You’re absolutely right to ask.
I’ve not got that far yet - I only have one ferrite to play with, so fingers crossed.
Update - it’s not looking promising. 
I can hear my own distorted audio in my headphones (TX into a dummy load). Have I bought an antenna? Tried my LDO/LiPo setup side-by-side and there were no problems.
Keeping the power bank regardless (my ageing 5Ah bank is rather anaemic and doesn’t support USB-C), but I’ll have to risk taking my LiPo in hand luggage it seems.
Edit: @ZL4NVW
I experimented further today - no hash detected with a dummy load in my office, neither in M7GFJ’s well sorted shack (connected to an external antenna).
But, not only was there a tiny bit of distorted audio feedback in my headphones, TX audio appeared to have a tiny bit of abruptness to it (difficult to describe).
I thought I’d plug my LDO board in-line with the power supply for the heck of it - that got rid of the feedback and cleaned up the TX audio. I expect the same could be achieved with capacitors across the PD cable output.
With power banks being designed for a constant current draw (i.e. charging devices which are probably less fussy about wiggles), I guess it isn’t surprising? Perhaps other power banks are better - I don’t know.
I’ll take this solution with me overseas on this occasion (to avoid being relieved of batteries) - but for use back home, I’m sticking to a 3s LiPo (regulated down to under 12V). 
1 Like
A few days ago, I saw pictures of an exploded BS170 and a self-desoldered switching power supply FET in a QMX.
The maximum source-drain voltage of the BS170 is 60V. With a high SWR and Uss = 12V in class E, this is quickly reached.
This reinforces my decision to use the 9V version with two LiPo cells for more over voltage headroom
73 Chris
1 Like
The QMX appears to use a push-pull Class D output stage (though that could possibly be a P-P-Class E)
The key point is that being a transformer coupled, push-pull, the voltage across the off fet should not be able to go significantly higher than 2-3 x VDD, or lower than -Vdiode, regardless of Zload.
This common assertion of an overvoltage failure mode seems incorrect.
A fundamental issue of class D,E is that the very high efficiency into a nominal load, means a very low nominal dissipation in the output devices, perhaps 0.5 - 1W. Thus Hans can use to-92 small signal fets to get ~5W. Under the same conditions a linear amplifier dissipates 5-10W.
However when you present a severe mismatch, a linear amplifier might dissipate 7-15W, a small difference from 5-10, while the class D/E amplifier can also end up at 5-15W dissipation (from 1). TO92 fets do not survive.
If a Class D/E design uses devices that are only capable of the nominal dissipation, then it when presented with an unfavorable impedance it has to either massively reduce the output power to stay inside the dissipation limit, or blow up. Neither of these is really very satisfactory.
Class D push pull probably fails from excess device dissipation caused by excess current.
Class E can fail as the high efficiency is a product of presenting the correct phase angle (impedance) to get low switching losses. Wrong impedance →switching out of quadrature→ high switching losses.
Single ended Class E can also fail from overvoltage.
2 Likes
The episode of the Ham Radio Workbench with Hans is interesting to listen to as he talks about this amp. They also talk about over voltage.
4 Likes
I agree 100%. The maximum power dissipation of 350mW @ 25°C of the BS170 will quickly be exceeded, with 3W output and 50% efficiency.
The problem of the maximum drain voltage of 60V is an additional issue, as I discovered in the following clip.
I don’t want to presume to offer advice to Hans, considering his enormous creativity.
But perhaps using two IRF510s would be worth considering.
So I simulated the QMX power amplifier using two IRF510 FETs and an HC540 bus driver for the exiter. The whole thing is a Class C design with a 2:3 transformer.
Running at 8V, I got 5W, but a peak voltage of 50V at the drains. That would be too much for the BS170.
More on this in clip 1.
The dangerous peak voltage of 50V can be significantly reduced by adding a 220pF capacitor between the drains.
Clip 2
73 Chris
5 Likes
Interesting video. The spikes speak to the transformer+connection leakage reactance, as the opposite (off) fet should be trying to go below ground if the transformer is perfect. (Perhaps they do if the scope gain was higher?) . I notice that Hans uses external (schottky) clamp diodes D503/4 to stop the slow internal fet diodes conducting.
Perhaps you could try reducing the lead lengths to absolute minimum and using a bifilar (e.g. twisted pair) primary to reduce the leakage reactance, and see what that does to the 50V peaks. Bear in mind that the spike is narrow and perhaps represents ~100MHz i.e. the stray inductance needs to be VHF appropriate in class D not HF appropriate like class B/C.
2 Likes
If this is something you are experimenting with more …
For efficiency, you might like to try increasing the driver HC540 VDD-VSS to 6.5V ie maximum for the driver.
It can also be interesting to try adjusting VSS above 0V. This becomes a tradeoff between getting a higher Vgs (i.e lower Ron and faster Ton, when the fet is on (when VSS is higher), vs not turning the fet off quite as fast (slower Toff when VSS is higher).
This is completely dependent on the fet’s Vgs characteristic, it’s not generic at all. I can be beneficial for “traditional” fets with higher VgsON, and probably no use for low VgsON fets.
As far as the FETs, I was thinking to just try spare mitsubishi 5W RF fets a la FT817/FX4, as they only cost a couple of bucks.
Several of my transmitter products used the wonderful 74AC11004 as drivers, which has centre power pins, so you can run the whole chip single ended well into the VHF. Sadly now an expensive legacy part. Something to bear in mind is that when you use the driver single ended (i.e. all stages go up and down together) then the RF current goes out the VDD pin.
If you run it push pull (i.e half the pins go up while the other half go down) then most of the RF current is cancelled at the power pins. This is far better for end-power-pin IC’s (pins 20,10).
When you are using push-pull at high frequencies, there is a difference between having pins as blocks i.e. pins 18-15 =Q and 14-11=_Q and interleaving them Q=18,16,14,12 _Q=17,15,13,11. Within the chip the rf circulating current loops are smaller when interleaved, but some stray C is smaller as when blocked. It can be worth comparing both arrangements
4 Likes
Interesting demonstration of a push-pull PA and subsequent discussion. How it maintains its RF output up to 28MHz without much drop off is surprising as single IRF510s are notorious for dropping off fairy quickly above 10MHz. I suspect its due to the 74hc540 drive holding up.
As for using gates side by side or interleaved, Ive never thought about the trade offs but it makes sense!
Simon, those RF FETs you refer to are presumably RD16HHF1. Sorry to say they’re no longer a few $ and generally no longer available.
On the topic of T0220 FETs for QRP PAs, has anyone tried FQP13N10?
(Apologies, this content has drifted off the main concern of not cooking a QMX but the root cause of such failures is the designer’s choice of FET PAs).
2 Likes
RD07M family eg RD07MUS2B.
They are 25 or 30V fets, so you really couldn’t be having any spikes…
3 Likes
I like PD because I can charge the powerbank from a solar panel. Whilst not as light as the Powerfilm Lightsaver Max (now very difficult to get hold of in the EU), it can give you a 100% solar setup for trekking.
73 de OE6FEG/G3FEG
1 Like
I agree 100%. T
Hi Chris, very nice video presentation 
From Hans QMX schematic, I think it is a class B push pull as the output transformer is directly connected to load (via a 5 pole LPF)
Class D has a series LC resonator between transformer and load, defined as “Transformer Coupled Voltage Switching Class D Amplifier” or a parallel LC as “Current Switching Class D”.
Pascal VK2IHL
3 Likes
Hans has been working on a firmware update - a PA voltage limit (11.5V default IIRC) that can be adjusted. Looks promising - thread below:
https://groups.io/g/QRPLabs/topic/qmx_voltage/117638199
Input voltage is already flexible (my QMX works fine from 9-12V), but the upper limit might have caused a bit of worry for some.
This might just be the icing on the cake. 
Watching with interest.
PS I foresee reduced demand for voltage-to-noise converters (Aliexpress).
PPS Voltage-to-heat converters are fine… but I’m biased. 
1 Like
I can’t wait to load 1.3.002 when we get back from the Lakes.
It’s great that a diode is no longer needed with 3S LiPo pack.
Also interesting that Hans has tested the new firmware with 14 volts input.
The other interesting feature, is now being able to turn do the power, for use with a PA50 amplifier.
1 Like
Still early days (it’s new, and there are some unknowns), but I’d be happy to use a bareback 3s LiPo based on initial feedback. 
I always put shottky diode. It protects qmx against my mistakes. And protects my wallet against my mistakes for the little price od few hundreds milivolts
Where do you insert these batteries?
1 Like
Something like this… though the wires look bit too skinny for my liking:
Having used a number of 18650 and 26650 cell holders I would strongly advise people against buying the holders using spring terminals. I have seen numerous spring contact holders fail/ become intermittent for all kinds of cells. Whilst it ultimately depends on the max current consumed, people are much better buying the holders with the heavier duty connectors such as these: 1 2 3 4 Cell 18650 Black Battery Holder Storage Box Case for PCB Soldering DIY | eBay UK
6 Likes