The goal of this project is to implement a PCB which will allow the user to implement a vacuum tube filament supply using National Semiconductor’s LM2267x-series of switchmode regulator ICs. The board is to be universal in the sense that the user by proper choice of a few external components can design the regulator to power filaments in tubes ranging from directly heated tubes such as 26, 2A3, 4P1L, 300B, etc. to indirectly heated tubes such as 6J5, 6L6, and 12AX7. PCBs for this project are available for sale on the PCB Sales page.
The advantages of a DC filament supply are fairly obvious. Using DC to power the filament eliminates a source of hum. With DC filament supplies, the choice is between constant current and constant voltage. I have personally never been able to tell a difference in sound quality between an amp with filaments heated by a constant current and one with filaments heated by a constant voltage. Theoretically, there really shouldn’t be any difference — assuming the current/voltage sources are reasonably close to ideal. The characteristics of a well-designed, regulated supply approach those of an ideal source — at least within the audio band. A constant voltage supply is less expensive to implement, so that’s where I ended up.
To generate a regulated constant voltage, two choices are available; linear regulator and switchmode regulator. A traditional linear regulator is by far the easiest to implement. An LM317 IC, a couple of passive components, and you’re set. However, all linear regulators work by reducing the input voltage across a pass device. Hence, they dissipate a fair amount of power. After doing the math of the filament regulators for my 300B-based stereo amp I concluded that the power dissipated in the filament regulators under worst case conditions was about 20 W. This is more power than the amplifier can deliver to the speakers. I found this rather disturbing. Dissipating this amount of power in a heat sink inside an amplifier chassis without reaching scorching temperatures is quite a challenge. Hence, I started looking at switchmode regulators.
After looking at a handful of different switchers — LM2734, LMZ12002, LM3102 to name a few, I came across the LM2267x-series. These ICs come in various flavors that differ in options and output current. I ended up settling on the LM22673 as this offers a set of options that are very suitable for a universal filament regulator:
- Programmable current limit (up to 3 A for the LM22673; 5 A for LM22679).
- Output voltages from 1.285 V to well beyond 12.6 V.
- Soft start.
- Package options exist that allow for humans to solder these ICs onto a circuit board.
|R1, R2, R3||0805 SMD|
|C1||Electrolytic SMD||10 mm diameter|
|C2||1206 SMD||X7R MLCC|
|C4, C5, C7||1210 SMD||X7R MLCC|
|C6||400 mil (10 mm) diameter, 200 mil (5 mm) lead spacing||Electrolytic can, through-hole|
|C8||0805 SMD||X5R or X7R MLCC|
|C9||250 mil (6.3 mm) diameter, 100 mil (2.54 mm) lead spacing||Electrolytic can, through-hole|
|C11||Electrolytic SMD||Footprint supports 5 mm ~ 8 mm diameter|
|C12||Tantalum SMD||Footprint supports case sizes 6032-28 and 7343-31|
|L1||500 mil square SMD||Example: Coiltronics DR127-series|
|J1, J2||Male connector, leaded, right angle||Digikey P/N: 455-1648-ND|
|J1, J2||Female connector housing||Digikey P/N: 455-1183-ND|
|J1, J2||Crimp terminals||Digikey P/N: 455-1319-1-ND|
As stated above, the critical components, L1 and Cout depend on the operating conditions (Vin, Vout, Iout). In addition, to ensure the best transient response and best regulator loop stability, the output capacitor must be chosen carefully. I highly recommend using an OSCON or polymer type as these provide the most consistent performance as they age. I have settled on a half-inch square inductor type as this footprint is quite common. Shielded inductors provide the best bang for the buck. If you do not populate the current limit resistor, R3, you need to choose an inductor with a rated saturation current greater than 5.5 A.
While supporting all combinations of output voltages and output currents with one inductor-capacitor combination is highly unrealistic, I do intend to keep the number of L-C combinations to a minimum. Therefore, the values in the Known Good BOM spreadsheet reflect values that I find to provide good performance, while keeping the inventory to a minimum.
|Output Voltage, Current||BOM|
|5.0 V @ 1.4 A||FilReg_3P2_BOM_5V0.pdf|
|6.3 V @ 0.3~1.2 A||FilReg_3P2_BOM_6V3_300mA.pdf|
|6.3 V @ 1.2~2.5 A||FilReg_3P2_BOM_6V3_1A2.pdf|
This section is intended as a starting point for those who aim to eek every bit of performance out of this regulator and those who wonder how I arrived at the various component values. If you prefer a more “plug & play” solution, feel free to skip this section.
Input and Output Voltages
As stated above, the LM22673 was chosen as it provides soft-start, programmable output current limiter, can provide output voltages in excess of 12.6 V, and can supply the 2.5 A needed to light a 2A3 filament with some margin to spare. The choice of the LM22673 primarily drives the choice of catch diode, D1, which — worst case — will need to be able to handle the short circuit output current. For the LM22673, the typical short circuit current is limited to 4.2 A. Hence, a 5 A diode was chosen. One could argue that a 6 A diode should have been chosen, as the maximum short circuit current over temperature and process variation is 5.5 A. So worst case, the diode will not survive an indefinite short circuit on the output. It will survive a momentary short circuit, though. The choice of the CMS04 diode was also driven by its very low forward voltage drop. This minimizes the power dissipated in the diode, hence, maximizes regulator efficiency. The CMS04 can handle a reverse voltage of 30 V. As a rule of thumb, the rated diode reverse voltage should be 1.3× higher than the input voltage. Hence, the choice of the 30 V, 4 A CMS04 limits the input voltages to 23 V. Note that this is the absolute maximum input voltage. The design target should be around 18~20 V max.
Now that the input voltage is known, C1, C2 may be chosen. C1 provides local energy storage to avoid huge ripple currents in the wires that feed the regulator input. C2 provides most of the switching currents — this capacitor provides the load current when the regulator switch is on. It is critical to use a high-quality multi-layer ceramic capacitor (MLCC) for C2. X5R and X7R dielectric types provide that level of quality (low ESR, high self-resonance frequency) and can handle the required ripple current.
In addition to the input voltage limits imposed by the voltage ratings of C1, C2, and D1, limits on Vin and Vout are imposed by the minimum and maximum duty cycle the switcher can support. The minimum duty cycle, D, is 7.5 %. The maximum duty cycle is typically 90 %. Note that as
the duty cycle limits impose limits on the input and output voltages. In other words, the output voltage must be greater than 7.5 % of the input voltage. And the output voltage cannot be greater than 90 % of the input voltage. I suggest staying away from the extremes and ensuring that the regulator will operate within the range of; 0.25 < D < 0.75.
Choice of Output Inductor and Capacitor
The choice of L1 and Cout impacts overall regulator performance the most. L1 is selected from the duty cycle (Vout/Vin) and the ripple current. Conveniently, National Semiconductor has done the math for us and distilled it to this equation:
F is the switching frequency (500 kHz for the LM22673). Basically, the larger the inductance, L, the lower the ripple current for a given combination of Vout, Vin, and switching frequency.
As the output voltage ripple of the regulator is determined by the inductor ripple and the output capacitor equivalent series resistance (ESR), it would seem logical to choose L to be as large as possible. However, larger inductors are expensive and physically large. A reasonable compromise (according to National) is to choose an inductance which allows for a ripple current of approximately 30 % of the DC output current. Hence, IRIPPLE = 0.3 * Iout. The inductor also needs to be chosen to handle the peak current of the regulator without saturating. As the worst-case peak output current is 5.5 A, the inductor should be rated for a saturation current greater than 5.5 A. It should also be capable of handling the RMS output current without overheating.
The output capacitor needs to be chosen carefully as it plays a large role for regulator stability and load transient response. WebBench seems to pick L-C combinations that result in a L-C resonance frequency of 10~15 kHz. Except for the lower currents, where the resonance frequency is placed significantly lower, but the zero formed by the output cap and its ESR is placed in the 10~15 kHz range. Unless you’re up for the full stability analysis (in which case the pole, zero frequencies for the regulator loop are listed in the data sheet), I strongly suggest using WebBench and simulating the loop response to find a good starting point for L1 and Cout. When you specify these components, note, that not only are their values critical, their parasitics (ESR for the cap and to a lesser extent DCR for the inductor) also critical. In my “Known Good” designs, I have used WebBench to find components that would allow for a phase margin of 60° or, preferably, better.
The bottom line is; use the specified components. If you cannot find exactly the components specified, substitute components with equivalent characteristics in terms of inductance for L1, capacitance and equivalent series resistance (ESR) for Cout.
C3 should be a high-quality, low ESR ceramic cap. Any of the X5R or X7R dielectric types of ceramic caps should do the trick.
R1 and R2 set the output voltage. Pick a combination of resistors that allows for a few mA to flow in the resistors. This will serve as the minimum load on the regulator. The equation for these resistors is grabbed from the LM22673 datasheet:
R3 sets the peak current limit. This is a handy feature if an output current of less than 3 A is needed. As described above, the output inductor, L1, must be capable of handling the peak output current without saturating. Hence, by choosing R3 appropriately, an inductor with a lower saturation current may be chosen. This allows for the use of inductors with higher inductances while still fitting on the half-inch square footprint. The relationship between R3 and the typical peak current limit is found in Figure 2 of the LM22673 datasheet.
Notice that the current limit set by R3 is the peak current limit. Hence, to support an RMS current of Iout, the peak current must be at least Ipeak = Iout + IRIPPLE/2.
C9 provides the soft-start feature. U1 supplies a charging current of 50 µA to charge C9. The output voltage tracks the voltage on C9, hence, once C9 is fully charged, the output voltage has reached its target value. With C9 = 220 µF, a start-up time of approximately five seconds in achieved. This minimizes the in-rush currents in the tube filaments, hence, minimizes any turn-on related thermal stresses on the filaments.
This board is intensionally designed to be rather flexible. While I use the LM22673 for the filament regulator, it is possible to use the other members of the LM2267x family on this board as well. Jumper header JP1 is included for this purpose. JP1 allows access to pins 5 and 7 of the IC. These pins are used for various features of the LM2267x family. For example, the LM22670 uses these pins for the Enable and Sync features. See the relevant data sheets for how to use these features.
Assembly is fairly straight forward provided that one has access to a good soldering iron. I normally use a 1.6 mm chisel tip and 0.5 mm diameter solder for the smaller SMD components. For the leaded parts and bigger SMD’s, I use 0.7 mm diameter solder. The difference in diameter of the solder makes it easier to apply the necessary amount of solder without applying too much.
The footprint for C11 is a bit tricky in that it supports capacitor sizes from 5 mm diameter to 8 mm diameter. Hence, when mounting a small-diameter cap on this footprint, care must be taken to ensure that the capacitor is centered on the footprint. Otherwise, it is likely that the output of the regulator will be shorted. In addition, L1 can provide a challenge for a wimpy soldering iron as its pads have a fair amount of thermal mass. However, with a 6.3 mm chisel tip, I find that L1 solders down without incident.
The regulator IC, U1 provides the main soldering challenge as its DAP (metal pad on the bottom of the IC package) will need to be soldered to the thermal plane of the board. I find it easiest to solder the DAP first, then the IC pins. The following sequence is what I use. Click on the images for a larger view.
Step 1: Populate all surface mount components except U1.
Step 2: Apply a thin coat of solder to the DAP of the IC. Apply enough solder to the exposed part of the thermal plane to fill the thermal vias and form a thin layer of solder on the top of the board. Apply flux to the thermal pad as well as the DAP. I use a Kester brand RMA flux pen for this (Mouser P/N: 533-0186). The flux is probably not mission critical as there is some flux in the solder, but as the flux increases the surface tension of the solder, it makes it more likely that the IC will center itself on the pad as the solder is heated.
Step 3: Place the IC on the footprint. Suspend the board in the air either by a clamp/”third hand” device or by holding it by hand. Apply heat with the soldering iron to the exposed part of the thermal plane on the bottom of the board. Once the solder melts, the IC should gently settle in the center of the footprint. Remove the soldering iron and allow the board to cool.
Step 4: Solder the IC pins. The detail-oriented perfectionist will notice that I’ve used a touch too much solder here… I could also clean up the excess flux. Details, details…
Step 5: Mount the remaining components and test the board.
It should go without saying, but just in case… Test the power supply before hooking it up to your expensive vacuum tubes. Verify that the output voltage is within a few percent of the design target. Preferably, test the supply by applying a resistive load to the output. Note that the output voltage needs to be measured at the pins of connector J2 — not at the load. The voltage at J2 should be within a few percent of the design target, the voltage at the load will vary as a result of the resistance of the wires connecting the supply to the load.
When measuring the output voltage ripple using an oscilloscope, ensure that the probe leads are as short as possible and measure directly across the output connector, J2.
Measurements will be posted as they become available.
- Rev. 3.1: First production version.
- Rev. 3.2: Added footprint for tantalum capacitors, C12.