Project 70 is based on this design, but specifically for headphones. This is where you can really get the benefits of Class-A with none of the drawbacks.
Because the quiescent current can be quite unstable with variations in the supply voltage. Normal changes in the AC mains can cause Iq to shift above and below the preset value. A simple modification can be added that virtually eliminates the problem (or reduces it to the point where it is immaterial). This, plus another optional modification to help stabilise the bias current are included on the new Revision-A circuit board.
The Zen – along with Zen improved, son of Zen, Bride of Zen, Second cousin of Zen (or did I imagine that one?) Class-A amp designed by Nelson Pass seems to have become popular. (See references.) I cannot imagine why, since the very concept is flawed in many ways. It has minimal feedback, but that is because it has minimal gain to start with, and appears very simple. Perhaps this is the attraction – but at what price? The capacitors needed for the power supply are massive to try to get rid of hum, and massive means expensive. The “improved” Zen is a little better, since it uses an inductor (or choke) in the supply – obviously the hum drove someone mad. Inductors are expensive too, and also hard to get, and the capacitance has been doubled in at least one version I have seen – ouch, this is seriously expensive!
Well, actually I can see why it is popular. It satisfies the requirement of many amplifier builders, in that it is simple, stable, and very tolerant of layout and component variations. The sonic characteristics will also appeal to many, due to the valve-like sound (or tube-like, if you prefer). Having looked at the original and many of the “improvements” currently on the web, I did a few tests of my own and frankly, found the amp lacking in the fidelity department. Hi-Fi this most certainly is not. But …. does it sound good? Apparently so, based on the number of people using (and praising) the Zen, but the feedback I have had on the DoZ so far (and my tests) is also very positive and encouraging. At the time of writing, hundreds of DoZ amplifiers have been made, with comparatively few reported problems. The issues that have been encountered have been addressed in the Revision-A circuit boards which are now shipping.
Nelson Pass quotes Einstein as saying “Everything should be as simple as possible, but no simpler”. I agree with this entirely, and quickly realised that the Zen is simpler than it should be for its intended purpose.
Therefore, I have done some serious work on “Death of Zen”, a new Class-A power amplifier that will blow the Zen and all its kin into the weeds, without busting the budget or sacrificing sound quality. Minimal global feedback and lots of local feedback to ensure a very fast and linear amplifier, using the smallest number of components possible. This is the goal, and the remainder of this section explains why.
Photo of Assembled Rev-A Board
Lets look at the basic Zen concept, as shown in Figure 1. A power MOSFET is biased using a pot (needed to correct for different device characteristics) so that the voltage at the drain is about 1/2 the supply voltage. Current is limited using a constant current source, and this needs to be set to provide a current that is higher than the maximum peak current to the speaker. Since the amp is not DC coupled, an output capacitor is needed to keep the DC out of the loudspeakers. An input cap is also needed to stop the source (the preamp, or for my tests, an audio oscillator) from stealing the bias voltage.
Figure 1 – Basic Zen Concept
Now at first sight the idea looks sound (pardon the pun). We do need to do some basic maths to determine the current needed, but this is easy. Using a 35V supply, the bias point will need to be about 1/2 supply (17.5V), and this means that for ideal devices the peak speaker current is +/-17.5 / 8 = 2.19A (say 2.2A). It is necessary to add a little more current to ensure that the active device current remains high enough to stay within the linear region, so lets say 2.5A
In theory, the +/- 17.5V should allow a peak power of 19W, but this is not possible due to the losses in the devices.
As a result, the amp is rated at 10W, and this is reasonable. The output resistance (at DC – this is not the same as impedance) of this output stage is easily determined from Ohm’s Law, so R = V / I = 17.5 / 2.5 = 7 Ohms. Although this is the resistance, the impedance will be similar, although generally slightly lower. According to some, this is the first fault of the design, since damping factor will be at best 1.14 – this is a little shy of the 100 or more that most audiophiles strive for, but more on this later. With the addition of feedback (and yes, the Zen uses some feedback), the output impedance is quoted as about 1 Ohm.
Most readers of my pages will know by now that I am not a fan of switching MOSFETs for audio, since they are far less linear than bipolar transistors or lateral MOSFETs. To me, this is the first failing, since I fully expected the distortion to be somewhat higher than I would consider acceptable for a ghetto blaster, let alone a hi-fi system. Note that lateral MOSFETs are different from vertical types – the former are intended for audio, the latter for switching.
The Big Test
I proceeded to set up a test, using a suitable MOSFET and comparing it with a transistor in the same circuit configuration. The test setup is shown in Figure 2, and I was able to directly substitute the transistor and MOSFET into the circuit, adjust the bias and run the test. Since I wanted to see distortion components alone, I simply used an 8 Ohm drain / collector load, as this is approximately equivalent to the circuit operating with a current source load and driving a speaker. I kept the operating level lower than normal to ensure that a suitable current reserve remained.
Figure 2 – The Test Setup
In the above, the D.U.T. is the device under test. Emitter, base and collector (or source, gate and drain) are connected as shown. For the power supply, I used my “monster” supply, which is variable only because I use a Variac (variable voltage transformer) to supply the incoming mains. I used a 22,000uF capacitor for added filtering (it was not enough!), and proceeded to take some measurements.
First step was to set the quiescent voltage with the pot, so I had 1/2 the supply voltage at the drain (I tested the MOSFET circuit first). With an applied DC of 30V, this meant a voltage of 15V, so the current was 1.875A or 28W dissipation (both in transistor and load – for a total of 56W). The hum was higher than I would have liked, but I can make this disappear using the averaging capability on the digital oscilloscope.
Applying a 1kHz sine wave, I could see that the distortion was quite visible at close to clipping, and I was able to operate at a maximum of 6V RMS output before the distortion became too noticeable. I then hooked up my trusty distortion meter to see just how much there was. Remember from the test circuit that I have included a 0.5 Ohm resistor in the source to help linearise the circuit – not to too much avail it seems, since the distortion was measured at 1.58% (after hum removal), and it increased very rapidly if I increased the voltage. Hmmm. This verified my suspicions, but now I needed to test a bipolar transistor in exactly the same test setup to compare the two.
I used a Darlington transistor (I dislike these too, but it was convenient and the extra gain is essential with bipolars), and was able to bias the transistor using the same circuit as before. Again, I applied a signal, and was not at all surprised to see that the maximum voltage before distortion was visible on the oscilloscope was considerably greater, and the output generally looked cleaner right up to the point of clipping. I would expect that a discrete complementary pair (the configuration I always use) will be better, but I was rushing to get this into print, so used the most convenient device to hand. At least this means that I can improve on these figures without too much trouble.
To be completely fair, I tested the distortion at 6V RMS again, and measured 1.03% – a worthwhile improvement I thought. Increasing the output to 8V, the distortion climbed to 1.18% – still less than the MOSFET, and with a much improved voltage swing. At this level, the MOSFET was delivering outrageous amounts of distortion, as it started to clip.
The measured distortions are not entirely fair, because of the distortion waveform. With the MOSFET, the distortion waveform was peaky, with quite sharp transitions (indicating high order harmonics). The RMS value is probably too low, and certainly does not indicate accurately the audible effect of the distortion. By comparison, the bipolar transistor had a very smooth and almost perfect 2nd harmonic, with very little evidence of any high order harmonics at all. Asymmetry in the residual distortion waveform showed that there was also 3rd harmonic distortion, but at a lower level – I think I will have to have a listen to the residual signal to determine the “musicality” or otherwise of the distortion I measured.
29 Oct – Further Tests
The following day, I decided to buy a MOSFET rather than use the one I had, and selected a MTP3055 as a budget device designed for audio and switching. I still don’t know exactly what the other one is supposed to be for, but it shouldn’t matter – bipolar transistors can be selected for linearity, but it isn’t that big a deal. Not so with MOSFETs as I discovered – the new one was markedly better than the original, but still fails to touch the bipolar. Incidentally, the Darlington bipolar I used was a TIP141, and is designed for switching (lest I be accused of fiddling my results by device selection). I did not retest the 60N06 at the lowest level, but given the other results I could see no point.
Since I now have 3 separate test results, I have tabulated them below.
Output Voltage (RMS)TIP141 Darlington
MTP3055E MOSFET60N06 MOSFET
It would be useful to carry out these tests with a completely hum free supply so that the distortion is not affected by the supply ripple, but by averaging the measured result with the oscilloscope I believe the results are accurate enough for comparison, especially since the same configuration was used for all tests.
Quite obviously, the bipolar is a winner at low levels (where the distortion is most noticeable), and I am sure that using a nice linear transistor such as the complementary pair, these results would show the superiority of the bipolar transistor even more clearly. Again, as the supply limits are approached and the current through the devices varies the most, the bipolar is again well ahead.
UPDATE: The MTP3055 distortion figures actually are very close to those published by Pass Laboratories for the Zen, so this validates the test circuit I used, and that the figures are not exaggerated in any way. (27 Nov 99)
Full Test Of Death Of Zen (DoZ) Concept
The next step was to test something close to the final configuration, to see what things had an effect (profound or otherwise) on the performance. I still used the TIP141, knowing that I can improve on this greatly as I progress, although as the final circuit shows I eventually chose not to use a compound pair after all. Figure 3 shows the test circuit, still using the 8 Ohm resistor as a load, but I ran these tests using my bench supply to eliminate the hum problems. All tests were performed at an output of 6V RMS (equivalent to the 8V tests above, due to the lower supply voltage).
Figure 3 – DoZ Test Circuit
This is the basic configuration I will be using for the final design, although there will be some resistor value changes as I get closer to the final circuit, and the load resistor will be replaced by a constant current source. For those who want to try the circuit with a high output impedance, I will also include the modified feedback network.
Some interesting things came to light during testing, especially when I included the resistor (R6) from base to earth on Q2. With no resistor, I measured a distortion of 0.15%, and this was almost completely 2nd harmonic. There was a very noticeable degradation of the positive going slope on a 10kHz square wave, and a fairly low slew rate resulted. Adding the resistor improved this dramatically, and reduced the distortion to 0.05% – but it was now almost completely 3rd harmonic.
This will create a conundrum for some – would you rather have very low levels of 3rd harmonic distortion, or considerably larger amounts of 2nd harmonics (bearing in mind that the 3rd harmonics are still there). I cannot see any good reason to tolerate any more distortion than is absolutely necessary, so considering the much better slew rate (and therefore high frequency performance), I will be including this in the final design. You might want to leave it out if you want the 2nd harmonics, but I don’t think the end result will be very satisfactory.
This is due to the transistor’s turn-on and turn-off characteristics becoming more symmetrical by providing a base discharge path, but I did not expect such a large difference. The frequency response extends to over 100kHz at full power (6V RMS for these tests), and square wave response shows that the amp is both fast and stable – and this with a very ordinary switching Darlington. I saw no evidence of measurable distortion above the 3rd – there must be some, but I have no way of measuring it. The 3rd harmonic appears to be an almost perfect sine wave, with some very small variations.
Slew rate is better than 6V/us (positive going) and over 20V/us negative going – not as good as some, but I blame the TIP141 for this. I have checked the specs on it, and it is a fairly slow device (like most Darlingtons) as confirmed by these tests.
None of this testing has been done with a circuit board. In all cases I simply bolted the device to a heatsink, and attached the other components as required. Power connections were all made using alligator clip leads. Since I have used exactly the same “rats-nest” wiring for all testing (including these last tests), and I have not been able to induce additional (or reduced) distortion by moving leads about, the amp looks as if it will be fairly tolerant of assembly methods (all known assembly methods will be superior to what I have done so far).