«Arch Nemesis By Nelson Pass Introduction A poster of Einstein once said, “Things should be made a simple as possible, but no simpler”. This can ...»
By Nelson Pass
A poster of Einstein once said, “Things should be made a simple as possible,
but no simpler”. This can apply to audio amplifiers, but if they are evaluated
subjectively, the simplicity thing can get a little of of hand. Of itself, minimalism
exerts a strong aesthetic attraction, and there is a reasonable belief that fewer
components in the signal path allows more information to get through with less
If like me you are interested in understanding of how we hear distortions with our brains (instead of our meters), you might appreciate that simple circuits help isolate these phenomena. I listen to all sorts of flawed circuits because I enjoy hearing the differences, and it helps to train my ears. In this regard, reducing the number and types of flaws makes it easier to tweak a single parameter and hear the difference. I think it's also true that simple distortions are often more forgivable in a listening situation and create less fatigue.
The Nemesis In 1985 Jean Hiraga wrote an article in two parts presenting, among other things, a design for a very simple Mosfet amplifer called the Nemesis. Subtitled “An Homage to the WE 25 B”, the piece celebrated classic simplicity in amplifier design, specifically a Western Electric amplifier that used a single gain triode driven by an input transformer and driving an output transformer, as shown (simplified) in Figure 1.
Hiraga also discussed a 1982 amplifier apparently done as an application note for Siliconix using the VN64GA N channel power Mosfet driven by a J106 Jfet shown in Figure 2.
He went on to simplify this circuit by eliminating the input Jfet, driving the Gate of the power Mosfet directly (Figure 3).
A second version had an interesting connection from the Source of the transistor to the secondary winding, providing both feedback for the transistor and some “auto-former support” to the output secondary (Figure 4).
But Part 2 of L'amplificateur Némésis showed the final schematic where the transistor feedback connection to the secondary winding was dropped, the simplified circuit reverting back to Figure 3. It appears that he was more interested in the sound without the feedback, even though the measured performance suffered.
In 1994 I played with similar concepts in the Zen Amplifier (Figure 5) and followed up with a series of variations on the theme which explored single- transistor designs, some with feedback and some without, but none of them employing output transformers. These articles can be downloaded from www.passdiy.com. At the time I was solely interested in the performance obtainable from single Class A gain stages alone and didn't want to also consider the additional distortions of passive components (including transformers) in the signal path.
But there was another reason for not using transformers in the Zen amplifiers - the power Mosfets involved are already pretty happy at the voltages and currents needed by loudspeakers. Tube circuits operate at higher voltages and lower currents by a factor of about 10, so tube power amplifiers really need a transformer to efficiently transform signal energy to higher current and lower voltage when it comes to driving 8 ohms.
The Zen amplifier philosophy (“What is the sound of one transistor clapping?”) calls for a minimum of parts. A component has to be needed to be included, but if you alter the need criterion from “measuring better” to “sounding better” then a potentially different perspective opens up.
It is a common belief among audiophiles that measurements don't correlate all that well with subjective experience This is not very surprising – the ear/brain is immensely complicated, and there are many experiments to demonstrate that our understanding of hearing is not much better than our understanding of consciousness, which is not good. Simply the fact that different cultures and individuals hear known “audio illusions” differently gives us a clue while making the problem seem more intractable. I don't expect it to be well understood in my lifetime.
Some “objectivists” think that audiophile subjectivism is delusional, and they are often right, but that doesn't mean that people hear the same way as test equipment. In the first half of the 20th century, there was a reasonably clear association between measured performance and perceived performance, but probably this was due to the rather high distortion of early equipment, where 1% distortion was considered quite good. These days it's common to see amplifiers measuring.001% or even less, but the audio marketplace doesn't seem to particularly reward such achievement.
Back to our Program...
One of the charms of simple circuits is that they have a better correlation between objective (measured) and subjective (heard) performance. It seems that a simple circuit that measures good is more likely to sound good than a complicated circuit that measures good. Moreover, It appears that simple amplifiers like Nemesis and the Zen make it easier to hear differences between single components and compare these subjective differences to measurements.
So Why an Output Transformer?
All components have distortion. We can rank them pretty easily based on simple measurements like THD (total harmonic distortion) and variations they cause in frequency response. Wire and resistors are at the top of the list because as a rule they measure quite low. Next are the capacitors, which give us low but easily measured distortions. At the bottom are active gain devices such as tubes and transistors. And transformers.
Signal transformers don't tend to get a lot of respect from objectivist solid state guys due to bandwidth and distortion issues. You can build a good transistor amplifier without them, and so most do. But Hiraga was (is) not a fool, and in addition there is a small audio cult that likes transformers, even when they aren't essential. They use them for output coupling, input coupling, volume controls and passive crossovers. These are often the same people who disdain capacitors with nearly the same emotion they reserve for MP3 compression.
What's wrong with these people and what is it with transformers?
Jan Didden's Nemesis Who knows what's wrong with Jan? Whatever it is, apparently he addressed it by building himself a copy of the Nemesis. I heard about it because he also seems to have wanted to make more of them and talked to Jack Elliano at www.electra-print.com about getting some transformers made. Ultimately he decided that shipping to Europe was too expensive - did anyone on the forum at www.diyaudio.com want to take up the project?...That would be me.
I have spent quality time with coupling transformers before, but I had never really warmed up to them, possibly because I had not yet reached the 10,000 hour level of listening required to achieve audiophile expertise. In any case, a couple years ago I began experimenting with transformers to solve some problems in a couple of future Zen amplifier projects, and got some fairly good results (good enough for Zens, anyway). Having worked out some circuits, I acquired an assortment of transformers and began evaluating their performance with an eye toward picking the best one. They represented a wide range of cost and materials, and some clearly measured better than others, but when I listened to them I found myself drawn to the sound of one that didn't measure so well.
The dissonance that measures bad/sounds good created called for an unbiased test. So I built two identical amplifiers except for transformers – the very expensive one which measured best, and the unpretentious one that didn't measure so well. I packed them off for a reliable blind test with Joe Sammut, who has 10,000 hours more listening time than me.
“This one is really musical, and that one is not very good.” Well, that's another data point – a transformer that measures better loses to one that does not. Perhaps if my French was any good, Hiraga would have explained it to me long ago.
The Arch Jack sent me a nice pair of transformers, and I set about making a simple recreation of the Nemesis but with variable values for supply voltage, input DC bias, Source resistance, and resistance across both the primary and secondary coils of the transformer, as shown in Figure 6.
As the schematic reveals, there is a lot of opportunity to play around. The input bias voltage ranges over +/- 10V DC. For enhancement-mode Mosfets and Jfets, it may require a positive voltage as high as 8 volts or so. For depletionmode devices the bias voltage will range from as low as -5 volts to as high as +2 volt. The 10 Kohm resistor between the BIAS voltage and the Gate of the transistor is arbitrary. I used this value because Hiraga did, but you can consider values as high as 100 Kohm for Mosfets and depletion-mode Jfets. If you see more than 100 mV DC across it with an ehancement-mode Jfet, then you might want to reduce the value, but it's not a big deal.
Not shown, but you may want to consider a 50 to 100 ohm resistor in series with the Gate of the transistor. This is customary, but I did not experience issues in this amplifier without it. If you have issues with high frequency oscillation, you will want to insert one. Of course if you are using Mosfets, you need to avoid zapping the Gate with a static shock. Elementary caution is usually more than adequate.
Typically the main power supply will range from about 30 to 40 volts, but you can go lower or higher if you want, within the dissipation limits of the transistors.
As it is, I ended up dissipating about 40 watts in a single transistor with 35 volts, which is pretty close to the limit. Since 30 volt supplies are common, you should feel free to use that value if it's convenient.
First, I wanted to explore the limitations of the transformer. It is a single-ended design with a 64 ohm primary and 8 ohm secondary, which is about a 2.8 to 1 turns ratio. The maximum primary DC current is rated at 1.3 amps. Electra Print's specification for bandwidth and distortion was taken with a 25 ohm source impedance driving the primary.
The bias current and source impedance are important factors in this circuit as they have a strong influence on the distortion and frequency response, particularly at low frequencies. If the bias current is too high, the transformer saturates at low frequencies and the distortion goes up and the frequency response suffers, as seen in Figure 7. Here we see an example of this circuit where only the bias is varied, and where higher current through the primary creates greater roll-off at the bottom end.
Figure 8 shows an example of distortion as a function of bias current, and we see that for circuits of this type lower bias improves the bottom end, but higher bias improves the midrange. Higher bias improves the performance of the gain device, and incidentally allows for greater power. You can appreciate that performance trade-offs will be involved.
The source impedance of the circuit driving the primary of the transformer has a similar effect. This transformer was designed around a 25 ohm source. A typical Mosfet operated single-ended Class A as in Figure 6 has an intrinsic output (Drain) impedance of a couple hundred ohms or so. In Figure 6 you will see a variable resistor R2 which can be used to adjust the source impedance seen by the transformer primary. In an example test with a 1 amp bias current we see that the low frequency roll-off (-3dB) is at 40 Hz. With R2 at 75 ohms it's 25 Hz, and with 36 ohms it's 18 Hz. Distortion degradation with higher source impedance is comparable to the example of Figure 8.
As with bias current, adjusting source impedance gives us the opportunity to examine potential performance compromises. Lowering the source impedance via R2 improves transformer performance, and lowers the output impedance of the amplifier as a whole (more damping factor for the loudspeaker), but it loads the gain device, making it work harder to deliver the voltage we want and creating more distortion as a result. The reason that R1, R2, and R3 are variable in this circuit is to afford the opportunity to adjust and optimize the performance against different gain devices, loudspeakers, and listener preference. There is no single right answer, but later we will look at an example that worked well in my system.
For gain devices I had the old standby IRFP240 N channel enhancement-mode Mosfet, plus Ixys IXTH6N50D2 and IXTH20N50D depletion-mode Mosfets left over from the De-Lite amplifier (www.diyaudio.com).
In addition to Mosfets, I had three examples of SemiSouth power Jfets, the enhancement-mode SJEP120R100 and SJEP170R550, and the depletion-mode SJDP120R085. This last part almost didn't make it into this project, as I was not prepared to talk about it until it was publicly disclosed. The SemiSouth parts are made of Silicon Carbide (SiC) and while designed for fast high efficiency switching, they turn out to have superior linearity, resulting in lower distortion.
As a first step, I decided to create an apples-to-apples comparison of the performance of the gain devices. Using the circuit of Figure 6, I set the supply voltage at 32 volts, R1 at 1 ohm, R2 at 36 ohms, and R3 open. The voltage at the Bias pin was varied for each device to give a 1.2 amp bias current.
Each device was measured for response, distortion vs output power, and distortion vs frequency. These figures were taken into an 8 ohm load, and the response and distortion vs frequency were taken at 1 watt and with a 25 ohm and a 600 ohm source impedance from the input signal generator.
Figure 9 shows a table summarizing the results. Not shown is the frequency response, which was consistently -3dB at 25 Hz and 25 Khz with both the 25 and 600 ohm source impedance. All the parts had essentially the same 1 watt distortion at 1 Khz and 20 Khz with a 25 ohm source impedance, but all had more distortion at 20 Khz with 600 ohms, and so I include that data.
All of these parts work well enough to use, and we note that the ubiquitous and cheap IRFP240 is no slouch, however two parts stand out. The SJDP120R085 Jfet has the lowest distortion at all power levels and frequencies, except at 20 Khz (600 ohm source), where the SJEP170R550 beats it due to it's low input capacitance.