Nixie Tubes are cool retro looking decimal digit displays useful for many modern DIY projects like the venerable Nixie tube digital clock. The Nixie tube, invented in the 1950’s, can provide a great fusion of old display technology with new innovations. Unfortunately, one major difficulty in using them is that Nixie tubes need voltages up to 170V to energize. While this voltage can be made several ways, a convenient way and the subject of this blog is to generate this voltage from a 5v supply. This will allow the use of the same 5v supply used to power the Raspberry Pi, Esp8266, Arduino or other microcontroller that controls the display or IOT project. Most solutions already available on the internet required a 9-16v input supply. This design will operate from a 5v supply input supply, minimize the PCB footprint and use low profile and readily available parts to maximize the space available to the Nixie tubes. In this blog, the schematic, Layout, design equations, and simulations of the design are provided. Not counting the circuit board, the maximum component height of the proposed design is 10mm (the inductor) and the footprint size is just 14.5cm squared. The calculated efficiency is over 90%. Of course, no design is complete until you confirm with experimental results. In Part 2, we will discuss the final results and experimental verification.
PCB Layout Design Considerations
The PCB Layout was designed to be a building block component with four mounting holes, a USB micro B input connector and a terminal block output connector. Attention was made to maximize the ground plane, use a shielded power inductor, EMI beads (L2 and L5), and maximize bulk capacitance to minimize conducted and Radiated EMI noise.
- Input voltage: 5v.
- Output Voltage: 170v
- Maximum Design Current: 20mA
- Input capacitance: 200uF
- Output Capacitance: 2uF
- Switching Frequency: 50kHz
- 10mm Max Component Height (by Inductor)
- 14.5cm^2 Footprint size.
- Input Connector: USB Micro b
- Output Connector: Terminal Block
- M2 Sized Mounting Holes
- Shielded Inductor
- EMI Beads in series with 5v Input and 160v output Nets
All the files for the KiCad Schematic, Kicad Layout, Spice Simulations, and efficiency Calculations are available on Github.
- Kicad Schematic, PCB Layout, BOM
- Spice Simulations (LTSpice (TM))
- Boost Converter Efficiency Calculations
DCM Boost Converter Topology Design
This design uses the discontinuous conduction mode (DCM) boost converter topology to achieve the high 170v output voltage, high efficiency, fast transient, and small footprint. DCM gets its name because the boost inductor current resets to zero amps every switching cycle. Two other possible design approaches are the continuous conduction mode (CCM) boost converter and the flyback converter. With only a 5v input, the CCM boost converter is extremely limited in transient performance and has an extremely hard time to reach a 170v output voltage (the duty ratio must be more than 97%), while the flyback requires a custom transformer and you need to deal with leakage inductance losses. From an efficiency point of view, the DCM boost converter looks quite good and is comparable and possibly better than both of these approaches. for a low profile power supply, it is hard to beat the DCM boost converter and the simulations and equations shown later back this up.
The design focused on achieving 170v output voltage with the highest efficiency possible and with the smallest footprint. The 5v to 170v DCM boost converter design is managed by the TI UCC3803 controller. This controller has the advantage of operating from a 5v supply. The discontinuous conduction mode (DCM) of operation was chosen for this design to ensure we could reach the high output voltage and low profile without the need for a custom flyback transformer winding. The 33uH power inductor ensures DCM at 50kHz switching frequency up to the maximum full power (3.6Watts). The peak current is limited to around 4amps by resistors R8, R10 and R12. The DCM mode of operation also allows for a fast response and simple compensation method. The input and output capacitors are all surface mount ceramic capacitors for ultra-low footprint, long lifetime, Good filtering and minimal ESR. Additional ferrite beads L2 and L5 reduce input and output ripple further and help with EMI. The soft start time was increased by C14 from the default 4ms value programmed into the UCC3808 to approximately 20ms.
Key Design Specifications
(boost flyback design.xlsx file at Github site shows the complete calculations)
- Input Voltage (Vin): 5v
- Output Voltage (Vout): 170v
- Output Load Current (Io): 6 x 3ma = 18ma
- Maximum input power (Pmax): 3.6Watts
- Power Inductor: 33uH
- Typical Duty ratio at full power: 63%
- Inductor Ripple Current (Pmax): 1.9A
- RMS Power FET current : 870mA
Calculating Duty Ratio
A DCM boost converter does not follow the same conversion ratio found with a CCM mode boost, namely that D = 1- Vin/Vout, where D is the power FET duty ratio. Instead, in DCM, the duty ratio is governed by the following equation:
Where Io is the output current, L is the Power Inductor, Ts is the switching period (1/F). With the above equation and the proper inductor value, the duty ratio of the DCM boost required to reach 170v will be less than 97% (the value of Duty for the CCM boost) and will bring with it some nice advantages. In this design, the nominal duty ratio at full power is close to 60%.
Design starts with the Inductor Equation
The first step of the design was to choose a power Inductor so that the DCM mode of operation was maintained all the way to the maximum power. The following equation determines the maximum inductor value. In general, the higher the inductance, the lower the system losses. However, in practice, a value about 50% of this value is needed to provide fast loop regulation and account for component losses that counter the ability to generate the high output voltage of 170v.
Where Po is the output power. The calculations from the equation above give a value of approximately 80uH. The final value chosen was 33uH, about 50% of the calculated value.
Ultra Fast Regulation
A spice simulation shows the fast transient response, stable compensation, and good regulation with this DCM topology. The DCM mode of operation designed to operate around 60% duty ratio at full power provides the fast regulation. At 60% duty ratio, there is plenty of control range (i.e. 60% to 100%) to maintain the output voltage. Also, since the inductor energy is fully transferred to the output each cycle, the plant mode of the converter is reduced and the duty ratio can be rapidly changed without having to worry about the inductor dynamics since you start at zero current each cycle. Capacitor C6 of 100pF is used to place a zero before the loop gain crossover frequency to add phase margin. With a full load step response, the output voltage drops only 1v (less than 1% from 170v).
The step response above also shows a well-damped recovery response, representing a phase margin of more than 45 degrees.
Calculated 91% Efficiency at full power
The DCM boost converter has zero Diode reverse recovery losses and does not have any flyback transformer leakage inductance loses, offsetting the higher RMS losses in the power FET and sense resistor. While the complete efficiency calculation worksheet is available at Github, the four losses from the power FET, power diode and power inductor equations are shown below.
- Power Fet conduction Loss: 68mW
- Sense Resistor Loss: 75mW
- Diode Conduction Loss: 11mW
- Power Inductor Conduction Loss: 35mW
- Total Calculated losses including snubber loss and gate drive: 264mW
Power FET Conduction Loss:
The equation below shows the Power Fet Losses are directly related to the power FET Rdson value. In this design, the chosen power FET, FDS2672CT, has a nominal value of 90mOhm to minimize this value.
Sense Resistor Conduction Loss:
The sense resistor is the low ohmic resistor in series with the power FET. This resistor is used by the UCC3803 to control the peak current of the power switch. The equation is just like the Power Fet Loss with the Rdson replaced by the sense resistance, Rsense. In this design the Rsense value is 100mOhms.
Power Diode Conduction Loss:
The power diode losses equal the diode on voltage multiplied by the average diode current. For the power diode in this design, the DFLS1200-7 is a Schottky type and has an on-voltage of 0.4V.
Power Inductor Conductive loss:
The conduction losses in the power inductor are also directly related to the equivalent series resistance. In this design, the power inductor has a Resr = 45mOhm.
The next step is a verification of the design. The board is being manufactured and I’ll follow this up with experimental efficiency, step response results. I would like get your questions or comments so please comment below.
- Wikipedia entry on Nixie tubes: https://en.wikipedia.org/wiki/Nixie_tube
- Datasheet of the IN-4 Nixie tube: http://www.tube-tester.com/sites/nixie/data/in-4/in-4-sh2.htm
- A nice design of a 25 cm^2 footprint 9v to 160v CCM boost power supply with 85% efficiency. Layout and schematic provided: http://desmith.net/NMdS/Electronics/NixiePSU.html
- A simple 9v input power supply: http://www.instructables.com/id/High-Voltage-Power-Supply-for-Nixie-and-Valve-Tube/
- More techniques and discussion for generating Nixie power supply: https://threeneurons.wordpress.com/nixie-power-supply/
- LTSpice is a registered trademark of the Linear Technology Corporation
Just put together your board, but now moving over to troubleshooting Im kind of stuck.
My output shows -218mV instead of the 160-170V I was expecting. The inductor also gets hot really quick.
Could you give me a nudge in the right direction of how I could go about trying to figure this out?
After some more fiddling around I got it. Somehow managed to solder the diode (D1) the wrong way. Rocking 4 x IN12 tubes atm:) Thank you for a neat design Mark!
I’m super glad you got it working. Really cool!
First I want to say I think this is a really great project and I feel like I am really close to having it working. However, like the person who posted before, I could also use a nudge in the right direction. I ordered and populated three of your Rev3_1812 NixieSupplyMini boards to use in conjunction with your ETA_Nixie_Clock_Rev2_1810 tube board.
Both are populated using the components listed in the current BOM available on GitHub.
Unfortunately, none of the three power supply boards will output any level of HV. I can see anywhere from 4.3VDC – 5.6VDC depending on the outputs probed. Any help would be appreciated.
Also, just to confirm the recent design revision, which pins are supposed to be GND, HV, or +5. I did not notice that this was shown on the silkscreen. The previous design of 4 pins appeared to have two grounds on pins 2 & 3 while 1 or 4 were +5VDC or +170VDC. Just for confirmation
I will do my best to help. The symptom you are describing is that the power supply is not oscillating or working. When the supply is not operating the output voltage will remain approximately the value of the input voltage (i.e. 5v). With this info, my first guess is that you need to pull the N_ENABLE pin low (i.e. short to ground). This is Pin 2 of J6. If this is floating the power supply will remain off. I do this to ensure the 170v does not start until the Raspberry Pi will purposely pull the pin low.
On Rev 3 of the supply, I modified the connectors from 4 pins to J6 (3 Pins) with J1 (Two Pins) to increase the High Voltage spacing between the 170v and ground. Over time, I did not want the 170v to arc to the ground due to carbon buildup. So with this new configuration, the +170v spacing is greater than befoe. So with Rev 3 of thee PCB, J6 has the following PINOUT connection:
Pin 1: GROUND
Pin 2: N_ENABLE ( you need to connect to Ground to enable supply)
Pin 3: 5v
PLEASE DOUBLE CHECK WITH KICAD LAYOUT WHAT IS PIN 1 versus PIN 3. the PIN 1 is NOT shown on the silkscreen (something that can be fixed), and looks like it is actually pin 3 if you just look at the PCB.
J1 has two pins that are BOTH +HV or +170v.
I added an updated picture to the END of the blog (part 3) showing this PIN out.
Ok, so I double checked D1. It is oriented correct. D2 on the other hand I believe was not. I flipped D2 around and I do not have 170VDC but I do have 38.5VDC. Does that help?
I was looking to the square pad to indicate pin1 but was not sure if the second two pin was just to extend J6 to have 5 pins. I understand now regarding the separation of the ground.
I guess now I am curious if flipping D2 was the right thing since I was not grounding the enable ping. Also, I just recognized that R90 (1K) was for the enable.
You had mentioned something about R90 possibly not being needed any longer?
Never mind that was on the clock board enable I think.
Good note, the square pad indicates pin 1. You should still use R90 for protection .
Good note, the square pad does indicate pin 1. You should still use R90
Ok so with the enable jumper in place I get rouge 40VDC this seems to be the case with D2 in either orientation. I was pretty sure I saw that the cathode faced Q26, but I feel like the line on the case at times has indicated the opposite of what it should.
The best way to check diode direction is to use the diode checker with multimeter. In either case 40v is good news :), But still some work to do. What output do you get if you raise input to 6v? Do u have a very small current load or none at all?
Here are some other hints to try, but not sure they are cause. If u are using a simple USB supply, Double check current. Is should be 1+ amps. Also a short cable helps, Because some cables have high resistance.
Here are some more HINTs I hope should help:
1. Make sure Vcc (pin 7 of U1) is at 5v or higher. You can raise this above 5v, up to 12v, as long as you used 16v capacitor ratings.
2. 40v means that something is loading the supply, or a protection mode is getting engaged.
3. So make sure there is no load, then the load is just R15 and R16 or about 640k.
4. Typical protection modes are UVLO : Undervoltage lock out where the input voltage drops below 4v.
5. Another is overcurrent, possibly because R10 is the wrong value. If pin 3 of the U1 goes over 1.5v, the over current protection will trigger and shut the supply down for a little while. Then it tries to restart.
6. Overcurrent could trigger if the soft start capacitor is too small (C14 should be 47uF and R7 = 2k).
7. Make sure C14 voltage raises near 5v and that the voltage at Pin 1 of U1 is lower. this will ensure that D2 is installed correctly. If you are not sure of Q26 or always want the supply to start when 5v is applied, you could remove Q26. this will not allow the Raspberry Pi to control the power supply, so you should only do this during debug.
8. check the switching frequency by putting a high voltage probe on the drain of Q1. It should be about 50kHz
Hi Mark. I really appreciate your design work and nice write-ups.
I also have just built this supply and see an output of ~40V. I switched from a usb supply to a bench supply but see the same result. I can raise the output to ~100V by raising the supply to 12V. At 5V supply the circuit is pulling ~40mA (without load).
Probing around 1 obvious problem is that R/C input is oscillating at ~540kHz, not the designed 38kHz. The R and C and the correct values.
Under no load, 5 volt bench supply the Ucc3803 has the following measurements:
1. comp = 3.32V
2. fb = 0.441V
3. CS = 0.68V
4. ref = 3.98V
5. gnd = 0 (~100mV p-p ripple)
6. out = 540kHz high duty-cycle square wave
7. Vcc = 4.92V
8. r/c = 540kHz square wave
Looks nominal to me except for fb and r/c. I welcome any further hints.
Hi max, Good work on building it up and providing details. From your data, the most likely root cause is the switching frequency difference of 540khz versus the correct 38kHz. Because the boost converter operates in the discontinuous mode of operation, Your symptoms match what would happen if the switching frequency is too high.
This frequency is set by the passive devices R2, R12 and C7 around the UCC3803. Make sure R2 and r12 are placed correctly. The silkscreen is not clear and may be swapped. A previous reader had a similar issue. I’ll try to add picture of r2 and R12 to blog later this week.
Thanks and let me know.
R2 and R12 were indeed swapped. I was mislead by the silk screen and my poor eyesight.
Thank you very much.