London Power ad

[-]
Search the Forum








(Advanced Search)

Fixed to Cathode Bias switching problems...
#1
So I built an amp with a fixed/cathode bias switch like that for the Standard project in TUT5. The only difference was that I used a relay in place of the switch and a cap across the cathode resistor, but I've had some problems.

While testing without tubes, I found that in cathode bias mode, 1/2 of whatever the bias voltage was set to showed up at pin 5 of the tubes...fixed bias was fine. Simulations confirmed this behavior, and I found that changing the 2x 30k1 resistors around Q10 to 331k (c-b) and 2.21k (b-sw) ensured the bias didn't leak through anymore. Is that normal?

The second thing is the relay keeps sticking after a couple of switches. I assume this is the contacts welding from the cap discharging when the relay shorts it. Oddly, when I unsolder it and activate it out of circuit with a battery, it no longer remains stuck and works again. This relay looks heftier than the PA66-CR ones, so I expected it to be fine. Is there a way to stop this from happening like soldering a cap across the contacts to take the initial discharge or something?

My solution for now was to just get rid of the cap, but the problem is testing with a sine wave was putting 10W into that resistor (470R 7W). The ones in TUT5 are all 5W, and I realize that a constant sine wave is not like actual playing, but is this a big enough resistor? I'm going to assume that with the cap, all we need to worry about is more or less the idle power...any AC will go through the cap?

Thank you all!
Reply
#2
Hi

You have an early printing of TUT5 - the two 30k1's for each of Q10,11 were replaced by a single 30k1 from C to B on each BJT.Also note that the NPNs drawn earlier was a mistake and that these should be PNPs with C to V- and E to the bias pots. Everything works as it should.

Relay sticking can be caused by various things, including the contact welding you mentioned. What have you done on the coil side?

The coil voltage is zero when not energised, with both ends sitting at the supply voltage. Upon closing a switch for the switched end of the coil, the supply is across the coil and the relay contacts have switched. When you open the coil switch, there will be a very high transient voltage on the free end of the coil caused by inductive flyback, which for a 12V coil could easily be 80Vpk.

Normally a diode is placed across the coil to absorb this energy and clamp the voltage across the coil. However, even though this is the most common snubber method used, it is far from the best as it restricts how fast the contacts can move. Using a zener diode across the coil control element allows one to allow a specific transient height. For example, a 36V zener would allow a 24V transient given a 12V supply.

One can also use a cap or series-RC as a snubber across the control element, but these do not work as well as the zener.

Regarding Rk and Ck: Rk has to be sized for the average heat over the audio cycle. An unbypassed Rk will have transient heat higher than its rating but that is okay if the avrage is within the rating. Personally, i use Rk values that set the idle power to much less than the full rating of the tubes and TUT5 maintains that protocol except when one tube of the pair is pulled, then the remaining tube will run at its full rating. DO NOT TRY THIS IN OEM CATHODE BIASED AMPS UNLESS YOU INCREASE Rk AS IS DONE HERE !!

Ck is there to keep bias stable, and as TUT4 informs, is the first form of "fixed bias".

have fun
Reply
#3
Hi Kevin, thank you for your reply! Yeah I have all the TUT series as well as the speaker book and TOT...love them all.

Ok, saving a component is a better solution than mine! Plus adding that resistor back that opens up an easy way to do mixed bias, huh?

On the coil I used a diode snubber, but I can easily try the zener...I believe I have some 33V. I assume since we are looking at the coil that it means that reduced speed of the contact could be the issue? It's pretty weird when it happens, you can hear the relay click but its far weaker than normal.

So for Rk, how do we determine what average is? What I tried to do was bring the amp to max clean power with a sine, then measured the voltage across Rk. In this case it was 70V and the resistor was 470 Ohm, which calculated as 10.5W. But this is where it gets foggy for me. Plus, that sucker got hot fast, which scared me a little.

That's really smart why you use the higher value for Rk. If one tube died, I'd imagine the other would shortly follow. I can't speak to the merits of warmer bias yet, so higher values are fine with me, lol.
Reply
#4
Hi

In the bias circuit, we are not adding a resistor back unless you meant that you removed all four 30k1s? There has to be a base current path for the fixed-bias position and the R between C and B for each transistor does this. The bias control is split between tube pairs and each BJT switch has to be independent, so maybe for you this means adding back two Rs.

How you measured Rk during the full power test is correct and it is dissipating the power you calculated, so yes it is going to get hot fast and should be a higher wattage part. Maybe the TUT5 Rks should be 10W, but the original player this amp was designed for rarely used the cathode-bias mode and likely never full tilt since Power Scaling is on-board.

I am not actually much of a fan of cathode-biased power stages and use it only for a tone difference. I've used Rk values up to 1k for standard tubes and this certainly keeps the heat controlled for both the tube and Rk.

The unsnubbered coil allows maximum speed for the relay contacts to open and close but places high-voltage stress on the coil control element. The standard diode is a brute force damper and does not allow the fastest relay transitions. Too-slow open/close causes the welding effect if switched under load. The currents in the Standard are low, but it is enough to hold the contacts together.
Reply
#5
So what I meant was, let's call Rc-b R1 and Rb-sw R2. The simulations helped me figure out that R1 needed to be 100x R2, so I went with R1 = 330k and R2 = 2.21k (what I had on hand). But it sounds like you kept R1 = 30k1 and eliminated R2, making it 0 ohms. So you save a part because 30k1 is surely 100x 0 ohms, lol. Those are things that are easy to overlook.

Then my mixed bias comment was that if you kept R2 30k1, it could be any value to suit really, and hooked up a switch to short it, you'd effectively have mixed bias provided you also switched an appropriate resistor in the cathode at the same time. R2 shorted = cathode bias, R2 in circuit = mixed.

If the amp was the same circuit in the earlier TUT5 like I have, then you likely wouldn't need a 10W part because you'd have mixed bias (if R1=R2), which would cut down on the power. I actually thought that's what you meant to do. Since the cap makes it a form of fixed bias, but removing the cap requires a larger part and generates a lot of heat, you could add a portion of the fixed bias voltage back. This would allow the cathode resistor to still bounce around like you'd get with it unbypassed, but you'd get the heat savings because of the fixed bias portion.

I can see it being good for some styles with a lot of expression in the fingers. This was my first experience with cathode bias...I really thought it shined with single ended settings.

Thank you again!
Reply
#6
Hey

Maybe it's a coincidence that your Rc-b =330k and the screen voltage sampling resistor in the bis reg is 330k? The two are actually unrelated as in fact is the case for the Rs around the BJT bias switches. You have to think of the output of the bias reg as a voltage source and the BJTs simply as switches, tying that voltage source to the bias-set network, or isolating the voltage source from the bias-set network.

The Rs around Q9,10 are selected to allow sufficient base current that the BJTs can pull the bias pots as close to the bias reg output as possible.

There are lots of ways to have mixed bias. TUT4 shows swept fixed to cathode bias using RmX methods. Another approach is to have a variable Rk and a tracking bias reg with adjustable tracking ratio. You could also make things completely manual but safe by using much higher Rk values. Older STUDIO amps had such a feature controlled via the Fiixed-Cathode switch and a Heat switch, later combined on one 3-way switch as Tweed-Mix-Modern.
Reply
#7
(08-17-2019, 08:13 PM)K O\Connor Wrote: Hey

Maybe it's a coincidence that your Rc-b =330k and the screen voltage sampling resistor in the bis reg is 330k? The two are actually unrelated as in fact is the case for the Rs around the BJT bias switches. You have to think of the output of the bias reg as a voltage source and the BJTs simply as switches, tying that voltage source to the bias-set network, or isolating the voltage source from the bias-set network.

The Rs around Q9,10 are selected to allow sufficient base current that the BJTs can pull the bias pots as close to the bias reg output as possible.

There are lots of ways to have mixed bias. TUT4 shows swept fixed to cathode bias using RmX methods. Another approach is to have a variable Rk and a tracking bias reg with adjustable tracking ratio. You could also make things completely manual but safe by using much higher Rk values. Older STUDIO amps had such a feature controlled via the Fiixed-Cathode switch and a Heat switch, later combined on one 3-way switch as Tweed-Mix-Modern.

My apologies for taking so long to get back on this. Every time I wanted to come tell you and the forum that the problem was solved, the relay would stick again. I started with a 15V zener in series with a diode on the coil and that worked for a few days. I increased the zener to 33V and it worked for a while more, and a 50V zener kept it from sticking even longer, but ultimately it always stuck. I gave up on it for a while and then recently decided to just modify the existing PCB so that instead of shorting the Rk/Ck combo for fixed bias, the relay's center is grounded and the relay selects Rk/Ck or the tube cathode. This way we shouldn't have to worry about Ck's discharge current as Rk will bleed it off quickly. This has worked ever since, even with 100uF caps!

Re: Rc-b = 330k, merely a coincidence. Once I saw the ratio was ~100:1, I was concerned (unfounded) about Rb-sw being small so I just multiplied the usual 33k by 10 to allow for Rb-sw to be a bit bigger. And actually the 330k was causing an issue with power scaling where once I would dial down even just a little the distortion would sound pretty bad. Going with the 33k/jumper combo got rid of a lot of that, so thank you for mentioning that....I might not have found that as easy!

The only problem I have left in this build is this raspy distortion that comes about when driving the output tubes any higher than a mild overdrive and is most apparent when notes are ringing out. It's as if it stacks on top of the natural overdrive from the tube and has this odd swirl as it decays, and gets worse when power scaling. The output stage is a 4 tube where 1 pair can be dialed out (like in the G-M-X chapter of TUT4) and I've found that physically removing EITHER pair gets rid of it, OR the 2nd pair must be a full volume or that raspy distortion is there. I see no oscillations on the scope, no odd voltages or signals, but simulations appear to show that the "on" pair can drive the "off" pair through the plates. Peak power in the "off" tubes appear to be a considerable % of dissipation (16W), and the current waveforms appear to be FW rectified...Any clue what that is or what can stop it?
Reply
#8
Hi liquidair

That is an interesting set of observations.

First, the currents in a push-pull stage will look half-wave rectified, but add together in the OT as a complete sine wave, so that part is normal. FW rectified on one side would be odd as that would mean that it is not just flyback energy being expended in the 'off' tubes, but maybe you meant FW rectified considering both halves of the circuit?

Ideally, the idling tubes should appear like current sources, although being tubes, they will be far from ideal CSs. Still, I would think that the tube's conduction is essentially set by the bias condition and it should only conduct the idle current despite the change of voltage on the plate, just as the plate curves show. There might be something amiss in the layout or signal leakage around the pots? Do you see/hear this effect regardless of the bias method?

I've built several 4-tube output stages with signal level controls in place and do not recall hearing the distortion you cite, using mixes of 6L6, 6V6 and EL-34. To me the 34s always sound a little distorted.

Regarding the relay coil snubbing. The zener is used across whatever CONTROLS the coil. Say the coil is switched by a mechanical switch. The zener would be placed across that switch. With the switch open, both ends of the coil are at +12V. Close the switch and you see one end of the coil go to zero volts as expected. Open the switch now and see the end of the coil fly up above the supply but be clamped at the zener voltage minus the supply voltage, then return to the supply voltage.
Reply
#9
Hi Kevin

So I made a mistake in the simulation and had the screens tied to Va (in addition to the OT). Once I changed that the current in the off tubes looked like it was following the plate signal up and down about the idle current, albeit with some really nasty looking waveforms. So we're good there.

But right, that technique is shown multiple times in the TUT series, I know this is something you've done, and we see this in a few commercial designs. And it's funny, I had read the TUT series so much prior to actually hearing/playing through EL34s, so I know how you feel about them. But when i first played with them, their tone was exactly as I imagined them sounding reading your books prior...so great job describing them! And ya, I hear that little bit of distortion too.

When I observed the build after posting, I did see some voltage on the grids and thus cathodes, and I do have a tiny bit of feedthrough, but I do see voltage on the grids and the cathode that looks like the ones in the sims when I open the wiper to gird connection. Interestingly, changing Fixed/Cathode bias and Triode/Pentode on the "off" tubes definitely affects the "on" tubes with the wiper lifted too. Triode mode seems to cut the output power of the on tubes as if I had set their screens to triode mode.

And because I can't observe this distortion on the scope with a function generator, I tried scoping while playing. At first I didn't see anything, but I turned down the power scaling control slightly and it looked like crossover distortion appeared. I can tell that's it by the way it decays; it sort of oscillates like you're turning the rate down on a trem/vibrato and when it ends it sort of collapses like a gated dead battery. That's exactly how the signal behaved on the scope.

This is as far as I've gotten but it's curious because I'm biased to ~60% fixed, and I dropped the cathode resistor to 330R (~75%) just to experiment; I wouldn't think the tracking regulator would shift into cold bias that quickly (using the 'Standard' circuit). But we're on to something at least!

Re: Coil Snubbing. Ooooohhhhh, ya, I put it across the coil, not that way! That makes sense. I remember wondering while I was adding the zener where that 'extra' 12V came from in your first suggestion, lol.

Thank you so much Kevin!
Reply
#10
Hi liquidair

Simulations can show unexpected things for sure!

Having the screens tied to plate is not a mistake per se; rather, just shows triode mode during the mixed on/off-tubes.Certainly changing the mode between triode,UL and tetrode/pentode will highlight the different plate curve slops of these modes, which indicates the changing internal impedance of the tube according to the wiring configuration. The sim would then show varying signal currents in the 'off' tubes, which is still a little surprising.

I don't recall ever changing the mode of the 'off' tubes to hear what effect it might have on the sound, so not as thorough a test as you have done. Despite that, I don't recall hearing the distortion you report and none of my customers reported such an issue. Whether they heard it and didn't care or thought it was part of the design I don't know?

The Power Scale pot in the Standard is linear, so the voltage and current each change linearly resulting in an exponential change of power as you dial down. This is not linear to your ear. A log pot would provide a square of the log change which is faster than linear to your ear. The bias track is reasonable but not perfect but exhibits a compensating effect at low voltages, a serendipitous effect of the low internal gain of the bias regulator.
Reply
#11
They sure do! But then when things are right, I'm amazed at how accurate they are.

And no, I don't think this is something that is inherent in the design or that I have super ears or anything, this is not typical.

While the "off" tubes do seem to be driven and their configuration seems to affect things, I don't think this is the root cause. Now that I figured out how to see it on a scope, it's clearly crossover distortion.

I'm convinced we're onto it and that it's caused by the bias regulator somehow. I took some data plotting Vs, -Vb and Ik while scaling downward and Ik seems to drop below the ideal line (Imax to 0) on the transfer characteristics...which is worse considering Ik includes the screen current (which should skew the points upwards, not down)!

And one other odd observation was that with the power scaling set to minimum (Vs=52.24V), I tried to see if I could bias out the crossover distortion but found I had a range of only 1mA (4-5mA). That seems odd to me. While I'd expect the available range to be restricted, 1mA is tiny. Got a feeling that is part of this.
Reply
#12
Hi liquidair

The Power Scale control should dial down Va and Vs to essentially zero volts as shown in TUT5. I have to assume you changed the 33k to 120k or higher to limit the sweep as you've listed above?
Reply
#13
Actually, I tried DC power scaling, so I have a 22k low limit resistor on a 1M + 100k pot Vref divider (which is exactly what you predicted in 120k, lol). As of now, this makes ~50V the lower limit.

I think I figured out what is going on; it's two things in combination. First, for whatever reason I used a 68k resistor in the bias regulator feedback instead of the 47k. Coupled to this, I added a 100k to ground on the -Vbout node. I can't find any notes as to why I did either of these. After changing to 47k and removing that resistor, things appeared much better on the scope and there's about 3-4mA range now at the lowest power scale setting (this may not matter but it's a difference). Most importantly, things sound much better, everything is spankier and more lively.

I'm not sure which resistor made the difference, but my best guess is one of them was choking the bias regulator when the screen voltage dropped? This would explain why I'd would hear it at full power sometimes and when the second set of tubes is added, the screen voltage would sag.

Second, is that my testing is atypical in that I'm using a homemade load box and into an IR loader. In this case, it's no big deal to play at full power cause it will be completely silent. It's also no big deal for me to turn the power scale down to drive the power tubes much harder, which is when that distortion is at its worst. However, with a speaker cabinet, you'd never play at full power, and would likely be more sensible as to how much signal you're throwing at the output tubes. So this is likely just due to me using these circuits in extreme ways.

I'm thinking this could be alleviated somewhat by semi-scaling the phase inverter like that graph in TUT6 and/or increasing the grid stoppers on the tube grids.

And one more note, the off tubes don't seem to affect the on tubes as much. Triode still does noticably, but this is likely normal...according the the datasheet the plate resistance in triode mode is about the same as the primary impedance seen by the tubes, whereas as a pentode its around 10-20x higher.
Reply
#14
Hi liquidair

The 68k in the bias reg is the culprit as it sets the ratio of raw bias voltage to the bias-set network. We generally never use more than 56k, with 47k being the lowest noise point and quite useful, although 51k is a bit better for large-bottle tubes.

The 100k on the output of the bias reg is of no concern and is included in our kits.

Regarding using DC Power Scaling of the Super Budget style: It is good to have 330k-1W off each mosfet source to ground. This provides a tether that helps keep the mosfets stable.

The bias adjust range at minimum Power Scale setting is also of no concern except that in your case the whole tracking proportion was off. The output stage is biased at full voltage and then left alone for the Power Scale circuit to do what it's designed to do.

Note that were you using different tubes, some luw-mu triode or other, then a bias reg feeback resistor of 68k or higher could be appropriate. For example, 6080 et al are mu=2 and need typically -120V of grid control to keep from melting down. Even that sets a reasonably warm condition for the plate.
Reply
#15
Over the weekend I plotted the new Ia vs Vs and with the 47k we follow that ideal line on the transfer characteristics graph almost perfectly! I do have those 330k 2W installed as per TUT6 (or 4, i can't remember) so we're safe there!

Thank you so much for your help Kevin, I sincerely appreciate it!
Reply


Forum Jump:

[-]
Come in where it's warm!
A warm welcome to tube amp modding fans and those interested in hi-fi audio! Readers of Kevin O'Connor's The Ultimate Tone (TUT) book series form a part of our population. Kevin O'Connor is the creator of the popular Power Scaling methodology for amplifiers.
Please remember these three principles: respect, sharing, community.
Not familiar with The Ultimate Tone book series? See discussion topics, or click here to visit London Power/Power Press Publishing.

[-]
Tube Amp Forum Hosted by London Power
London Power logo