On the clean PA/exciter thing, I have a prototype all HF band transmitter on the bench downstairs that a 2 tone IMD3 of -63dB ref one tone (-69 ref PEP) @ 250W, via cartesian envelope feedback. It still needs the output filter board designing and building, but that is just a case of winding 20 or so large toroids.

The driver and PA stages are actually very conventional for the most part, RD16/BLW50F/BLW96 pairs, held to about 10db per stage with local feedback, with the overall envelope feedback error amplifier pair having 40db of gain, which, when you close the loop, makes the distortion just vanish.
Holding the stage gains down with local feedback means I need about 2 more gain stages then I would if I ran the things wide open, but  means that even open loop the performance is very much better then a lot of current commercial offerings.

The complexity is adding a couple of extra DDS chips, some ultra low noise opamps and a few more mixers, but I think it is worth it. 

My current project is to improve the efficiency of the PA by playing games with the supply voltage and bias using feed forward control from the DSP that produces the modulation signals. If I can get the average transmit power efficiency up far enough then the savings in heatsink metal may pay for the cartesian loop electronics.

One day I might write it up for Radcom/QEX/Dubus or just possibly just for the web.

Regards, Dan.

Hi all,
There is a trick used by AM broadcasters in increase mean modulation levels by using an allpass filter to cause a change in the group delay of the audio chain at about 1Khz which reduces the asymmetry common to the human voice waveform and thus allows an increase in drive to the clipper without increasing the actual amount of clipping.

As the DSP for speech processing seems to be seldom discussed, I was wondering if anyone know of a case of this trick being used in amateur service SSB rigs.

In fact a more general discussion of DSP speech processing may well be interesting, as it seems to me that increasing average mod levels by a few dB could well be more cost effective then beefing up the PAs and power supplies by the same amount.

Regards, Dan.

Hi all,
Just wondering how folks here would play this one.

I have a prototype cartesian loop transmitter that is currently doing ~100W on all the HF bands with IMD3 better then -63dB ref one tone (-69 ref PEP), which is very cool, but the transmitter unfortunately is not, it runs only about 50% efficiency at full output and rather less as the power output falls.

Now two things occur to me:
Firstly that -63db is a stupidly good IMD3 spec, and is probably better then the in channel IMD of most receivers, so it might be worth trading maybe as much as 10dB of that performance for less heat?
Secondly that actually with typical SSB modulation having a crest factor of at least 6db (with the processor set to stun), it might be worth trying to optimize efficiency at some level less then full output? 

I have two notions that I would like some feedback on:

1: As long as the drive is within limits, it might make sense to actually drop the finals standing bias as the transmitted power increases so as to move toward class C operation at medium modulation levels (and back to AB at low power, or towards  B at high power to keep the drive required reasonable).
This does of course hurt linearity, but as long as the change is slow compared to the feedback systems loop filter the feedback will adjust the drive power to suit.

2: I can rig the finals power supply to support being ramped up and down to vary the drain supply, by delaying the I,Q pair to the modulator by a few ms, and sending sqrt (I^2 + Q^2) to the power supply board. This supply is a switchmode affair, and doing this will provide a better drain match at all modulation levels and so reduce the heat in the finals. Sort of like a polar loop, but with the actual feedback being cartesian, with the power supply modulation being open loop.

Ideally I would like to manage ~70% efficiency while retaining two tone IMD at a level I would consider respectable, as that would reduce the heatsink requirements by enough to pay for the extra electronics in a rig at the few hundred watt level.

Comments on the pros and cons of the above heretical notions WRT the PA strip would be appreciated from those who have experience with semiconductor PA strips.

Regards, Dan.

True, but for some of us building the gear is its own reward.

Ham radio is a broad church and while I personally would rather spend hours tweaking something I designed and built then spending the same time sat in front of a FTdx5000 or whatever, it takes all sorts.

73 Dan.

Well you can knock electrons loose from all those materials, the key phrase is 'work function', but so what, a few dozen, or even a few hundred electrons is an utterly negligible amount of current in this context, and even when accelerated across a few KV, does not amount to a meaningful amount of energy compared to that stored in the tank circuit in normal operation (or even that in the VHF resonance in normal operation).

The grid has capacitance, ergo current MUST flow to change the grid voltage, dV=dQ/C. I have already demonstrated that the grid capacitance is large enough to make the charge on any reasonable number of photoelectrons that could be freed utterly irrelevant.

Further in standby the tube is biased well off, the grid would have to become quite a lot more positive to allow gain, and would have to stay there for long enough for oscillation to build up. If the amp was unstable at VHF it would show in normal transmit mode way before the instability got sufficiently severe (I don't think it could) to show with the tube based off.

I am out of here.

73 Dan.

But nowhere near enough to be in any way significant without gas to provide for avalanche breakdown......

Do the math already, this is not hard to demonstrate, dV=dQ/C, grid capacitance and the charge on the electron, work it from there, not hard. This is high school level physics, not exactly university level stuff (I wish this forum software supported LaTEX markup).

Hell getting a non avalanche breakdown tube to work as a particle detector takes MAJOR effort with the preamplifier and that is with a tube designed for proportional counting, look at the pain folks have getting helium three neutron detector tubes to work without excess noise.

You really need to look to ordinary physics here, think forces between current carrying conductors, magnetorestriction, that kind of stuff, not weird shit happening in the switching device (Be it conventional tube, spark gap, thyratron or large solid state switch), bangs in high energy electronics are not always the result of a arc occurring where it can be directly seen or heard.

Regards, Dan.

Wait, think this through....

Disconnecting the TR line makes the problem go away?
So the problem is NOT in the amplifier, it is in your TR wiring!
In particular I would guess the cable is shorted and is jamming the amplifier into transmit mode.

Check the TR cable with an ohmmeter and check the rigs TR switching works the way you expect.

Regards, Dan.

Well you can get an entire vector network analyzer for less then that, so I would be surprised if it was more then a few hundred bucks for just a single port TDR capable rig.

Regards, Dan.

Not directly, but I could see magnetorestriction in the wiring and possibly even the power transformer making a bang as the current pulse sets up fields between conductors.

I have actually heard this effect in pulsed power applications, the intake room had a sort of ghostly impossible to locate ticking sound coming from everywhere as the wiring in the conduits reacted to fast risetime pulses.

Regards, Dan.

Time domain reflectometry will identify where any impedance discontinuities are for you.
Usually easier just to examine the cable closely, problems are often visible.

If you need to go to TDR, use a scope, a signal generator with a 50 ohm output and a T connector to connect the scope, generator and one end of the coax.

Set the generator to make a square wave at a volt or so, and a few hundred Khz, and open the far end of the coax. Set the scope to trigger on the falling edge and adjust until you can see the reflection from the open end coming back down the cable on the scope screen.
Verify that you are seeing the reflection from the far end by shorting the far end and making sure the reflection is in the same place but with different phase.
The other reflections you can see earlier on the screen are now your damage, scale appropriately to find the distance.

If the cable is open circuit then you will not be able to use the open/short method to calibrate things, in which case assume a velocity factor of 0.66 (solid dielectric) or  0.85 (foam or air spaced), and calculate the distance to the discontinuity that way, remembering that the timing on the scope screen is round trip!).

Me I would just cut it in half, you will stand a good chance of getting one good half, then cut the dud in half, giving you also a good 1/4 section, then cut the dud quarter in half..... Keep going until you have all the various short jumper cables you will ever need, or until the remains get too tiny.

TDR is only worth doing if it is LDF450 or something equally expensive, or the thing is miles long and difficult to get out.

Regards, Dan.

 

Ok so I have been doing too much with fets lately....
The math still applies to the grid just the same as it did to the fet gate.

Faulty tube/socket/insulation failure/bias supply fault, something of that type would be my guess.

When fault finding assume the simple faults until you really do need to introduce weird physics (the second type of fault are **RARE**), 99.99%  of the time even the weird ones turn out to have a (with hindsight) trivial cause.

Also, it is often a bad idea to take user reports at face value, "Nothing has changed" then after a week of headscratching it turns out that "Nothing" includes a new PTT cable or upgrading the software or whatever. Been there, the judge ruled it justifiable homicide...... 

Regards, Dan.

Work the numbers, gate capacitance, charge on an electron... delta V on the gate.

Lets assume a 1pf gate/cathode capacitance (more makes the effect smaller), then knocking an electron off and having it accelerate to a large distance will change the grid voltage by delta V = delta Q /  C = 10^-19 / 10^-12 = 10^-7 V  (Working to orders of magnitude).
It takes 10^7 electrons to change the bias by a volt given a completely open circuit grid @ 1pf...... The only way you are getting that is if the tube avalanches, and if there is enough gas for that you have worse problems (Or you have a hunk of Cs137 on the shelf behind the amp in which case you also have worse problems).
Interesting, but I would be looking elsewhere.

Regards, Dan.

The GM tubes have a ***very*** fine wire used for the anode, so there is a very high electric field in the vicinity.

Thus an electron released at the cathode by the photoelectric effect or just a beta wondering into the tube at low energy will have a high probability of being accelerated to an energy sufficient to trigger an avalanche in the tube. Incidentally this is why GM tubes are poor for spectrographic studies, the pulse height is largely independent of the energy of the particle.

A 3-500Z or whatever by contrast has a heated cathode that runs (normally) in a permanent cloud of electrons, with the grid screening them from reaching the anode. There is nothing like the electric field strength and nothing like enough gas (except in some modern tubes ) to cause avalanche breakdown. 

Consider a electron being knocked off the grid by a cosmic ray, well given the grid capacitance and the size of the charge on the electron you can work out the change in voltage on the grid (hint is is small). The same applies to an electron knocked off a gas molecule as long as there is insufficient gas and electric field strength to cause an avalanche discharge (Which is a tube design consideration).

Tube breakdown in a sufficiently gassy tube might be triggered by a cosmic ray I suppose, but realistically that tube would fail in short order anyway if it is that gassy..
 
I don't really buy it as a significant effect.

Regards, Dan.

I am pretty sure that I have NEVER seen an amateur HF amp that met state-of-the-art signal quality (And I include the stuff I design and build myself in that).
Yep tuning the receiver 5-10K up the band is a far more useful test then anything in channel. 
Of course the fact that the US considers 43dBc harmonic suppression acceptable at any power level does not help (what happened to the + 10 log P term?)
I have never understood the instinctive hatred of CB amps from US hams, they make a reasonable source of bits if you can get them cheap enough, and the CB market does much to keep things like modest HF power devices available (What do you think the RD16HHF1 is intended for)?

That thing in fact looks like an excellent starter kit for building a modest single band valve linear, it needs a few meters, and some new electronics, but the metalwork and power supply are probably reusable, and I can think of worse ways to get into valve homebrew.

Regards, Dan.

The AM class E  guys seem to manage switchers that cope with full modulation bandwidth with low distortion, and they do not in general use envelope feedback.
Now quite possibly for my use case, something even cruder would do as a little too much DC will just result in lower drive power as the loop reduces the drive to maintain the correct output envelope, less efficiency, but it is still going to be an improvement on a fixed 50V rail when the modulation only needs 10W of output.

I am well aware of the extreme linearity requirements for the reverse path, any non linearity here gets converted directly into distortion in the output, the output frequency mixer is a H mode job, and the final down conversion mixer is a video switch, mixer IP3 is just not the issue it once was.

Broadband noise in the error amplifiers is a problem, and picking the pole frequencies for th e loop filter is something of a tradeoff, fortunately there are some VERY quiet opamps available these days for low source impedances.

40dB of feedback cures most ills!

Agreed that there is something nice about big grounded gate glass fet amplifiers, but for modest power levels I think SS wins these days.

Regards, Dan.