Quick repair project - Marconi 2024

[peter-h]

I will machine some up from ~ 3mm aluminium. I didn't realise they were plastic. On a ~ $8000 (new price) box that's not good.

[G0HZU]

Yes, they are just cheap plastic handles. Not good at all.

I did a quick google and found an image of a broken one. This is what they usually look like after someone has either dropped it on the handle or the handle snapped as they were carrying it.

[G0HZU]

They are very good generators though... The frequency plan inside lends itself to achieving very low phase noise over most of the generator's range. This is because of the multi VCO + divider+ multiplier system they use. I've measured the phase noise on quite a few of the ones at work and they are always very good on the lower ranges.

I have one myself here at home with option 4 and 11 fitted and I've done quite a few performance tests on it. The quality of the internal modulation is excellent. Much, much lower distortion than the spec sheet suggests.

The user interface can be a bit fiddly and annoying but the performance makes up for this. The only thing that really bugs me about this one is the fan noise is quite annoying in a quiet room. I'm not sure if it has a beefier fan for the options fitted as I don't notice the fan on the ones at work. But they don't have these options fitted. it only had 120hours' use on it when I bought it and there was no dust inside so I don't think the fan is noisy due to wear.

[peter-h]

I have looked for one with Option 4 but (here in the UK) they cost about 70% more.

Option 4 brings it up to the stability spec of the 2030 which is a much bigger unit, albeit oddly enough a lot cheaper when looking on Ebay.

I don't actually need the extra spec though.

IFR make a lot of avionics test gear and you need good quality modulation for that - e.g. the 90/150Hz ILS modulation needs to be closely depth-matched, to something like 1%. And Marconi before them did a lot of that stuff too. So maybe they designed it with that in mind. Certainly the 2030 is heavily pushed by their marketing stuff into that market.

Noisy fans are a real hassle and on non-warranty units I tend to replace them with SAN ACE fans, 12V ones powered from 5V Very silent 

[G0HZU]

I might see if it's possible to fit an alternative fan
Here's a plot of the phase noise of one of the units at work. It had an odd blip in the phase noise at a couple of kHz and the E5052A found it.

If you ignore the strange blip, the phase noise at 28MHz is pretty decent for a mid range generator. I have a library of phase noise plots at work of all the various generator models we have. They all get measured on the E5052A at about 12 points in their frequency coverage and it is a useful reference.

I've added another plot of the phase noise at the top frequency 2400MHz and the little blip is still there. When I took these plots a year or so ago the blip was only there on this particular example.

[grumpydoc]

I have looked for one with Option 4 but (here in the UK) they cost about 70% more.

Option 4 brings it up to the stability spec of the 2030 which is a much bigger unit, albeit oddly enough a lot cheaper when looking on Ebay.

I got lucky and mine has the OCXO. It would probably be possible to fit an ovened oscillator fairly easily - the one in mine is an NDK unit. I thought  someone had posted a llink to the datasheet but can't find it now.

[peter-h]

Is the oven the only difference?

They take quite a while to warm up and you get loads of drift during that time. Many years ago I was designing precision HV power supplies and we were aiming to get the reference to a few ppm per degC. I don't recall the device we used but it was a TO5 case and was in a 90 degC oven which used a lot of power. The oven heater was implemented as a PTC thermistor so it warmed up and just sat there. Now you can get 1ppm/degC for a fiver, in a SO-8

[grumpydoc]

Is the oven the only difference?

As far as I know.

The one in mine is an NDK END3032A (see photos earlier in the thread). If anyone does have adatasheet it would be handy.

There's one for sale on the dutch ebay.

[peter-h]

There is a pile of 10MHz OXCOs on Ebay - all relatively dirt cheap e.g. item # 300861292471.

Or, working on the basis of going for the biggest and most expensive, if you wanted one which has already had 18 years of ageing you could got for item # 111271809705

Would the Dutch one actually fit directly into a 2024, with no config changes etc?

[grumpydoc]

Would the Dutch one actually fit directly into a 2024, with no config changes etc?

I don't have access to ebay at work but while having another google trawl last night to see if I could find a datasheet for the NDK OXCO found an auction for one attached to a PCB (possibly from a 2026). As it's the same OCXO as in my 2024 it should drop straight in**.

Obviously you would need to re-run the OCXO cal because the control voltage will need to change.

** OK you need to remove the CPU board to take out the old TCXO and solder in the new OCXO but you know what I mean

[peter-h]

I bought that Ebay OCXO...

I will keep it in case it is needed later.

I also collected the 2024 today. A very nice piece of gear.

[G0HZU]

Some time ago I logged the warmup profile of my 2024 with the HS OCXO option (it has options 4 and 11)


It takes about 5 minutes to get to a decent level of accuracy. Not bad and much quicker than some OCXOs.
I've attached the raw text file below.

The first column is time in seconds and the second is frequency measured by an Anritsu frequency counter and logged onto a netbook PC via USB/GPIB.

[G0HZU]

I also logged the amplitude flatness via GPIB with two power meters I have here.

One is an old HP431C/HP478A thermistor type power meter from about 50 years ago and the other is a more modern Anritsu thermocouple type.
The meters agree quite well and the ripple is definitely in the 2024. Not a particularly spectacular performance for flatness but more than acceptable

Note: the zigzaggy fuzzy noise on the traces isn't noise. It's the 2024. It has a regular fine sawtooth response in the level of about 0.02dB. The old analogue HP431C meter picks this up really well but the Anritsu can only give 0.01dB resolution over its GPIB interface so it looks grainier.

[peter-h]

It must be extremely difficult to achieve flat output power (voltage into 50 ohms) over that frequency range. They use every trick in the book to avoid using a large number of reed relays and are having to linearise various bits of the attenuator. Somebody must have spent ages getting that circuitry to work right.

[G0HZU]

I have to disagree a bit on the levelling performance I would class the levelling performance as more than adequate for most users but I think it lets the rest of the generator down a bit.

I think the levelling circuit is based on a low cost, low parts count peak detector + ALC system and the ripply performance in my tests shows that this generator isn't really lab grade in this respect. Maybe mine's a bad example but the detector/levelling circuit looks pretty basic to me in the  manual and there aren't many calibration points to make up for its detector and attenuator/path flatness. I expected slightly better performance from Marconi here, especially in the 10-800MHz region.

What is amazing to me is the performance of the old HP431C power meter. I have had this old power meter since the mid 1990s and I have never adjusted/calibrated it in 15-20 years. Unlike newer thermocouple meters it's a fully closed loop system and it just seems to stay accurate year after year

[peter-h]

Some of the old kit was very good.

Yesterday, when I collected the 2024 and an Artitsu 2661 spectrum analyser, I tested a 1980s HP DVM (6.5 digit - bought it for £100 in 1998) against a £50k Fluke calibrator they have. The DVM was under-reading by 0.006% (60ppm).

Funnily enough I also tested a 5V reference I built with a Maxim £5 "5V 0.02%" voltage reference chip. That was within 0.001% of the Fluke calibrator.

I wonder why Marconi didn't use a simple resistive attenuator, with 20p MOSFETs doing the resistor switching? Maybe the parasitic capacitances at GHz frequencies make that a problem. Nowadays, this
http://www.psemi.com/content/products/productSubCat.php?cat=Digital+Step+Attenuators&product=50+Ohm
would probably be an obvious way to do it.

[G0HZU]

I've only skimmed the manual for the 2024 (and I've never lifted the lid on one) but I think the strategy for the levelling is to use an ALC system and a peak detector. Then they use a series of switched resistive attenuators.

My guess is they use calibration lookup tables of frequency vs attenuation vs correction to provide ALC correction factors for each combination of attenuation for each frequency. That way they can use basic/cheaper parts and still get fairly accurate performance.

It looks like the switched attenuators use lots of resistors per section and they use a restricted range of resistances in the sections. eg the values appear to be in a narrow range and they get stacked in series and parallel as building blocks of resistance. Presumably these values and arrangements give the flattest response over LF to 2400MHz and most accurate 50R source impedance.

On my high power version of the 2024 the performance of the levelling isn't very good at high power levels and I only really trust the levels below about +13dBm. It can go up to +25dBm but I've limited mine to +20dBm in the menus as +25dBm is enough to damage some of my other test gear.

Some of the newer Agilent sig gens do offer solid state attenuators as an option... I think the idea is you can use the sig gen all day every day in an ATE rack and have many millions of attenuator setting changes without wearing out any relays. I guess the RF accuracy will come from clever correction vs level vs frequency lookup tables in the sig gen memory?

I've used those Peregrine RF attenuators in the stuff I design at work (also Skyworks and Hittite) and the performance is pretty good. However, they aren't classed as precision devices in terms of VSWR or accuracy and they are damage prone if hit with big signals. Plus they tend to suffer a bit with reduced signal linearity down at very low frequencies. But it's amazing how small these devices are. Some of the Hittite stuff is really tiny.

[peter-h]

I won that OXCO board on Ebay - here is a close-up

[danmcb]

this is a great repair!I have a different piece of test gear (Audio Precision test set) that has the exact same problem - horizontal lines. I am going to see if I can do the same thing (as the TMF2010 display is about 200$ these days).

[VK5RC]

Thanks to Grumpydoc for inspiration. found a "unique do-a-up opportunity (Marconi 2024 with buggered screen) for not crazy money with good output. Managed to find the LCD used and 30mins later, hardest part was soldering on some new 0.1" header strip.
Viola. Thanks Grumpydoc et al
2 before photos 1 after, I haven't yet attempted repair of the crack in the perspex as it is off to one side and doesn't worry me (yet)

[CJay]

I won that OXCO board on Ebay - here is a close-up

Ah, those are the boards out of the Aeroflex Cellphone tester boxes, headless things that need a PC to run.

Did you find a datasheet on it?

[chicken]

I hope it's not bad form to revive this old thread to add the repair log for my own Marconi 2024.

After remodeling my lab, I finally came around to take a closer look at the Marconi 2024 I bought late last year (thanks CatalinaWOW). It has four issues:

• 1 blank pixel line on the display
• The fan makes a racket after running for a while
• Error 508 Amplitude Mod unlevelled on most frequencies below 1.2 GHz when warmed up
• Inconveniently, all outputs are on the rear-panel (Option 5)

Luckily, the display is a very minor nuisance. I won't bother to fix that.

For the fan, I ordered a replacement (5V, 60x60x15mm) which should arrive soon.

The 508 error appears on and off across several bands, which rules out the band pass filters. I suspect that diode D511 is the culprit. It tests ok with a similar voltage drop as D510. But when I cool it down the error disappears, and when I heat it up, the error appears at more frequencies. I ordered a replacement (Broadcom HSMS-8101) and will report back.

Judging from big solder blobs, it looks like someone already replaced D510 and the adjacent resistor and capacitor. And while poking around, I panicked a bit when I came across this:

But looking at the pictures from Dave's teardown, this seems to be a factory original bodge. I think it bypasses what the service manual calls "printed 2.5 GHz L.P.F." (Sheet 7).

To convert the Marconi back to front-panel outputs, I ordered custom cables from coaxrf.com. There might be cheaper options, but they are reasonably priced, decent quality and fast (if you're located in the US).


For the RF output I used a 8" RG178 cable with angled SMA male and angled N blukhead connectors. The length worked out well. 7" might work but could be tight. Better long than sorry.

I left the rigid coax to the back in place. Anyone has a rigid coax for the front panel that they want to trade? Also, any thoughts on the impact of this change on calibration?


For the LF Output and Ext Mod Input, I ordered 13" and 25" RG178 cables with straight BNC bulkheads on one end. Both are a bit on the long side, but see above. There are pin headers on the PCB to mount these inputs, so the bare ends of the coax cables were stripped and soldered to 3-pin male 0.1" headers.


A bit of heat shrink and everything looks nice and tidy. The BNCs for the rear outputs are soldered to the board and seem to be routed directly to these headers, so I don't think there are any further changes needed on that end.


Stay tuned for an update on the 508 error.

[TK]

Interesting fix & MOD.  Did D510/511 replacement fix the 508 unlevelled error?

[TK]

I have the same error 508 on my Marconi 2024.  It appears at many frequencies.  I set a frequency where I do not get the error, I heat Diode D511 and it appears, remove the heat and it goes away... but I replaced the diode and it keeps doing the same.

I suspected bad caps on the PSU filtering section of the CPU board, recapped, replaced TIP-41, TIP-42 (11V and -11V regulators), replaced both LT1086... I keep getting the same error.

Then I switched to a suspect TL074C (a) opamp (IC801) as output changes from -5V to -9V when the error 508 appears, replaced it... same error.

Service guide indicates input for this section of the opamp IC801 should be around 1.4V... I measured 0.9V, so I traced the origin of the signal back to the CPU board, IC33 (another TL074C opamp), replaced it...  same result.

With or without the error 508, output seems to be working fine.  FM modulation works, but AM modulation (some settings) makes the keypad freeze and only the rotary knob works... There must be something affecting the 5V supply to the CPU board...

And when I enable / disable the RF output or modulation, I get briefly the error 508 and goes away

So next step is to recap the power supply.

Sorry for hijacking this thread, but I am running out of options and wanted to share my experience just in case someone had a similar repair experience.

[chicken]

I have the same error 508 on my Marconi 2024.  It appears at many frequencies.  I set a frequency where I do not get the error, I heat Diode D511 and it appears, remove the heat and it goes away... but I replaced the diode and it keeps doing the same.

Sorry for not posting a follow-up, it was a busy month.

Like you, I replaced the D511 without any change to the 508 error. Also replace D510 for good measure, same result.

But installing the quieter fan certainly helped thinking things through more calmly. 

Service guide indicates input for this section of the opamp IC801 should be around 1.4V... I measured 0.9V, so I traced the origin of the signal back to the CPU board, IC33 (another TL074C opamp), replaced it...  same result.

I also noticed the low voltage on the AM input, so I ventured upstream to the digital board. Being a bit more wary of premature soldering, I took lots of measurements and did a spice simulation of the AM predistortion stage. As measurements and simulation more or less agreed, I concluded that everything is working on that path.

My current theories:
- Something wrong with the feedback loop. However, I get AM modulated output from my Marconi, and if I read the block diagram correctly, AM modulation is added to the carrier through this loop. Anyone please correct me if that's not the case.
- An issue in the filter stage that impacts all/most frequencies below 1.2 GHz. My next step would be to check all diodes in that area.

With or without the error 508, output seems to be working fine.  FM modulation works, but AM modulation (some settings) makes the keypad freeze and only the rotary knob works... There must be something affecting the 5V supply to the CPU board...

Same symptoms here. Also, the keypad does no freeze up when using a low frequency for AM modulation. My theory is, that the RF level is just below the error threshold and with an AM signal the waveform repeatedly triggers an interrupt, overwhelming the main CPU.

And when I enable / disable the RF output or modulation, I get briefly the error 508 and goes away

On mine, when changing frequency it goes away and comes back in various spots. But I can't really nail it down to a specific range as it changes over time and with RF power level.

So next step is to recap the power supply.

I don't think it's a power supply issues. The rails that I checked look fine and noise free.