Headphone amplifier power supply

The amplifier can run on a single 300V power supply. A downside of most tube amplifiers, this one included is the poor power supply rejection rate, or how much noise on the power supply ends up in the output signal. The White cathode follower has a fairly good PSRR for a tube amplifier but the ECC83 input stage has a poor PSRR. To test this in simulation I used a 310V power supply with a 1V 100hz or 10Khz sine wave on it. With no input voltage applies a 10mV sine was visible on the output at both 100Hz and 10Khz. This is about 100x less then the power supply noise or an PSRR of 20db. Compared to the 100db PSRR opamps can reach 20db is fairly low.


To make sure there won’t be a audible 100Hz noise from the power supply on the headphone the power supply has to be well designed. With 32 ohm 100db/mW headphones an 1mV 100Hz signal still generates 0.3nW, 30nW sounds like nothing but with 100db/mW this equals almost 40db, enough that it is audible. With 20db PSRR this means that 100mV of noise on a 300V power supply is hearable in some headphones. A standard tube power supply with a few big caps and resistors is not going to cut it here.

The better option is a regulated power supply, a bit like an 7805 or LM317 power supply but then running at 300V. As a normal LM317 is not capable of going up to those volatages without some tricks it’s time for an MOSFET follower. There are more then enough cheap MOSFETs that have no problems with 300V or more, even at 100 or more mA’s. Add a few high voltage zener diodes and a simple regulator is born.


After rectification and the big 47uF capasitor there still is 15Vpp of 100Hz noise at the input of the MOSFET. At the output with an 100mA load, just 40mV. As the tube amplifier uses about 50mA’s this power supply is good enough for the job. With different zener diodes different output voltages can be used making it a versatile schematic.

Tubes need to get warm and cozy inside before they start working so an 30 second delay was added using an NE555 timer and a relay that run of the 6.3V filament power supply.

An LT1085 was added to regulate the 6.3V filament power supply. After all the effort for a clean 300V the last thing we want is the tube filaments ruining our amplifier.


PCB’s where made for the power supply and ordered at elecrow in china. A few weeks later they arrived and building and testing a power supply was a piece of cake.

Everything worked as expected, the only issue is that I forgot a resistor that empties the big caps over time. Luckily the tubes use a constant 20mA per channel so when the power supply is in use this shouldn’t be an issue. The Eagle files can be downloaded here and are CC-BY-NC-SA licensed.

All tube headphone amplifier

As a hobby project I decided to build a tube headphone amplifier.

Why tubes you might ask, well, because they are awesome. Looking at efficiency, distortion etc transistors are vastly superior. But tubes are fairly simple to work with, look cool and it’s something different then all the opamp designs.


  • Can drive 32 ohm headphones
    • Some tube designs without an output transformer has issues driving low ohm headphones, as I also use normal 32 ohm headphones I need to have an amp that can drive them
  • Can drive 250/600 ohm headphones
    • As I also use a Beyerdynamic 990 pro with an 250 ohm impedance the amp also needs to be able to drive these headphones.
  • Low(ish) distortion
    • As it will be a tube headphone it’s impossible to get the ghastly low distortion figures that are possible with an opamp or transistor design. Still the distortion has to be low
  • Low noise
    • headphones tend to be extremely sensitive, 100db of sound with just 1mW of input power is nothing special. So the noise on the output has to be low, even 1mV of noise will be heard through some headphones.
  • Low(ish) price
    • It’s still a hobby project and I’m still a student. I can’t really spend hundreds of euros on an amplifier.

Circuit topology:

I want a fully tube design, not a hybrid design. Getting tubes to drive low low impedances like 32 ohm is not a problem with an output transformer. The downsides of output transformers is that multiple tabs are needed when you want to drive 32 and 250 ohm headphones, meaning they will have to be custom made. This means they will be expensive and hard to get.

An output less tube amplifier is quite common for headphones, designs using an 6AS7 or 6080 tube as cathode follower with half a tube per output are common. The downside is that the 6AS7 needs quite a bit of juice to get glowing, 2.5 amps to be precise. The other downside is the distortion. Simulated it went over to 1% at a 1Vpp output.

A better approach is using a so called “White cathode follower”. This is a topology that uses 2 tubes per output in a push pull configuration. This topology can deliver more current then an cathode follower and has a lower output impedance. It is a bit complexer and it requires two tubes per output but it has no problems driving an 32 ohm headphone. After a bit of searching online I found this schematic using an ECC99 tube per channel:

They claimed it can deliver 2V rms into an 32 ohm load. This is more then enough to generate over 120db of music. The other nice thing is that I had a few 6N6P tubes laying around. These cheap Russian tubes are very similar to the more expensive ECC99 tubes. I didn’t have an ECC82, only an ECC83, a tube with a much higher amplification factor.

Luckily this design uses some negative feedback to lower the output resistance, meaning the amplification factor is adjustable. After messing around a bit in LTSpice I came up with the following schematic:


It can deliver 45mA’s or about 3Vpp into an 32 ohm load or a massive 20Vpp into an 250 ohm headphone. 600 ohm headphones can expect a silly 42Vpp, which will fry most headphones with ease. Simulated distortion levels are low for a tube amp. About 0.3% @ 2Vpp in 32 ohm or 0.1% @ 10Vpp in 250 ohm. That looks good enough to try in real life.

The ECC99 design is designed by Menno van der Veen and found on his website which has a really nice tube preamplifier: http://www.mennovanderveen.nl/eng/ts-vv-2006.html

The real deal

Time to build up the circuit idea from the last post on a breadboard. I decided to use an LM317 and a LM2670. These are both fairly easy to get, inexpensive and available in a TO220 package.  There are multiple versions of the LM2670, for this application the –adj version is needed.

The schematic is as following:


The schematic is fairly simple and is based on the schematic from the LT3083 datasheet.

And a breadboard picture:


As Farnell forgot to ship me the low ESR capacitors I ordered I used normal ones, so I expect quite some noise. The schematic works as expected, I can change the output voltage between 1.2 and 7V and the output of the SMPS stays roughly 2V higher. All photos are takes with 3.3V as output in a 10 ohm load, 330mA current flowing. As I said in my previous blog the noise using this should be lower than just using the SMPS, let’s have a look if I am right:


The noise at 3.3V 330mA output are small peaks that are roughly 200mV peak peak. The peaks are very narrow and are in the MHz range. When we look at the frequency it’s about 260Khz, which is the switching frequency of the LM2670, no surprises there. These peaks are caused when the LM2670 switches the internal MOSFET on. Now have a look at the noise from the SMPS.


The peaks are visible again but there is something extra, a triangle waveform, this is the capacitor charging and discharging. This is so clearly visible because of the fairly high ESR from my rubbish capacitors

As this is a lower frequency noise then the short peaks they are not visible after the LM317. This is because of the ripple rejection of the LM317. At 120 Hz the ripple rejection is 80db, this means a 1V ripple of 120 Hz will be lowered by 80db, or 10.000 times. Any ripple of 120 Hz will be almost completely removed by the LM317. But the higher the frequency the lower the ripple rejection. At 250 KHz it is roughly 30db or about 31 times. This would still lower a 100mV ripple to just 3mV, but the LM317 is just not fast enough to do something about the peaks.

As I said in the beginning of the blog I used normal capacitors, 100uF electrolytic capacitors. I also added a 1uF film capacitor as they have a low ESR. If I remove the 1uF the noise almost doubled, adding a second peak, a big peak where the LM2670 switched the internal MOSFET on and one where it switched off. With better capacitors the noise should get lower but this already shows the difference a small good 1uF capacitor can make. The fact that it’s build on a breadboard doesn’t help for the noise either, but all in all a successful experiment.


Power supply ideas

Time for the first actual blog. I’m working on a lab power supply at the moment and I’d like it to be efficient but still low noise. The best way to make an efficient power supply is a switch mode power supply or SMPS. Making one with a variable output voltage is not too hard. The big downside is noise. If the SMPS uses a switching frequency of 1 MHz, something quite common with efficient SMPS’s, making it low noise is difficult. Not only you need a fairly big capacitor to get the ripple low, the capacitor needs to have a very low ESR, if the ESR is too high the capacitor won’t be able to charge in the short time it has with the 1 MHz switching frequency. A lower frequency helps a bit but achieving a low noise of something around a 1mV is almost impossible.

Cheaper power supplies use a linear regulator like the LM317. These devices have a low ripple and noise output without using expensive low ERS capacitors. The downside is the efficiency, with a 20V input and a 5V output at 1A a loss of 15 Watt is happening, meaning a big heatsink is required.

Why not combine these two, giving the low ripple of a linear regulator but the efficiency of a SMPS. making an SMPS that has an output voltage of 2 to 3 Volt higher than the requested output voltage would be great, this would only mean a 2 to 3 Watt loss in the linear regulator.

This kind of circuit is possible and called a tracking pre-regulator. An example of this is mentioned in the datasheet of the lt3080 or lt3083 device from Linear Technology.


Using the 10K and 200K resistor the output voltage is set to roughly 16.5V. But they also put a MOSFET in there. This schematic actually uses the threshold voltage of the MOSFET to set the output of the SMPS. The output of the SMPS is roughly Voutlinear regulator + Vthreshold. Using this schematic there will be a loss of 1.5 to 3V depending on the used MOSFET, much better then with just the linear regulator. The downside is the extra costs and more complex schematic because of the extra parts. The noise will be more then with just the linear regulator but much better then with only a SMPS.

It also works wonders in LTSpice, the LTSpice can be found here: http://sdrv.ms/13vRUDb

The SMPS in this schematic is an LT part as it’s from an LT datasheet. Linear Technology makes great IC’s but most of them are a bit expensive, the LT3680 is not different at almost 10 euro’s in singles. The schematic should work with quite some SMPS IC’s like the cheaper LM2670 from Texas Instruments. The LM2670 is also available in a TO220 package, great for a hobbyist. Time to order one and build it up!

The first post :)

Look at this, my first post on my first blog, who knew. Something about myself, I am a 20 year old student from the Netherlands studying Embedded System Engineering. I also started my own company with 2 other students called BRC-Electronics which you can read more about on www.brc-electronics.nl.

I will try and use this blog to post about my electronics experiments and thoughts as a form of reference for myself and for others.