OM345 module swapout in my Marconi 2022C signal generator

[G0HZU]

I've owned a Marconi 2022C sig gen for a few years now and the previous owner replaced some of the failed OM345 modules with MMIC modules that were presumably purchased on Ebay. The sig gen has worked OK with these MMICs although the harmonic distortion was a bit higher than normal.

I originally planned to replace these Ebay MMICs with original OM345s but these are difficult to find and are often very expensive. So a few years ago I designed a clone of the original OM345 using a suitable RF BJT. I modelled the original OM345 using Sonnet EM and Genesys and designed a small PCB about the same size as the original OM345. I actually milled the PCBs a long time ago but never got around to building any prototypes. I've finally got around to building and testing some. The aim was to build something that performs better than the various MMIC alternatives available on Ebay.

Note that I can't simply do the 4k7 (or piggy 3k9) mod on this sig gen because quite a few of the OM345s have been swapped for some dodgy MMIC alternatives by the previous owner. So I either have to find some OM345s or fit my own alternative design.

My first attempt at a OM345 clone seems to have very similar performance to the original OM345. The large signal performance is the same (as expected) and the gain and reverse isolation is very similar. When measured in a 50R system, the noise figure of the clone is slightly lower at about 3dB (vs about 4dB for the original OM345).

Note that the OM345 circuit is a fair bit more complex that the simple circuit given in the OM345 datasheet. To get close to the original performance in terms of frequency response and I/O matching I've had to use about a dozen passive SMD components and these are crammed onto a tiny PCB about 8mm x 13mm in size. The model I've used includes the PCB layout and I've used decent models for all the passive parts and I've used the manufacturer's s-parameter data for the BJT. The model seems to agree very well with the results I get from my first built prototype.

[G0HZU]

I've measured a few OM345s on my VNA and compared the performance to the model and my prototype clone as in the image below. You can see that the gain and reverse isolation is in good agreement for the OM345 vs the model and the fully built clone. There's not much point trying to get a closer fit to the gain response as I noted a similar spread in gain vs frequency between several samples of a genuine OM345. So I think I'm close enough here.

There's also a plot of the input and output impedance of the OM345 vs the clone and the agreement is quite good for a first attempt. It's very difficult to accurately replicate the RF characteristics of a thin film module like the OM345 when only using SMD parts. The OM345 has lots of complicated RF traces and tiny shapes on it and these will have distributed properties that can't easily be replicated with a simple PCB like mine. I don't think I'm going to bother trying to improve this. It's close enough (in my opinion) to fit to my 2022C as it is. I've built three so far and they are all the same. I've got to build another five in order to be able to fill all the vacant OM345 slots in my 2022C. Luckily, I milled ten boards.

I've now removed all the OM345 modules and MMICs from my Marconi 2022C and plan to replace them all with my new clone board. I'll post up a few more performance plots in future posts and I'll also show the performance of the MMICs that were fitted to my 2022C. The gain vs frequency was quite different to the OM345 (several dB too much gain across VHF and into UHF) and the large signal handling was notably worse than the original OM345.

I also noted some differences between the OM345s that I extracted from the 2022C and I think one of them was close to expiring as it was only drawing about 8mA instead of the expected 11-12mA. However, the others all had similar gain and reverse isolation to my plot below.

[G0HZU]

Here's the gain vs frequency for two of the ebay MMICs that were in my 2022C. There were four of them fitted but I've only measured two of them.

Both looked very similar on the outside. They were both sealed with pale grey resin. However, the RF performance was quite different between them. One of them had a blue dot on it and one had an orange dot. So maybe they are different versions?

Both were presumably purchased as OM345 replacements on Ebay?

You can see that the gain vs frequency is very different when compared to a genuine OM345. One of them has far too much gain at VHF compared to a genuine OM345. The input and output matching is quite different too. Both also have inferior large signal performance compared to the original OM345 despite them drawing 15mA each compared to about 11mA for a genuine OM345.
Note that I've compared against a different genuine OM345 for this plot. If you look closely, you can see that the gain response of this OM345 is slightly different to the one in my earlier plot. That's why there's not much point me trying to get the gain response of my clone to perfectly match a single example of an OM345.

I suspect that one of these Ebay MMICs is a Minicircuits MAR-1 (or equivalent) and the other one is an MAR-7 (or equivalent).

I'm now quite pleased that I've finally removed all these dodgy devices from my signal generator!

[G0HZU]

I've built up a little MAR-7SM PCB to mimic the various MAR-7 MMICs that are often used in place of an OM345. In the plot below I've compared the gain and reverse isolation against a genuine OM345.

I've biased the MAR-7 at 12mA and also at 16mA and there are plots for both cases in the graphs below. Also included is the predicted response of my board if I use the Minicircuits factory model data for the MAR-7 at 16mA and I include the loading effect of the bias resistor and also the PCB microstrip in the overall model of the complete test board. This is the green trace and you can see it agrees very closely with my pink trace of the actual MAR-7 when biased at 16mA in my test board. This should give some confidence in my measurement methods using the VNA as it agrees with the gain response when using the factory data for the MAR-7. I've not modelled the SMA connector losses in the model.

You can see that the best agreement for gain occurs if the MAR-7 is biased at just 12mA and the gain response is quite similar to that of the OM345. However, the reverse isolation degrades and it is about 4.4dB worse than the reverse isolation of a genuine OM345 in this respect. Also, the large signal handling will be quite poor when the MAR-7 is biased at only 12mA.

Note that all of the MAR-7 plots include the shunting/loading effect of having a bias resistor of about 600R with no bias choke. Some Ebay MMICs include a bias choke and some don't.

Tomorrow, I'll post up some plots of what happens (in terms of distortion) when an AM modulated signal is fed through the OM345 and also through the MAR-7. I think IC14 in the 2022C is an OM345 that has to cope with AM signals. I hope this stuff is useful and/or interesting

[djsb]

Can this be used in a 2022E? Will you be sharing the PCB design (KiCad or Altium Designer?) and a BOM for home construction? Thanks.

[G0HZU]

Hi
Sadly, I'm not really planning to release a data pack on this design. I did it all a long time ago and only got around to building the modules in the last couple of days.
My main aim for posting stuff here is to show the risk of using ebay (MMIC) alternatives to the original OM345.

I think it's generally easier and safer to either do the external 4k7 axial resistor mod or the 'scrape and fit' 3k9 SMD resistor mod to revive a faulty OM345. I can post up plots that show the difference between the two ways to revive the OM345. The external 4k7 resistor does introduce some negative (RF) feedback and this causes a slight reduction (0.5dB?) in gain and a slight reduction in reverse isolation (0.5dB?) compared to the 'scrape and fit' mod using the SMD 3k9 resistor.

See the plot below to see what I mean. Up at UHF the feedback via the 4k7 resistor is different due to phase shifting so the gain reduction is less up at UHF. The input and output impedance is about the same in both cases.

To make this comparison I used a genuine OM345 that had failed in the usual way. I compared the performance when it was repaired using the external 4k7 resistor vs the 'scrape and fit' 3k9 mod.

Fortunately, the small difference in gain should not affect the amplitude calibration of the 2022 sig gen because the ALC system will auto correct for the gain difference and it will therefore maintain the correct calibrated output level. I hope these plots are useful

[G0HZU]

I've just done some large signal comparisons between the OM345 vs my clone vs a MAR7 and also the dodgy grey Ebay MMIC.

I now think that the grey MMIC is really something like a MAR-1 with some loss added after it to reduce the gain. This would explain the awful large signal performance below.

Here's the output 1dB compression point of the various modules at 240MHz:

Genuine OM345  +3.6dBm
My Clone            +4.0dBm
MAR-7 12mA       -2.3dBm
MAR-7 16mA       +2.1dBm
Grey MMIC          -5.0dBm

I also measured the AM performance and the harmonic distortion performance.
The genuine OM345 and my clone performed equally well and were best for both AM and harmonic distortion.

The MAR-7 was slightly worse when biased at 16mA and it was really poor when biased at just 12mA.

The Grey Ebay MMIC was worst of all. It generated lots of AM distortion and harmonic distortion. I'm hoping the harmonic distortion will be improved in my 2022C when I fit my clone parts. The clone really is just an attempt to copy the original OM345 circuit but I've used modern SMD parts instead of thin film technology.

[rainbow]

Very interesting write up. Thank you very much!
Do you also have the possibility to test the MAR-3SM+? It has higher input current but the circuit of the 2022 can support it.
Anyway, any MAR* has worst reverse isolation (~ -3dB) and this can be an issue in the OM345 that buffers the 160MHz where the AM modulator can cause frequency deviation in the VCO.

[G0HZU]

Very interesting write up. Thank you very much!
Do you also have the possibility to test the MAR-3SM+? It has higher input current but the circuit of the 2022 can support it.
Anyway, any MAR* has worst reverse isolation (~ -3dB) and this can be an issue in the OM345 that buffers the 160MHz where the AM modulator can cause frequency deviation in the VCO.

Hi, yes I agree that there is a potential issue with reverse isolation for IC1 ahead of the 160MHz AM modulator. I mentioned about it here a while back in post #42:

https://www.eevblog.com/forum/testgear/marconi-2022e-repair-and-refurb/25/

I do have a couple of boards I milled for a high reverse isolation buffer to replace IC1. Again, I never got around to building them. I've found the boards and could try fitting one to my 2022C at some point. This buffer has about 50dB reverse isolation as opposed to about 21dB for the OM345 and about 18.5dB for the MAR-7.

I do have sample bags of most of the common MMICs here including the MAR-3SM (as below) but I won't be able to test one for a while. I could swap out the MAR-7 for an MAR-3 but I want to keep on testing the MAR-7 for now. So far, the MAR-7 looks to be a reasonable replacement for the OM345 but it doesn't have the same large signal performance or the same reverse isolation as the original OM345. The gain response isn't quite the same either. I'm not yet sure if this matters much for a 2022C.

[rainbow]

ok. just because MAR3 should be a bit better in gain linearity e reverse isolation than MAR7

[G0HZU]

I've built up all eight OM345 modules and tested them. The spread in gain between them is really small. I have used tight tolerance parts and the results look good so far. See the plot below. It shows the gain response of all 8 prototype clones of the OM345.

I've also built up the high isolation version and I've tested that too. It currently runs at about 22mA and hopefully this can be reduced. However, it managed 42dB reverse isolation at 160MHz and 35dB at 500MHz. Not quite the 50dB I was expecting at 160MHz (based on the model data) but good enough I think.

The gain was 12.2dB at 160MHz and 11.5dB at 500MHz. So it could possibly also be used at IC7?

The noise figure is a flat 5dB across VHF into UHF and this is about 1dB worse than a genuine OM345. However, I don't think this matters for the IC1 location. The signal handling is much better than a genuine OM345 but I think I'll leave it at 22mA for now.

The noise figure of the clone is about 3dB and this is slightly better than a genuine OM345 in a 50R system.

[807]

Can we have a close-up photo of your modules?

What transistor did you use?

[G0HZU]

It's all back together and I've been testing it today. There is a really big improvement in the harmonic distortion levels (on the >62.5MHz ranges) now that the grey Ebay MMICs are gone.

I'm not sure there's much difference otherwise. I also replaced the electrolytic caps in the VCO area and reversed some of a previous owners mods to the RF board on the BFO range. The harmonic distortion on the <62.5MHz range also seems improved, but this might also be due to me removing some 'mods'.

What transistor did you use?

I'm fairly certain the original BJT inside a genuine OM345 is a Philips BFR92 with a P1 logo on the SOT-23 package. I've carefully tried to remove the coating on a couple of OM345s to try and establish this. Sadly, the Philips logo/label on the BJT is formed from a really soft coating so it is easily smudged. However, I can see a fairly clear P1 on both. I think other people have done similar investigations.

I used a new Infineon BFR92P as the Philips part is long obsolete. I think the newer part is the reason I get a slightly improved noise figure. Presumably the die processing is better on modern parts. The gain is also slightly higher above 860MHz with the new part. I tried fitting the BJT from an OM345 to my clone and the gain was about 0.5dB lower up at 1GHz for example. Otherwise the response was the same across VHF and into UHF.

[G0HZU]

It's difficult for me to demo the difference in harmonic distortion now that I've changed across to the cloned OM345 in all 8 locations in the 2022C. However, one OM345 that will definitely influence harmonics is IC14. This is part of the three stage final amplifier for signals >62.5MHz.

If I crudely look at the cathode of PIN diode D41 with a zero loss probe, this corresponds to the RF input of IC14.
With the 2022C set to 90MHz and -40dBm, I can then increase the level up and down in 1dB steps and see this level change within a window of about -22dBm down to -34dBm.
This won't drive the OM345 very hard so it won't generate much in the way of harmonic distortion as the output would presumably be -22dBm + the 12dB gain of the OM345 = -10dBm.

However, if I keep on increasing the level in 1dB steps right up to the +13dBm limit of the sig gen, the input level to this OM345 increases to about -12dBm. This suggests that the output power from this OM345 could be about 0dBm maximum.

This is about 4dB below P1dB for a genuine OM345. My 2022C shows about -36dBc harmonic distortion with an output frequency of 90MHz at the full +13dBm from the generator.

A MAR-7 biased at 16mA (with a 600R bias resistor shunting the output) has a P1dB of about +2.1dBm.
A MAR-7 biased at 12mA (with a 600R bias resistor shunting the output) has a P1dB of about -2.3dBm.
The grey MMIC had a P1dB of -5dBm.

Luckily for me, IC14 still had a genuine OM345 fitted. Therefore, I don't know how bad it would have been if a grey MMIC had been fitted at IC14.

I'd expect the harmonic distortion to be slightly worse than standard if an MAR-7 was fitted (with a 600R bias resistor shunting the output) and biased at 16mA. It would be notably worse if the MAR-7 was biased at just 12mA.

From what I can see on the internet, many MAR-7 modules (designed as OM345 replacements) are biased at about 13.5mA.

So it would be interesting to see what the harmonic distortion would be like at 90MHz at 13dBm output with an MAR-7 fitted at IC14. IC14 also has to cope with AM modulation, so there is the risk of slightly increased AM distortion with an MAR-7 biased at just 13.5mA.

I'll try and do some tests at these power levels for AM and harmonic distortion for a genuine OM345 and a MAR-7.

[G0HZU]

Here's some plots of harmonic distortion at 150MHz at 0dBm output power for 'just' the OM345 on its own and this is compared to the alternative devices.

You can see that the genuine OM345 (at 12V 11.5mA) generated distortion at about -32dBc.

The MAR-7 generated distortion at about -22dBc when only biased at 13.5mA. This is with a 600R bias resistor and no bias choke.
The MAR-7 generated distortion at about -29dBc when biased at 16mA. This is with a 600R bias resistor and no bias choke.
The clone managed about -35dBc when biased at 12mA. This improved to -43dBc when biased at 16mA.

The grey MMIC is also plotted and this device struggled to produce -1.8dBm even though it was biased at 15mA. The harmonic distortion is really poor.

It looks like the MAR-7 needs to be biased at 16-17mA to match the OM345 in this test. The MAR-7 could be biased more softly in some other locations in the 2022C and this would minimise the total current requirement for eight MAR-7 devices if all eight were replaced in the 2022C.

 If anything, this test demonstrates the risk of trusting sellers who offer OM345 alternatives on Ebay. The grey coated MMIC performed really badly in this test. It would probably be OK in some other locations in the 2022C but not at the IC14 location.

[G0HZU]

I also measured the noise figure of the OM345 across 40-1000MHz and compared it to the clone and also to an MAR-7 MMIC biased at 13.5mA.

There wasn't much in it. The noise figure for the various genuine OM345s I have here varies from about 4.0dB to 4.5dB from device to device and it is quite flat across the full frequency range. The MAR-7SM measured about 3.2dB noise figure across the frequency range when biased at 13.5mA. This agrees quite well with the MAR-7SM datasheet at 16mA bias.

My OM345 clone showed a flat noise figure of about 2.9dB to 3.0dB across the full frequency range.

When used in a Marconi 2022C these tiny differences in noise figure between the devices don't really matter. Both the MAR-7 and the clone beat the OM345 noise figure but this won't really offer any advantage when used in the 2022C sig gen.

I'm lucky to have some really nice noise figure testing gear here. I use a fairly modern Agilent spectrum analyser with an external low noise preamp and an Agilent 346A noise source. The external preamp is better than the preamp in the analyser in terms of noise figure and port VSWR and this setup works really well. It should easily outperform the classic old 8970B noise figure meter for example.

[G0HZU]

One minor niggle with my 2022C is that the internal 1kHz AF source has some distortion on it. See the analyser plot below.
This is at about -50dBc which should be fine for most users. However, the FM modulator in the 2022C can be quite good and so I think it deserves a cleaner AF source.

One way around this is to use external modulation. Another option is to add an extra stage of AF filtering to the internal AF oscillator. The internal oscillator is derived from the 10MHz OCXO by using division and filtering to clean up a 1kHz square wave. The filtering is done with an active LPF using a dual op amp.

By adding an extra active LPF this can be cleaned up to get better than -75dBc distortion. This is a bit OTT but I think that the -50dBc distortion from the internal AF source is a bit disappointing.

There's a plot of the response of the extra LPF below. it would just require a tiny piggy PCB to accommodate the extra dual opamp. The next time I take the 2022C apart I'm going to try and fit this LPF. I think i will also try the high reverse isolation buffer amp in place of IC1 and IC7 to see if it reduces the unwanted FM that occurs on top of the AM modulation from the 2022C.