• These commercial threads are for private transactions. diyAudio.com provides these forums for the convenience of our members, but makes no warranty nor assumes any responsibility. We do not vet any members, use of this facility is at your own risk. Customers can post any issues in those threads as long as it is done in a civil manner. All diyAudio rules about conduct apply and will be enforced.

Universal filament regulator

An interesting thread... I'm also in the process of designing a 300B amp using Western Electric (300B) tubes. So this (thread) has been a good read and some food for thought. What I've not seen is a discussion of the various DHT types, i.e., the filament structure, alignment within the tube itself and what coverage the filament has within the confines of the grid/plate structure. Not all (DHTs) share the same filament/cathode construction and as such some heater drive circuits may work better with one type vs another.

As an example, I'll use the 45 triode and then compare the original single plate 2A3. The filament design of the 45 is very simple. It uses a single filament wire which is arranged in a "M" configuration. The two top peaks of the "M" are tethered above the grid/plate structure and the two bottom ends of the "M" plus the center peak (or dip) are tied to wire supports on the bottom of the tube below the grid/plate structure. The filament voltage is applied to the two bottom ends of the "M". This arrangement gives the 45 a symmetrically aligned filament within the grid/plate structure. Also note that less than the entire filament wire is contained within the grid/plate structure, so any calculations and/or theories which consider the entire length of the filament wire and it's associated heating voltage as interacting with the grid/plate structure (current flow) would be inaccurate. With with the 45, you have 4 segments of the filament wire which can interact with the grid/plate structure. As the 45 has a symmetrical filament, using an AC heating voltage "should" be a preferred method, as the the AC signal for heating should cancel itself out as the ends are out of phase with each other. Using a fixed DC balance (as the filament wire should have an even cross-section and linear resistance) is logical and effectively balances the filament as a cathode. A balance pot to null any remaining AC component (which is caused by uneven coating of the filament wire, a slight physical alignment error, etc.) should result in quiet operation, assuming good quality tubes. By contrast, using a DC heating voltage skews the bias towards one end of the filament and any ripple component can not be reduced by the symmetrical filament. In the case of a 45, I would lean towards an AC heater supply as it looks to be a better fit and feasible to obtain quiet operation.

The single plate 2A3 has a different arrangement. The filament is a "series-parallel" arrangement per the RCA manual. It's filament has two "sets" of wires. Each set is arranged to make 4 vertical passes thru the grid/plate structure held by a tension bar across the top with a pair of pusher springs. Note it's not a "M" structure like the 45. The heater voltage is applied to the ends of the wire (like the 45). The two sets are sitting side by side within the grid/plate structure, i.e., one set covers the left side of the grid/plate structure and the other set covers the right side. Once again, the entire run of the filament wires are not contained within the grid/plate structure, so any calculations/theories are as above with the 45. The parallel reference is how the two sets are tied together. Where the two sets meet in the middle of the tube, the ends are tied together as one filament contact. The two outer ends of the filament sets are then tied together as the other filament contact. As a result, it's impossible to heat a SP 2A3 with AC and get anything close to an acceptable output noise level.

The Western Electric 300B has a similar filament arrangement and can not be heated with AC and achieve an acceptable output noise level. So, heating either tube will require a DC filament supply for quiet operation. As the purity of the DC can affect the performance of the tube, attempting to calculate it is difficult, as it's hard to measure what percentage of the filament is within the grid/plate structure. This makes it difficult to accurately calculate distortions, noise, etc. which are the result of filament supply noise. Whether constant current or constant voltage is used, you can probably make a good case for either but results may be more subjective than measurable.


With ac-heating I imagine that the filament structure and anode/grid overlap may influence any assessment of modulation effects - specifically any effects due to time varying cathode-anode voltage (due to the filament voltage alternations).
But with dc heat, the fields remain static, so will there be any special considerations in this case? I can't see any reason for it.

If the dc is noisy, then that's a different matter of course. But it is not too difficult to achieve ripple levels of <1mV even without any voltage feedback loop at all.

If you try your 45s on ac heat, take some spectrum from simple tones and look out for the 50/60 or 100/120Hz sidebands - the amount seems to vary from type to type, so maybe your structure observations can be witnessed here, acting together with the low voltage [2.5V] across the filament in the case of the 45.

Even with acceptable levels of hum, though, you still face the other bugbears of ac-heat:

- exposure to the emissions at the transformer secondary. Screened windings mandatory, common-mode chokes, and ferrite sleeves are helpful. (but these do not compare with the insertion loss offered by 50pF of NPN transistor, in the case of a good transistor solution);

- the transformer secondary (<0.5 ohm for a 2.5A winding) shunts the filament [2 ohm], so you get some circulation of cathode currents around this long loop, subjected to noise along the way, dielectric exposure, etc. Trying out different transformer construction types shows widely different sound, if you want to quantify these effects.

on top of the modulation effects!

I have not tried the 45 personally, but with the DHTs I have experimented on, the sound of ac-heat is murky and inarticulate compared to properly implemented dc. In fact, the quickest way to tell whether you have made a good filament regulator, is to quickly switch between ac and your creation, and listen. The difference should be unmistakeably positive.
 
Administrator
Joined 2004
Paid Member
In this vein tubelab has commented significantly on IM distortion generated by 50/60Hz AC heating in the 45. It's quite significant, might want to ask him about it..

Noting that all of this could be regarded as off topic from the original purpose of the thread, perhaps it would be a good idea to discuss this somewhere other than Tom's thread? (Unless he is OK with it.)
 
Noting that all of this could be regarded as off topic from the original purpose of the thread, perhaps it would be a good idea to discuss this somewhere other than Tom's thread? (Unless he is OK with it.)

I would prefer that this thread is returned to a build thread dedicated to switchmode, constant voltage, DC filament supplies. That's why I started the thread in the first place.

While the discussion of various filament types and how different types of filament supply topologies impact the performance of a DHT amplifier stage is interesting, I really do believe that it would be better served in a separate thread. I would like the SNR or on-topic/off-topic ratio of this thread to be higher than it currently is.

Thanks,

~Tom
 
Agreed... was not my intent to start another thread discussion here, only to point out that much is being discussed about effects of the cathode wire, types of DC heating and deeper aspects, but nobody was considering the actual physical construction of the DHT in use and I think it matters. So back to our normal scheduled programming....

Regards, KM
 
Agreed... was not my intent to start another thread discussion here, only to point out that much is being discussed about effects of the cathode wire, types of DC heating and deeper aspects, but nobody was considering the actual physical construction of the DHT in use and I think it matters. So back to our normal scheduled programming....

No worries, no hurt feelings here. I do enjoy a good technical discussion, but I also want the SNR of this thread to be reasonably high.

Anyway. Today's musings involve the LM20333. That's a nice little regulator. Compared with the LM3102 it offers higher output current (3 A vs 2.5 A) which would allow some margin when used with a 2A3 for example. It also has a tad better line and load regulation. The main drawback seems to be that it's a traditional current/voltage feedback controller, hence, stability vs a wide range of output currents may be a bit tough to achieve. I'll have to do the math and see...

~Tom
 
Last edited:
Mucked some more with the circuit. Reducing the switching frequency from 800-ish kHz to 600-ish kHz (R2 = 68 kOhm rather than the 47 kOhm I was using) made the IC run some 15~20 degrees (C) cooler. It did not affect the ripple voltage, only its frequency.

I also made another observation (which is probably old hat to most switchmode designers). There is considerable voltage drop across the ground plane. In some places I can measure upward of 20 mV of drop. Needless to say, in the final implementation, the top and bottom plane will have to be stitched together in several places to ensure a low-impedance ground plane.

~Tom
 
Trying the LM20333 to get a little more output current capability (3.0 A vs 2.5 A) and I'm also hoping to gain a little more efficiency. The LM20333 has the option of using an external schottky diode in parallel with the synchronous MOSFET. This should yield slightly higher efficiency (--> less heat dissipated).

I just threw the schematic together. I'll need to get crackin' on the board layout and test.

~Tom
 

Attachments

  • FilamentReg_2p0.png
    FilamentReg_2p0.png
    33.5 KB · Views: 722
And the winner is......

Folks,

So thus far I've tried the LM3102, LM20333, and LM22673.

The LM3102 and LM20333 are both in 20-pin eTSSOP with exposed DAP that need to be soldered to the ground plane for heat dissipation. The LM3102 is a constant on-time (COT) regulator, the LM20333 is a traditional current mode regulator.

As I've reported earlier, the LM3102 is rock solid. But I find it to be a tad marginal for use with, say, a 2A3 that requires 2.5 A for the filament as the LM3102 is a 2.5 A regulator. Hence, I've been on the hunt for a 3 A regulator.

The LM20333 is a 3 A regulator. It works just fine, but the ripple voltage on the output is on the order of 30 mVpp (2 A load). The pin-out of this chip is not the most efficient for a 2-layer board layout. Both the input voltage, ground, and switch node are pinned out on both sides of the chip. Hence, the most efficient layout is to route Vin down from one end of the chip and the switch node up from the other end. However, that reduces the thermal path to the DAP to around 5 mm in width. So the chip gets rather hot in normal operation (about 90 C after 30 minutes of delivering 2 A @ 5 V). Granted, I don't have thermal vias in place on my prototype, but still...

Then I stumbled upon the LM22673. 3 A output current, TO-263 Thin package with exposed DAP (very hand-solderable), a very intelligent pin-out, and traditional voltage control with internal compensation. What's not to like?
So far I like it. After 30~45 minutes running flat out at 2 A, 5 V out, the top of the chip is 45 degrees C. The board and surrounding components are barely above lukewarm. The DC load and line regulation are quite impressive.

Here's a brief recap. Attached pictures show the LM22673 schematic, setup, and measured ripple (20 MHz BW).

LM3102:
Efficiency (12.8 V in, 5 V @ 2 A out): 87.4 %
Ripple voltage (5 V, 2 A out): Approx 8 mVpp
Line rejection (change in Vout for a 10 V change in Vin): 36 mV (49 dB)

LM20333:
Efficiency (12.8 V in, 5 V @ 2 A out): 89.5 %
Ripple voltage (5 V, 2 A out): Approx 30 mVpp
Line rejection (change in Vout for a 10 V change in Vin): 4 mV (68 dB)

LM22673
Efficiency (12.8 V in, 5 V @ 2 A out): 90.7 %
Ripple voltage (5 V, 2 A out): Approx 6 mVpp
Line rejection (change in Vout for a 10 V change in Vin): 1.1 mV (79 dB)
Load regulation (change in Vout for a 2.37 A change in Iload): 600 uV (72 dB)

For those doing the math at home, the 600 uV change in Vout for a 2.37 A change in the load current amounts to an effective output impedance of 253 uOhm. Yes. uOhm...!

And the winner is ..... the LM22673.

~Tom
 

Attachments

  • FilamentReg_3p0_Board.jpg
    FilamentReg_3p0_Board.jpg
    230 KB · Views: 682
  • FilamentReg_3p0_Ripple.jpg
    FilamentReg_3p0_Ripple.jpg
    139.3 KB · Views: 659
  • FilamentReg_3p0.png
    FilamentReg_3p0.png
    23.5 KB · Views: 670
Nice work Tom!

Would you happen to know what the difference between the LM22673 and the LM2673 is? I designed one with the LM2673 but haven't built it yet.

Thanks!! I'm quite pleased with it as well. The Nichicon R7-series caps rock! 7 mOhm ESR. I use it for the 150 uF/16 V output cap. I need to do a couple of layout tweaks and try a few other standard voltages, then I'm ready to commit to a board run.

I don't have access to any special information other than what's on the National website. But from my knowledge of the business, it looks like the LM22673 is an updated version with higher switching frequency and improved line/load regulation. Not to mention updated package. The LM2376 uses the thick version of the TO-263 package. The thin version used in the LM22673 is nice but a little more difficult to hand solder as it doesn't have a tab. You have to flow the solder for the DAP through thermal vias. With a good soldering iron, this shouldn't be an issue, though.

500 kHz switching frequency seems to be a good compromise between efficiency and output ripple. It's nice when the switching frequency falls reasonably close to the SRF of the output cap.

~Tom
 
I finally had a couple of hours of free time to work on this filament regulator project. I've measured the line regulation (ripple rejection) for three different output currents at 5.0 V out. See attached.

At 60 Hz, the ripple rejection comes in at 75 dB. Not shabby...

Now I need to go and build a test rig to measure the load regulation.

~Tom
 

Attachments

  • LM22673_LineReg_5V.gif
    LM22673_LineReg_5V.gif
    14.3 KB · Views: 653
Output impedance

Now that the rainy season has officially started, it's time to start up on the filament regulator project again.

I put together a test rig to measure output impedance. The results are attached. < 3 mOhm @ DC. The peak at 10 kHz is the resonance of the output LC filter. The Y axis is the output impedance in ohms. (300 mOhm at the top of the chart, 2 mOhm at the bottom).

~Tom
 

Attachments

  • LM22673_OutputImpedance.gif
    LM22673_OutputImpedance.gif
    12.1 KB · Views: 499
Would a single regulator handle a pair of 841s, which would be a total of 7.5V @ 2.5A, or is that pushing it too far?

That should be just fine. I've run the regulator flat out at 2.5~3 A and it gets slightly warmer than lukewarm. That's on my prototype board which is not thermally optimized. The output voltage in my case is 5.0 V. So I think it should be just fine with 7.5 V @ 2.5 A.

For the final version, I'm getting a "real" board made with thermal vias under the IC and a thermal plane on the bottom. To make it economical I'm getting a good handful of boards made and will be selling the excess through my website.

Actually, the 2.5 A drawn by the 841 and 2A3 was the reason I went with the LM22673 over the smaller regulators. The goal is to make this regulator as flexible as possible and usable with as many tubes as possible.

~Tom
 
The max load is 3 A regardless of output voltage.

Although, the regulator is rated at 3 A continuous, I'd probably leave a 10 % design margin. The regulator doesn't start current limiting until well past 3 A (5.1 A peak as I recall, but I forget what that translates to for average or RMS - depends on the inductor current).

My intent is to have a universal design that hits all the popular tube voltages from 1.5 V to 12.6 V (including 4.1 V for the Russians). Ideally by swapping only one resistor, but I may have to have a few different inductors to choose from as well.

~Tom
 
I'll probably be in for a couple when you get done.

Sweet!

Perhaps you can provide an opinion here... The regulator itself takes up very little circuit board area. The input rectifier and reservoir cap take up about as much room as the regulator. Would you prefer a board with a rectifier or without?

Onboard rectifier would require only AC to run. The board without rectifier would require rectification to be done elsewhere and the board to be fed DC. The board with onboard rectifier would be roughly twice the board cost of a regulator-only board.

Others are welcome to chime in as well, of course.

I'm leaning in the direction of a regulator-only board. There are many different packages for rectifiers and reservoir caps, so arriving at a "one size fits all" is a challenge. And both rectifier bridges and reservoir caps have traditionally been chassis-mounted anyway.
The other advantage of regulator-only boards is that they stack nicely. I could probably stack 3~4 of them before running out of room in a 3" tall chassis.

~Tom
 
WebBench shows operation down to a load current of 0.3A, so I expect it could be used for a single 4P1L without issues. A pair puts it near peak efficency.

I'll be interested in a couple to several. Since this is for DHTs, etc, does one need one power supply for each filament? It seems one should not short the filamentss together unless it is a PP pair.

I should go back and look at some of Wavebourn's 4P1L designs.