C'est Compliqué - Broskie Hybrid Tube Circlotron

NickKUK

Member
2019-12-28 9:16 pm
So it ran - however the BJT has a thermal runaway with the heatsinks being toasty.

screenshot-2022-07-03-at-11-30-34-png.1068979

Result is that I think the LED in one of the HV supplies (let's call that side "A" which is the left) started to toast but the 10R didn't nor did the others. So naturally I thought there must be a short in the BJT. The tube is fine and has been running at 9mA idle. 55R load resistor is fine but that's 5W.

It looks like the BJT is reading 0R between collector and emitter vs the other BJT that reads about 65R.
 

NickKUK

Member
2019-12-28 9:16 pm
Looks like both NPNs are toasted :D The upper is 0R between all the pins and the bottom is not much better. I bought additional BJTs and mosfets for this reason.

I stand corrected the LED is fine (cut the BJT pins and powered up). I did a calculation it's 30mA max would have required 500V which would have blown the 350V caps. I noted something sparked but unsure what that was or was my imagination. All the power supplies have a big MURS diode across them to prevent voltages from reversing should they power down at different rates.

Now it would be dumb not to learn from the experience. Looking at the data sheet I think the 68R is using the bias too hard (44 or 50R may be better) but also there's a few options for temperature compensation using feedback and dropping the bias based on the current increase. The DC feedback should, in theory, manage a DC offset by chancing the tube bias - that would essentially feedback the NPN bias in the event that heat was increasing current. A better bet may be adding a bias temperature compensation based on the current.

On starting the amp I did note that both sides were sitting at -31Vdc on each side of the the 55R Load. A very different position for the modelled amp and without the BJT and LV power installed. I think the next step is to look at this - no BJT but run with the LV supply in place. I've been very careful with the AC and rectification to ensure all the supplies share the same phasing so any residual ripple is common.
The two sides are floating hence they will find their own equilibrium with the ground points being 1K away from the earth point.

i’ll investigate later.
 

NickKUK

Member
2019-12-28 9:16 pm
So I've left the BJTs out of the equation and started to focus on simply the valve and the power supplies. The LV supplies don't have the 1uF caps nor the 220nF X cap on the rectifier side plus the entire amp doesn't have snubbers - hence the noise spread from the AC diodes:

SDS00156.png


So the new AC top enclosure has helped reduce the noise. The remaining noise should be hit by the snubbers etc. I'm still getting a 40dB 100Hz mains noise:
SDS00158.png




Measuring the voltages without the BJTs in place shows the almost the same voltages seen when the BLTs are operational:
Screenshot 2022-07-31 at 20.36.07.png

The tubes are sitting at -5.5V grid bias and about 10mA which is fine however the mosfets don't seem to be doing much but the voltages are all sat floating around the -24 to -30Vdc mark (at the Load), That differs completely to LTSpice. The only thing I think could happen is that the pads I'm using for BJT thermal connection to the heatsinks is grounding the amp some how causing the negative offset. Let me cut the remaining collector pins (which the back of the NPN is a collector pin 4).

Removing the BJTs from the grounded heatsink the FFT changes to this:
SDS00159.png


Note the missing 100Hz noise.. Holmes Interesting Dr Watson

Also the offset the LV power supplies are now sat at 0V ref with a 32.5V positive (not surprising as the BJT is hanging in the air without any over connections to the positive except the HV negative. the HV is now sat at 200V (ie HV+LV) with the tubes idling at 10mA. The 55R load is now sees 0V on either side.

So the BJTs were grounding (at least enough to cause a 2V drop) which drew enough to re-reference the entire load to -32Vdc! I'd removed the BJTs by cutting and separating the emitter and base pins leaving the collector connected to the LV power positive. That means the only physical connections were the collector and the back of the BJT which is a collector pad.

So that's was fun - it may also indicate why I had a runaway. I need new non- electrically conductive thermal pads. I used a similar series of pads with the MF A220 but the BUZ package is plastic and the pads are meant to be non-electrically conductive unless they break down when cooked or have some microscopic metallic particles puncturing the pad.

The amp now seems to be behaving like the LTSpice model. I was touching the heatsink and chasing over the time and didn't feel a thing (30V would have been felt) so that care doing the grounding was working nicely.
 
Last edited:

NickKUK

Member
2019-12-28 9:16 pm
I used 1200 grit emery paper to sand the heatsink mounting sites, used larger pads, tested for non-continuity between the HS and collector and then after the first run I retightened the screws.

I have a voltmeter across one of the 1R LV power supply resistors which tells me the current the NPNs are pulling. Derating for 30V at higher temps this drops from 4A to 200mA for DC operation. So it's useful to watch as the amp's NPNs heat up so the current draw increase slowly up 180mA at which point I cut the power. A cool draught through causes the NPNs to cool and the power draw drops.

Originally this was designed for more bias current but my concern now is that would result in hot heatsinks and thus we have a thermal runaway again. Reducing the operating point reduces the power output of the amp and so it looks like a different NPN is in order with less derating etc. Also if the input wave form is large enough currently it could hit Vbe(on) which wouldn't be a pleasant experience.

There's also about a 900mV offset across the load which could be due to the tube so a small trimmer may be needed to balance out the tube bias.

So I'm happy this is running - the PSUs seem really good (at least unregulated etc), just my selection of NPN sucks :D

So in the next weeks I will look at NPNs that would fit the bill. Although I'm modelling a lower NPN bias point I think they will not work well with the tube cathode peak outputs.
 

NickKUK

Member
2019-12-28 9:16 pm
This shows the ripple noise cancellation - the upper and bottom traces are either side of the load, the maths (white trace) shows the difference:
SDS00160.png



Part of the thermal issue is that the LV power supply actually operates at 32Vdc and not 24Vdc which reduces the SOA below what's needed. So burning about 4W with a resistor and LM317 would work to reduce the noise and lower the the thermal issue back to the model's 24V.

The same is true with the HV although a little more complicated.
 

NickKUK

Member
2019-12-28 9:16 pm
So the bode plot - With some tightening of bolts it can typically last long enough to get a bode plot or two before we start getting close to 190-200mA which I've set as my SOA max for now.

SDS00187.png


That's one side of the channel as I can't bode plot the difference of two probes. That's a pretty flat curve at -0.7dB across 10Hz to 20KHz. Also note that with a balanced source, the output would be louder still. (also note no snubber present currently with just the RCRC filters)

I had a listen using the test speaker in parallel with the 55R load (Ie it's driving 26R load :D). There's still mains hum and some DC offset but the sound is good. I think the next step is to target the heating issue and the DC offset.

The heating issue is starting to calm down a little - it takes longer for the NPNs to heat up. The phase change compound required 60degC to sort itself out and the MJE243G has a max of 150degC. The issue here is the SOA for 30Vdc is 200mA for dc operation which may not allow the temperature for long enough to phase change.
I'm thinking a LM317 to create a 20V regulated load would be good at that point the SOA can deal with 600mA max for DC operation. That would also reduce the mains hum noise.

Next up would be the HV supply with a 317 although it would drop the voltage a little we have enough for the 12BH7A-STR to produce a good sound.

The 317 would be cheaper in the initial stages compared to to a 3075 for this initial phase given I need a voltage regulator for each of the floating power supplies.
 
Last edited:

NickKUK

Member
2019-12-28 9:16 pm
I've started to think more about targeting the ripple with some Maida LM317 style regulators. This is my initial design I'm just starting to model. The downside is that L2 mosfet is £30 a piece (RDSon is in the mOhm), however for modelling this will do as the model seems relatively well behaved.

Screenshot 2022-08-06 at 08.27.17.png


I'll see how the LTSpice FFT turns out. The current source I think will be altered to be a simple CCS running 5-10mA for efficiency.
 

NickKUK

Member
2019-12-28 9:16 pm
I've almost got my maida working - it's been a learning experience.

With different components, the required minimum operating voltage drop across each component varies which changes all the values. Also as we only have a small headphone sized voltage swing and a RCRC ahead of the regulator the amount of voltage drop doesn't need to be too high. The main drops are 4.5V min for the mosfet and 3V for the LM317.

The top UF diode prevents reverse polarity across the regulator.

R43 (or a CCS) ensures that the zener D37 has enough current to operate. It also sets the position with that zener of the bubble that the regulator sits in under the input voltage. (you'll also need to sort out the voltage reference divider R63,R64 and R66 for the output voltage to match up within that bubble). The diode D73 should be specced to operate at 10-25% of it's rating. I can see this being requiring a bit of testing with the physical diode.
On startup and a short R43 sees current and the modelling shows this - on a short at full voltage that's going to be quite a current.
R60 is a grid stopper, to filter out the high frequency noise.
D47 is a Vgs positive safety - a LTSpice current source starts at power on and therefore provides a positive base voltage initially. I'm modelling this as a belt and braces (I'll add another across the mosfet source and drain).
D35 pulls the Vgs down to limit on short (startup too)

Mosfet - 4.5V starting voltage, a range of ~5Vgs to 9Vgs means we can use R58 as a voltage regulator, currently this is set to 10R. The mosfet has a <100mRon. It's a 500V 60A piece but for now it will do for modelling given it's hellishly expensive.
So we need at least 4.5V (including any audio noise) across the mosfet.

The LM317 sits in a 10V bubble created by the maximum voltage difference between Vin and Vout set by D36.
There is a stability 1uF cap on the Vout and a bypass cap C14, to stabilise the regulation. For this reason the Vadj needs to protected - this why we have D39 so when the system shuts down C14 does remain charged at high voltage - destroying the LM317. It should be said the same occurs for C16 but D36 also provides protective draining for that.
The voltage divider R36, R64 and R66 (variable) provide the output reference at the required Vadj = Vout-1.25Vdc. Things to note a current source provides the internal LM317 1.25V voltage reference. This is only reliable if the LM317 is on thus the 10uF is charged by the ADJ output and the current coming across R63.

The model below still has some flaws - namely the voltage drop out of the LM317 isn't correct yet but I wanted to show the results so far from LTSpice:

Screenshot 2022-08-07 at 09.10.05.png


Below the drop out voltage impact - so I need to short out the voltage:
Screenshot 2022-08-07 at 09.52.05.png
 

NickKUK

Member
2019-12-28 9:16 pm
So with zero signal, the FFT now looks like this in LTSpice:

Screenshot 2022-08-09 at 15.05.41.png

That's better with the V(b)-V(a) rather than I(R1). Now both the mosfet and LM317 are operating when you look at the sim. That's without the current source. It does look like I'm seeing some diode switch harmonic noise still.

Also some good news this morning - I've been offered a new job which should help this move forward a little.