[Amps] Tubes, transistors, and 'abuse'

Sun Apr 23 16:07:29 EDT 2017

Cathy,

> I'd welcome your opinion on the design.

I will provide it in this mail.

> Mine looks to be this design:
> 
> http://www.kg9e.net/projects/hfpacker/
> 
> It may be telling that K5OOR has resigned the amp since my kit was
> built. The current one being sold appears to be physically larger than mine.
> 
> Documentation:
> 
> https://storage.googleapis.com/wzukusers/user-17403798/documents/58c313c4618a4lexQCfw/Construction%20Manual%20miniHFPA%20with%20schematics%20and%20BOM%202017%2003%2010.pdf

Okay. Since there isn't full info on the version you have, I will base 
my analysis on the second version, and you have to see how much of it 
applies to the amplifier you have.

While this is a tiny amplifier, it's affected by some of the same 
problems that afflict legal limit amps. So the discussion might be 
interesting to other forum members too. Really interested readers should 
download the info from Cathy's links, to see what we are talking about.

So, this is basically a broadband push-pull amplifier using two IRF510 
MOSFETs in class AB, powered from 29V obtained from 12V via a DC-DC 
converter, to deliver roughly 20 to 45W RF output, depending on the 
band. I will comment about it in two aspects: One is based on the specs 
and data provided by the designer, the other is a partial explanation of 
why the results are so poor.

The designer gives a table of output power and current consumption, on 
each band. The worst case is the 17m band, where the amplifier delivers 
only 29W output for 8.2A consumption at 12V. That's a total input power 
of 98.4W! Of course we have to subtract the power lost in the DC-DC 
converter, and the power used in auxiliary circuitry, but that won't be 
much. Probably around 85W are going into the amplifier proper. 29W 
output is then a 34% efficiency, and leaves 56W to be dissipated in the 
MOSFETs. The power loss in the transformers and low pass filters is 
likely to be irrelevant.

56W dissipation is 28W in each IRF510. These FETs are rated for 175°C 
maximum junction temperature, and 3.5 K/W internal thermal resistance. 
The FETs are mounted on a heatsink that has a thermal resistance of 
around 2.5 K/W at the working conditions of this amplifier (without a 
fan). Each FET is mounted on a ceramic insulator that has 0.86 K/W 
thermal resistance, plus the two greased interfaces, which have about 
0.1 K/W each if well done with high quality thermal compound.

I will disregard the thermal resistance of lateral conduction through 
the heatsink base plate, because it should be small in this amplifier.

So the total thermal resistance we have, for both FETs together, is

2.5+(0.1+0.86+0.1+3.5)/2 = 4.78 K/W

Thus when transmitting a full power carrier on 18 meters for a time long 
enough to reach final temperature, the MOSFET junctions will be 4.78*56 
= 268°C hotter than the ambient air!!! Needless to say that they will 
burn out far earlier.

Of course it's not realistic to transmit a full power carrier for 
several minutes, unless transmitting an RTTY bulletin. But even while 
the heatsink is stone cold, the junctions will very quickly heat up to 
around 150°C above ambient as soon as full power is being transmitted on 
that band, and this places the MOSFETS at their absolute maximum limit 
temperature - and from there they will keep heating while operation 
continues and the heatsink slowly warms up! This is truly a programmed 
disaster.

Now this is for the worst-case band. On the other hand it's for 1:1 SWR! 
At a typical, practical SWR, there will be more heating. Even on a 
"benign" band, the MOSFETs will overheat in plain normal operation.

In short, Cathy, your poor IRF510 MOSFETs are being fried by this 
amplifier. The failures are caused by simple and plain overheating due 
to excessive power dissipation, which comes from a combination of poor 
efficiency and excessive output power demand from these small MOSFETs.

And now let's get to the second part of the analysis: Why is the 
efficiency so low? Calculating it, from data provided by the designer 
and estimating the power used in auxiliary circuitry, the drain 
efficiency of this amplifier is roughly:

160m: 60%
80m: 57%
40m: 50%
30m: 41%
20m: 48%
17m: 34%
15m: 44%
12m: 45%
10m: 44%

Instead a good class AB amp should have an effiency of around 65%, but 
very few achieve this.

17m in this amp is an anomaly, probably caused by some resonance in the 
output circuit that causes an extremely disfavorable phase ratio between 
voltage and current. But the other bands show a general tendency: Almost 
normal, correct efficiency on 160m, tapering down to sub-50% by 30m and 
staying roughly constant. Many solid state push pull amps do the exact 
same. And the reason is quite simple:

A push pull amplifier needs very tight coupling between the two drains, 
up into the harmonics of the highest operation frequency, to achieve 
correct class AB operation. In class A this good coupling isn't 
required, nor is it required in saturated (near square wave) operation. 
But in linear class AB operation it's essential. Without this tight 
coupling, around the zero crossings the two FETs together cannot conduct 
the full total drain current forced in by the feed chokes, resulting in 
high voltage peaks on the drains. These either just cause added losses 
through additional charging and lossy discharging of the drain 
capacitances, while also driving the FETs into a sort of dynamic class A 
operation through drain-gate feedback, which happens through the FET's 
drain-gate capacitance even if there is no external feedback path; or 
the peaks get so high that they make the FETs enter avalanche breakdown, 
sharply increasing power dissipation.

The required tight drain-to-drain coupling can either be implemented via 
a center-tapped output transformer, but it MUST be a real, magnetically 
coupled center tap, _not_ a half-turn tap; or a bifiliar feed choke must 
be used. This amplifier uses the latter option. But unfortunately the 
bifilar choke used isn't up to the task, and judging from the published 
text, the designer hasn't fully understood the requirements for this choke.

The choke he used has 10 bifiliar turns on 2 stacked FT-50-43 cores. 
This must give roughly 100µH per side, which is FAR more inductance than 
needed. This much inductance wouldn't really hurt here, but the problem 
is that along with the high inductance comes a significant leakage 
inductance. And the leakage inductance acts like two independent 
(uncoupled) chokes in series with the (coupled) bifiliar choke, 
destroying the coupling between the FET drains if this leakage 
inductance is too high.

I would have to build that bifiliar choke to measure the actual leakage 
inductance, and I'm too lazy now to do that. But judging from 
experience, I would expect around 0.5µH leakage inductance per side. 
That's an impedance from 6 ohm on 160 meters, up to 91 ohm on 10 meters. 
Meanwhile the drain load impedance in this amplifier is 11 ohm per side. 
So the coupling between drains is modest on 160m, poor on 80m, and gets 
worse on higher bands, to the point that from 30m upwards there is 
essentially no drain-to-drain coupling. And this is the fundamental 
explanation for the poor efficiency of this amplifier! The bifiliar 
choke would need to be re-designed so that it produces tight coupling 
between the two drains, all the way into VHF, to cover at least the 
lower harmonics of 10m. At 11 ohm drain impedance this can be done, 
although it's not really easy. Instead with a high power LDMOSFET amp, 
or a 100W 12V powered amp, both of which have drain load impedances 
around 1 ohm, it cannot be done, to the best of my present knowledge - 
and that's the explanation for the "typical" 45% efficiency of most HF 
push-pull broadband amps we see!

Other than the problem with the choke, I dislike the fact that this 
amplifier has a highly reactive, insufficiently damped gate drive 
circuit. This might cause self-oscillation outside the HF range. It 
looks like the designer tried to resonate the gates in the center of the 
HF spectrum. A better approach is usually to swamp the MOSFET gate 
capacitances with low value resistors, then apply drive through 
additional low value resistors. This appoach gives much flatter 
response. It costs some drive power, but then this amp has a big 
attenuator at the input! Instead of using that attenuator, it would have 
been wiser to place that attenuation directly in the gate circuitry to 
get a flatter response and reduce the chances for self-oscillation.

And some amount of negative feedback would further enhance stability, 
further even out the frequency response, and improve the linearity.

Satisfied with the analysis, Cathy?  Any questions, comments, anyone?

Manfred

========================
Visit my hobby homepage!
http://ludens.cl
========================