Marconi TF144H CT452A RF Generator

[G0HZU_JMR]

Thanks, that looks much better. I've spent some time doing temperature ramp tests using a Stuart hotplate and with the test PCB placed inside a diecast box on top of the hotplate.

If I set the hotplate to 75degC and wait an hour or so the inside air temp of the box is 55degC and the detector drifts about -0.17dB from a 24degC start to this higher temperature. Obviously, this is only a very crude form of temperature testing but the die cast box gets as hot on the outside as a typical piece of vintage test gear in this test.

This test is done with the detector set with 12mV rms RF at TP1 in the diagram below. The diagram below is pretty much what is on the board at the moment although I'm playing with the 20M resistor values and the 1800R bias resistor values. The two diodes D1 and D2 are contained in a tiny HSMS-286K IC from HP/Avago and this costs about £1. But it seems to track temperature better than a pair of hand matched 1N5711 diodes.

The circuit first needs to be balanced by making sure that the DC voltage at TP1 and TP2 is the same down to a fraction of 1mV with no RF drive. The 1K trimmer is used to set TP2 to match TP1 and then the circuit is ready for use. If this isn't set correctly then the slope changes slightly. It can deliberately be offset slightly steeper or shallower (by trimming slightly either side of this) but the ideal is with the DC voltages the same because this gives the square law response at the output of the MAX4209.

Once the circuit gets finalised I think it will be possible to reduce the component count because some of the resistors can be absorbed into one component.

[vintage_8bit]

Jeremy. As I have no surface mount stock I have sent for a kit of mixed resisters / capacitors they are "0805" size. Do you think these will be ok? They were very cheap so no problem if not suitable.
My DVM here will only measure down to around a mV, so I would probably do the balancing at work (we use keithly 2000 as well). Colin

[Radio Wrangler]

You just have to watch out for the different grades of ceramic capacitor. NP0 and C0G are close to zero tempco, while decoupler grades like X7R can change significantly with temperature

David

[G0HZU_JMR]

Sorry, I've been a bit busy this last week or so. I do have an update though as I did spend some more time on this circuit last week.

I had a bit of a setback in that the circuit stopped working correctly and this was traced to the little twin diode IC. The diodes are only rated to 4V and this is partly why I reg'd the MAX4209 down at 3.3V. I didn't want any risk of the Hi Z input integrating up near 4V because this would mean a potentially damaging reverse voltage at the diode(s). But I can only think that this is what has damaged the diode IC. Maybe it's possible to get the voltage to integrate up to a risky level just by probing here with a high impedance DMM?

Alternatively, it may have happened when I was soldering different load resistors but then again I hoped that the 1nF cap would help protect the diodes from ESD/transients here. But the diode IC failed 'somehow' and it definitely wasn't RF overload because it has a big attenuator ahead of it at all times.

So I'm left scratching my head a bit. I tried going back to the 1N5711 diodes but these don't track as well over temperature.

It's worth trying the little 0805 caps but as David says you definitely need to select NP0 / C0G. Any tempco issues or leakage issues will degrade the system performance causing drift and/or a jittery needle on the meter display.

[G0HZU_JMR]

Whilst reading through the datasheet I also spotted an HP application circuit for the 286K diode. If you look on the application circuit they show a similar type of tempco 'concept' circuit. But they have the diodes and the biasing backwards compared to my circuit. This would presumably mean I'd have to swap the inputs around on the instrumentation amp because the sense will be inverted.

I need to look at this circuit in more detail because it may be more rugged than my circuit? Presumably there will be circuitry in the MAX4209 that will prevent the voltage integrating (much) down below 0V. If so, this may be a better way to do it in terms of making something that is rugged enough to withstand a bit of scope/DMM probing etc.
If anything, the HP circuit shows that we might be on the right path to making a stable detector in terms of temperature.

I think I might swap the diodes and the MAX4209 inputs around to see if it works OK like this.

[vintage_8bit]

Don't worry about time Jeremy. I appreciate the hours you have allready spent on this project. When dealing with vintage equipment you learn to be patient. I have just found another piece of conductive rubber sleeving which started to play havoc with the ALC loop, but only for a few minutes after switch on. It was hidden behind a tag strip so I had missed it when I first had this sleeving problem.
Anyway, meanwhile I will find some better caps. Colin

[G0HZU_JMR]

Yes, it's taking a while to optimise this circuit such that it works well and remains fairly simple. But I'm having fun trying I had a go at swapping the diodes around as per the HP app note and I had mixed results. I couldn't get the temperature drift below 2dB in my harsh temperature test on the Stuart hotplate. I tried various bias currents as per their app note but no joy. Their circuit 'must' work because they quote really good performance in terms of square law range. But I couldn't get good tempco performance. It was much worse than my original circuit and I'm not sure why. So I'm definitely missing a trick with the HP circuit and the app note schematic is probably missing some detail.

For comparison, I then rebuilt my original circuit used the same diode IC part as used in the HP version and the temperature drift was back down at 0.2dB or less. It behaves really well even during cooldown and even if I speed up the cooldown by lifting it off the hotplate. My original circuit maintains square law performance at all temperatures with/without modulation. But my first attempt at the HP circuit couldn't match this. I think I'll build up the second PCB with my original circuit and then mod the first one back to the HP version and try a few things to see if I can get them both the same.

[G0HZU_JMR]

This morning I'm in the process of retesting the original circuit on the hotplate with my latest bias setting and the temperature performance is extremely good. So far less than 0.15dB drift during the initial ramp up on the Stuart hotplate. Once it gets fully warm (PCB >55degC) it will probably drift by 0.2dB but this is much better than I originally hoped for from this circuit. Crucially, it
maintains the correct square law response when hot. If there is drift of a couple of dB then the square law performance falls away markedly and this is bad news in terms of achieving consistent dial calibration across temperature.

In the pictures below you can see a couple of quick tests with a thermal camera. Because the hotplate and the box are both made of metal, they are reflective (like a mirror) and to get a reasonably accurate measure of temperature I have to stick little squares of matt black tape on the surface to be tested. You can see in the thermal image of the PCB that the metal part of the SMA-BNC connector looks 'cold' at 30degC but this is a false reading caused by the mirror effect. So a cooler part of the room is being reflected into the camera by the shiny BNC connector body.

You can see that the hotplate runs at about 78degC and the box cover at about 42degC but with the box cover quickly removed the little PCB is up at around 53degC. There's about 0.2dB drift in the detector from a 22degC start up to this PCB temperature. 53degC is quite hot to the touch and I'm guessing that it might get this hot inside the sig gen when in normal use?

[vintage_8bit]

Very impressive reults.
Although the new detecter would be fitted in the sealed RF box, it would be at the bottom of the box. The valves are all at the top of the box on the outside, fitted with cans. So I imagine the heat generated in the box would be minimal. I think with the stability of your last results l can't see temperature now causing any issues. Colin

[G0HZU_JMR]

Yes, it's beginning to look quite promising. I've now built up the second PCB with my original circuit. Note that on both of these PCBs I have changed a couple of resistor values since my last schematic was posted up. I've changed the bias feed values and also the load resistors and this seems to work best for stability.

The copy circuit is spookily similar to the first PCB. Even the detector output is within a tiny fraction of a dB. The drift over temperature is exactly the same at 0.2dB. One niggle is why the diode IC failed in the very first version. It could have been ESD damage during rework or some other issue with overvoltage during probing. It wasn't overdrive damage or overheating from my soldering iron as I've been careful not to 'age' the diodes into imbalance through excessive use of the iron.

[vintage_8bit]

Jeremy. Just cleared away the last project so thoughts returning to the genny.

If you think we are going to go with the MAX4209 & HSMS-286K, I will order these. The "NP0 / C0G" cap, I found is in a 0805 package (Farnell 1856213) Will this type be ok? Any recommendations welcomed.

I looked at the data sheet for the 286K, the SOT-363 package, being only 2mm square with 6 pins!, if I've got it right. I see the centre pins are for grounding. That will need a microscope .
Anyway the work shop ambient is over 30 degrees at the moment so It probably gets hotter in side the genny's case than I first thought. All the best, Colin.

[G0HZU_JMR]

Sadly, I've not had much free time to look at this beyond doing a few more tests with HP's 'reverse' bias method.

I tried doing some absolute drift tests on a single diode when biased the HP way and the drift is much, much worse in terms of how much the bias voltage changes (at the node that feeds to the instrumentation amplifier input) if I blast the diode with a hairdryer.
If I revert to my original circuit the drift is an order of magnitude less and this probably explains why I got such poor results with the circuit from the HP app note. There's obviously more to it than the simple concept circuit they published. They didn't provide any component values so it's hard to know what they actually did. Clearly, I'm missing a trick here.

But my original circuit seems to work very well, especially without the 20meg load resistors and with quite high values where the 1.8k bias resistors are shown in my last circuit. But the long term concern here would be if the PCB material absorbed moisture over time and became leaky. So the circuit would have to be built on decent FR4 material and not a cheaper material that might be very porous. So I may revert to fitting the 20 meg load to define things a bit better.

I'll try and have another play over the weekend.

If you think we are going to go with the MAX4209 & HSMS-286K, I will order these. The "NP0 / C0G" cap, I found is in a 0805 package (Farnell 1856213) Will this type be ok? Any recommendations welcomed.

The 4209 does seem to be working very well, even from a single supply and the 286K diode is working well although there is a slight worry about zapping it when probing etc.

Assuming that there is no rush it's probably worth waiting a bit longer and I'll post up a complete circuit with some formal test results. Note that we still have to work out the attenuator values and I think it may be wise to do away with the pot at the opamp output and replace it with a fixed resistor. Otherwise there are too many things to adjust. I suspect that it won't be necessary to have a pot here if there is a way to set the level further back in the circuit. But for this to be the case, we need to get the attenuator value correct

I found is in a 0805 package (Farnell 1856213) Will this type be ok?

At the minute I'm still using high quality caps from ATC which is almost certainly overkill so I'm going to try several types of 1nF COG cap I have here. I can't check the cap you suggest because the Farnell site is doing its maintenance (again) and it is closed at the moment.

[vintage_8bit]

If the capacitor you used is readily available I wouldn't mind paying a little more. But if not or if curiosity needs to be satisfied, which is good enough reason, then I don't mind waiting for this and the other tweeks.

Once the circuit is finalised and if you wanted to prevent moisture ingress then a conformal coating could be applied to protect from damp shacks, which perhaps is where a lot of this vintage equipment is kept.

I'm happy about removing the output pot. This leaves the adjustment to just RV103.
In a previous post I reported the voltage across the thermocouple heater to be around 0.278V rms. With the "Direct output" at 2.0V rms. I will try and confirm this value with RV103 set to its mid point as this result will affect the attenuator values. This will be with 24 Ohm in place of the heater. I assume the input impedance of your circuit will have to be halved to match? Colin.

[Radio Wrangler]

The ATC porcelain capacitors aren't particularly low in ESR, their tolerances are quite normal. What is special about them is their ability to carry amazingly high currents. So they are definitely the device of choice for transmitters. I have single 100A size (close to 0805) ones carrying pulsed RF at 300 Watts.

A TF144H isn't going to need them. If you need particularly low loss, the AVX "U' range are good and a lot cheaper, but they aren't capable of the same level of brute power.

David

[vintage_8bit]

Thanks David. Jeremy I have just had a re look to see what range of adjustment RV103 gives us for a 2v rms signal at the direct output.
It was initially set, I believe to give .278mv rms when fitted with a good thermocouple and the needle on the set carrier mark.
With a 22 ohm load (didn't have 24 ohm) simulating the heater. I can adjust the voltage across the 22 ohm load from 707mv rms to 120mv rms.

Hope this helps with the maths
Colin.

[G0HZU_JMR]

I had another play over the weekend and the schematic below includes the input attenuator resistors that suit 280mV rms at the input. The schematic below is the one that gives the best performance over temperature in terms of drift and square law response over temperature.

The attenuator is now designed for 24R input impedance and 280mV rms drive There is a 1uF cap C6 inside the attenuator network and this is there to prevent any bias pulling of the diodes. Without this cap there is a DC path to the input which could slightly pull the bias off the optimal setting if the resistance ever changes at this node due to range changing etc.

Note that I can't test it very accurately/thoroughly when built like this because the 24R input impedance causes subtle issues with mismatch uncertainty in my sig gens. The Marconi 2024 is the most affected here. Because I'm testing for a very accurate detector slope, any misbehaviour in the sig gen performance will make the detector look bad.

The little Marconi 2024 gives an error in 1dB step accuracy on certain attenuator settings when fed into a mismatch of 2:1. I know it's the generator because I can use the 'attenuator hold' feature to preserve the range over which it can deliver accurate 1dB steps without having to click in a major attenuator (relay) change. The error is quite subtle, just a small fraction of 1dB but it's enough to spoil or confuse the testing process.

The Agilent ESGD sig gens I have perform slightly better but the AM modulation performance is quite poor on these generators as the carrier level sags slightly on high mod % settings.

Probably the best way forward is for me to finish playing with one of these two detector boards I have here and send it to Colin to try out? My biggest concern at the moment is to do with the frequency response. My detector board is very flat from kHz to 70MHz but I'm worried that the sig gen will have some form of pre/de emphasis built in to cancel it's own detector/path responses.

So a 'flat' detector might not be the right thing to produce. It may need to have a compensation network added to it. So there's probably a few more hurdles in the way...

[vintage_8bit]

Jeremy.
There will probably not be a better way to test than to try for real, so I would of course like to try one of your boards if you are happy to send one. Let me know & I will PM you. Colin

[G0HZU_JMR]

OK, but I'm still playing with the bias and load settings to try and get the best detector performance. I'll let you know when I've finished playing and then I'll post you one of the boards. You can keep it as a reference, I don't want it back

What I have found so far is that none of my sig gens are very precise when it comes to generating accurate AM modulation at 80% levels. They all have slight issues and this depends on the modulating frequency and also the RF level and RF frequency.

The ideal test source would be able to generate a unmodulated carrier at (say) 0dBm and then modulate it at 80% AM and this should give an increase in average power level of 1.21dB. It should be able to do this at all mod frequencies and carrier frequencies.

But all of my generators fail to deliver this reliably. Looking at the specs for them, the accuracy of the mod depth at 80% is approx +/- 5%.

So I'm now looking to generate accurate AM (with mod both on and off ) using the IQ modulator in the ESGD sig gen with the ALC turned off. My first fumblings at this are giving good results when I look in the time domain on a scope and also in the frequency domain on my E4406A signal analyser. When I look on a power meter the power rises 1.23dB on my Anritsu power meter and the little detector board shows 1.25dB. So this looks to be pretty good. It seems to be consistent across all mod frequencies tested so far using this technique. If I try and do this using the regular AM modulators in the sig gens the results are not as consistent and it probably reaches the +/-5% tolerance on the spec sheet. It also looks poor on the E4406A in terms of carrier level vs mod sideband level and doesn't give the expected 1.21dB on the power meter like this either.

It does make me wonder how accurate the Marconi TF144H modulation will be?
For its internal metering to system to be valid the sig gen needs to maintain the relationship of the unmodulated carrier (for the SET CARRIER position) and it then needs to generate an AM envelope of 80% modulation depth as well as maintaining the carrier power without sagging. It also needs to deliver fairly linear modulation. Otherwise the 1.21dB relationship is lost. My modern sig gens struggle to do this at 80% mod depth.

Sadly, I don't have much free time to look at this project. I just get the odd hour or two at the weekends so progress is a bit slow

[G0HZU_JMR]

The IQ mod method works quite well, much better than the raw sig gen modulator, but I think I have found a quicker way to test this detector for accuracy wrt mod rate.

This new method below offers much more control although it doesn't actually produce a regular AM waveform. What I have tried this evening is to combine both of my ESGD sig gens via a high isolation combiner and some attenuators and arranged both carriers to be 1kHz apart. This produces a DSBSC signal, the classic SSB test signal. I can now verify the difference between unmodulated and modulated by turning off one of the tones. On my accurate Anritsu power meter the power falls exactly 3.00dB +/- 0.01dB on the display if I turn off either carrier. I can also see a perfect two tone modulation envelope on a scope with precise/sharp crossover when both tones are on.

When I test this across mod rates (carrier spacing) of 200Hz to 20kHz the result is the same -3.00dB change on the power meter because I'm only changing a carrier frequency by a tiny amount. There isn't a 'modulator' to worry about anymore. When I feed the same waveform into the little detector board the worst error I see across all of these mod rates is 0.07dB away from a 3dB change when I turn off one of the carriers to remove the modulation. This is a very good result.

Obviously, this isn't a classic AM test signal and it looks a bit different on a scope compared to an 80% modulated waveform but I don't think it matters. It's the power change accuracy that matters. This technique makes the testing very consistent. I repeated the test making one of the carriers slightly smaller than the other such that the change in power with both carriers was exactly 1.21dB. Again, this isn't a classic AM test signal and it looks a bit different on a scope compared to an 80% modulated waveform but I don't think it matters.

The little detector board showed 1.23dB across the whole of 200Hz through to 20kHz carrier spacing. I wish I'd thought of this method earlier... I can now test the mod rate performance in a few seconds with much reduced uncertainty in my modulated/unmodulated test signals. Note that I've changed the 10k resistors to 4k7 and I've included 10meg ohm load resistors in these latest tests. But these values may still change.

I do think this little detector board has a very good chance of performing very well in the TF144H

[vintage_8bit]

Some more good work and a new method for modulation!

I am grateful Jeremy for your offer of the detector . I now have an amplifier and the twin diode package, I could at least send these to you.

Anyway I will wait to hear more. I wonder how many generators will benefit from this, & perhaps other uses will be found. Colin