Adjusting 300B filament voltage

  • Thread starter Thread starter Deke609
  • Start date Start date
So 168C, or maybe D.  The Kaiju and SII are quite a bit more tolerant to line voltage fluctuations and have the universal power transformers that allow you to wire for lower line voltage if that comes up, so you may want to consider making this a BeePre only effort, as you could really dial things in if you only used the BeePre on the autoformer.
 
Many thanks PB. Good to know that it should at least work in principle. I will weigh my options. I do like the idea of having everything plug into one unit. Maybe even my dac as well. The bigger plan is to incorporate 3-4 combo common/differential mode filters (already built), each rated 3A, and have 3-4 pairs of outlets - with switches for bypassing filtering and possibly even the autotransformer. I'm even contemplating using some high current/voltage schottky's with 1.5V forward voltage drop on switches to better calibrate the voltage. Plus an analog voltmeter dial to monitor what's going on.


cheers and thanks, Derek
 
I'll add that much of fun for me is being able to think something up myself, build it, and have it work reasonably close to as planned. I 'm ok with the prospect that years from now (or perhaps much sooner) I'll look at the thing and realize that I could have achieved the same or better results for 1/5 the price and 1/4 the effort. And if I later choose to dismantle it, I'm pretty confident that the parts will get used in other projects. 


cheers, Derek
 
I have a working breadboard rebuild of the BeePre and am working on dialing in the filament regulator (LM1085 board).  I have a bunch of questions that I'm hoping @PB and @PJ can help me with.  Any directions/pointers you can provide will be greatly appreciated. I am taking my time with the BeePre rebuild and really want to find an implementation that works well with my EML 300B tubes.

In stock configuration the reg board is set up so that Vin - Vout = approx. 3.5V.  Based on my best attempt at making sense of the TI datasheet for the LM1085 (attached), it looks to me like the LM1085 only needs about 1.5V difference between Vin and Vout, and even @1.5V there's should be at least a 200mV cushion at typical operating parameters.  But the stock config has an additional 2V of difference - which would otherwise suggest to me that there is huge cushion to prevent reg dropout,  but my past experience suggests that this isn't necessarily the case and that the reg can drop out even where Vin - Vout > 1.5V.

So my first question is: is some other part of the fil reg circuit dropping the additional 2V?  As best as I can tell, the first half of the fil reg circuit is identical to the TI "ripple rejection enhancement" model found at 8.2.9 of the datasheet, except for the use of 33uF caps where TI has 10uF caps.  But the last half of the reg circuit looks like its based on what I've learned is called a "Sziklai pair" -  close cousin to the "Darlington pair" but with reported thermal regulation, impedance and possible lower voltage drop benefits over the latter -- I dunno about any of this: trying to read about it made my head spin.  Related second question: What does the second half of the fil reg circuit do? / why is it there?

My next set of questions is about noise and ripple rejection.  When the regulator drops out, where does the resulting hum come from - is it power supply ripple being passed to the filament? or is it generated by the reg circuit itself?  If the former (passed power supply ripple), then my separate filament trafos with CLC filters that reduce ripple to a modeled (PSUD2) 8 mV pk-pk should help to keep things quiet if the reg drops out (for comparison, my PSUD model of the BP stock voltage doubler/rectifier shows about 2.25V of ripple being fed to the inputs of the reg board - all of which is eliminated by the reg circuit).  But if the reg circuit is the source of the dropout hum, then no amount of ripple rejection before the reg circuit will make a difference.

To minimize the incidence of dropout, I'm thinking of increasing Vin to 14V instead of 13.5, while keeping Vout at 10V.  I can't tell from the datasheet whether this decreases ripple rejection / increases noise. Does it?  And if it does, do you think some or all of the increased noise would be offset by the increased ripple rejection of my CLC filter? If my basic arithmetic and use of an online dB calculator is correct, then 2.25V pk-pk (stock) : 8 mV pk-pk (CLC) is 281:1, or about -24.5 dB.

I anticipate that increasing Vin by 0.5V will put more thermal stress on the regulator. I plan to deal with heat stress to the LM1085 via (a) replacing the stock mica spacers with "kerafol keratherm" insulators that are reported to be more heat conductive than mica and comparable to a combination of mica and thermal grease; (b) forced air cooling @ 400 cfm.  But this only takes care of the LM1085 (hopefully). Might other parts of the board - e.g., the transistors, need additional cooling if I bump up Vin?

Apologies for the numerous questions, but I really want to get this right.

cheers and many thanks in advance, Derek
 

Attachments

Congratulations! You are discovering the large amount of work involved in developing a detailed design! I'l offer a few comments,, but you'll still have to do the work.

There is some headroom on the dropout to allow for variations in powerline voltage, plus more to allow for ripple on the input line.

The second part of the BeePre regulator  is a broadband noise cancellation circuit. You should be able to estimate the voltage loss.

Power dissipation is easily calculated if you know the voltage and current in a component. Temperature is more challenging due to variation of the environment.

In my experience, you want to keep the junction temperature of silicon components below 100C for longevity, even though a specified maximum temperature is usually around 150C.
 
Many thanks for the pointers PJ. I am happy to do the work.math myself - otherwise I won't have a clue about what's going on.

Paul Joppa said:
Congratulations! You are discovering the large amount of work involved in developing a detailed design!

Well, what I'm trying to do is a far cry from designing a filament regulator - but even trying to tweak the values of your regulator is proving to be a sizeable research and learning endeavor.  :)

I'll look into the voaltge drop of the broadband filter and do some power.heat dissipation calculations as suggested.

Just one repeat question: when the reg drops out, is the resulting hum power supply ripple passed to the filaments or noise produced by the regulator itself? This is important b/c if it is power supply ripple, I will be sure to find room for my CLC filter; but if the hum is from the reg itself, then I'll focus even more on preventing reg dropout and consider leaving out the CLC filter if space is limited.

cheers and many thanks, Derek
 
Well ... I couldn't figure it out from the datasheet, so I did a listening test on the breadboarded BP. The filament heater circuit of the right channel has a CLC filter that delivers 13.9V to the LM1085 board, and the equivalent of a 3.87R cathode resistor that results in 4.85V dropped across the EML 300B filament (I hope to bump that up to 4.9 by tweaking Rset2 to bring Vout to 10V from existing 9.91V). The left filament heater circuit is stock, and delivers approx 13.45V to the LM1085 board, and has a 4R cathode resistor, resulting in 4.75V dropped across the EML 300B filament.

I listened to one channel at a time - starting at 120VAC mains via a variac and then dialed down the mains voltage until I got hum (dropout).

Right channel: hum started at 107VAC - hum is fairly quiet and doesn't get much louder if I dial down the mains voltage even further (e.g., to 100VAC).

Left channel: hum started at about 113.5VAC - hum is loud and buzzy, and gets much louder as mains voltage is dialed down more (e..g, to 108VAC, at which point the hum was really loud) [This is a bit surprising b/c in the summer I was getting dropout at mains VAC < 117]

So my provisional conclusion is that at least a good part of the hum from reg dropout is either (a) power supply ripple itself, or (b) caused by power supply ripple.  But I can't rule out the reg producing some of the hum, independent of power supply ripple.

And for building purposes I've learned two things: (1) a modest bump in voltage feeding the LM1085 board (in this case, just less than 0.5V) gives me a huge buffer for dealing with variances in mains voltage and preventing dropout ; and (2) in the event of dropout, the CLC filter tames the hum considerably.  So I plan to incorporate both.

I haven't done any heat dissipation calc's yet, but I'm hopeful and somewhat confident that the combination of new heat transfer compound (in place of the mica) and forced air will do the trick. I seem to recall the Ohmite datasheet for the heatsink showing that forced air results in significantly cooler temps. I'll have another look.

cheers, Derek
 
I would also mention that it took PJ (and me to a lesser degree) months to figure out how to make the BeePre sufficiently quiet.  Don't be surprised if it's a long road.
 
It's already been a long road!  ;D  I've been at this since July.

Based on today's listening test, I think the filament reg setup is looking good and all I need to worry about is thermal overload. I'm turning my attention to that now.
 
Okay. I've done some very rough heat dissipation calculations.

I started from the following formula set out in the Ohmite heatsink datasheet:

TJ = TA + PD (θJC + θCH + θHA)

where,

          TJ is junction temperature,
          TA is ambient temperature,
          PD is the power consumption,
          θJC is thermal resistance of junction to case (stated on datasheet to be 0.7 C/W)
          θCH is thermal resistance from case to heatsink - I'm assuming this is the thermal resistance of the insulator
          θHA is the thermal resistance of heatsink to ambient air


I've also made three simplifying assumptions:

(1) that TA (ambient air temp) is the same in the stock BP as in my future rebuild. I think this is a conservative assumption since I will be adding a lot of vent holes on the top plate around the tubes and the pc boards, the new chassis is much larger than stock and will have more open space, there will be at least an inch air gap at the bottom of the chassis (min 1 in feet), and the two 140mm x 140 mm fans will draw cooler air from outside the chassis

(2) that the total difference between Vin and Vout is dropped by the LM1085

(3) that the reg board draws a current of 0.5A [correction: 0.05A] in addition to the current drawn by the filament (this seems high - but I figure it will do for rough calculations) and that all current passes through the LM1085


Assuming the above, this means I need only compare PD *  θ-total for stock and rebuild.

Stock w/ Regular 300B w/1.2A fil current
       
θJC (junction to case) = 0.7 C/W (from datasheet)

θCH (case to heatsink) = 0.5 C/W  (can’t find hard figures – 0.5 was the lowest stated I found for a mica insulator w/o thermal grease/paste)

θHA (heatsink to ambient) = 3.5 C/W @ 100 linear feet per minute (can’t find estimate for natural convection, so I am using 100 lfm from the heatsink datasheet, which is the lowest forced air rate indicated – this seems high to me)

So PD (θJC + θCH + θHA) = 3.5 * 1.25 * (0.7 + 0.5 + 3.5)
                              = 20.56


Rebuild w/ EML 300B w/ 1.3A fil current

θJC (junction to case) = 0.7 C/W

θCH (case to heatsink) = 0.09 C/W  (thermal resistance of Keratherm Red. This seemed insanely low to me until I check the thermal resistance of Kerafol's other types of Keratherm insulator pads. Red is their best. Their worst thermal performer is "Keratherm Brown" @ 0.44 C/W, - so I am inclined to trust the specs)

θHA (heatsink to ambient) = 2.5 C/W @ 200 lfm (reduced from my calculated 291 linear feet per minute which gives 2 C/W - just to be conservative)

PD (θJC + θCH + θHA) = 4 * 1.35 * (0.7 + 0.1 + 2.5)
                          = 17.82

So, based on this admittedly rough estimate it looks to me like it might work. 

If I've mucked this up, I'd appreciate someone telling me  :)

cheers and thanks, Derek
 
Just a note - the National Semiconductor data sheet specifies thermal resistance as 0.7 C/W for the control section, and 3.0 C/W for the output section. TI seems to have changed how they do thermal calculations, and in a quick look I don't see them making the distinction. I've always used the 3.0 figure, which is compatible with most other TO-220 packages.
 
Many thanks PB and PJ.

Using 3.0 C/W results in the following estimations: stock 30.65, rebuild 30.25. Since I think I was fairly conservative in estimating cooling for the rebuild and more liberal in estimating cooling of stock, this gives me enough confidence to proceed with the rebuild.  I have the great advantage of having a proven comparator (stock) that I know was designed to perform within the proper thermal operating range of the LM1085.  So all I need is a temp rise ≤ stock and I should be good.  At the very least, my rough estimate suggests that actual thermal performance of the rebuild will not be significantly poorer than stock. Once it's built I can use my cheap thermocouple to measure case temperature and use the reading to estimate junction temp.  If it's below 125C, all is good. If it's below 100C, all is great.

If more cooling is needed, I figure I can bump up airflow by adding a funnel to each fan to direct more/faster air at the LM1085 heatsinks.

And now back to trying to figure out layout - you'd think that having more than twice as much area to play with (320 in-sqr versus 144 in-sqr) would make layout a breeze, but it's not!  The CLCLC HV and CLC LV filters take up a quite a bit of room. And I still need to figure out how I want to power the fans - I have a separate trafo that would do the trick, but that will take up even more space. So I'm considering taking power from the Pt-7 LV windings. And either way, I'll need a small CLC filter to get proper DC.

But I'm having fun. I got my second wind after hearing the hum-free breadboarded BP. Man ... that hum issue drove me insane - example: removing, filing, cleaning and reinstalling the house ground connection to the water supply pipe at 2AM on a Saturday night!  Did nothing, of course, but it was badly corroded and loose, so in the end I'm glad I did it.

cheers and thanks, Derek

 
Deke609 said:
... I did a listening test on the breadboarded BP. The filament heater circuit of the right channel has a CLC filter that delivers 13.9V to the LM1085 board, and the equivalent of a 3.87R cathode resistor that results in 4.85V dropped across the EML 300B filament (I hope to bump that up to 4.9 by tweaking Rset2 to bring Vout to 10V from existing 9.91V). The left filament heater circuit is stock, and delivers approx 13.45V to the LM1085 board, and has a 4R cathode resistor, resulting in 4.75V dropped across the EML 300B filament.

I listened to one channel at a time - starting at 120VAC mains via a variac and then dialed down the mains voltage until I got hum (dropout).

Right channel: hum started at 107VAC - hum is fairly quiet and doesn't get much louder if I dial down the mains voltage even further (e.g., to 100VAC).

Left channel: hum started at about 113.5VAC - hum is loud and buzzy, and gets much louder as mains voltage is dialed down more (e..g, to 108VAC, at which point the hum was really loud) [This is a bit surprising b/c in the summer I was getting dropout at mains VAC < 117]

So my provisional conclusion is that at least a good part of the hum from reg dropout is either (a) power supply ripple itself, or (b) caused by power supply ripple.  But I can't rule out the reg producing some of the hum, independent of power supply ripple.

And for building purposes I've learned two things: (1) a modest bump in voltage feeding the LM1085 board (in this case, just less than 0.5V) gives me a huge buffer for dealing with variances in mains voltage and preventing dropout ; and (2) in the event of dropout, the CLC filter tames the hum considerably.  So I plan to incorporate both.

@PJ or @PB - I may have found an error in my reasoning. I previously concluded (above) that the extra 0.5V DC input to the LM1085 reg board greatly lowered the mains VAC dropout point. But is it just the extra VDC? Or does a significant reduction in incoming ripple account for some or much of it?  My previous conclusion was premised on the idea that the reg board acts as a filter that smooths the voltage supply - but it just occurred to me that maybe it "shunts" the ripple (still not sure how the reg circuit works), in which case incoming ripple voltage  reduces max regulated DC output.  If it "shunts" the ripple, then I I think it follows that I can reduce the input VDC to 13.5 (as provided on PJ's schematic), and still have a lot of dropout headroom, AND run the reg board cooler - b/c I'm using a CLC filter in front of the reg board that knocks down ripple considerably.

Am I right about this?  I'd test it myself, but the wood chassis mock-up rebuild is almost done and I don't want to risk damaging parts with any more desoldering/resoldering than is necessary. But if the above makes sense, I will tweak the incoming VDC of the reg board once the amp is up and running again.

many thanks in advance, Derek
 
When I calculate regulator performance, I look at the input (DC with ripple) and take the lowest voltage - i.e. average DC minus peak ripple - then allow further margins for dropout and line variations.
 
Paul Joppa said:
... I look at the input (DC with ripple) and take the lowest voltage - i.e. average DC minus peak ripple ...

Many thanks PJ. I think that confirms my suspicion. Once the mock-up of the amp is functional, I'll do a bit of experimenting with dropout. If I'm lucky, the CLC filter in front of the reg board may even allow me to go lower than 13.5V input and still be dropout-proof down to 115VAC mains.

cheers and thanks, Derek

 
Back
Top