Introduction

The main reason for building a new set of amplifiers came with purchasing a set of old studio monitors, the legendary JBL 4333. These 75 Watt speakers with 15″ bass drivers needed a amplifier that could deliver some more punch than my 20 Watt EL34 amplifier.

For a long time I have had a large quantity of 6P45S (PL519 equivalent) sweep power tetrodes lying around and have therefore looked for a amplifier design using these tubes.

Safety

WARNING!: Working with electricity is dangerous, all information found on my site is for educational purpose and I accept no responsibility for others actions using the information found on this site.

Read this document about safety! http://www.pupman.com/safety.htm

Considerations

When searching for a amplifier design to follow, I came across a Hungarian article on rebuilding a old amplifier called the APX-100. It’s was based on the PL509 tubes. I added in some good ideas from a user on diyaudio.com to use self balancing preamplifier and phase splitter. All modified in regard to the design of the EL34 amplifier by Claus Byrith.

The User Kruesi on diyaudio.com gave a good explanation of his design ideas.

After considering many splitter topologies, I finally settled on using a Long Tailed Pair, since

a) it is able to swing to the full supply rail unlike the split-load “accordian” splitter”

b) Both outputs are equal and opposite, unlike the floating paraphase types in which one output has almost no second-order THD and the other output does – and the outputs also have different clipping behavior

The LTP avoids both these issues, but performs acceptably only when operating into a constant current sink. This is often approximated with a large cathode resistor, but that’s usually a long way from an active constant current source. Of course a tetrode or pentode could also be used for this at the expense of great complication. A constant voltage at the base of a bipolar device translates into a constant current at its collector, given a high beta. Seems like just the thing…

Rather than derive the base voltage of the bipolar current sink from a fixed (regulated) voltage source, it’s derived from the combined plate voltages of both halves of the 12AX7.

DC analysis:
The two 820k resistors are equivalent to a single 410k resistance fed ftom the plate voltage of either section of the 12AX7. the value 820k is chosen to be much higher than the 82k plate loads of the 12AX7 so they should have minimal effect on plate loading.

I’d like about 1mA Ib for each section, running the plates at about 200V (as we’ll see). The two 820k resistors combine to 410k, and in series with the 2200 ohm resistor form a divider to produce about 1.2V at the base of the NPN device. Subtracting 0.6V for Vbe we have 0.6 V across the 330 ohm Re. Thus the emitter current is 1.8 mA. For devices with high beta, the collector current is about this same value, so each half of the 12AX7 has a cathode current of 0.9 mA. Since Ip=Ik, the 82k plate load has 0.9 mA through it, dropping 75V from 300, leaving the plate voltage at 225. (It’s not exactly 200 V due to the fact that I’m using 0.6 V as the value for Vbe in this example -the actual value is slightly higher).

The fun starts when we look at the AC signal:
If the two halves of the pair are perfectly balanced, one plate will be swinging more positive while the other is swinging more negative, and the combined AC voltage at the junction of the two 820k will be zero, leaving only the 200 VDC component.

Let’s say the two halves don’t have identical mu, and the input side has a higher gain than the feedback side of the pair. In this case, the voltage at the junction of the 820k will be an AC signal, out of phase with the input signal. This causes an AC variation on the base voltage which in turn modulates the collector current in such a way as to place an AC signal on the cathodes in-phase with the input signal, of exactly the right amplitude to cancel out the excessive gain of the input side of the pair.

It can be seen that the AC balance of the differential pair is now primarily dependent on the match of the two 820k resistors, and is now much less dependent on the intrinsic mu of each triode section. Using standard 1% resistors with no special matching, I measured a 65 dB Common Mode Rejection Ratio (both halves of the splitter driven from the same source). Very good balance indeed!

So now we have a self-balancing circuit without the need to hand-select 12AX7s, also a very high impedance current sink in the cathode circuit, and also a form of local feedback within this stage to improve balance.

Since the 6SN7 driver also operates as a differential amplifier, we may as well employ this same technique there as well, to preserve good balance going into the KT88s.

Specifications

Schematic

Power supply for one mono block

Mono block amplifier

Datasheets

Phase splitter 6N8S

Construction

An estimation of the power output this amplifier is capable of is to look at the full output tube plate voltage swinging across the primary side of the output transformer.

470 Volt peak is 332 Volt RMS over half of the primary resistance of 3500 Ohm giving us around 190 mA. So the power through is around 63 Watt and taking losses and rounding into account, it is fair to say this is a 50 Watt amplifier at low distortion. The output transformers can however not handle this output power so the bias will be adjusted for lowest power but still in the linear range. The amplifier will just have to be driven in a sane manner and never played with maximum input voltage.

9th April 2013

I came across two 50 Watt output transformers and one power transformer to a very good price. As I wanted to build mono blocks I contacted the company that originally made the transformers and had a second identical power transformer constructed at a very reasonable price. The transformers is made by Dagnall electronics located in Britain and with production on Malta.

I decided to build a prototype without a PCB so changes was easier to make and there would be room for experiments and complete rebuilds.

21st August 2013

The test housing is all made from scrap metal and so will the final version be. I am planning to order nicely painted front covers to have a professional finish to it.

17th October 2013

Tube sockets and transformers are placed to minimize influence between components and the possibility for lots of airflow around the tubes.

24th October 2013

The first version of the firmware for the ATMega16 micro controller is written, it is basically a 4 page menu system on a 16×2 LCD display that can be flipped through by the push of a button. Code examples will be made public later when the software is thoroughly tested.

9th November 2013

The amplifier circuit itself is soldered directly on to the sockets and a ground bar runs through the middle of the amplifier. The heater wiring is done in stiff, thick and twisted wire with good clearance and 90 degree angle to the signal wires.

The filter capacitors on the power supply board are mounted on the normal side and all diodes and resistors are mounted on the backside. With the PCB facing downwards the capacitors are shielded from all intense heat sources and will only experience the ambient temperature.

12th November 2013

The power supply resistor values are chosen to give the right voltage under load, the load is represented by large power resistors, voltages will have to be double checked with the tubes as load instead.

Before turning on the amplifier for the first time the bias balance potentiometer is adjusted to the middle position and bias voltage adjusted for most negative voltage possible. This first adjustment can be done with amplifier on at full voltage but with the output tubes taken out.

Very low voltage testing of the amplifier, only 115 VAC through a variac, showed that it worked fine and could amplify a sine wave from the signal generator. As soon as input voltage came above 180 VAC, the speaker would suddenly click and the fuse for the high voltage would blow.

A sure sign of high frequency parasitic oscillations. What comes next is a long journey to locate the source of these oscillations. As I only had old used tubes, I tried to change the output tubes but without any improvement, not even after four different.

Grid resistors on the 6P45S tubes was changed from 2K2 Ohm to 10K Ohm to follow the more conservative high frequency stopper design of the APX-100 amplifier. No noticeable change.

The 175 VDC supply for the screen grid was in my first layout tapped through a resistor from one of the capacitors in series for the high voltage, this unbalanced the power supply greatly and I made a 175 VDC linear MOSFET regulation directly off of the high voltage. Parasitic oscillations still occur.

The feedback signal from the secondary side of the output transformer had a long signal path in a single wire, I changed it to a screened cable with screen connected to ground. Parasitic oscillations still occur.

I had used wire wound resistors for the screen grid, exchanged them for carbon resistors without any noticeable improvement.

High frequency bypass capacitors, value 4.7 nF, was installed from filament supply legs to ground on the output tubes. Parasitic oscillations still occur.

Pulling the phase splitter tube out when the parasitic oscillations are running showed that the oscillation kept on going and therefore is located in the circuit of the output tubes and not in the preamplifier, phase splitter or negative feedback.

10 Ohm 11 Watt wire wound power resistors was installed as plate stoppers between the output tubes and the output transformer. This damped the signal by a great magnitude but the parasitic oscillations would still occur.

Now being very close to rebuilding the whole amplifier, as I had been unable to locate a faulty component, I brought the whole box of 6P45S tubes and tried one after another. I tried another five tubes before having a couple that actually worked.

So the problem all along was old used, some broken, some gassy, some very worn and some almost new together, this was also the point where I at once started construction of the tube tracer kit I had bought, next time I test the tubes in advance and not just think they are working just because I have the same fault with 7 different tubes 🙂

Here is a video of the first time the amplifier is working at full input voltage and negative bias adjusted for 1000 mV over the cathode resistors. This is almost double of what it will be running with, as these high values would exceed maximum plate dissipation if it was running at maximum input signal amplitude.

19th December 2013

The first measurements on the output power and quality of the amplifier have been done.

The first test is looking at 1 kHz square wave and by looking at it and comparing with charts of square wave forms from old radio books, it can be determined what kind of short comings or faults that are present in the system.

The slight sloping of the square waves shows that the low frequency response is good and that the response of the amplifier is pretty flat.

As frequency rises it can be seen that rounding occurs, rounding of the square wave is a sign of bad high frequency response.

As square waves are a sine wave with all its harmonic frequencies, looking at 400 Hz and 1 kHz square waves is enough, as the harmonic frequencies passed by the transfer is in the order of 10 times the frequency. So massive rounding is expected at 10 and 20 kHz.

The next test is done with a sine wave to find the -3dB points. First the clipping point is found at a 1 kHz sine wave and the output voltage noted down. To measure the bandwidth of the amplifier, this is done at half output power of the clipping power. That corresponds to 0.7 * clipping voltage. That voltage will be out reference voltage. To find the lower -3db point, the frequency is turned down until the output voltage is 0.7 * reference voltage. Upper -3dB point is found by turning the frequency up until the output voltage is 0.7 * reference voltage.

Clipping here occurs at 32.8V across a 7R3 resistor load with a sine wave, this is 147 Watt peak power.

Lower -3dB point is at 9.45 Hz and upper at 45.45 kHz.

4th January 2014

To improve the high frequency response, C7 in the negative feedback network was changed from 1 nF to 0.47 nF, moving the cut-off frequency from 41 kHz to 87 kHz.

A slight kink on the 10 kHz and 20 kHz square waves show that the high frequency response have improved.

Clipping here occurs at 39.2 V across a 7R3 resistor load with a sine wave, this is 210 Watt peak power. The resulting -3dB point, half the power, is just above the design goal of 100 Watt output power.

Lower -3dB point is at 11 Hz and upper at 72 kHz.

23rd June 2014

I made new printed circuit boards, both for the power supply and amplifier. There was some changes to the power supply from the prototype. I added 150 V stabiliser tubes for the 300 V supply and the screen supply is also on the board.

Everything is installed in a enclosure from the Italian company HIFI2000.

25th November 2014

The first power up and test with signal generator as the amplifier is installed in its enclosure. There is some problems with hum that will have to be investigated.

Further investigation of hum issues was conducted by waving a isolated 1000 VDC rated screw driver around in proximity of different components while watching the secondary side of the output transformer on my oscilloscope.

I identified two vulnerable places where a great deal of noise could be induced through capacitive coupling and there is also sensitive to noise through induction from magnetic fields.

The first issue was a small part of the signal line in coaxial cable that was not shielded. Explanations are written on each screenshot from the oscilloscope. The first pictures show the output without any interference with the circuit. The second shows the effect of touching the isolation on the part of the signal line in cable that was not shielded.

The second issue was the coupling capacitor in the input circuit before the pre-amplifier. The yellow wave form with the highest amplitude show the induced noise by touching it as it was installed.

The two blue wave form screenshots show the test to locate the pin connected to the outer foil layer in the capacitor, the capacitor is simply connected to the signal and ground of the oscilloscope probe and squeezed around with your fingers. Switch the connections around to perform it at reverse polarity.

The wave form with the lowest amplitude tells us that the pin currently connected to the ground clip is the pin connected internally to the outer foil layer in the capacitor. This outer layer will also function as a shield in high impedance circuits and that pin should be connected to ground or the path with lowest impedance towards ground.

This shows that film capacitor can have a sort of polarity when it comes to very sensitive circuits. A film capacitor in a audio amplifier can actually be mounted backwards.

10th June 2015

All wave forms are from the secondary side of the output transformer.

The first oscilloscope screenshot shows a Fast Fourier Transform (FFT) analysis of the noise generated by the normal diodes for the 340VAC high voltage supply 1N5408 and 40VAC bias supply 1N4007.

The second oscilloscope screenshot shows the difference between normal diodes like 1N5408/1N4007 that have reverse recovery times around 2uS and fast diodes like MUR480/MUR420 that have reverse recovery times around 50nS is shown in the oscilloscope screenshot with yellow and green wave forms. The spike amplitude is around 15% less but the overall 50Hz hum at the positive half cycle is a little more prominent. Changing the diodes gave a difference in the sound from the switching spikes.

The third oscilloscope screenshot shows the much reduced noise levels after a ground loop formed from star ground point to signal input plug was removed and along with the much shorter switching spikes from the new fast diodes.

Audible it appeared like 90% of the hum disappeared. The greatest performance gain was however from removing a ground loop, the faster diodes did not have such a dramatic effect, it was hear able, but not on the magnitude of removing the ground loop.

11th June 2015

Short demonstration of the amplifier playing music.

26th September 2015

I ran measurements on my HP 8903A audio analyzer. Dummy load was a 8.6 Ohm 200 Watt resistor and thus the output power from the output level test gives some 70 Watt out at 0.5 V in. The other tests are done at 0.5 V input too.

I had the amplifier hooked up to my JBL 4333s for the first time and it is now obvious that there is a reason for the high thd+n measurements, there is a great deal of noise, still not sure which kind, but sounds like white and harmonic. Next step is to analyse the noise in a spectrum analyzer.

Suspects of the noise could be the 50 Watt output transformers running at 70 Watt, so maybe bias is set too high or AC balance is not good enough.

A sad side note to this testing is that I had the audio analyzer looped to itself for testing and output voltage was set to the maximum 5V. I forgot about this setting and hooked the amplifier up to the audio analyzer and just as test began there was sparks flying from the output transformer and some smoke. The output transformer is damaged from internal arcing and I will have to buy a new one.

I changed the output transformer with the one I had for the 2nd amplifier. The output transformer was also shifted 90 degrees on two axis’s in order to cancel any possible magnetic coupling to the power transformer. It did however not show any difference in measurements on the audio analyzer.

Conclusion

The amplifier is still under construction and testing.

Demonstration

The amplifier is still under construction and testing.