Marconi TF144H CT452A RF Generator

[vintage_8bit]

Richard. The reason for the low frequency was, after removing the turret I had tacked in a spare coil pack "B". This gave me the space to more easily fault find and work around the ALC detector. I will remove it tonight and refit the turret, time permitting.
David / Richard. We already have an output from the ALC detector that rises in step with the output, if this was fed in to a summing op amp circuit with the other input fed from just a simple diode detector for the modulation could this work. I realise that it won’t be detecting the power in the carrier but would it not be sufficient to allow the 2V rms to be set accurately enough. When the modulation is applied the gain of the simple detector could be calibrated to bring the needle of the meter up to the 80% mod mark. Colin

[Radio Wrangler]

For a straight carrier, Yes.

With modulation applied, the simple detector will follow the peak voltage.. the thermal detector followed the total RMS power which rises much less, so the modulation percentage marks would be wrong.

If the simple detector was filtered slower than the lowest mod freq, then it would stay at the carrier level and not show any effect of modulation.

David

[G0HZU_JMR]

The other thing I'd be looking into is the 'law' of the two devices. Tid the thermocoupe give a voltage out proportional to the power input, or proportional to the RMS of the input voltage? There is a square-law function difference between these.

Yes, this is a good point.

I think a classic thermocouple should give a voltage proportional to temperature and therefore it should be a square law device for measuring RF power. i.e. the detector output voltage should be proportional to input power. I'd expect the Marconi device to be the same but it might not be.

However, you also have to be careful with some modern 'true rms' detector chips because they don't produce a square law detector output. Some output a straight line vs Vrms (rather than vs power). So this might be the problem here.

Back when I was looking at the averaging detector used in the ALC loop of this generator I did have a quick play at making a crude temperature compensated diode detector that operates in square law (to replace the dead thermocouple).

This would work like the thermocouple but only at much lower power levels because you have to keep the RF voltage levels below the thermal voltage (26mV) to stay in square law. So it would need an extra attenuator ahead of it.

I could dig it out and play with it again? The main risk with it is the poor stability wrt temperature changes and I guess the temperature changes/gradients inside this generator could be quite significant during warm up alone. The detector output is only going to be a few millivolts so it would also need a decent precision opamp driver on the output to cancel issues with offset voltages.

I kind of abandoned this approach because it would probably only qualify as a 'better than nothing' fudge. I can dig it out and post up a few test results if that helps? I used a hairdryer and a Stuart hotplate to do some crude temperature testing here at home but I could easily stick it in a Thermotron oven at work to test it more formally? Over frequency it was extremely flat from a few kHz right up to 70MHz so no problems in this respect.

I think the best performance would come from choosing the best (LBSD) diode in terms of immunity to temperature changes as it would be easiest to compensate?

[vintage_8bit]

Richard. Re post 78. I have just checked that the results I saw with a carrier of 30Khz are the same at 1Mhz. With respect to the vrms output. With and without the cap.

Jeremy, I see from your experiments that a diode detector is possible but with some complications regarding low input levels and thermal considerations, that I wouldn't of thought about.

I dont know if this is significant or helps. The difference between the 2.0V and 2.3V dc that you inject when calibrating, is 15%.
I measured the difference at the meter terminals between the "Set carrier" mark at 5.2mV, & the "set modulation to 80%" mark at 6.7mV. This gives a difference of about 30%. Colin.

[G0HZU_JMR]

The power increase when 80% modulation is applied should be a factor of 1.32.

So I would expect a square law detector to give a DC voltage change at the detector output that changed by a factor of 1.32 (for a 1.2dB increase in average power due to the 80% modulation).

You measured 6.7/5.2 which is fairly close at 1.29.

If I just test my uncompensated Schottky diode detector with a sig gen I get these figures from it.

20MHz -25dBm 2.90mV
with 80% modulation at 1kHz I get 3.90mV

This is a ratio of 1.34 which is close to theory for a square law detector device.

If I turn off the modulation and increase the sig gen by 1.2dB then I measure 3.77mV which is a power increase of about 1.14dB (for a square law device). It should have been an increase of 1.2dB so the detector is slightly out here. In theory, the 3.90mV and the 3.77mV numbers above should have agreed but the diode is probably only just into the square law region at -25dBm.

When tested across frequency using an Agilent ESGD sig gen the detector stayed flat to within about 0.1dB from 300kHz to 70MHz. To give confidence in the actual performance/flatness of the test sig gen, see below for a typical plot of the flatness of the sig gen. This plot is taken from the datasheet for the sig gen. My sig gen is a couple of months out of calibration but it still checks out as being remarkably flat on a decent power meter. i.e. similar to the plot below.

[trh01uk]

Originally Posted by vintage_8bit

Richard. Re post 78. I have just checked that the results I saw with a carrier of 30Khz are the same at 1Mhz. With respect to the vrms output. With and without the cap.

Colin,

yes, that's good. It was a long shot, but it looked a bit suspect according to the spec. Your geny is probably better than spec, so it doesn't matter.

Going back to the main problem here, I think Jeremy has put his finger on the problem. The AD8361 puts out a DC voltage proportional the incoming RMS RF voltage coming in. We need the square of that voltage to get a signal proportional to power.

It looks like Jeremy's diode detector idea, running at a low power level so its stays in the square law region, is probably the way to go. It should be easy to make a small surface mount module, with chip resistors to make an attenuator flat out to the top frequency of 72MHz.

An alternative approach would be to stick with the AD8361 approach, and just put a new "modulation calibration" point on the meter scale. That's simple, but probably not favoured by those who want to retain the originality.

Richard

[G0HZU_JMR]

It looks like Jeremy's diode detector idea, running at a low power level so its stays in the square law region, is probably the way to go. It should be easy to make a small surface mount module, with chip resistors to make an attenuator flat out to the top frequency of 72MHz.

Here's the circuit with an opamp added. The LF355N was the only suitable opamp I had here in DIL8 so I tried it and it worked very well considering how basic the circuit is.

The opamp needs to have very low input bias current (few pA) and also it needs to have an adjustable offset feature as per the image below. I was able to trim the offset out using the 10K pot and get very good performance.

it was pretty good over temperature as well. Much better than I thought. But there will be better diodes and better opamps than this. Note that the leakage current is sufficient to discharge the 100nF cap quite quickly once the drive is removed.

One downside is that it needs split +5 and -5V supplies. The LF355N offset is quite immune to supply changes so the +5V and -5V reg circuits don't have to be special in any way. Maybe a basic LDO reg for +5V and a 7660 based chip for the -5V?

The circuit below is just a lashup and needs to be optimised and it really is just a 'better than nothing' option if you can't get a thermocouple.

As Richard says, it will need a decent broadband RF attenuator ahead of it made (ideally) with SMD parts. This is needed in order to get the drive level to this circuit down at approx -25dBm.

I tested it with a hairdryer to heat it up and got the opamp at 75degC and the diode at 68degC and the output stayed close to 300mV DC. The offset in the opamp degraded a bit by maybe 20mV but I don't think this is significant. I tested it for 80% mod at 1kHz right down to 100kHz and it seemed to work fine.

This is very much a 'cheap as chips' circuit needing dev/improvement. It ideally needs temperature compensation and a better opamp and diode to further improve the performance over temperature. I just made it with old spare parts I had lying around

[G0HZU_JMR]

In case anyone is unfamiliar with using a diode way down in the square law region you can read up a bit about diode detectors in this classic old HP app note here:

http://hparchive.com/seminar_notes/P..._detectors.pdf


It may seem a bit strange expecting a diode detector to 'work' with an RF drive level of just 12mV rms but this is the drive level you need to get fairly well inside the square law region.

I tried some of the temperature compensating techniques in that note but I didn't manage to get a significant improvement. Probably because my test setup for setting/controlling temperature is a bit erratic with poor control and repeatability. I even tried using a second diode to compensate and this did seem to give some improvement.

I've also got one of the classic HP8473C Low Barrier Schottky diode (LBSD) detectors here as in the image below and this diode is well characterised for use up to 26GHz as per the datasheet below.

http://cp.literature.agilent.com/lit.../5952-8299.pdf

This can be used down in the square law region and I've used this detector to help provide a confidence check for the flatness of my various microwave power heads for my various lab power meters. This detector is remarkably flat up to 4GHz but the low frequency performance is limited by the 30pF? shunt capacitor inside. I don't know if I'm lucky to get a 'good one' but it performs well inside the amplitude vs frequency limits in the datasheet. right up to 4GHz. The return loss is remarkably good up to 4GHz as well.

So this is my reference detector for flatness. However, in the crude schematic in my previous post I just used a cheap 1N5711 diode for the detector. These are just a few pence each. Not ideal but it worked very well.

[G0HZU_JMR]

Hope nobody minds me posting up the image below but David might like to see this one...

A while back I managed to find and buy a 'new old stock' HP478A RF power detector as below. This uses a thermistor arrangement inside and these sensors date back to the 1960s. I've already got one of these but I couldn't resist buying this one as it was so cheap.

It still looks to be shrink sealed and the brown washer for the N type connector isn't fitted yet as you can see it below the sensor.

I'm not sure if the sensor really is genuinely new old stock (or how old it is or even if it still works) but it looks sparkly new and has obviously never been used. I'd consider this to be a very rare find and it was very cheap. It does seem odd to see it 'shrink wrap' packaged to cardboard like this. I expected it to be boxed but maybe these were supplied like this at one time?

[vintage_8bit]

Ok. I'll put the AD8361 back in drawer.
As you have had some good results, I would like to try and build that circuit. I can find some IN5711's could you suggest an alternative op amp? I couldent find an LF355N so easily.
My concern will be the attenuator. Im used to building on just verboard and standard components which I assume will be a problem particular at the higher frequances, and with such a small signall. Colin

[Radio Wrangler]

Odd to see a 478 shrink wrapped. I bought a pair in Nato cardboard boxes and double plastic (heat sealed) bagsfor 50p from a lucky dip stall in the 80's. Lucky, I'll say. New.

I was showing some folk from NPL around when we were negotiating contracts for reference noise source calibration and passing the 432A assembly line (the meter unit that goes with the 478 bolometer head) and someone said "For god's sake don't stop making those!" with the rest joining in.

The 478/432 are the RF to DC transfer standard that all the other stuff traces back to. It can do a few tricks that even the Class-IV NIST meter can't do, though the Class-IV has the edge on some DC offsets. To go with your 478 you need a 432A, and a nice pair of six digit voltmeters....

Ah! Them were the days.

David

[Radio Wrangler]

The HP semiconductor division went with Agilent, then got flogged-off to a consortium of financiers. Recently they bought out Broadcom and now everything is under the broadcom name.

Farnell stock the real original HP low barrier schottkies, but in SOT23 packages. Look up HSMS- prefixed part numbers.

David

[G0HZU_JMR]

Odd to see a 478 shrink wrapped. I bought a pair in Nato cardboard boxes and double plastic (heat sealed) bagsfor 50p from a lucky dip stall in the 80's. Lucky, I'll say. New.

Wow, that really was lucky... I've also been successful in finding a tidy example of a HP432A power meter to sit alongside my old HP431C. There was no rush on my part to find one because the HP431C works well enough to do what I want. However, after a few years of patiently waiting I finally found a HP432A that is in very good condition. It's probably not that old because it has the later type of power switch but it cost me about £50 delivered. It was the cleanest one I'd seen in all this time and it was also one of the cheapest so I'm doubly pleased.

[G0HZU_JMR]

As you have had some good results, I would like to try and build that circuit. I can find some IN5711's could you suggest an alternative op amp?

An AD820 in 8 pin DIL package would probably work even better than the LF355N but they typically cost £3. Also, there's no guarantee that this circuit will work correctly in the TF144H because there is still a chance that the dial calibration for the thermocouple will be subtly different to the diode and opamp. Also, the temperature drift may be too great.

The AD820 in DIL8 package has the option for the external trim/adjust feature for the offset voltage but I don't think this feature is fitted in the SMD version. So I would recommend using the classic DIL 8 package.

What we could do between us all (for a bit of fun) is design a little PCB module that fits snugly into the TF144H with this circuit and the attenuator fitted.

The ideal PCB would have mounting tabs/holes to marry up to the TF144H in the relevant location. How much space is available in the area vacated by the thermocouple and how does the RF connect from the existing resistor network to this module? Does it use short wire links?

It would be nice if the AD820 opamp would perform from a single +5V supply in this application but I suspect that the offset voltage will be too significant.

One option for the voltage inverter for the -5V supply is the tiny little Maxim MAX1697U. These are cheap and work well but they are very fiddly to solder if you are not used to working with SMD. This would need a dedicated PCB design with the SMD artwork.

I've got a T-Tech 7000S PCB milling/routing machine here at home so making PCB prototypes for stuff like that is easy for me. See the link below. So I could mill and make a PCB and post it to you? I've got several LF355N opamps here and hundreds of 1N5711 diodes and also quite a few (used) MAX1697U chips and 5V LDO regs in SMD.

http://t-techtools.com/store/index.p...roducts_id=288

If we could agree on a PCB outline I could mill something quite quickly? I'd include the SMD resistive attenuator in the PCB artwork as well. This will use small chip resistors that cost a fraction of a penny each.

[G0HZU_JMR]

Just a quick update... I managed to find some AD820A devices in SMD and I tried this device in place of the LF355N. This was after I downloaded the AD820 spice model from AD and tried simulating the circuit.

I also tried running it from a single supply as the AD820 is a single supply rail to rail device. The results were quite encouraging. The DC offset can't be zeroed with this circuit because the SMD package doesn't have the NULL pins active in pins 1 and 5.

http://www.analog.com/en/products/am...oduct-overview

Also, the x101 amplifier doesn't amplify by an accurate x101 factor but the important part of the circuit seemed to work well. i.e. it still gave a square law response to power changes or to modulation. So it may well be possible to make a cut down version of my previous circuit with just a single supply rail. It probably won't need the 5V regulator chip either. It could be run direct from the 6.5V supply in the sig gen. eg via an RC filter in the supply feed. So maybe there is no need for the 5V reg or the -5V supply rail or the DC offset nulling pot?

If the offset null performance across a batch of these devices isn't as good as this one then there are two options available:

Try the AD820B version of the opamp or try configuring the detector to use two diodes in a doubler config. This will still work in square law but you get a x2 increase in detector output to counter the spread in offset performance in the opamp. I guess the next thing to work out is the attenuation factor required in the SMD attenuator ahead of the square law detector.

[vintage_8bit]

Jeremy. Thank you for your time and thoughts. I have found an LF355N and a IN5711 and I am just putting it all together to see if I can repeat your results, but this time driving the meter. I was going to make the input resistance 24 ohms so it simulated the load of the thermocouple heater. With a 24 ohm load my notes say this results in 280mv RMS across the heater. I might check this next time I power up. I was struggling with the maths to work out what the attenuation should be, but its late now.

There is quite a bit of room behind the turret. With the thermocouple removed there is also a hole in the chassis to put a standard mounting pillar.
I would have thought the valve heater supply would be usefull for power.
The resistors and adjustment pot are all just mounted on 2 parallel standard terminal strips. I could take a photo and post if required but the layout is in the manual if you have it. Colin

[G0HZU_JMR]

Colin, I'll see if I can find a copy of the manual with a good copy of the pictures and have a closer look.

One thing I noted during my early testing was that the capacitor used on the output of the detector needs to be quite stable over temperature in terms of leakage and value. The first cap I tried was a cheapo leaded ceramic disc cap as you would find in a basic radio.

This gave a detector output that was quite jittery. I tried using my old Altai multimeter as an analogue dial and the ceramic cap definitely spoiled the performance as the needle on the dial was unsteady and jittery.

Also, the detector output went offscale if I blasted with the hairdryer. I ended up using a decent SMD cap here and the improvement in performance/stability over temperature was marked.

I'm still worried that the overall temperature stability of the diode, the cap and the opamp will mean that this solution will be a bit too drifty and will be judged as 'close, but no cigar...'

[G0HZU_JMR]

To give an indication of how much thermal drift I now see with the hairdryer, I tried testing the latest version with the AD820A and the single supply with a decent ceramic cap and no offset trimmer fitted.

The hairdryer is set to hot and medium airflow and I blasted the whole circuit in the hot airflow. This is a fierce test. The most drift I see is in the first split second and I see about 0.6dB of drift. This is due to extreme temperature gradients. If I keep blasting the circuit stabilises back to about 0.3dB change. So the circuit seems pretty good although it obviously isn't modern lab grade performance. I suspect that you might see this much drift in detector performance over normal sig gen warmup.

To repeat, with a cheapo ceramic disk cap fitted at the detector output the temperature stability is awful. It will change over a dB just by breathing on it and the hairdryer sends it violently offscale by many dB when first blasted. Also, it's never stable no matter how long you leave it at room temperature with short term jitter of +/- 0.2dB over a few seconds. i.e. there is an annoying amount of drifty jitter on the needle.

So you need a decent quality cap here. The other thing to note is that the 100nF value means the detector response is very slow. It might be better to use a smaller value here (for stability and response time?) and accept some degradation in performance at lower frequencies?

[vintage_8bit]

Jeremy. Have you a copy of the manual? I have the military model CT452A.
The tagstrip layout is on page 33 fig 4.6
The circuit is the 1st sheet of fig 4.11. Hope this helps. I tried in a P.M. to attach but no joy.

Colin.

[G0HZU_JMR]

I've got a fairly fuzzy manual and I think I can see how it is arranged now.

I spent a bit of time comparing the single supply AD820 against the split supply LF355 in DIL and the LF355 is definitely superior in terms of square law performance. I think this is because the split supply DIL opamp can have the offset nulled.

The simpler 820 circuit only operates over a narrow window and it doesn't quite deliver the accuracy of the other circuit.

So I'm now trying the single supply AD820 with a 'doubler' detector using two 1N5711 diodes and this seems to have improved it to the point of being a contender. The doubler circuit helps to offset the inferior offset null as it delivers a higher DC voltage into the opamp for a given RF drive level. It now gives very good square law performance (modulated or unmodulated) over a wider range.

I've designed a basic PCB 'core' based on this circuit and I think I'll mill this one and test it. The current circuit is built dead bug style and it would be nice to put it on a PCB.

I've also ended up changing the cap to a 1nF for now. I'm using an exotic ATC cap here in SMD and this is very stable. But I think a cheapo NP0 cap will be OK here.