CCS Loaded Parafeed Output - Design Considerations?

[L0rdGwyn]

Hi Bottleheaders,

I owe a lot to Bottlehead for starting me on my DIY journey, having started off with the Crack + SB and Crackatwoa (this was my Crackatwoa build log: https://forum.bottlehead.com/index.php?topic=10708.0).  Hard to believe that was only two years ago.

I've started working on my own personal from-scratch designs and builds.  I hope it is okay to ask about personal DIY projects on the forum?  The first is nearly complete, a MH4/REN904 MOSFET CCS loaded input, 6A5G SET output design using Lundahl iron (mid-build photo attached).

I am now working on my second design.  Nothing innovative, but at a high level, it is a MOSFET CCS loaded 6J5 input and MOSFET CCS loaded 45 parafeed output headphone amp.  I am currently working through the various quirks of a parafeed design

I have schematics and whatnot, but I will not bore you with the details unless asked.  I really just wanted to ask the forum if my own conceptual understanding of a MOSFET CCS loaded parafeed stage is accurate, if you would be so kind (gosh there are a lot of parafeed conceptualizers out there lately, huh?).  These are somewhat bouncing off of the questions asked here by Deke609.

My parafeed stage cascode CCS uses the IXYS IXTP08N100D2 (top) and IXTP08N50D2 (bottom), which from my reading provides an 500Mohm or more AC impedance at low frequencies (I haven't worked it out myself).  With that being said, my understanding is that the load the tube "sees" can essentially be treated as the OPT primary impedance alone.  No matter how heavily the secondary is loaded, the magnitude of the CCS AC load will dwarf the OPT primary in parallel.  The CCS is basically a brick wall to AC signal at all frequencies.  I made this assumpton when drawing the load lines for my design.

With the heavily loaded OPT primary provided by a high-impedance headphone (I use 300ohm) and the relatively high inductance provided by a parafeed OPT primary (137H of the Sowter 8665 I am considering), the low frequency response does not seem to be a cause for concern from my manual calculations and verified using Bode plot of my circuit model in LTSpice.

While this advantage of CCS loading the output tube is very appealing, there are two sacrifices: required B+ and power dissipation.  I am using a low-voltage bias point for my 45 tubes of 180V, so a reasonable 400V power supply can be used.  However, the 45 CCS top device must dissipate 6W at a 180V 31mA bias point, which is a bit of a challenge to dissipate.

The last sacrifice is determining the parafeed capacitor value with a CCS load.  I have seen anode choke calculations out there, some provided by Paul, but there does not seem to be an accurate way to calculate the cap size when CCS loading.  Is the best approach to get a swatch of different high voltage electrolytic capacitors, pop them in the circuit, get a Bode plot to see the affect on frequency response, then substitute an equal value film cap when the proper capacitance is determined?  Using LTSpice, I can view the effect of the parafeed cap value on the low frequency reponse and resultant subsonic LC resonance with the OPT primary, but I can't say whether or not it is a reliable way to determine the value...

The last practical concern for this design is not so much related to parafeed as it is the 45 tube.  I have been told that because the 45 can have high amounts of internal noise, a high-turns-ratio output transformer should be used.  I was originally looking at the Sowter 8665, wired in either 12:1 or 6:1, but it was suggested to me to use a full-sized speaker parafeed OPT like the Sowter 8983 with a 25:1 or 17:1 winding ratio.  Otherwise, this internal noise would be an issue with headphones.  This high turns ratio results in a flat load line for the output tube with a 300ohm load, and power output from 30-60mW.  Is the internal noise of the 45 something anyone can comment on from personal experience?  If a high-turn-ratio OPT was used, an option would be to put a low value resistor in parallel with the headphone output, say 9.1ohm, to lower the load seen by the tube to 8ohms.

Thank you kindly for input

[Paul Birkeland]

The 5K:8/16 transformer recommendation is a good one.   I just finished a similar project and found that the high step-down ratio was a good idea.  It wasn't heater hum or power supply buzz that was audible, but rather tube rush.  Since you're on our forum, I'll plug our SEX Iron ugprade package for this, with is a 40H plate choke and 4K OPT that would be perfect for this project.

Yes, CCS dissipation is pretty brutal at 6W.  You'd want a big heatsink, maybe 7C/W or lower to whisk that heat away.  You could also load the 45 with a pentode to move that heat above the chassis plate, which would help with the lifetime of all the components inside the amp. 

I would recommend buying a few values of Solen caps to try out.  4.7uF, 6.8uF, 8.2uF, and 10uF would be worth trying with headphones and speakers.  My intuition says that anything between 5 and 10uF would probably work just fine.

[L0rdGwyn]

Thanks for the recommendation on the iron upgrade kit, I will definitely check it out.

That is a beautiful amplifier, really like the color scheme and accents, has a sort of "steam punk" vibe.  Is it getting a production run how fortunate you had the same experience with the transformers - duly noted, I suppose it is all but certain then I will us a high-turns-ratio OPT then.  Maybe a ~17:1, the high turns will help my output impedance too for lower impedance headphones.

For the CCS dissipation, my options are A) large TO-220 PCB mount heatsink (Ohmite has one that does 3C/W, if I can fit it inside) paired with a reasonable amount of ventilation for the amplifier chassis, B) a large fin type heat sink mounted to the side of the chassis (ground isolated) with a drilled hole to mount the device directly onto it, or C) the pentode CCS.  The design of a pentode CCS is not something I am very familiar with, but I suppose I have to consider it.  I would really prefer to follow through with my MOSFET CCS, but like you said, I might be cooking my capacitors...

And absolutely brilliant suggestion on the Solen caps, I forgot they were so reasonably priced.  I'm sure I will find some use for the extras when the project is completed too.  Thanks for knocking out a few of my questions, PB.

[Paul Joppa]

As I said in my reply to Deke609 (posted March 8 at 04:45:05 PM), the parafeed output transformer is not a simple inductance. To model it accurately, you need to know a number of things, like eddy currents in the laminations and hysteresis of the core material. These things are nearly impossible to find out with sufficient accuracy to rely on simulations. Best bet is to measure and listen with several capacitances.

Here's what I can say about CCS-fed parafeed outputs that are lightly loaded:

The simple analysis gives a capacitance of L / R-squared, but the R is the tube's plate resistance rather than the primary impedance. This usually calls for very large caps. With more usual capacitors, you would expect a substantial resonance. I tried, but I couldn't measure that, because the transformer parasitic losses damp that resonance.

[Paul Birkeland]

You're very welcome for the advice.  On the Ohmite heatsink, be extra careful that they aren't giving you a 3C/W rating with some given amount of airflow.  I've nearly been tricked into taking those ratings as gospel without noticing that they are calling for a hurricane to be blowing over the heatsink in operation.

The pentode CCS requires setting a stable screen to cathode voltage (zener diodes are fine for this), then you can set the G1 to cathode voltage to set your standing current.  The challenge with these is that the cathode of that pentode will be at several hundred volts, so you will need to bias up the heater winding quite a bit, and this may preclude using the heater winding you would use on the driver tube with the pentode CCS. 

[L0rdGwyn]

As I said in my reply to Deke609 (posted March 8 at 04:45:05 PM), the parafeed output transformer is not a simple inductance. To model it accurately, you need to know a number of things, like eddy currents in the laminations and hysteresis of the core material. These things are nearly impossible to find out with sufficient accuracy to rely on simulations. Best bet is to measure and listen with several capacitances.

Here's what I can say about CCS-fed parafeed outputs that are lightly loaded:

The simple analysis gives a capacitance of L / R-squared, but the R is the tube's plate resistance rather than the primary impedance. This usually calls for very large caps. With more usual capacitors, you would expect a substantial resonance. I tried, but I couldn't measure that, because the transformer parasitic losses damp that resonance.

I see.  In that case, it seems an experimental approach really is best.  Thank you Paul, that is very helpful.  I'll get an array of Solen capacitors and alligator clips to swap the caps and be prepared to take measurements, both subjective and objective.  I'll just have to budget out an extra, extra large space for the capacitors, just in case.  I certainly hope larger the 10uF is not necessary.

You're very welcome for the advice.  On the Ohmite heatsink, be extra careful that they aren't giving you a 3C/W rating with some given amount of airflow.  I've nearly been tricked into taking those ratings as gospel without noticing that they are calling for a hurricane to be blowing over the heatsink in operation.

The pentode CCS requires setting a stable screen to cathode voltage (zener diodes are fine for this), then you can set the G1 to cathode voltage to set your standing current.  The challenge with these is that the cathode of that pentode will be at several hundred volts, so you will need to bias up the heater winding quite a bit, and this may preclude using the heater winding you would use on the driver tube with the pentode CCS. 

Good point on the airflow.  The Ohmite datasheet for their largest R series heat sink states 3C/W at "natural convection", so that's promising.  They have some heat dissipation vs. airflow data as well.  I see the dilemma with the high cathode voltage on the pentode CCS.  I think my first approach is going to be assessing the feasibility of the heatsinked MOSFET.  If all else fails, I will get serious about using the pentode.  I'm never against adding more tube feng shui, but it does add a bit more complexity and footprint.  I'll do some reading regardless, even if they aren't used in this instance, I'm sure I will utilize them in the future.

[Paul Birkeland]

Here's a set of pentode curves for the E55L with the screen grid 125V above the cathode.  For what you're proposing to do, you could use a pair of 60V diodes in series connected to the cathode, then a dropping resistor from B+ to the zeners to get yourself a screen supply that's reasonably well regulated with respect to the cathode.  Screen current is about 4mA, and you could look at the zener datasheets to see what kind of current they would like to maintain decent regulation in order to calculate the value for your dropping resistor.

40mA of current corresponds to 3.25V of bias on the E55L.  Screen current plus plate current is 44mA, so that's a 73 ohm resistor that you would place between the #45 plate and the E55L cathode.  Then you would connect the #45 plate to E55L G1.  The 73 ohm resistor could be a 100 ohm pot that you could adjust to get your desired plate current with some precision.

If you wanted to be a little lazier, if you have 400V of B+ and 180V on the plate, then you'd want the E55L screen at about 300V.  400V-300V=100V, and we have about 4mA of screen current, so that's a 25K resistor between B+ and the screen (throw in a 1uF cap from screen to cathode as a bypass).  This gives you your pentode CCS with two resistors and one capacitor, and you'll have an impedance of about 30K, so not too shabby.  I picked the E55L because the cathode can be 200V above the heater according to the datasheet, so you can heat it from the 6J5 winding.

The E55L would require a minimum of 25V of compliance, so there is an advantage there with the solid state CCS.

[L0rdGwyn]

Here's a set of pentode curves for the E55L with the screen grid 125V above the cathode.  For what you're proposing to do, you could use a pair of 60V diodes in series connected to the cathode, then a dropping resistor from B+ to the zeners to get yourself a screen supply that's reasonably well regulated with respect to the cathode.  Screen current is about 4mA, and you could look at the zener datasheets to see what kind of current they would like to maintain decent regulation in order to calculate the value for your dropping resistor.

40mA of current corresponds to 3.25V of bias on the E55L.  Screen current plus plate current is 44mA, so that's a 73 ohm resistor that you would place between the #45 plate and the E55L cathode.  Then you would connect the #45 plate to E55L G1.  The 73 ohm resistor could be a 100 ohm pot that you could adjust to get your desired plate current with some precision.

If you wanted to be a little lazier, if you have 400V of B+ and 180V on the plate, then you'd want the E55L screen at about 300V.  400V-300V=100V, and we have about 4mA of screen current, so that's a 25K resistor between B+ and the screen (throw in a 1uF cap from screen to cathode as a bypass).  This gives you your pentode CCS with two resistors and one capacitor, and you'll have an impedance of about 30K, so not too shabby.  I picked the E55L because the cathode can be 200V above the heater according to the datasheet, so you can heat it from the 6J5 winding.

The E55L would require a minimum of 25V of compliance, so there is an advantage there with the solid state CCS.

Thanks for the example, PB.  Let me take a day to dig into the pentode CCS topic, I will be back!

[L0rdGwyn]

Hey PB,

Been doing some reading on pentode CCS.  My big take away is the major advantage of using a pentode lies in the effect of greatly increased internal resistance due to leaving Rk unbypassed.  It should be equal to:

r(ccs) = ra + Rk*(mu+1)

So if I am understanding correctly, I would expect the resistance of the E55L CCS to be higher?  Maybe I am overlooking something, let me know if this makes sense:

mu = gm * ra = 0.045A/V*20,000ohm = 900

r(ccs) = 20,000ohm + 73ohm*(900+1) = 85,773ohm

I've been looking over datasheets for some other potential pentodes for this circuit targeting ~31mA g2+cathode current, came up with what I think might be the best option (with some drawbacks, of course), courtesy of Morgan Jones, the EL83.  Attached are the pentode curves at 170V on g2, would target -3V on g1 for a 28mA bias, with Ig2 at ~4ma, would give me close to 31ma out.  Check out these flat curves!

If my understanding/math above is accurate, here is what I would expect for a EL83 r(ccs):

mu = 130,000ohm * 0.010A/V = 1,300

Rk = 3V / 0.028A = 107ohm cathode resistor

r(ccs) = 130,000ohm + 107*(1300+1) = 269K

Obviously a lot less than what is achievable with a MOSFET device.  With a 5K:16 wiring and a 300ohm secondary, would give a reflected primary of 93,750ohm.  A 16ohm resistor in parallel with the 300ohm headphone on the output would drop that back down to 5K, which is an option.  Still, the EL83 has a Pa of 9W, so should be able to handle the voltage drop and get that heat above the chassis.

Two drawbacks with the EL83 are the requirement for elevated heaters (Vkf max is 100V), and the drop out voltage of around 70V.  If my load lines are accurate, I think the 70V dropout will work with my 400V B+.  As an example, I've attached a load line for a 5K primary 8ohm secondary.  The max voltage on the negative grid swing is limited by the symmetrical 0V grid point on the positive swing.  With a 400V B+ and 31V on the 45 grid, overhead is:

400-266-31 = 103V

Also attached a schematic of what the CCS might look like, need to figure out the value of the g2 to cathode bypass cap.  I included grid stoppers, and the heaters would need to be elevated at least 100V.

Gotta do my due diligence and see if this would work, but if my numbers are accurate, the performance hit is pretty massive vs. the cascode MOSFET solution.  Have to weigh it against figuring out effective cooling.  More tubes is always fun
though

Edit: fixed the orientation of the schematic below.

[Paul Birkeland]

In the model I worked through, the impedance was pretty dominated by that screen dropping resistor and bypass cap.  These components would still be in parallel with your CCS, but of course the screen bypass cap is reactive so the CCS impedance becomes reactive as well.

IMO there is definitely a point of diminshing returns when you have something like a #45 with an RP of 1.6K, which is relatively low.  I wouldn't get too excited about going much beyond 16K at 20Hz, which would have performance like a 125H plate choke.  I've also attached some measurements of a random output transformer showing what impedance is actually reflected when unloaded vs. loaded vs. shorted.  This is the end of the world as we know it when a pentode is driving the output transformer, but not so much for a triode that has an amplification factor that's rather stable with changing loading impedance.  It also illustrates that your 5K:16 transformer won't reflect a 100K load in the real world.   

When you move away from the ideal transformer model, you have to deal with all the abnormalities and limits in the transformer design itself that determine the shape of that unloaded curve and the shorted curve, as those are operational boundaries.

[L0rdGwyn]

In the model I worked through, the impedance was pretty dominated by that screen dropping resistor and bypass cap.  These components would still be in parallel with your CCS, but of course the screen bypass cap is reactive so the CCS impedance becomes reactive as well.

IMO there is definitely a point of diminshing returns when you have something like a #45 with an RP of 1.6K, which is relatively low.  I wouldn't get too excited about going much beyond 16K at 20Hz, which would have performance like a 125H plate choke.  I've also attached some measurements of a random output transformer showing what impedance is actually reflected when unloaded vs. loaded vs. shorted.  This is the end of the world as we know it when a pentode is driving the output transformer, but not so much for a triode that has an amplification factor that's rather stable with changing loading impedance.  It also illustrates that your 5K:16 transformer won't reflect a 100K load in the real world.   

When you move away from the ideal transformer model, you have to deal with all the abnormalities and limits in the transformer design itself that determine the shape of that unloaded curve and the shorted curve, as those are operational boundaries.

Very interesting...thanks for the measurements, that does put the transformer limitations in context.  And of course the bypass cap and dropping resistor are in parallel, I overlooked it.  I suppose they limit the achievable performance of a pentode CCS in this configuration.  Well good performance is still very achievable with the EL83, especially if aiming for 16K at 20Hz.

Well thanks for humoring my questions, I'll leave you be.  Maybe I'll make a post here when the amp is complete and update on the design I ultimately went with.  Much appreciated!

[Paul Birkeland]

Sure, keep us updated on your project!

[L0rdGwyn]

Hey PB and Paul - I have been digging deep into the pentode CCS topic for a few days, there is surprisingly little information out there on using them.  Anyhow, I have all but committed to using the EL83 as my parafeed CCS.  Hoping I can ask you a question about it first, have a little picture of it below.

Before my question, relating to PSRR at 120Hz I have found this design has 31dB of attenuation.  This will work for my power supply since I will have about 2mV rms ripple at the top of the EL83, that will be knocked down to around 55uV before it hits the OPT, so that's okay.

My question is on AC impedance vs frequency.  As you pointed out, with the 1uF bypass cap, the CCS is reactive, with the impedance rising as frequency decreases, sort of the opposite problem than the choke loaded parafeed design.  With a large g2 dropping resistor on a small signal pentode CCS, the contribution from the cap to the overall impedance might be pretty small, but not so much the case here.

For example, at 20Hz, the impedance is ~61.6K, at 100Hz ~58K, and things start to level off around 200Hz.  A friend of mine did a Spice sim of PSRR vs frequency, shown below that shows the trend.

Okay finally my question: will this increasing impedance at low frequencies have an audible effect on the frequency response?  Or a subtle shift on the AC load line with higher impedance at low frequencies?  I guess I am just wondering if this will be an audible issue, or if I should forget about it and move forward.  Increasing the value of the bypass cap would flatten things out a bit, but I am sure that will have some other consequence.

Trying to figure all of this stuff out ain't easy, a lot of respect for what you guys do, every new design change opens a host of new issues to tackle!  Thanks guys.

[Paul Birkeland]

Okay finally my question: will this increasing impedance at low frequencies have an audible effect on the frequency response? 

The math for gain of a grounded cathode amplifier with a cathode bypass cap is:

Av = (mu*RL)/(RL+rp) where rp is the plate impedance of the tube an RL is the load seen by the tube.

In your case you can get a snapshot of gain at a given frequency to make this determination.  Your RL will be the CCS impedance in parallel with the parafeed cap reactance plus the OT reactance.  The output of this is in terms of amplification factor, so then you can take the log of that number and multiply it by 20 to find the gain.  The mu of the #45 is 3.5 and I would use 1600 ohms as the #45 rp, which will be close enough (to save you some time looking it up).

Av = (3.5*RL)/(RL+1600)  To go down 3dB from an infinite load, you would need to drop from that infinite load to about a 4K load. 

[L0rdGwyn]

Okay, let me give this a shot...

My calculated impedance at 20Hz is ~61.71K, with a 5K primary, corresponds to a RL of 4625, gain of 8.30dB

Calculated impedance at 1kHz is ~57.13K, with a 5K primary, corresponds to a RL of 4597, 8.28dB

So a boost of 0.02dB at 20Hz relative to 1kHz......yeah, this doesn't matter at all.  Thanks for showing me, PB, I'll save for formula for next time.