SMPS for vacuum tube power amplifiers.(status: back at it)

[Circlotron]

On your schematic the TL494 outputs are open emitter so you need to put a pull down resistor. Probably something like 470R, someone could suggest a value?

[T3sl4co1l]

Gotta have a pull-down for that to work!  The "emitter" outputs are just that, pull-ups only (or, well, pulling to the respective "collector" for the most part, but as shown, yeah).

Flyback isn't very good for TL494, and especially not voltage mode.  Why not another equally classic, easy-to-use controller like UC3842?

Tim

[SK_Caterpilar_SK]

Gotta have a pull-down for that to work!  The "emitter" outputs are just that, pull-ups only (or, well, pulling to the respective "collector" for the most part, but as shown, yeah).

Flyback isn't very good for TL494, and especially not voltage mode.  Why not another equally classic, easy-to-use controller like UC3842?

Tim

I did actually make a SMPS based on the UC3842, but the output was rather noisy, very noisy electrically and audiably. I think it is due to my design but for the intended use it was not a big deal. With the TL494 I had so far only positives compared to everything else. I dont really like the 3842 or any 384x because I dont really see the function of a current mode controller. Its also non ideal for light loads. At least thats what I can tell from my findings.

I SAID ITS NOT COMPLETE Yeah I gotta look at it a couple more times before im sure that everythings as suposed.. Would 3k be enough for the pulldown?

[T3sl4co1l]

The fundamental reason for current mode, is because the inductor's state variable is current.

Control the current, and you never have to worry about exploding the switch, overheating anything (maybe), or fusing the input or output, because current is limited, period.

Then, you control that current, to regulate voltage.  There's no LC filter (as in a voltage mode forward converter) to make your loop impossible to stabilize, compensation is straightforward.  Instead, the L is in the current loop, and the current loop and C is in the voltage loop.

The biggest drawback to voltage mode controls is exploding transistors due to unlimited transient response.

Tim

[SK_Caterpilar_SK]

The fundamental reason for current mode, is because the inductor's state variable is current.

Control the current, and you never have to worry about exploding the switch, overheating anything (maybe), or fusing the input or output, because current is limited, period.

Then, you control that current, to regulate voltage.  There's no LC filter (as in a voltage mode forward converter) to make your loop impossible to stabilize, compensation is straightforward.  Instead, the L is in the current loop, and the current loop and C is in the voltage loop.

The biggest drawback to voltage mode controls is exploding transistors due to unlimited transient response.

Tim

"The biggest drawback to voltage mode controls is exploding transistors due to unlimited transient response."
That statment makes sense...actually its what im experiencing with my converter. I threw out the 3842 as a no go because I have used it in the past and wasnt really the best ever I seen. I have blown so far 5 fets on my pre pre pre prototype converter. It was getting as hot as the blazes of hell maybe even hotter (MrCarlson). Then that got fixed by puting it on a beefy heatsink and still died after a couple hours. The only transistor that survived all so far is the G30N60. IGBT. And that does heat up but but not so dramatically and id rather blame my PCB for that heating than anything.

Im also not as familiar with the  UC3842 as with the TL494 but I guess a look at the datasheet can change a lot of things withing a couple of minutes. The biggest problem I see with the UC3842 is current function. This may not be a problem with the mains powered PSU, but it sure is with the 12-24 range. There are simply too high currents for a standard resistor to sense the current. I supose I could calculate resistance to get 70W max out of the controller and get the resitance and create a resistor on the PCB by calculating the trace resistance to create a quite precise shunt. Just have to remmeber not to solder coat it . I dont have a lot of the same switching transformers so I can make only 3 more prototypes, one is the TL494 one other one could be the UC3842 and I can decide wheter to fit it with the GA3459 that has a slightly higher inductance and get 20A peak currents at 50kHz at 5V or the GA3460 that has a lower inductance and gets 50A of peak current at 50kHz at 12V.

Yeah I will look into the UC3842 and try that one as well.

One thing I cant understand at first glance is a lot of the schematics I see on the interwebz like the one I attached dont have direct feedback. Its totally feedbackless? I guess it knows the voltages by the sensing the transient? That would be my only guess. And the losses within the transformer are low enought to not disturb any other rails?

Now seeing how it kind of works, I can see what the problem could have been with my original design. Tho I was 14 at the timne of designing that 10-24 input to input voltage-60V 50W converter  . It has a lot of flaws xD , but interestingly I do remeber that the transistor was not heating as much as the diode did, so the fet did give a damn unlike the diode, that was burning hot without a heatsing and even with the heatsink it gets to a modest temp. Im getting off track  .

PS: I just noticed one thing in that schematic, the voltage is hard set I guess? How can I vary the voltage because the output should be settable anywhere between 240V up to 500V?

[SK_Caterpilar_SK]

Alright, so I went trough some Flyback controller datasheets, new ones old ones, and I figured out how the feedback works on the schematic has no optocoupler. Very interesting indeed how it works using the aux winding to power the chip and sense the voltage at the output at the same time. But sadly that is not an option for me since I have a transformer that has no aux windings HOWEVER, I found a lot of great controllers like the LM5022. Requires a miniscule amounth of external parts, its a low side controller specifically for boost applications and it does SEPIC and Flyback controller. It has an extremely wide boltage operating range (6-60V!) and it really does all the things by itself. All I would need to do is to determine the CS resistor, frequency, the feedback and a really stable and simple SMPS is born..hopefully..We all know its never that simple  .

Looking at TIs library collection of flyback controllers, there isnt really a need to go with old classics other than if it costs a lot more than the classic, but really the features on these things should be better overall.

So far these are the chips I have checked out:
LM5155
LM5022
And those seem to be the best I can get for my specific needs. The schematics include easy Opto isolation options too or go non insulated, but the application I could see that in is when you power the converter from low voltage inputs like a PC powersupply which is known safe. But the big deal with PC powersupplies is that the output ground is mains ground reffernced, so if you connect a computer to it it automatically becomes a ground loop and there is no point of trying to supress that, its way easier to just isolate the input from the output so I might just go with either the LM5155 or 5022. First I have to make sure if its even available for a regular person like me. Great i searched for it because both these chips can do me good with the direct mains input and also the low voltage input converter.

(Datasheets)
LM5155
http://www.ti.com/lit/ds/symlink/lm5155-q1.pdf

LM5022
http://www.ti.com/lit/ds/symlink/lm5022-q1.pdf

[oPossum]

One thing I cant understand at first glance is a lot of the schematics I see on the interwebz like the one I attached dont have direct feedback. Its totally feedbackless?

The winding used to power the UC3842 also provides feedback. The various output voltages are mostly a matter of turns ratio.

[SK_Caterpilar_SK]

One thing I cant understand at first glance is a lot of the schematics I see on the interwebz like the one I attached dont have direct feedback. Its totally feedbackless?

The winding used to power the UC3842 also provides feedback. The various output voltages are mostly a matter of turns ratio.

Yeah I just said that in my last post before this one, thanks for pointinf out tho, if you could explain me how it works a little more in depth.

[T3sl4co1l]

"The biggest drawback to voltage mode controls is exploding transistors due to unlimited transient response."
That statment makes sense...actually its what im experiencing with my converter. I threw out the 3842 as a no go because I have used it in the past and wasnt really the best ever I seen. I have blown so far 5 fets on my pre pre pre prototype converter. It was getting as hot as the blazes of hell maybe even hotter (MrCarlson). Then that got fixed by puting it on a beefy heatsink and still died after a couple hours. The only transistor that survived all so far is the G30N60. IGBT. And that does heat up but but not so dramatically and id rather blame my PCB for that heating than anything.

You were probably going the wrong direction, leading to positive feedback of frustration.

Bigger transistors have more capacitance = more switching loss, both in the controller/driver and the transistor.  IGBTs are also only effective at higher voltages.  If this was another low-voltage application, that would be a problem!

You can of course mis-design a current-mode switcher to explode.  It's a much more straightforward design process to design one correctly, in comparison to a voltage-mode control that needs several hacks (soft start, current fault, lead-lag compensation with a zero in the filter) to behave acceptably.

(Of those, you're already familiar with soft start; it is what it says.  This ramps up PWM slowly, which works when the output is not shorted and when the input is applied suddenly (the usual case).  This fails if the output is overloaded or shorted, and this also fails if the input drops momentarily, or a transient load is applied.  For that case, a current fault can detect high peak currents (usually with a current transformer and comparator), and retard or stop switching, and maybe restart from a soft-start condition.  PITA if your load was expecting stable voltage under transient load, but whatever.  The compensation is jargon for an output filter that has ESR as a critical part of its behavior (typically somewhat lossy electrolytic capacitors are used, for their useful ESR), and the error amp has an R+C across both feedback resistors, to give a slight edge in phase margin near cutoff.  Effectively, we try to speed up the error amp's response just a little, but also shelf its response just right, to avoid oscillation.  Well, until the electrolytics dry up at least...)

Im also not as familiar with the  UC3842 as with the TL494 but I guess a look at the datasheet can change a lot of things withing a couple of minutes.

If you need further convincing (and don't have any on hand), build the block diagram yourself, with a comparator (LM393 will do), logic (discrete transistors or a CD4001 or 4011 will do) and driver (complementary emitter follower 2N4401/3 will do).  Uh, and oscillator of course.  555 in a pinch, but a... well, anything that can make spikes will do, so among RC oscillators you probably want to use a diode across the timing resistor to get one short edge; 555, CD4093, CD4049, CD40106, etc. will do.  So, overall, 4093 probably the best since you can use two for osc, two for RS f-f.  Or a spare comparator, that's fine too.

The biggest problem I see with the UC3842 is current function. This may not be a problem with the mains powered PSU, but it sure is with the 12-24 range. There are simply too high currents for a standard resistor to sense the current.

Well not so much current as voltage.  Newer controllers will happily sense more like 100mV there (LM5001 comes to mind, although that's a regulator (integrated switch), and I forget what's similar that's a controller (external switch), but that LM5022 is not a bad example).  Which makes 10, 20A, or more, practical.

So, two things:
1. You can bias the current-sense node up with a voltage divider from VREF to ISENSE to shunt resistor.  This raises the DC voltage at ISENSE, while lowering the AC gain slightly.  Instead of a 0-1000mV range, you might have say 700-1000mV range (basically you lop off the bottom 700mV of useful ISENSE range), while only needing, I don't know, 350mV or so of shunt voltage.
1a. You can combine this with slope compensation, which adds DC anyway, but also some AC (namely, from the RtCt pin), which compensates the current loop for operation in CCM (continuous conduction mode -- inductor current doesn't return to zero every cycle).  So it reduces the active range on the ISENSE pin (also reducing the shunt voltage drop), and has the benefit of higher inductance (lower core losses).

A note on peak current mode control: you need to operate in DCM (not CCM), otherwise chaotic behavior results*.  Slope compensation allows a less-than-100% ripple fraction (i.e., Ip-p / 2 < Idc), although not too much less (50-80% ripple fraction at rated output is typical).

*The underlying reason for this is quite cool: it turns out, a peak current mode controller is an electronic implementation of the Logistic Map, an iterated chaotic function.  This function takes a parameter, which as the parameter increases, at first the behavior is nonlinear but reasonable, but suddenly it splits into a multitude of values that cycle between themselves from iteration to iteration (period doubling, limit cycles).  The corresponding circuit parameter is, guess what -- inverse ripple fraction.  So, keeping the ripple fraction high, prevents chaos.  Chaos is undesirable in circuit, mostly because it causes increased ripple and a hissing sound.

2. We don't need to use a resistor.  Anything that senses the same current will do.  A current transformer is a typical choice:
https://www.seventransistorlabs.com/Images/Mag_Amp_PSU.png
(incidentally, speaking of slope compensation -- with the values and components shown, this circuit needs slope compensation!  It's an old circuit.)

Note that transformers won't sense DC.  We can sense partial DC, as long as it returns to zero every cycle ("pulsed DC").  Which we can ensure here, by using a diode on both sides of the transformer -- well, not explicitly so on the primary, but we can ensure that current is only drawn in pulses, when the transistor is switching, so it works out the same way.  A large resistor (1k here) across the CT winding dampens the "reset" pulse, reducing stress on the output diode (FR102).

That looks like this,
https://www.seventransistorlabs.com/Images/Snubber_103Z.jpg
(Actually a different module, with a lower ratio CT; same idea, in any case.)

Both of these also show a key improvement, a dV/dt slope snubber.  Your attached also shows this (C8, D4, R11).  This allows the transistor to turn off, say in 40ns, while the drain voltage rises, say over 80ns (depends on current, because the peak inductor current is charging the capacitor at dV/dt = Ipk / C), greatly reducing turn-off losses.  (Turn-on losses aren't usually a big deal, because there's not a lot of inductor current to pull down on at that time.)

The RC value can also be chosen to dampen the free ringdown (when the inductor's current is discharged, before the next cycle begins), saving some more EMI.  (The reduced dV/dt already saves some EMI.)

One thing I cant understand at first glance is a lot of the schematics I see on the interwebz like the one I attached dont have direct feedback. Its totally feedbackless? I guess it knows the voltages by the sensing the transient? That would be my only guess. And the losses within the transformer are low enought to not disturb any other rails?

Depends on what's required.  Normally, a TL431 and opto is used for feedback.  The 3842 is wired as an inverting amp (no compensation), so the opto commands whatever peak current from the 3842 it desires.  The internal comparator handles peak current, so that's perfectly fine.  The 431 then regulates voltage by controlling that current.

This works fine for one or two outputs, but the problem with multiple outputs is cross-regulation.

Regulating from the aux winding is the same thing: the aux output will be damn accurate, but the rest will be soft and vary up and down with load.

Cross-regulation is driven by leakage inductance between secondaries.  Ideally, you want them all very tightly coupled, so they all receive the same flyback voltage, no matter the current draw on each.  Practically, 10 or 20% cross-regulation is reasonable.  Better is achievable with a carefully designed transformer.  Worse is expected from a naively designed transformer...

If you need better cross-regulation, and the outputs are common-ground (as your heater and B+ outputs might be), you can use joint regulation.  One TL431 controls the throttle, and its feedback resistor divider is fed by both rails.  The regulation is the weighted average (weighted by the ratio of voltages and resistances) of the two outputs.

For a tube amp, this means B+ doesn't fucking explode while the heaters are cold, or conversely, that the heaters just never even begin to heat up while the B+ is completely unloaded.  Instead, B+ overshoots some, but not by an insane amount, while the heaters warm up a bit slower than normal, but eventually everything gets there.

Like, I had to do that on this one,
https://www.seventransistorlabs.com/Images/Discrete_Tube_Supply.png
instead of a single 680k from +100V to TL431, it's actually more like 1Meg, and there's a, I don't know, 47k or something, from 6.2V to TL431.  Same 18k from TL431 to GND.

I built two of these; one is in an all-tube set, https://www.seventransistorlabs.com/Radio_20m/ and one is in my Theremin (which is solid state, with the +100V for varactor bias and optional tube-based timbre circuitry).  The latter may be regulated on 6.2V only, I don't remember (which is the main load, so that would be fine).

Now seeing how it kind of works, I can see what the problem could have been with my original design. Tho I was 14 at the timne of designing that 10-24 input to input voltage-60V 50W converter  . It has a lot of flaws xD , but interestingly I do remeber that the transistor was not heating as much as the diode did, so the fet did give a damn unlike the diode, that was burning hot without a heatsing and even with the heatsink it gets to a modest temp. Im getting off track  .

Yeah, like I said -- there are fewer things you need to get right in a peak-current-mode controller, and they are easier to get right, but you still have to get them right in the first place.   You probably just didn't know at the time.

Tim

[SK_Caterpilar_SK]

You were probably going the wrong direction, leading to positive feedback of frustration........

Some of the things you mentioned in your post I am familiar with. The soft start is easy to implement but in practice, yeah youre right if input is suddenly lost and then reapplied and the output is loaded, the entire slow start is just  like if it wasnt there at all.

No thanks I dont really want to build the entire thing xD. It would sure uncover things but I can see how its working on paper, its just been a while since I looked in the UC3842 datashet as I said I was 14 at the time and I kind of developed a stigma arround it. Allaround current mode does not look so bad now to me. But the LM5022 seems awfully similar to the UC3842 in the datasheets.

I would like to go with the LM5022, looks simple and seems functionally great too. I only have to do some datasheet searching and get the values done. Also the way I see it, I could technically swap the inductor for a switching transformer. I dont really know how to get the LM5022 into DCM. Also having read an article about DCM VS CCM, just to let you know 500V 200mA is not low power so CCM is unavoidable if I understood the thing correctly. (Inductor in the schematic in the datasheet of the LM5022)

Are you telling me that I can change the sensitivity of the the current sense by injecting DC to the Isense pin of UC3842?

Transformers sense AC thats pretty clear to me . I also do know how current sensing transformers work. So you basically worked arround not having an AUX winding and hacking one together with a current sensing transformer? Should work as far as I can imagine.

About the Feedbacks you mentioned :
For regulating two outputs you just average the signals? Thats kind of how I got that.

In your case the most load will be from the 6,2V output, but in my case for now I have only one single output and that is for the HV. My powersupply can already as is do 65W output power at 450V, and im looking for a way to improve it to get the most power out of it with the least noise. Regulation can be tight because were only regulating one rail and yeah I dont know what else to say.

The ideal powersupply I would like to design at some point has 3 outputs. 6,3V DC heater voltage which would get me at least 6A of current (about 40W), 50V DC biasing voltage (low power, really a couple of miliamps will do I would say 5W would be enought any day), and the hopefully adjustable 300V-550V HV output that would be at least 70W capable. Ofcourse this is assuming youre powering a single powe amplifier, if id like to go stereo which would almost be like a life goal to do for me ( ), would have to basically double the power of all. Maybe even go as far as 170W for HV if it would be used for one big fat mono amplifier.

The power rating I guess all comes down to the switching transformer and the fet. I have no idea what tipe of a switching transformer I would use for a 115W total output power powersupply let alone the big fat stereo brother that would do double that. At that point I would probably consider just splitting the powersupply into the 6,3V heater supply combined with the 50V bias and the HV supply to be discrete alltogether.
But im really getting ahead of myself. Previously the big goal was to make any HV powersupply that would be useable. Now the most recent was to create one that can give me quite the power and I also suceeded. So now I just want to make it more stable useable and create the mains powered version aswell. And then the next thing to do is to look into designing a switching transformer, hopefully get a hold of the knowledge and design a transformer that would perform as the ideal psu I want to present as a complete thing. 6.3V 40W, 300-550V HV at 70/80W, 50V bias voltage really low power 5W would be enough.  And I dont know if such a PSU would be adjustable in that much of a range (200V range for HV that would not disturb the other rails so dramatically seems unrealistic) and even if not, making powersupplies for specific voltages would not bother me at all.

But im REAAAAALLY getting ahead of myself.

Now Im going to manufacture the converter with the TL494 to see the outcome, Im already looking into the LM5022 because I want to use that controller in my next design that should be more stable and less noisy.. hopefully more efficient too. Then the multiple output rail powersupply and then I might just feel complete enought to also make a speaker cabinet which is on my plans to do xD . All of this next to me repairing stuff, but without repairs I would not have the finances to even begin with a project like this. But im afraid its going to take me too long to get this one done.

(why do I sound so dumb to myself when I read this back lol)

PS: Thanks Tim for your massive help tho. Really appreciate it. And sorry for my grammar as you could have guessed im not native english speaking  .

EDIT: BTW would love to build that radio at some point but instead of subminiature tubes either with 7PIN miniatures or 9pins..makes em easy to source if they die. I probably have OCD because I was tripping hard when I saw the wiring xD . No offense but PCBs just look a lot more clean lol. I would love to built it at some point so I may aswel design a PCB for it..IF I will ever get to that day  .

[T3sl4co1l]

DCM vs. CCM, or more broadly the ripple fraction, is determined by the ratio of inductor voltage to inductance and frequency.  So, to move towards DCM, use smaller inductor, lower frequency or higher voltage.

Tim

[Circlotron]

A note on peak current mode control: you need to operate in DCM (not CCM), otherwise chaotic behavior results*.

CCM with a flyback converter is somewhat tricky. Say you decide to reduce the output so you reduce the mosfet on-time pulse width and this has the effect of *increasing* the converter output pulse width for a number of cycles until the transformer flux ratchets down to a level that matches the desired output. Much harder to stabilise than DCM.

[SK_Caterpilar_SK]

DCM vs. CCM, or more broadly the ripple fraction, is determined by the ratio of inductor voltage to inductance and frequency.  So, to move towards DCM, use smaller inductor, lower frequency or higher voltage.

Tim

The LM5022 seems to be capable to do both CCM and DCM. It seems to work as DCM up until a certain load. Using the Coilcraft GA3459, which has low inductance or even the GA3460 which has even lower inductance, it might just stay in DCM mode. (GA3459 5uH, GA3460 2,5uH inductance). Also the primary resistance is very low, so this all may just be enought for the controller to stay in DCM.

OR I could use a controller you would recommend. Because I simply cant know every single bit of info existing out there. So yeah I would look into other directions no problem just show me which way . Mos common controllers I know is the 384X the TL494,TL497 and these that I just found (LM5022 and others I mentioned in my post somewehre above).

Im also looking at coilcrafts custom transformer request document, maybe try it out, see what the outcome will be. https://www.coilcraft.com/custom_trans.cfm

[SK_Caterpilar_SK]

So Im back at this again and now that I see how the powersupply preforms, I want to take it up another level.

Here is what I want to achieve:

Input:
90VAC-265VAC

Output:
6,3VDC 40W (3% load stability, necessary for tube heater life.)
-60VDC biasing output (not much power required so ill just say max 3W, its critical to be stable as much as possible!)
if possible- adjustable output 350V-500VDC 80-100W ( Regulated stable.)

So far I dont know how am I going to achieve this. Im even thinking about using two powersupplies. One for the HV which now could be adjustable, and one for the heaaters 6,3V and the biasing -60V.

Furthermore I dont know what topology to use. Flyback as far I know is only good up until 100W and to add on top of that, I know nothing about switching transformer design so I will have to check out a lot of literature. If someone could point me to some really usefull literature about designing a switchmode transformer and what topology to utilize I would be thankfull.

 I have high hopes for you Tim hope you have something to send me a link to.
Btw I have been looking at your radio project and damn I like it, I want to build it but with 9pin tubes, I think it will be better in the long run. The only drawback is that its not FM so in my location if you listen to the same radio station over at AM and FM only the news are the same and everything other over the AM band is for old people and the FM band is the mainstream. For that purpose I want to build a pulse counting FM tube radio. I think you already saw it somewhere. Found the thing over here http://diytuberadio.blogspot.com/2011/02/love-grandfather-tube-radios-diy.html

Thanks
Adam.

[T3sl4co1l]

Flyback for both is fine.  Or forward for heaters, but not for HV because flyback is better there.

I have an FM radio project too that I haven't documented yet: it uses a 6688 RF preamp (double tuned grid and plate circuits, BW = 20MHz), 6J6 single balanced mixer, 6C4 LO.  Some of the goals were developing a feel for grid impedances, which are significantly lower than the cold capacitance, when heated and biased.  In particular, the 6688 is around 500Ω || 18pF at 100MHz, I think it was.  6J6 is higher impedance (grid to grid), and its cathode input impedance (which is where the LO is introduced) is basically cathode resistance (i.e., ~1/Gm, and that's total Gm of both halves, since they act in parallel for a cathode-LO-input SBM).  That constitutes the front end, with an IF output at 33MHz into 50Ω and a final BW of a few MHz.

The IF strip is three 5702s, double tuned for about 200kHz bandwidth and a shitton of gain; except for the first stage which is single tuned (matching 50Ω to the grid), but combined with the front end's IF output network, it's actually triple tuned.

The grid impedances are around 10kΩ plus some pF at 33MHz.  Combined with about 5mS transconductance, that's about 50 gain each, but it's more like 100 because the plate load impedance can be matched higher (closer to 20kΩ I think).

Keeping them from oscillating was a bit tricky.  The Cg1a for 5702 is relatively high, for high gain at 33MHz.  I'm pushing the maximum stable gain, a figure which has to do with the gain and phase shift of the input and output networks, and the feedback between them.  There is some neutralization, where the grid network is returned to ground through 100pF, and a small fraction of the plate signal is cap-coupled to that node.  The plates also needed shielding, for which I just rolled up some tinned steel, since good luck finding originals?  (The 5702s are again mounted in fuse clips, which hold the shield which hold the tube.)

The last 5702 is biased harder, and loaded with about 5kΩ (single tuned) for maximum power output.  This is tapped down for a ~1kΩ output impedance to drive the detector.  (The reason being, the higher plate impedance maximizes gain and output power, while the lower output impedance gives more drive for the detector.)  This drives a 6AL5 detector (half wave doubler configuration, so its average impedance is fairly low), which generates negative detected AM (where applicable).  This also serves as ALC, as the grid bypass caps (the 100pF's) are chained together with ~10kΩ ferrite beads and biased with this voltage.  For FM, this gives some limiting action already, and for AM, this gives a somewhat logarithmic response, which is fine for detection purposes.

The final IF is further coupled to a 6BN6 sheet-beam discriminator.  These are only practical at quite low frequencies (~4MHz IF), due to how the mutual impedances work out, and the width of BCB FM.  But IF is 33MHz?  For this reason, I have a 2nd LO at 29MHz (a 5744 submini triode), and a diplexing filter network (two different L||C traps, coupling IF to 6BN6 grid, and LO2 to 6BN6 grid, but isolating LO2 from IF final, and vice versa).  This gives about 10V of both at the 6BN6 grid, which due to its sharp transfer function, acts as a mixer itself, generating the 4.5MHz IF3 internally as it were.

The 6BN6 is a neat tube, but limited on options, given commercial FM modulation depth, and the beam conductance.

A word about that: the electron beam itself is made of matter, with mass -- electrons.  To physically move it around, requires work.  That work manifests as a resistance component.  The work increases as frequency increases, so the resistance decreases.  (In fact, apparently it goes as 1/f^2.)  There is also a factor for the beam's space charge working against the driving electrode, which is conservative so manifests as capacitance.  It happens that the capacitance is dominant, and for most "high performance" tubes (the best ones of the 40s and 50s, and most from the 60s), the dynamic grid capacitance (at nominal bias) is about 80% of the cold capacitance.  The capacitance varies with bias (because of how much beam is passing the grid), and with emission (because the space charge between grid and cathode acts as a spring).

The resistance is much higher than that reactance, so you can still get reasonable tuned Q, and the tube's GBW is still more or less constant with respect to center frequency.

Another aside -- the band width of a tuned amplifier is limited by the source or load resistance, and the minimum reactance of the circuit.  It seems all known devices (BJTs, FETs, tubes) have a transconductance-with-capacitors characteristic, with some loss element for the input and output, and none have a fundamental voltage-source-with-inductors characteristic.  So, while the same physics applies for both, it seems we only need be concerned by the case with capacitance as the limiting reactance.

Then: the impedance of a tuned network is Zo = 1 / (2*pi*Fc*C), and if C has some fixed minimum, then at a given center frequency Fc, the characteristic impedance Zo has a fixed maximum.  Where we have some source resistance Rs or load impedance RL, we have a Q of R / Zo, and a bandwidth BW = Fc / Q.  (When Q < 1, it may be more meaningful to think of it as a combined low-pass and high-pass network, and you can consider replacing the inductors with resistors, or with proceeding stage outputs, i.e., making a DC-coupled amp with no LF cutoff (no highpass).  In that case, BW is the baseband bandwidth, i.e., the HF cutoff.)

If we have more grid C, we need lower L for the same Fc, and get higher q and lower BW.  If we reduce R, we get higher BW, but proportionally lower gain.

So, at a given Fc, an amplifier exhibits constant GBW.  Since grid conductance varies with frequency, we should expect GBW to change with Fc, but in practice it's not a big factor.

So, back to the 6BN6 discriminator.

An already-modulated electron beam carries momentum along its path.  This manifests as a transconductance between grids.  In a tetrode, there is a G_g1g2 which is nonzero -- but the inverse is not typically true, i.e., G_g2g1 is extremely small.  This is called a nonreciprocal effect, and happens because the beam acts like a conveyor belt.  (Nonreciprocity is very difficult to realize without spending power -- of course, we're talking about an amplifier tube here, so we're doing quite a lot of work, biasing this tube, and this is physically consistent.)

This is used in the 6BN6 to excite the second grid.  An impedance is connected to g2, the mutual conductance develops a voltage in it, and that voltage modulates the plate current.  When that impedance is a tuned circuit, the phase shift is frequency dependent, and we get a plate current that looks pulse-width modulated with respect to input frequency.  This is filtered down to "DC", and hence FM is detected.  Slick as hell, isn't it?

The downside is, that mutual conductance can only be controlled by cathode current, and only over a modest range.  The Q factor of the network needs to be quite low, so that its bandwidth is comparable to the modulation depth -- 200kHz width at 4.5MHz Fc is pretty low, meanwhile the impedance needs to be very high, because the transconductance is very low (in the uS).  I ended up using a lot of turns of fine wire (I forget how many, maybe 50-100 of 37AWG on a 9mm form), with a ferrite slug to tune it, to get that little extra impedance out of the network.

The distortion still isn't terribly great.  The limiting is quite good -- no fading is apparent, and the response is just as you expect from a modern receiver, as you tune across stations.

The overall structure -- you might notice that there's a lot more tubes in here than a commercial FM receiver, of course partly just because I don't want to have to optimize it that much; but also because I want some more options with this receiver.  In the front end, all the tuned networks are pluggable, so they can be switched out for, say, 120MHz (aviation) or 144MHz (amateur 2m) bands.  The IF strip is a separate module connected by BNC jacks; if I want a lower or narrower IF (say for AM voice or FM NB), I can construct one.  Or add a 50Ω output to the 33MHz final IF, and connect that to a 2nd IF strip, to the same end.

Pictures!

Power supplies -- back on topic in this thread, I use a pair of single-output flyback supplies.  They don't seem to generate much noise in the relevant bands, so reception is fine.  (They are cleaned up a bit, but I doubt they would pass FCC Part 15, particularly the HV supply which has some nastiness coming from the transformer -- bad windup.)
https://www.seventransistorlabs.com/Images/100WPowerSupplies.jpg
Middle and right, 150-300V 0.5A (set for 150V), and 6.3V 10A.

https://www.seventransistorlabs.com/Images/FMRadio2.jpg Front end, top view (shields off)
https://www.seventransistorlabs.com/Images/FMRadio3.jpg Bottom
https://www.seventransistorlabs.com/Images/FMRadio4.jpg Pluggable networks (FM BCB)

https://www.seventransistorlabs.com/Images/Radio_IF_Filter.png Analysis of 6J6 plate network (front end IF output; left network in above pic).  R9 is Ra-a (plate resistance), inductors are air core and coupled as listed, and capacitances of plate (Cga + strays) and trimmers shown separately.  Note output is series tuned into 50 ohms -- flexibility in topology is key for good optimization.

Cheers!
Tim

[SK_Caterpilar_SK]

Flyback for both is fine.  Or forward for heaters, but not for HV because flyback is better there.......
Cheers!
Tim

Man thats one hell of a big fat reply xD. Absolutely amazing Tim. And I love the radio stuff maybe end up some day building that pulse counting tube radio I showed because of its relative simplicity and yeah lol.

Here is the thing. I dont know how to design smps transformers properly. I see your powersupplies are quite powerfull and you probably did the math on your transformers yourself. So the deal is, I want to do my own math too lol.

At this point I decided to have two separate switching  supplies. One for the heaters and negative bias voltage, and the other one will be the big fat 100W 500V smps. This way I can really get tube lifetime out. One of the most important things with vacuum tubes is to keep the heater voltage at an exact level with the least heater voltage deviation. As low as possible. Like hopefully within 1-2% is achieveable at all. And this way the HV can be also regulated tightly which works out just about perfect this way.

And btw you kind of forgot about my problem of not having the literature for the job so if you could please point me to any literature. Or help me designign my transformer, I would be more than thankfull.

[T3sl4co1l]

I'm a terrible reference for literature unfortunately; most of what I know has been synthesized from piecemeal articles, and experiments, supported by a strong foundation of theory.  Hopefully others can volunteer something...

Offhand, these may be useful:
Magnetics Design Handbook, Unitrode/TI https://www.ti.com/seclit/ml/slup132/slup132.pdf
Transmission line transformers -- http://een.iust.ac.ir/profs/Tayarani/files/transmission%20line%20transformer.pdf
Handbook, Amidon http://www.introni.it/pdf/Amidon%20-%20Transmission%20Line%20Transformers%20Handbook.pdf
An original, in a sense; G. Guanella, New Method of Impedance Matching in Radio-Frequency Circuits https://hamwaves.com/chokes/doc/guanella.1944.pdf

I emphasize transmission lines, because they are the most general, and in my opinion not onerous to understand (but, prove me wrong; these things come naturally to me so I underestimate how much complexity goes into them).

The basics are this: a transmission line is formed by two pieces of wire being nearby.

That gives a characteristic impedance (which depends on the relative cross section only), and a characteristic length.  We instantly know all the other high-frequency properties of the structure (delay, cutoff frequency, leakage inductance, parasitic capacitance), and need only ask the impedance of the core to figure out the low frequency properties (magnetizing inductance, coupling factor).

In a transformer, we have a primary and secondary wound together, for example as a pair of single layer (solenoid) windings.  If these are wound in the same direction (say, clockwise, left to right, start to finish), then they act like bifilar wire wound edgewise, and the impedance is that of the bifilar pair (well... nearly).  The length is simply the wire length.

Or we can just wind bifilar wire flat as usual, and then we have two helical windings within each other, 180 degrees apart.  The impedance is again the bifilar pair (...nearly).

Or we could wind two layers, one left-to-right and the next right-to-left.  Now the helices are opposite handed, and the wires don't perfectly line up -- they're crossing over and under each other, every turn.  Well, in this case, we still have a given wire in proximity to others, it's just less uniform.  (As long as we aren't considering waveforms with edges as fast as the electrical length of a single turn, we don't care.)

What violates this sort of approach, is having multilayer windings.  Say we put down two layers of primary, then two layers of secondary.  The first primary layer "sees" its neighbor, not the secondary, so it doesn't couple directly to the secondary.  And likewise for the far secondary layer.  The inner layers see primary and secondary respectively, but also see the outer layers.  We see two things in this situation: the image currents from the outer layers flow on the inner layers, dramatically increasing their losses (proximity effect); and, the impedance between outer layers is a good, what, triple or so the impedance you might've been expecting given the layer-to-layer distance.  But the wire length is double, so the inductance is sextuple and the self-capacitance (capacitance between ends of a given winding) is equal to the isolation capacitance (between windings).

In short, multilayer windings have considerably lower bandwidth, and fairly higher impedance, than a proper single-layer winding in transmission line style.

What if you want a high impedance transformer?  Well, you can open up the distance between wires (higher winding pitch, thicker inter-layer insulation); but this does waste a lot of space.  You can at least use a longer bobbin (to get a wider single layer winding), or maybe a toroid if you don't mind winding one.  You also tend to need a lot of turns, though, especially at very high voltages.  So the reduction in bandwidth is inevitable, and at some point you'll need to use another approach, like a resonant supply instead of a switcher.

What if you want a low impedance transformer?  Use wider wires -- multifilar, lots of wires in parallel abreast -- or foil.  Planar transformers are very attractive here, where you can stack alternating layers in parallel (primary and secondary), more than halving the impedance compared to a single layer pair.

You can use some cheats, like connecting windings in series at DC, but stacking them (one diode per layer) so they act in parallel at AC (the capacitance between windings cancels out).  CRT flyback transformers are made this way.

I did a sort of this in this module,
https://www.seventransistorlabs.com/Images/DCDC_800V.jpg
which is a 12V input, 800V output (adjustable 100-800) module, and uh, 50W or so rated, I forget exactly.  The transformer has the windup,
https://www.seventransistorlabs.com/Images/DCDC_800V_FoilWindup.jpg
the primary is low impedance (12V DC input, some amps), so foil is appropriate.  The secondary is high impedance (400V and fractional amps).  The two sections are wired in series, and the rectifier is wired as a bipolar output (+/- 400V).  The "-400V" node is merely grounded, giving 800V total output.

Here, it's not so much that the secondary layers act together -- they can't, there's a foil turn between them -- but that they are in the same environment, it's symmetrical, so whatever happens to one, happens inversely on the other.  With the diodes similarly symmetrical, the output voltage is symmetrical, and the common mode noise is symmetrical (well, around the middle of the primary it is).

This is the key feature that my 100W offline supply was missing -- I didn't consider the relative voltages at the start and end of the windings, and ended up getting one backwards, so the full switching waveform gets induced across the isolation capacitance.  Huge fuckin' common mode noise.  This module, well, it doesn't actually matter any because it's common ground, but -- if it were isolated, it would be a far sight better, despite its much higher output voltage.

The windings are also pretty short, like 1-2m of wire, so the bandwidth remains high despite the impedance mismatch (the impedance of a round wire winding sandwiched between two plates, and a couple layers of tape, is around 50 ohms, whereas a kohm or so would be more appropriate).  In fact I see very little if any overshoot and ringing on the waveforms, I was quite impressed.  (Part of the short windings is the oversized core: it's good for about 100W in flyback, but only being used for 30, maybe 50W.)

Tim

[SK_Caterpilar_SK]

So it has been a while since I posted something and now its time for a question.

I have not had any time to make the powersupply at all. I still have the second version on paper not an actual working device but I have got some time toi sniff arround my prototype. the previous problem was that the entire PSU seems quite inefficient. The mosfet gets blistering hot and so does then switching transformer.

I have looked at the gate voltage and it seems fine to me, but the  transformer has an extreme amounth of oscillation. More specifically I was measuring ground to drain of the mosfet. That said on the high side of the G voltage it pulls short 2 ground on the drain (just saying so you dont have to figure out on your own.). Could this be causing the heating of the transformer and the mosfet? Interesting is that any capacitor in the input (bypass capacitor- electrolitic) right close to the switching transformer it gets very, very hot.

I dont really know why this is happening, so far my knowledge about SMPS are rather basic. I did design another powersuply completely, but since I did not know about this issue Im pretty sure it will happend again.

I have hope in your reply Tim . Would just a snubber network solve this or it is something that needs a more proper solution. I have a feeling it is going to require a proper solution.

On the oscilloscope where there is only one waveform I have measured between D and S  (G30N60-its not a mosfet-IGBT but a really expensive one that works really well, does not heat up as much as a IRFZ44. This heating issue stays a problem even for really low on resistance mosfets like for 100A above mosfets). The frequency of oscillation is 840kHz.

So how can I make it more relyable and efficient? Assuming the controller will be a TL494 in the future powersupply. Also I would not mind using a different chip if that makes it better.

[SK_Caterpilar_SK]

Also in reply to my previous reply, the first waveform where the oscialltion is way more pronounced im switching the transformer at 75kHz (unintentionally, I have set it to 50kHz just as specified by the manufacturer but seems under load it changed the whole thing, it is the fault of my crappy controller, the traces all act like antennas.

The second measurement was taken at 30khz switching frequency.

Also in the second measurement the 1st probe is in 10x , I forgot to set the scale properly on the scope.

[T3sl4co1l]

Looks like a shitton of leakage in the transformer.

Tim

[SK_Caterpilar_SK]

Okay...back at it but quite a bit dissapointed and hopeless this time. The latest circuit with the TL494 is not working the way I inteded it.. The controller spazzes out when the tubes begin pulling a whole lot of power.. It can cope with resistive loads great but cant handle dynamic loads. It simply wants to regulate the voltage too tight that it just goes nuts when the tubes want to pull more power (bass note or something). It spazzes out goes 90% duty and trips the OCP on the 13A 12V powersupply...optionally it burns a mosfet.

So its not working. I had 3 tries at this problem. LT3751 which did not work at all. Not even a single pulse from it. Custom controller which burned five mosfets before it worked at all, and the TL494, which keeps spazzing out tripping OCP and burning mosfets. My frustration has reached beyond borders. I have no idea what to do at this point.

I think this project was doomed the day I made the decision to go on. I spent too much money and time on the project that did not work so im kind of hopeless at this point.

For the ones supporting me and helping me thank you all, but this project inevitably met its end with a big failure sticker on its archive. Unless someone knows a proper SMPS design, because apparently I dont at all.

[FreddieChopin]

So it has been a while since I posted something and now its time for a question.

I have not had any time to make the powersupply at all. I still have the second version on paper not an actual working device but I have got some time toi sniff arround my prototype. the previous problem was that the entire PSU seems quite inefficient. The mosfet gets blistering hot and so does then switching transformer.

I have looked at the gate voltage and it seems fine to me, but the  transformer has an extreme amounth of oscillation. More specifically I was measuring ground to drain of the mosfet. That said on the high side of the G voltage it pulls short 2 ground on the drain (just saying so you dont have to figure out on your own.). Could this be causing the heating of the transformer and the mosfet? Interesting is that any capacitor in the input (bypass capacitor- electrolitic) right close to the switching transformer it gets very, very hot.

I dont really know why this is happening, so far my knowledge about SMPS are rather basic. I did design another powersuply completely, but since I did not know about this issue Im pretty sure it will happend again.

I have hope in your reply Tim . Would just a snubber network solve this or it is something that needs a more proper solution. I have a feeling it is going to require a proper solution.

On the oscilloscope where there is only one waveform I have measured between D and S  (G30N60-its not a mosfet-IGBT but a really expensive one that works really well, does not heat up as much as a IRFZ44. This heating issue stays a problem even for really low on resistance mosfets like for 100A above mosfets). The frequency of oscillation is 840kHz.

So how can I make it more relyable and efficient? Assuming the controller will be a TL494 in the future powersupply. Also I would not mind using a different chip if that makes it better.

Dumb question but is your probe compensated? Connect it to square wave test pins on scope and adjust until it's perfect rectangle.

[SK_Caterpilar_SK]

Checked against control and other square waves on the poversupply board.(fet drive signal from TC4420). The transformácie by that time was already through hell and back but i think I saw such ugly waveform because I was running a IGBT instead of a FET. That original power supply is dead by now. (Funny it wasn't the fet nor the circuit itself that died but the voltage regulation that was horribly underdone.)

[Audioguru again]

I was 19 years old in 1964 when I sold my vacuum tubes amplifier assembled kit (that needed its tubes replaced often) to an old geezer and replace it with a solid state amplifier that still works well today. I have not seen a working vacuum tube since then, but there is a store that sells a solid state amplifier with a couple of vacuum tubes glowing on top (not in the amplifier circuit) that are sold to old audiophools.

[mrjoda]

Hi  SK_Caterpilar_SK,

I found this topic quite interesting. The idea of SMPS for tube (headamp) is in my head a few weeks (actually more than weeks). I started slowly, bought some books, read a lot of literature. The whole schematic is in my head... briefly...

My ideas went down, when i found that isolated RF wire for primary winding is almost impossible to buy without selling a kidney. And insulation at least 4kV primary/secondary is another chapter. Chinese supplier for whole transformer is the cheapest and most available solution...

The optocoupler feedback is quite issue. A BIG issue. Fortunately, i have access to bode100 in our company so it can by done.

Maybe this topic is starter for me for comeback from ideas to project.


And last point :
your pcb from previous page is wrong and messy. EMC hell