Marconi TF144H CT452A RF Generator

[vintage_8bit]

I have now tried my LF355 and 57711 combo. I used a 1k and 47ohm in the pot-down. The feedback resistor ended up at 95.4k
As I increased in frequancy the output started to rise quite high.
But at 1 mhz and measuring the output with a DVM I got .276 dc (cw) and .356 dc with 80% mod. This ratio is exactly the same as the ratio given by the 2 witness marks on the Marconi meter, as per results in previous post.
I see that you have varsly improved the circuit now though.
I had not used surface mount so was pleased with my results. I do know some one who may be anle to print out some pcb's if we have a design. If we get a final circuit I could send you one to try. Colin.

[G0HZU_JMR]

That seems good. It's a relief to see that the meter cal marks agree. However, I think it's risky to use a 1k resistor in the attenuator as this will not give a flat response over frequency. This detector should be really flat right up to 72MHz.

I think the AD820A on a single supply rail and the diode doubler will be the way to go because it is so simple.

Out of interest, how does the RF waveform look on a scope as you set the modulation at 50% or 80%? Does it maintain a nice sine shaped mod envelope? I was unsure as to the quality (as in distortion level) of the AM waveform from this generator.

[vintage_8bit]

Ok, I will check the wave shape tonight. From memory I think its quite good.
What value do you think the overall pot down should be. I have a 24ohm load in parallel to simulate the thermocouple heater. Colin.

[G0HZU_JMR]

I'm still unsure how much attenuation is needed but I think the level needs to be centred at about -26dBm. If we centre it at -25dBm then I think the modulation peaks will start to creep to the boundary of the square law region. So this will give a slightly lower increase in detector voltage when the 80% modulation is applied.

So you might see a factor of 1.28 rather than the theoretical 1.32 because the mod peaks are not giving the ideal square law response. If we move it down to -27dBm then I think the performance will be closer to theory for a square law device. However, the set/mod marks on the meter may not agree with the 1.32 factor because you measured a ratio closer to 1.29. So it may be that we have to centre the detector right on the edge of the square law region to get it to agree with the meter dial?

I am also wondering if it's worth investigating a quadrupling detector using 4 diodes. This could be done with a pair of SOT23 dual diodes and it would push up the detector voltage even higher.

I'm still concerned about how repeatable this design will be, especially wrt the offset null performance of the AD820 when run from a single supply. As it stands my latest circuit would not survive long in a design review at my place of work because of repeatability/yield/temperature concerns. However, it does only have to work well over a very limited range of signals. I'm not sure what to do next. I'm tempted to try a pair of SOT23 dual diodes to make a compact quad detector. The 1N5711 performs remarkably well in this application but there will be better diodes available and this would all help with repeatability and performance over temperature.

[vintage_8bit]

Jeremy hopefully attached. Photo from scope. 10 Mhz CW 1KhZ mod. It looks good to me?
I tried to attach manual again but its two large. I can send it via email if you PM me your email add. Colin

[G0HZU_JMR]

That looks pretty good! It's not showing any significant distortion as in any distortion I can see. So the AM modulator system seems to be quite good on that sig gen. I was kind of expecting to see some distortion.

I did some quick tests on the detector today to look at square law conformance.

My sig gen is very accurate when tuned in 1dB steps and I can cross check this on several bits of test kit. So it should be linear to within 0.02dB across a 12dB range like I'm using below.

I think the detector needs to give quite good square law performance over about a 12dB window because it has to cope with the full modulation envelope and here is the result of a quick test of linearity.


-20dBm -20.15dBm
-21dBm -21.08dBm
-22dBm -22.05dBm
-23dBm -23.02dBm
-24dBm -24.01dBm
-25dBm -25.00dBm
-26dBm -26.00dBm
-27dBm -27.01dBm
-28dBm -28.03dBm
-29dBm -29.06dBm
-30dBm -30.13dBm
-31dBm -31.20dBm
-32dBm -32.25dBm

This is a cw test and the results are quite good but there isn't much more useful square law range than this.
So it's all a bit tight in terms of performance/range. You can see there is a sweet spot around -26dBm. I think the quad diode detector may give a bit more performance because I think the weak link is the opamp and the quad detector will make things easier for the opamp. If I turn on the 80% modulation in the sweetspot at -26dBm then the reported power goes up by 1.20dB (which agrees with theory) and I cross check the modulation depth on a couple of mod analysers I have here.

I also tried modulating the sig gen using its arbitrary waveform generator. I tried generating a chunk of noise over a 100kHz BW at -26dBm. The -26dBm level was checked on a lab power meter (thermocouple) and also a signal analyser.

The detector reported -26dBm within 0.05dB which was a nice surprise. The peaks and troughs of the noise will be exploring parts of the detector range way outside my chart above but I guess the high and low side errors must cancel out? The detector seems a bit deaf towards the top of the square law range (as expected) and a bit too sensitive as you dive deeper into it and this may be an issue with the less than ideal offset in the opamp.

[vintage_8bit]

If im thinking right....
Once you have set your 80% modulation, and you set to "Use", you can then rely on the scaling of the modulation depth control. This is only marked up to 80%. So you never look at the meter over 80% modulation. The meter is also marked at 30 & 50%. These presumably are to keep an eye on the calibration of the set mod depth control.
So I don't think performance over 80% mod will be an issue, Colin

[G0HZU_JMR]

Yes, I'm just trying to make sure it works over the significant parts of the modulation envelope. However, I have since had a think about ways to give improve the overall design integrity and I'm now wondering if it's worth going up a class from a basic opamp to an instrumentation amplifier. If I do this I (think) I can go back to a single diode detector and I think I can also introduce temperature compensation.

If I weakly bias the detector diode I can have an identical (weakly biased) diode on the other arm of the instrumentation amplifier to act as a tracking device for temperature. A decent instrumentation amplifier would solve a lot of niggly issues with the current circuit because of its superior offset and drift performance etc.

A typical candidate would be an LTC2053 from Linear Tech. This runs from a single supply and the datasheet almost looks too good to be true.

http://www.linear.com/product/LTC2053

I need to spend more time looking at this device and this alternative topology but this would hopefully provide a more robust solution. The LTC2053 costs about a fiver but the AD820 is only a bit cheaper so this would not add that much to the cost?

[vintage_8bit]

I think solving this problem is so worthwhile that spending a few pounds is justifiable. Hopefully it will help out quite a few. Colin

[G0HZU_JMR]

I've looked through the range of instrumentation amps on the Farnell website and I now think there are much cheaper devices than the LTC2053 that offer performance better suited to this application.

I didn't realise how complex and capable some of these devices are. I was expecting to just see three opamps arranged as a classic instrumentation amplifier. It will take me a while to choose the best one.

[vintage_8bit]

I had a look at your link for the LTC2053 and it did have an impressive spec. However if you think there are better ones I will await your next post. Thanks, Colin.

[vintage_8bit]

I see what you mean by the choice from Farnell. I whittled it down a bit by using the filter, but still to many. I see some have some form of chopping / switching circuit? in them. Colin

[G0HZU_JMR]

I'm still unsure about which one of those instrumentation amps to try. The complex auto zero versions (that use choppers etc?) are attractive because they would protect against long term ageing/drift in the circuit.

Here's a couple of hastily drawn circuits to show the ideas so far.

The first one is the uncompensated version that uses a doubler detector and an instrumentation amplifier. This will probably be good enough.

The second one shows the version with temperature compensation. I have no idea if this will work effectively but the idea is that the equal bias to each diode will result in an offset signal that will be rejected as a common mode signal in the instrumentation amplifier. So any changes due to temperature in the diodes will be negated. But I'm concerned that this crude implementation may cause more issues than it cures. It may spoil the square law performance for example.

Neither circuit shows any supply decoupling or a 3.3V reg but this would be added and would be common to both circuits. For example, the 3.3V SG1 node in the temp comp version needs to be decoupled to ensure good RF isolation between the two diodes. R1 and R3 will have very large resistance values and the SG1 node should have a decoupling cap fitted. No RF energy should reach D1 in this circuit and D1 and D2 should ideally be a matched pair of diodes.

[Radio Wrangler]

Some of the diode-based heads for HP RF power meters use simple FET choppers mounted in the heads, with an AC-coupled amplifier and a synchronous rectifier back in the meter unit itself.

Some clever ones use a 'backwards' diode for low level and then switch to monitoring a more normal schottky at higher levels. These sensors can cover -70 to +20dBm. The simpler single-gear ones do -70 to -20dBm which has the unfortunate effect that you can only use the 50MHz 0dBm calibrator if you have a good 20dB attenuator.

David

[G0HZU_JMR]

I wonder how complex the temperature compensation is on the HP gear? The uncompensated circuit I posted above quickly becomes unusable over temperature if the power level drops a few dB from -25dBm.

I did a few tests on the LF355N board this evening because I can null the offset on that opamp. I then tried biasing the 1N5711 up onto a pedestal to see if it still worked in square law for the AM modulation. The results were good! I tried with enough bias to give a 250mV (output) pedestal and also a 500mV output pedestal (i.e. quite big in both cases) and in both cases the AM modulation gave exactly 1.2dB change if I removed the pedestal from both readings before wrking out the dB change with the modulation.

Hopefully, this is how my temperature compensated circuit will work. The bias pedestal gets removed as a common mode signal so any changes in bias level due to temperature should cancel out. So I'm quite hopeful that the tempcomp circuit will prove to be the best one. I had a look at loads of instrumentation amplifiers and the favourite would have been the LTC1100 in DIL8 because it has fabulous performance and it comes in an old school DIL 8 package. But it costs about a tenner...

I think I'm going to try the little Maxim MAX4209 as this is only £2. The performance looks to be really good but it is housed in a tiny SMD package. This won't suit a lot of people. However, it might be possible to use one of the lower spec parts in the bigger SO8 package. But I want to try the 4209 first.

The uncompensated circuit is simpler but it only has a chance of working correctly in square law (over temperature) within a tiny window of about 2dB. So it will be very unforgiving if the drive level (in terms of average power) to the diodes isn't within this window.

[Radio Wrangler]

Many sensors use duplicate diode assemblies wth no rF input that are matched to the one in use as a reference for temperature compensation.

That was also done in the 478 bolometer head.

David

[vintage_8bit]

Jeremy. The smallest I've worked with is msop (3mm) and that went very well, I assume the Max 4209 can be obtained in the same package. As I said before I know someone who regularly designs and etches surface mount pcb's and he has offered to help. Colin.

[G0HZU_JMR]

Yes, I think the package size is about 3mm. I've ordered some parts including the MAX4209 and some alternative diodes and I should be able to build and test something in the next few days.

[G0HZU_JMR]

I milled a couple of test PCBs yesterday and I'm in the process of testing one of them today. You can see the PCBs being milled on an old T-Tech 7000S 'Quick Circuit' machine in the image below.

This board is the temp comp version with the MAX4209 instr amp and it supports both the 1N5711 diode package and also a dual diode package from HP/Avago.

Early test results seem quite good. I've tried both diode types and the tiny dual SMD diode is definitely better in terms of thermal tracking. There's a useful improvement in stability over temperature with this circuit.

I'll post up some test results later. At the moment I'm experimenting with different bias levels through the diodes to look for the optimum. Overall this circuit appears to be a bit better than the uncompensated version. More to follow...

[vintage_8bit]

Jeremy. That all sounds promising. Back in an earlier post you said the circuit diagram you had for the generator wasn't very clear. If you or anyone else would like a better copy, try the link that my son has created at ( https://m.**********/a/TzwLL ). Colin