an option.

Other avenues could be the IFR/Aeroflex/Marconi pizza box style generators - nice and compact, but they tend to be a bit more expensive for the bandwidth than the SMIQs.  Then there's R&S's SML and SMV which are more expensive, but are similar form factor to the IFRs, and right now I see an SM300 listed at $1500.


So to sort of sum it up, I think the SMIQ03 will be the most affordable option for the features/performance, but it's also a bit of a boat anchor, so it won't fit on every bench.

[Joel_l]

Thanks, I'll keep looking into Anritsu as I look for a good deal on a SMIQ. The SMIQ are big, the Anritsu not much smaller, but my bench could handle either.

[Joel_l]

After finding all the info I could on the MG3700A, it's an interesting unit but really too specialized towards digital communications. You can make it do AM,FM, PM and others but you have to generate a waveform file for the specific parametric you want to test, no on the fly adjustments. The only thing I haven't been able to figure out is how to sweep frequency.

I think for my needs there are other choices that are better. I'm not in a hurry, so I can sit, watch, and wait for an SME, SMIQ, or other to show up with the options I want.

[Joel_l]

Ended up buying a ESG-D E4432B, has all the options I want. Two possible failures that I think I will be able to address. Had reasonably low hours ( compared to others I was looking at ).

Just looking at manuals between the SMIQ and ESG-D, a SMIQ with all the options I wanted would have been nice, but they were far more money.

Some things I'll just have to see if the ESG is really an issue. Things I notice are the way the ESG does sweeps is not as good as the SMIQ but should be OK.

Should have it next week, hope I can fix it reasonably quick. The failures seem to do with the UND option, so for most my needs, it's usable as is ( I think ).

[DaJMasta]

I think the settling time for the SMIQ is ahead in sweeps, but the ESG series are still capable units.  If I remember right, I think Shahriar (TheSignalPathBlog) did a repair on an ESG series, though I think a different model and failure mode, but it could be useful to get your bearings in the instrument and/or whet your appetite for using it. 

[Joel_l]

I did see that repair post, was very informative. Seems the two issues I have read about were poor solder joints and/or bad static ram. I found some ram so bought some, was cheap enough. Also seems one possible failure mode was around some SHF0189 HFET, also cheap so bought some of those.

The biggest sweep nit I have is the ESG really only does list. Can chose a start, stop, and dwell, but only 401 points between, not really a sweep if  you are spanning any range. The SMIQ lets you chose a start, stop, dwell, and step. This seems much more practical to me. The other plus for the SMIQ seems to be analog modulation ( assuming option B5 ), for example FM deviation can be set up to 1MHz worst range, ESG is 50KHz.

I don't know that these things will matter to me in practice, but spec wise, the SMIQ seems ahead of the game for the analog portion.

The ESG is leaps ahead of my Wavetek, so I will be happy.

Discaimer, all I know is from looking at manuals. I don't know if latest FW updates may have changed things to make either worse or better from the manuals.

Looks like it is scheduled for delivery on Monday.

[PA0PBZ]

The biggest sweep nit I have is the ESG really only does list. Can chose a start, stop, and dwell, but only 401 points between, not really a sweep if  you are spanning any range.

Depending on why, what and how you want to sweep it can do very wide FM very slow, think 5 MHz with 0.1Hz modulation frequency. 

[Joel_l]

My E4432B arrived and I can say I am happy with what I got.

It does have the two issues I knew about, error 517, and muli-tone problems. Both are memory related though I don't know if it's the same memory. I do have the static RAMs on hand that should fix the multi-tone errors. In my initial playing with it, I could reproduce distorted waveforms out of the generator. I was happy to find it did not have any output issues. Using my SA, output levels seem good across a wide frequency range. AM, FM, PM seem to all work OK as well.

Some things I notice that I'm not sure are an issue or user error,

When loading a waveform file, I get a file not found error. Why do I see the file in the list if it's not there to load?

At power up, I see a couple of power on self test, but then the back light goes off for a moment then comes back on to see the FW version and some other info just before being ready to run. During the time it's off, it's running some memory tests. I was able to capture it using a camera and flash. It was still hard to see but looked like some memory address ranges were spit out and number of errors found ( 0X0 ). Not sure why the backlight is turning off

The factory preset ( set to normal ) turns RF off ( expected ) and modulation on ( not expected ). - Looking at the service manual, this is the expected behavior, seems odd to me, but not a problem

I'm still going through all the docs I can find on the unit. I will replace the static RAM which I expect will address the multi-tone issue and for now see if it coincidentally fixes the error 517. I don't know what memory is involved.

Any insights?

Thanks

[Joel_l]

I replaced the SRAM and one of the issues went away. When doing a multi-tone waveform, it's no longer distorted ( garbled ). But I don't think the SG is building them right. It still might be me not knowing how everything works yet but what I'm doing seemed straight forward enough.

I build a 2 tone waveform with 20KHz spacing. What I get is three primary peaks, the outers are 20KHz apart but I get the center SG frequency.

I build a 3 tone waveform and all three primary peaks are 20KHz apart, what I would expect.

Both waveforms have artifacts in decreasing power level. Not sure what those are about. - looking more at these, I think they are harmonics of the tones.

The levels are reduced from when the modulation is turned off. I think that is a side effect of the I/Q calibration being broken ( still ). They also don't match what's in the tone table.

My next step is to take a closer look at the whole board and reflow anything that looks off.

[Joel_l]

Ugh, now that it has been on for many hours, I'm starting to get unlevel errors in various spots, mostly between 200MHz and 500 MHz. More troubleshooting to do.

Now thinking the issue was always there, just did not pay attention to this.

[G0HZU]

I build a 2 tone waveform with 20KHz spacing. What I get is three primary peaks, the outers are 20KHz apart but I get the center SG frequency.

The centre (carrier) tone implies that there is a significant (unwanted) DC term fed to the IQ modulator. Maybe this is because your IQ modulator isn't calibrated? I have a couple of these ESGD sig gens and if I replicate your two tone waveform using the 'Multi Tone' mode I get a response as the image below.

When loading a waveform file, I get a file not found error. Why do I see the file in the list if it's not there to load?

On my 4433B I have to load separate 14bit I and Q bin files to it if I want to generate an arb waveform. If I (partially) mess this up it still shows the file as being there in the sig gen but it won't run. Also, I find that the system is very fussy and unforgiving when trying to accept/load IQ files and the whole process will only work if you get everything 'right' about your file format and also I found that it only works reliably with older PCs running an old OS like Win2000 or XP. I usually use an old Win2k laptop and Agilent's own loader SW to load arb files although I did create my own IQ generation SW and loader.

My advice would be to find/borrow an old win2000 laptop and install the Agilent ESGD tools onto it. The Agilent tools appear very flakey to me when run on a modern PC with a modern OS. For example, files that upload and run correctly on the Win 2k laptop will upload OK but (sometimes) won't run properly on the sig gen when loaded from a modern PC even though it appears to load error free on the same Agilent loader GUI. Also, if you get anything wrong in terms of file format (eg Endian) the sig gen will often accept the file but you will just get wideband mush when the sig gen runs the file.

If I get time over the weekend I'll upload some valid IQ bin files that work on my E4433B so you can try them. I'll have to dig out that old laptop and find the files so this may take a while...

[Joel_l]

Thanks for checking this, what you show is what I expect. I do have an old system set up just for running this old software. Still can't do an I/Q calibration, so hunting through things.

[Joel_l]

I decided to look at the A9 RF out board today. What I find,

First, it is not like most I have seen, it does not use the H1 part in the signal path. I have the 1GM14201 amplifier. It seems to be working unless there is a failure mode that allows it to be bad in just a frequency range ( @ 250MHz to 500MHz is off ). Anywhere else I can get 13dBm. The 4432B is rated for 10dBm at 3GHz, but I seem to be able to get 12.5 reliably. Makes me think the amp is OK, but could be wrong. I need to look at the block diagrams and see if different frequency ranges have different preamp paths. When I get to the frequencies the level is bad, it hits a wall. If I have 10dBm set then switch from 240MHz to 250MHz, the output drops to 0dBm. If I set the output to 0dBm, the unlevel goes away

U772 or U771? ( maybe a small VR ) is not mounted to board but glued to the top of an IC and wired to the board pads. It looks like a factory rework, maybe an ECN to use a slightly bigger part ( deliver more current ), though not having the board for a heat sink seems bad.

Theres an LM393 that has two pins lifted and one of the inputs wire to an input on the other side. Again, looks like a factory kludge. Looking at the layout, one comparator was not used but the layout shorted both inputs and the output to ground. They left one input grounded and wired the other input to one of the other comparators inputs. So fine, no issue ( other that there are many layout errors in this system ).

I find a red LED lit on my UND board, need to see if I can find info on what all the LEDs are.

There are 5 green LEDs on the fuzzy board, two go out when the system is finished booting.

Other things I have noticed, I get an unlevel message even though the output is where it's set to ( within .2 dBm ) other frequencies with the same offset are fine.
I also turned ALC off and ran a power search on a frequency that the level was OK on and it failed. Not sure why.

This unit is turning out to be a little more work than I wanted, but I'll likely learn a few things along the way.

In anticipation of fixing an unlevel issue and thinking I would find H1s in there, I bought some. Some how doubt the are genuine, but they still might be good. I expect them here in a couple days. If you need some, let me know.

[G0HZU]

On one of my ESGD sig gens I found I had to remove the hardware (cards) for one of the options to get the IQ calibration to run reliably. I think it was the real time mod option. It caused unreliable IQ calibration and removing the two cards for this option cured this problem instantly and reliably. Before this, the sig gen would often fail the IQ calibration routine.

I suspect I need to upgrade (or downgrade?) the firmware in order to get this option to work. But I don't need it so I just removed the cards. What was a bit disappointing was the amount of 'mod' bodgery on these cards by Agilent/Keysight. If you look closely at your RF output board you may find quite a few SMD parts piggy soldered in place by Agilent. The real time cards had loads of wobbly tacked on wires and SMD parts and it doesn't inspire a lot of confidence with respect to long term reliability.

[G0HZU]

My memory is fuzzy on this stuff now but when I created my IQ editing SW a few years ago I ended up monitoring/hacking Agilent's own SW to see how it sent data to the sig gen. I recall that their SW doesn't follow their own documented/published rules for the header info for this sig gen. Once I copied their 'working' method I got it to work and I ended up ignoring some of the official info in their ESGD programming documents.

See the youtube link below where I produced some SW that imports grabbed 16 bit IQ data from their 89601 software (and an Agilent E4406A analyser) and this allows the IQ data to be viewed and cropped and sig gen 'trigger event' markers can be added at critical parts of the waveform. It then converts it to an IQ file format that the E4433B accepts and then sends it to the sig gen. This allows offair signals to be grabbed by the E4406A analyser (eg from a wireless device) and then it can be loaded into the sig gen to replicate the modulated offair signal of the wireless device.

I remember having a lot of problems getting the sig gen to accept the IQ data files. I think there were issues with the marker bits and certain parts of the file format and I had to fudge the file data to make it 'valid'.

However, if you use their old loader program on an old PC running Win2k your should be OK to upload basic IQ files. But if you try and create your own loader program you may find the same issues as me?

[Joel_l]

I can't even think about how to transfer files yet. Seems if I can get the problem between 250MHz and 500MHz resolved, I can make other progress.

I did try an I/Q cal with all the cards removed. It actually passed on one of my user cals where I limited the range to 250KHz to 1 MHz. It would not do that before. I think the 250MHz mark is a wall because the output drops. I did play with power search at 250MHz, if I set the output to be at the level the signal dropped to, power search ran OK. I think nudged it up half a dBm at a time, I was able to get 1 - 2 dBm more before power search failed.

Today I'm going to probe up the output board and see if the drop in on it or coming from the synthesizer, then continue to trace base on results.

[G0HZU]

It might be best to draw out the frequency plan for the sig gen and then list where it fails to provide enough level.

It's a while since I studied the ESGD but I think the core of the sig gen is an octave tuning VCO that covers 500-1000MHz.

From this it is possible to produce any frequency from 250MHz to 4000MHz using a divide/2 (to get 250-500MHz).
500-1000MHz is covered by the raw VCO itself and 1000-2000MHz comes from a doubler and 2000-4000MHz (for the E4433B) comes from another doubler. All this happens on the synthesiser section(s).

0.1-250MHz is achieved with a down mix on the RF board.  i.e. I think it mixes 750-999.9MHz with a fixed 1000MHz LO to get 0.1-250MHz.

So if your unit is misbehaving across 250-500MHz you might want to look further back into the synthesiser stage at the div/2 section after the main VCO/PLL. Does the synth stage produce a consistent RF level across 250-3000MHz for your sig gen? Does it drop in level across 250-500MHz here? There are switched filters in the synthesiser sections that could also give problems.

[G0HZU]

What I can't see from this is why it all works again above 500MHz.

At a sig gen freq of 245MHz the synth feeds A9 with 755MHz and this goes through into A9 (eg through into the IQ modulator at 755MHz) and it only becomes 245MHz after it is mixed with 1000MHz just before the big final amplifier in A9.

When you select 255MHz, the VCO on the synth board runs at 510MHz and this gets divided by two to get 255MHz and this 255MHz gets filtered and fed to A9. I get the impression this gives a poor level.

When you select 490MHz, the VCO on the synth board runs at 980MHz and this gets divided by two to get 490MHz and this 490MHz gets filtered and fed to A9. I get the impression this gives a poor level.


However, when you select 510MHz, the VCO on the synth board runs at 510MHz and gets fed to/through A9 without any division or mixing. The difference here is that no division is needed even though the frequency is only 20MHz different from 490MHz.

So does the sig gen give a good level at 510MHz but a poor level at 490MHz? Both of these are straight through to the final amp on A9 but this is a powerful test because these two frequencies are so close together and they both go through the whole of A9 the same way. But they are generated in a different way back at the synth module. If there is a big difference then I'd assume the problem across 250-500MHz was back at the synth module.


If it gradually improves in level across 250-500MHz (but still is a bit low even by 500MHz) then this might be caused by a cracked (series DC blocking) SMD cap somewhere in the signal path in the synth or the early part of A9. Or maybe just a poor connection somewhere. Maybe even a loose RF cable connection assuming there is an RF cable connection between these modules.
 

[Joel_l]

yes, but what I think I saw was 255 coming into the A9 board at the right level, it was far down stream of J4 that the level dropped. I saw from an old thread of nctnico where the path was drawn out and I can see where the <250MHz path is and the signal is mixed with 1GHz and the difference comes out. I wonder what is happening through what ever switches the path.

I am going to go back and remeasure what is coming into the A9 board. If >= 250MHz is indeed lower in amplitude, then yes, I need to look at the synthesizer board.

Thanks!

[G0HZU]

yes, but what I think I saw was 255 coming into the A9 board at the right level, it was far down stream of J4 that the level dropped.

In that case, could there be an issue with the gain control tables? There might be corrupted data across the 250-500MHz range and A9 is (wrongly) attenuating the signal gain across this range? Can you monitor the gain DAC on A9 to see if it does anything strange across this range? I'd expect this to be fairly unlikely to be the problem but worth a look in case someone has tried and failed to calibrate this sig gen and they have messed up the tables. But I don't know how the calibration tables are arranged so there might not even be a 250-500MHz correction for this VCO div/2 range of 250-500MHz.

[Joel_l]

I had to do a double take. So I'm not sure what to make of this, opposite of what I was expecting if the issue was the synth board. 245/755MHz is -52 dBM, 255MHz is - 33 dBm. I expect a little more loss at 755MHz, but not that much. So I guess now I will measure with the SA at the final amp input.

[G0HZU]

What are you using as an RF probe?

I would recommend you use a small leaded (metal film) 470R series resistor into some skinny 50R coax and then feed the other end of the coax into the spectrum analyser. The analyser should hopefully provide your DC blocking but you could add a DC block inline to be safe. Use the free end of the 470R resistor as the probe tip. Use tight connections and a short ground connection to the PCB.

This probe will act as a reasonably high impedance -20dB probe for through signals in a 50R system up to maybe 1 or 2GHz if you make it neat and tight. So if there was 0dBm at the PCB then the analyser would show about -20dBm when probed.

[Joel_l]

I look at the input  to the amp, and now 255MHz is much lower. Somewhere 255MHz is getting attenuated on the RF out board. I need to find where the path gets switched.

[Joel_l]

I used 50 Ohm coax right on my probe points. I would think for relative measurements, good enough.

I also think I found the switches, it looks like they are using pin diodes to steer the signals on the board.

[Joel_l]

Looking at the block diagram and the PCB, I think the red circle on the on the block diagram is the red, blue, yellow circles on the PCB. I removed those components and just shorted pads.I was able to steer the <250MHz and >=250MHz. A part marked G41 feeds into final amp, the other two are marked G2V. I think they are all PIN diodes. At this end of the circuit, shorting the >=250MHz side ( red circle on PCB ) into the final amp did not change anything output wise, still get the unlevel error.

On the block diagram, I expect the blue circle just brings the signal level to what is was through the yellow circle.

What I'd like to check next is if the level coming out of the burst mode is the same as it is at the final amp. I also don't know what else might really be between the two switches.

I may make a better probe as suggested to minimize pulling down the signal. Drive capability might be different along the path.

Time to get more methodical. 

[Joel_l]

After many years you would think I would have learned that when it comes to RF you waste less time being methodical from the beginning rather than trying to save time doing things quick.

Anyway, I made a better probe and now things look more sane and consistent.

I have two shots at J4 showing that the 255MHz path is slightly lower than the 245MHz path. I picked off a few more points into the signal path and the results are consistent. As I go through more stages, the gap gets wider, but they seem reasonable.

I also learned some switching goes on in the synth board. For example, if my first measurment is at 245MHz and I then set 255MHz, I had to turn the RF on to get output from the synth at 255MHz.

So, I think the RF out board is mostly OK ( there is still the multi tone issue to address and not sure where that will be ).

Here are pics at J4 and "A" is I think the input to the burst modulator.



Now I'm off to the synth board.

Thanks for the help so far!

[Joel_l]

OK, I may have found something.

I was measuring around the divide by 2 circuit. Looking at the output of the synth, things looks pretty good. From 500MHz to 1GHz all within 2dBm ( lower as you get higher, seems reasonable ).

The output however is another story. Say I pick 300MHz, the output is about -20dBm with my probe. If I set 600MHz ( same synth output ), I see -40dBm. If I tune the SA back to 300MHz, I see -20dBm. Seems wrong.

So I took ( what I still think are PIN diodes )  out for both paths and shorted the non divided path. I got nothing at 300MHZ as expected, but still only -40dBm at 600MHz.

I noticed that the series resisor right before the PIN diode seems high in resistance ( red circle ). I measure 38KOhm. Everywhere else they are in the 10s of Ohms.

Does 38K seem good? the only other thing I can think of are some amps/buffers? marked A03.

Any thoughts?

I checked the level before the resistor and it is normal. I need to figure out what component an A03 is.

I figured ot the A03 to be a VAM-03. I replaced the resistor that measured 38K and replaced it with a 48Ohm ( closest I had ). the output of the divider is now good. I see about -16dBm at 600MHz. The final output is still not right.

Thanks

[Joel_l]

Well, the SG is now 99% functional. Outputs are all OK, it passes it's I/Q qual through the full range.

I measured all the outputs per the service manual an found some were off. To this point I was only looking at the top of the board, I finally pulled it from it's lower tray and found more caps in the signal path for the filters, touched all those up and the SG started working normally.

Can't say I really did anything purposeful to fix it, but learned a lot about how it works along the way.

@G0HZU - Your thoughts were pretty much right on, thanks for the insights!!

I still need to figure something out with the multi-tone, though now it's much better, still not 100% right, seems the CW freq is still to high, but at least somewhat functioning.

The only other thing I need to do is put that 38K resistor back in the divider circuit and see if things break again.

I put a 38K back it, it broke everything. So I will install a 17.8 Ohm which seems to be what I measure for other resistors in similar circuits on the board. I also noticed that the smaller the resistor, the smaller the unwanted signals when doing multi-tone.

I put a 17.8 Ohm in and things looks much better now. Calling the SG done. Back to being happy with what I got and in the end was not as bad to fix as it could have been. If I would have been more methodical and thought about what I was doing from the start, would have probably gone faster.

It seems there are at least two versions, the older HP that uses the SHF0189 ( H1 ) parts and a newer Agilent version that uses Agilent ICs in many spots. If an Agilent version has a bad Agilent IC, fixing might be an issue, though some here have been very creative in getting theirs working.