The Latest News, starting from 31 May, now continues on the new website.
The limited availability of the transistors specified for the MJR7 could be a problem, so I have been looking at alternatives. The lateral mosfets are available from a number of sources still, and there are a few alternative lateral types which are almost identical to the Renesas types, made by Exicon and Semelab, the main difference is the lack of internal gate protection zeners. Having used external zeners this is unimportant, but the usual precautions should be taken to avoid static damage. Keeping the gate and source connected with metal foil or something similar until it is firmly attached to the board should be sufficient. A relatively cheap option is the ACD102PSD p-channel and ACD100NSD n-channel from a company in the UK called class-d, who sell them for £3.00 each (Inc VAT). I have not tried their mosfets myself but someone who used them says they work ok. That company sell their own amplifier boards using these mosfets, so it is possible they will continue to have them available for some time.
The small signal transistors are more of a problem. I have plenty of 2SC2240BL and BC560C, but Farnell don't seem to have anything obvious to substitute for the 2SA1209 or 2SC2911. RS (UK) have a few possibilities. T3 and T4 need to be low feedack capacitance types, and could be 2SC3467D and 2SA1370E, both of which are TO92-mod case, which is the same as a standard TO92 but longer and with 1W power rating. Cricklewood (UK) appear to have stock of a few possibilities with TO126 case, including pnp types 2SA1381, 2SA1538, 2SA1360, BF472, BF872. In USA Mouser have the KSA1381E in large quantities at a low price, and these are rated 300V which may be some advantage, I mentioned in my 'Common-Mode Distortion' page that high voltage transistors can be expected to suffer less from base-width modulation. T5 has higher power dissipation than the others and something in the original TO126 case will be more reliable, maybe with a small heatsink if using much more than a 60V supply.
I have the first batch of MJR7 two channel amplifier boards finished and all the components I plan to include in the kits. I had intended to include the 2SA1209s and 2SC2911s but Farnell no longer have these, and one alternative supplier previously sent Chinese copies with low gain, so until I find a new reliable supplier I will leave these out of the kits and reduce the price to £35.50. I will add a fixed £2.50 postage charge to anywhere rather than a variable rate. Also not included are the lateral mosfets and the 63V electrolytics. I have adjusted the board layout a little to allow bigger 100V electrolytics for anyone wanting higher power, but remember then the driver stage transistors may need small heatsinks, particularly T5. The parts list on my new website lists the component types, and some of the resistor power ratings have been changed to match those in the kits, which were decided by availability from Farnell. The 470R may be confusing because R17 is 500mW and the other 470Rs are 250mW.
I have added a Paypal order page, which appears to also allow payment with a credit or debit card without needing to open a Paypal account. To limit sales I will send a link to that page via email to anyone wanting to buy, so first send a message asking for the link. My email address is:
The next batch of boards will be sold with just a few components such as fuseholder, 18g wire, and the input cfp transistors, but that will only be after the current kits have all sold.
There have been a few problems and delays making MJR7-Mk5 boards, but anyway, I have a few boards almost ready for tinplating. I spent a few hours drilling 2000 holes, which reminded me why I said 'never again' when I finished making Mk3 boards last year. I still have to order some of the components, almost all the replies to my survey were interested in the more complete kits, excluding only the electrolytics and the mosfets.
My new website renardson-audio.com includes some rewritten and updated versions of pages from this site, plus the latest versions of the MJR7 and headphone amp. Anyone who has read much of the material on this site will not find much new there, it is mostly a reorganisation of what had become a very disorganised website. Other sections may also eventually be rewritten and transfered. The home page is so far just an index listing the content, I want to improve that eventually.
The new site is hosted by JustHostMe, and they make information available about the countries of visitors. So far by far most visitors are from UK, but more surprisingly in second place is Finland, followed by USA. I had planned to finance the website with advertising (not the irritating pop-ups found here on Angelfire, just Google ads and maybe Amazon.com), but had assumed most visitors would be US based, so I will wait for the statistics to settle down before deciding about that. There are spaces at the side of each page for advertising, but at the end of each page is a 'full page' link which removes the side-bar, particularly useful for anyone wanting to print a page.
I will order the first batch of boards for the MJR7 this week, and should have a few made by the end of the following week. I will then add instructions for how to order.
Initial testing of the HP2 headphone amplifier revealed that the supply ripple rejection is not good enough. The supply rejection is affected by the choice of fet for the input stage 'current source', the rejection figure was only 26dB with either a 2N4392 or 2N3972, but improved to 40dB with the BF245B. Even then with 200mV supply ripple there is an obvious audible hum using my PX100 headphones. Adding a simple RC filter for the input stage should be adequate, for example 100R plus 1500uF will give another 40dB rejection at 100Hz. Trying this there was no longer any audible hum, with a measured level of 10uV. I used a half-wave rectified supply with only 2000uF for the testing, so a higher capacitance plus full-wave rectifier will help further, and anyway a regulator such as the LM317 would add little to the cost.
Initial distortion figure at 100mV output at 1kHz is around 0.01%, which is possibly good enough for a class-A amplifier, there is no real need to aim for ultra-low distortion, unlike class B or AB where crossover effects are potentially more audible than low order harmonics. The distortion does however rise at higher levels or higher frequency. The high power vertical mosfet fed from a high impedance is not ideal, its input capacitance will be relatively high and nonlinear, but I wanted to keep the circuit simple. Adding another jfet as a buffer stage to drive the mosfet the 100mV 1kHz distortion was still the same level, but only 2nd harmonic was visible, and at 10kHz there was now also 0.01%. still only 2nd harmonic. At 1V the figure increased to 0.1%, and higher harmonics up to the 5th became visible, but falling in level as the order increased. Again the result was almost identical at 10kHz. This seems to confirm that the mosfet capacitance was almost entirely to blame for the increased high frequency distortion.
My feedforward class-B output stage design in which one half of the output stage remains in class-A and acts as a feedforward error correction amplifier for the other half was as far as I know an original idea. I had seen a previous design in which one half remained in class-A, so that part of the idea was not something new. I have just found that circuit from 1968, and it is the Plessey SL403 IC, which was also sold as the Sinclair IC-10. The circuit can be found at the Paul Kemble site. It can be seen that there is only a single input to the output stage at TR13 base, and that transistor remains in class-A. When its current falls below a certain level it starts to drive the top half of the output stage, and current flows through TR12. There have been a few variations on this principle over the years, and a few of my own plus one from Wireless World are on my Output Stage Variations page, but in its class-B form it is usually not very linear and not ideal regarding stability because of the number of devices in a local output stage feedback loop. I was reminded of this technique while looking at headphone amplifier designs when I came across another version here. This sort of circuit was suggested to me some time ago, but I had not then realised how the current source works. This version is entirely class-A, the advantage in this case is that it has twice the maximum output current of a conventional single-ended class-A circuit with constant current source. A normal push-pull complementary output stage also has the same current advantage, but for a headphone amplifier only moderate current levels are needed.
The final MJR7-Mk5 is finished and in the process of being tested. I am finally happy with the design, and so there are unlikely to be any further changes, it even looks reasonably neat compared to earlier versions. The layout has been changed slightly, but there is no reason to expect any noticeable change in performance. The anti-thump circuit is the only real change, and I said earlier I was concerned about the specified leakage current of the 470u electrolytics. It turns out that there is no problem, the voltage drop across the 10k resistors charging the capacitors is no worse than 100mV, even though my supply voltage is currently 65V and the capacitors are rated 63V. I will turn up the quiescent current enough to pull the supply down to a safer level for long term use after a few more checks at 100mA. Switching on after the supply has had time to completely discharge gives the greatest switch-on pulse, but even then it is under 1V and slow enough to be almost inaudible with my own speakers, and the switch-off pulse as before is lower still.
I have a lot of updating to do for both MJR7 and HP2 pages. I am still thinking of starting a new website, this one is becoming too disorganised, some older pages have vanished with no links to reach them, and some of the older material needs either deleting or at least moving to an archive folder. There is still a Physics section somewhere, I have at least one update to add there when I have time.
The HP2 fet class-A headphone amplifier has been tried by Richard Thomas (resident in Portugal), and he has designed a board layout, for which I am very grateful. His tests found a peak to peak output of 11V before clipping, and no obvious switch on or off thumps. I have been busy with the final board for the MJR7 so I had neglected to try this design myself, and I was a little surprised to hear that it works well with a variety of fets. Getting it right first time is unusual. No doubt some adjustment can optimise distortion levels, so when I have some distortion test results there may be updates. I will add the board layout to the HP2 page soon.
Response to my survey has been slow, certainly not enough for me to launch a serious commercial enterprise, for now it can only be a small scale service for DIY enthusiasts like myself, to buy components in sufficient quantity to get a good price. One obvious example is the 18g copper wire for the output inductor which costs over £10 in a 47m reel when only 1m is needed. Also the resistors are available to anyone from Farnell, but they only supply them in packs of 50.
There are reasons why I am reluctant to supply either complete kits or finished products, one being the need for EMC certification and other similar regulations. If buyers use their own mosfets that limits my obligations in this respect, or so I believe. The few who have requested these options in the 'comments' box must, I regret, be disappointed. There is really no point selling complete boards, anyone who has difficulty assembling one themselves will also have difficulty making a power supply and fitting everything in a case, so anyone doubting their own skills should be advised to start with a simpler project to gain more experience. The MJR7 has one big advantage, which is that there is no reliance on specially selected or matched components and no accurate adjustments are needed, so the performance is repeatable. Using different brands of lateral mosfets in the two channels for example gave almost identical distortion measurements.
I added a page about optimising stability to explain why the MJR7 has such low distortion. A pair of lateral mosfets operating at 100mA quiescent current will have one or two percent distortion, and to get this down to more acceptable levels there are few options available. Feedforward can be very effective, but requires accurate selection of components or adjustment of distortion nulling for best results. Feedback, as I mentioned earlier on this page, has a few variations, but in all cases the loop gain round the output stage is what matters, and this can be maximised without sacrificing unconditional stability into capacitive loads by careful optimisation of the loop compensation as described. This is not to say that the MJR7 has the lowest possible distortion levels from a single pair of lateral mosfets at 100mA, I already did much better by adding feedforward in the MJR9, but to reduce distortion much further just using feedback I believe would require either pushing the unity gain frequency up to dangerously high levels or abandoning unconditional stability.
I am making a list of recommended components for the MJR7. The electrolytics are a problem because anyone wanting higher power could use up to 100V supply, so I need 100V electrolytics to fit on the board. The 4700u 63V I recommend are 25mm dia, but rated at 100V the Panasonic TSUP is 30mm dia, so I need to adjust the layout a little. Fortunately the lead diameters and spacing are the same, but this is not true for the 470u where 63V types have 5mm spacing and 100V have 7.5mm, so I will add extra holes for the larger size. I was also checking the leakage current specifications and was surprised to see some 470u 63V specified at over 800uA, which would be a problem for the anti-thump circuit where the capacitors are charged through 10k and need to charge close to the supply voltage. At 800uA they would be down 8V. Then I looked at another type from two different suppliers and found one specifying 3uA and the other 296uA. Checking the manufacturers data sheet there was a formula '0.01CV or 3uA whichever is highest'. There was no mention of the units, so I would expect C to be in Farads and V in volts, giving 296uA. This is higher than I would expect, so I wired up a 220u 63V to a 65V supply, hoping it would survive ok, and watched the leakage current fall. It started around 10uA and after a few minutes was down under 1uA. The temperature must have some effect, and the specifications are not entirely clear about this, maybe the leakage is specified at the maximun temperature, but anyway my test suggests the leakage of the 470u anti-thump capacitors will not be a problem.
Incidentally, if the supply is increased much beyond the recommended 60V the current source transistor will get hot, and a small heatsink should be added to this, and perhaps the emitter-follower stage also. The cascode stage transistor operates at only 5mA, so that should be safe at higher voltages with no heatsink. The specified maximum power level with no heatsink at 20deg ambient temperature is 1W, and none of the average powers exceed this, but reliability will be improved by keeping the operating temperatures down.
I checked for stability problems using the Magnatec 2SC2240BL and ON Semiconductors BC560C, using the lowest gain pair from my collection in one channel and the highest gain pair in the other channel. Using capacitive loads up to 4u4 with and without parallel 7R5, and with both shorted and open inputs, whatever I tried there was no sign of instability. I wasn't really expecting any problems, but I like to be sure. These are now the recommended input stage transistors, and I am hoping they continue to be easily available.
The 2SC2547E input transistors I have been recommending are becoming unobtainable and sometimes fakes, so I have been testing some 2SC2240BL which are still available from RS. The manufacturer is listed as 'Magnatec', rather than Toshiba who originally made them, so I had some doubts whether these were full specification originals or just copies of uncertain quality. Magnatec are a subsidiary of the manufacturer Semelab, but Magnatec are described as only a distributor, so it is possible they are still made by Toshiba. Rather than worry too much about this I decided to test a few samples for the characteristics important to the MJR7 application. What matters most is noise at 0.5mA Ic and 10k source impedance, and also current gain. I checked the gain and found all to be well above the minimum, ranging from 408 to 544.
I then tried some in my input stage and compared the noise at the amplifier output with the other channel using a 2SC2547E. Amplifying further with my soundcard and listening on headphones I could hear no obvious difference in either level or frequency distribution of the noise, but comparing the noise spectra using the averaging function of my PC spectrometer I was surprised to find the 2SC2240BL having lower noise at low frequencies. The 2SC2547E noise started to rise at about 20Hz, but the 2SC2240BL went down another octave before starting to rise. The data sheet mentions 'low 1/f noise', so this really does seem to be a feature, at least under my operating conditions. The 2SC2547E is well known for having lower base-spreading resistance, rbb', ( I have seen a figure of 2R mentioned a few times), and so is better for use with low source impedance such as moving coil phono inputs, but for the present purpose the 2SC2240BL appears to be just as good, or marginally better.
I also tested some pnp BC560C made by ON Semiconductors and found current gains between 550 and 600, apart from 2 out of 50 which were unusually 'low' at 460, which is still ok. The ON versions have a higher specified fT than some others, and this is useful because the pnp is the limiting factor in input stage bandwidth, which is relevant to overall feedback stability. The npn is effectively operating in common-base mode at very high frequencies as far as the local feedback loop is concerned, so it has a much higher bandwidth.
I have a final PCB design for the MJR7. I finally decided to omit the LM234 current source because it makes an almost unmeasurable improvement in distortion levels, and there are alternative options such as using dual-die mosfets at higher quiescent current which would have a far greater effect. I want to start a new website for my final designs so that I can sell boards and some of the harder to find components via Paypal. Anyway, that will have to wait until May.
I mentioned in the last update that I was looking at some Stereophile reviews. I have always been a DIY constructor, so commercial equipment has never been a great interest other than as a source of design ideas, but I wanted to see how well some of the 'high-end' amplifiers perform. I was particularly interested to see so many 'zero feedback' class-A amplifiers. One justification for not using feedback appears to be the idea that it adds high order harmonics, so it is revealing that none of those I saw had 7th harmonic as low as my high feedback class-AB design. Using many parallel output devices running at high current and bolted to a big heatsink seems to be the common approach.
My 7th Jan update mentioned that the CCIF intermodulation test should probably specify distortion levels relative to the 20kHz test frequency component. I have just been looking at some amplifier test results on the Stereophile website and see that their 19kHz and 20kHz test signals are both at -6dB which suggests they are using the peak level of the test signal as the reference level. I have seen another example in a book which appears to use a reference level 3dB above the 20kHz signal, which is what I was doing myself until recently. The 20kHz level reference I am now using will give the worst figures, but as far as I know this is the correct CCIF standard.
There are now links at the top of the MJR7-Mk5 page for board layout, parts list, setup and some photos of stages in the construction.
I have added the 20kHz distortion spectrum to the MJR7 tests and also a plot of 2nd and 3rd harmonics as a function of test frequency. I am quite pleased with this result, so I don't plan any great effort to improve on this. The primary aim in all my designs has been to reduce crossover distortion, so the 4V into 7R5 tests are sufficient to show how well this has been achieved, but for completeness I have also added a plot of harmonic levels as a function of output signal voltage, which shows that at 10kHz there is little increase in percentage distortion until we approach the clipping level. I also added the results at 1kHz at 4V and 10V output with a 3R load. The maximum output for just visible clipping seen using an oscilloscope was checked, this will be around the 1% distortion level which is a widely used standard for specifying maximum amplifier power. With 7R5 load the power is 31watts with 59V supply voltage (so with 8R and 60V supply it should be about 30watts), and with 3R load power is 50watts with 57V supply. This corresponds to a peak current 5.8A, so is well under the 7A maximum current specification of the mosfets. To avoid exceeding this limit the load impedance should not be under 2R5.
I am as yet undecided about making and selling boards and kits of parts again, it involves a lot of work for very little profit. Some of the components are not easy to buy in small quantities, for example Farnell only sell some of the resistors as packs of 50, and 18g wire as 47metre reels, so I know it is helpful to supply at least some of the components. I will decide about this maybe in a few weeks. For now the MJR7-Mk5 design is finished, and I have no plans for further power amplifier design work.
I must mention that this amplifier is not a commercial design, it is just for the benefit of DIY enthusiasts, and is recommended only to those with some previous experience. Anyone who has not already successfully built other electronic projects should consider starting with something easier. Using the same component types and values I used myself there should be no problems, but please check the cost and availability of the transistors in your location before deciding to make these amplifiers, they may be surprisingly expensive in some places. Buying from the cheapest sources may also be a problem, there are many fake transistors with inferior ratings. Stick to well known and dependable suppliers if their prices are not too excessive.
I have just realised I made a mistake regarding my 'DIM-60' tests. I used the -3dB frequency of the whole amplifier, which is about 60kHz, but the DIM specification refers to the input filter only, which is actually 120kHz with the 470p filter capacitor I now use (the second 5k6 resistors goes to a 'virtual earth', so it is in effect in parallel with the 470p for the purpose of calculating the -3dB frequency). It is of no real importance, I was just interested to see what sort of results were produced, but looking at a few published DIM-100 specifications for other amplifiers I am even more pleased with the results. The signal level is of course relevant, and I only tested with 1V rms output at 19kHz and a 4V square wave around 3kHz. I may try repeating this test sometime, at different levels and using the correct frequencies.
My designs are based on theory plus measurement, but of course amplifiers are made to reproduce sound and listening tests are at least useful to be sure there is no unexpected effect not revealed by the electronic tests. The listening impressions mentioned by those who have built my MJR7 have been entirely complimentary, and for what it is worth my own impression of listening to the MJR6 and MJR7 is that they appear entirely clear and transparent, but surprisingly different to my older mosfet amplifier which I had been using for many years. I eventually realised that most of the difference resulted from one being inverting and the other non-inverting, so this is something to watch out for if comparing my designs to others. A plausible explanation for the audibility is that if the music includes significant even order harmonics, as most music does, and the speaker then adds a similar level of even order distortion, then the two effects can add or cancel depending on whether or not the signal is inverted by the amplifier. Neither result is necessarily more 'correct', but one may be preferred to the other. Another listener agreed with me that there was an audible difference, but disagreed about which sounded best. The inversion can of course be avoided just by reversing the speaker cable connection at one end for each speaker.
I have been repeating the MJR7 tests with the reduced compensation capacitor and other changes. At first I was disappointed to find some distortion components higher than before, for example the 2nd harmonic of 10kHz is increased from -105dB to -100dB. Then I tried the CCIF intermodulation test with 19kHz plus 20kHz test signal, and was surprised to find the 1kHz product down at -126dB, which is better than I achieved with the MJR9. Higher order products at 1kHz intervals from the test frequencies were higher, as expected from the far lower feedback at those frequencies, but even these were only -108dB.
I wanted to try the 'DIM-30' test which supposedly reveals some forms of 'dynamic distortion', but I only have a 19kHz source in addition to the square wave output of my signal generator, and also the amplifier has its own 60kHz low-pass at the input (wrong:- see update 22-Jan), so I could only do a sort of non-standard 'DIM-60' test. There were only two visible products added up to 20kHz, and the highest was around -100dB relative to the 19kHz component, which I guess is good, but of course there is nothing to compare it to.
I have updated the MJR7-Mk5. page and the test results page. I am still trying a few ideas to try to find out why some distortion components increased in the latest version, and so far have eliminated some possibilities.
My 'CCIF' intermodulation test results have until now been given as the ratio of distortion rms levels to total signal rms level, and also as a ratio of peak levels, which is easier in practice if signal levels are checked using an oscilloscope. The CCIF method is now known as ITU-R, and looking at 'Recommendation ITU-R BS.644-1' I find a statement that the distortion products should be given as a ratio of just one of the test frequency levels. This will give results 3dB worse than using total test signal rms and 6dB worse than the peak ratios. I am not certain whether everyone adheres to this particular standard, or even if it really is an accepted standard or just a 'recommendation'. Provided it is made clear what is being compared there is of course no actual error in the figures, but this is rarely stated. An amusing article about audio standards here suggests that my search for a reliable statement of the relevant standard is probably doomed to failure, or at least not worth the trouble. One manufacturer of test equipment says that the IEC60268-3 standard is to specify intermodulation distortion relative to the level of the highest frequency component of the test signal. The DIM-30 specification uses the 15kHz component as the reference level, even though the 3.18kHz square wave is 4 times higher and has components up to some unspecified upper limit. I have now updated my CCIF specifications for the MJR7-Mk5 to comply with the IEC highest frequency component reference level.
Best wishes for a happy and peaceful new year to everyone.I tried a few changes to the MJR7 circuit as mentioned earlier. A simulation suggested that an extra 6dB loop gain was possible without seriously reducing stability margins with capacitive loads, so I confidently expected distortion to fall to half its original level. The actual result was that some distortion components fell, but some others increased. In the previous update I said that distortion is reduced in proportion to feedback loop gain, but of course this is true only if the circuit changes used to increase gain have no effect on the open-loop distortion, and even then it is not exact. My first guess is that there is some sort of cancellation effect involved, and also the nonlinear device capacitances at the cascode output could be a problem, but further investigation is needed. The changes do increase the theoretical maximum slew rate, but the low-pass filter at the input ensures that any fast square wave below the amplitude clipping level will still be far away from reaching the slew rate limit, so further increase is useful only to give a more impressive specification. One of the components I changed is the output inductor damping resistor, and with this reduced to 1R5 the square wave overshoot with a 2u2 load is much lower.
I think this version of the MJR7 will become the final version of what may also be my final power amplifier design. Endlessly chasing after ever lower distortion figures has little future after achieving figures under -120dB with the MJR9, and although other specifications could certainly be improved I have no interest in going far beyond my own requirements for domestic sound production. Some people really do seem to need several hundred watts of power, either because they use low sensitivity speakers, have huge listening rooms, or just enjoy the physical impact of high sound levels. Personally I use average sensitivity speakers in a smallish room, and the sort of music I like doesn't really benefit from earth shaking volume, so 30 watts is more than enough. Using dual-die mosfets and increasing supply voltage close to 100V the existing designs should be capable of over 100 watt output into 8R, but I only tested at 94V with single mosfets, and measured close to 80 watts output.
There are plenty of perfectly good amplifier designs around, but many appear to be unnecessarily complex. To sum up my own design philosophy, I look for ways to avoid problems, e.g. by using lateral mosfets to avoid bias regulation, inverting amplifiers to avoid common-mode effects, capacitor coupled output to avoid speaker protection relays, and so on, but the result is a simple and predictable amplifier with distortion levels better than many more complex designs.
I still want to try the fet headphone amplifier I included a while ago, and I still need to add board diagrams for the latest power amps, after which I have a few speaker design ideas I want to try, so I will still have plenty to do in the new year.
I have been asked whether I have tried any of the alternative feedback methods used by some other designers, such as error feedback, two pole compensation (TPC) and transitional Miller compensation (TMC). The answer is no, not because I think they don't work, but because I don't see any great advantage compared to my own single global feedback loop. In all cases there is nothing more than conventional negative feedback operating to reduce output stage distortion. Looking at the total feedback from the output of the output stage, round through the rest of the circuit and back to the input of the output stage, the reduction in output stage distortion is about equal to the gain round that loop in all cases.
Two of the methods mentioned achieve greater distortion reduction by starting to roll off the loop gain at higher frequency than the common -6dB/octave dominant pole method, but then must use a greater rate of reduction to reach the same unity gain frequency, and this more rapid rate increases the loop phase shift and may lead to a feedback loop which is only conditionally stable with reactive loads. I just tried a simulation of simplified examples of TPC and TMC, looking only at the loop mentioned above, (I am assuming the point of both methods is to reduce output stage distortion), and adjusting components for similar loop gain, then for both found almost identical excess phase lag over -170deg over a ten to one frequency range even with no load.
The circuits I tried are those given in Bob Cordell's excellent new book 'Designing Audio Power Amplifiers'. The circuits are on pages 178 and 182. Unfortunately there are loop gain plots on page 179 for TPC but none for TMC. Also I had to add 100k resistors to earth from input and output of the VAS to get the low frequency gain to level off. The only surprise was that I had to reduce the TPC compensation capacitors from 60p to 25p to make the loop gains match, which looks wrong, but I can't see any obvious error. To accurately match the gains I also had to add a bypass capacitor, 0.035pF, to the TPC network to reduce the gain peak, as suggested in the book. Using the same 20k in both circuits the TPC capacitors then return to the expected 60p for matching, so it seems the smaller resistor used for TMC is causing the problem.
The MJR7 compensation already has a higher average rate of roll-off than -6dB/octave. The feedback loop phase shift added by the compensation however is limited to -110deg with a 7R5 load, and when a 2uF load is added it only then reaches -170deg over a small frequency range. As in all feedback amplifiers there is a trade-off between distortion reduction and stability, and if the balance is already about right then no alternative feedback methods are going to make much improvement.
Using a 17 turn output inductor for the MJR9 the optimum feedforward resistor was 78R, close to my 80R simulation result. I have added one more distortion spectrum diagram on the MJR9 page, which was with the larger inductor, about 0.7uH, and here the 2nd harmonic of 10kHz is no longer clearly visible, somewhere below -132dB. Higher harmonics look worse, probably because the increased inductance should also have a higher parallel damping resistor. There is no point in further triming the feedforward to reduce these components which are far beyond the audible frequency range, and anyway at such a low level that they would be inaudible at any frequency.
I forgot to mention that for the MJR9 distortion tests I had 100uF capacitors across the 1n capacitors connected to the LEDs biasing the cascode stage and driver stage current source. I wanted to be sure there was no supply effect from these sources, although a calculation suggested this would not be a problem. When trying a 17 turn inductor I removed these capacitors and to my surprise found distortion far higher than previously. Further experiment shows that it is the cascode capacitor which needs bypassing, so now I need to check why this is a problem and the best solution. The 1n capacitors are kept small to improve recovery from overload, so a 100uF bypass is not a permanent option, maybe splitting the 56k LED current supply and adding a capacitor or zener is the answer, or returning to the Mk3 arrangement.
Although impressive low distortion levels were produced by the MJR9 I am discouraged by the high power dissipation needed in the buffer stage. Even with the largest heatsinks I could easily fit into the existing board layout the transistors are running hot, probably over 70 deg. C. The problem is that the lateral mosfets have a relatively low gm so the gate signal voltage can be high, several volts relative to the inductor output, so there can be a fairly high voltage across the feedforward resistor, and therefore a high current needed from the buffer stage and it needs a high quiescent current to keep it in class-A. The Quad 405 circuit using a similar feedforward technique has a bjt output stage with much higher gm, so less of a problem, but even so that used a 3uH inductor. The distortion without feedforward is already far lower than the 405, so there is no serious point and I plan to leave the MJR9 as just a demonstration of feedforward, and will add the board layout for anyone who wants to try their own experiments, but without any strong recommendation. Using a much larger inductor, maybe 3uH, could solve the problem, as could changing the layout to mount the buffer stage transistors on the main heatsink, but I don't much like either idea.
I have a few further tests and possible improvements for the MJR7-Mk5, so I will continue with that and design a board for the final circuit. It is possible I may make boards and some parts available later next year as I did for the Mk3 some time ago, but I have little enthusiasm for making many more boards myself. I may try getting them made, they would certainly be better, or at least have such refinements as soldermask and screen printing, but I then need to learn how to use a pcb cad program, the board designs are at present only bitmap image files.
I have added test results for the MJR9 feedforward amplifier. With 4V output into 7R5 I found distortion components in the audio range to be under -120dB (0.0001%), but there is a problem. My most recent simulations suggested a feedforward resistor about 55R for best nulling, but in reality I found best results at 46R. This is not good because the buffer stage would need to be run at higher current to keep it in class-A at high output into low impedance loads, and the buffer stage transistors are already running hot. I assumed the inductors to be 0.5uH based on my most recent measurement, but the original figure of 0.4uH would explain the lower resistor needed, so I will try increasing the inductors to 0.68uH (17 turns 20swg instead of 13 turns 18swg with 1cm inside diameter) with 80R feedforward resistor, and adding better heatsinks for the buffer stage. The distortion tests used the lower inductor, but a larger value should make little difference, or maybe a small improvement.
For completeness, when the design is finalised I may do a few distortion tests at higher levels into lower impedances. Lateral mosfets don't suffer from the gain reduction at high currents common to bipolar transistors, or the big increases in capacitance at low Vgd of vertical mosfets, so high level effects are unlikely to cause any serious deterioration in performance, and also at higher levels the crossover region can have less effect, so distortion may even be lower. My choice of 4V output for initial testing is because I have previously found that this level more clearly reveals crossover problems with a 100mA quiescent current, and this must be more important than harmonic levels at the rated output.
Test results have now been added for the MJR7-Mk5. As expected the 2nd harmonic of 1kHz has been reduced by including the LM234 current source, it is about 7dB lower than for the Mk4 at -119dB (0.00011%), but replacing the LM234 by a 470R resistor makes only a small difference, so unless feedforward is to be added to make the MJR9 I would suggest just using the resistor. At higher frequencies the distortion is typically 3dB lower than for the Mk3 version, and surprisingly the distortion at 20kHz input is lower than for the almost identical Mk4, with the 40kHz 2nd harmonic the highest at -95dB (0.0018%). Intermodulation with input frequencies 20kHz and 19kHz was measured at 1kHz at -109dB (rms ratio).
The MJR7-Mk5 circuit has now been modified to solve a clipping problem. At 1kHz there was no serious problem, and with light clipping it was still ok at 10kHz, but with heavy clipping at 10kHz it appeared that the LM234 switched off during positive clipping but turned back on too slowly leading to a sawtooth wave output instead of a sinewave. The solution found is to add a capacitor in parallel with the bias regulator LED and add a diode to the emitter-follower stage, which keeps the voltage across the LM234 from falling so far during clipping. The 10kHz clipping is then clean at any level, and also a small glitch which had been noticed at 1kHz when coming out of clipping is now gone. Fortunately the circuit board has enough space for the two extra components. I also tried just replacing the LM234 by a 470R resistor, and that was equally effective. I will leave one channel with the new modification and the other with just the resistor and compare distortion levels to determine whether the LM234 really makes enough difference to be worth using. Another less important change is to reduce the capacitor across the current source LED to 1n, which also improves recovery from clipping, but worsens the switch-on thump, making the anti-thump option more necessary.
A new page has been started for the latest MJR7-Mk5. I wanted to build and test the basic circuit first before adding feedforward, so there will be two final designs with and without feedforward, but both can use the same board. Construction is complete, and initial checks show no problems, apart from discovering that the choice of input stage transistor affects stability margins more than I had expected. The best types are difficult to find, and I need to investigate different sources, some suppliers I used in the past now appear to have problems with fakes. With the 100n plus 1R at the output the stability problem is solved, but if possible I would prefer to make the amplifier stable first without them so that these added components then give an extra safety margin. Further test results will be added over the next few weeks.
I added a new idea at the end of the Headphone Amplifier page. This is simpler than the previous version, and uses just four fets. I usually try to avoid this sort of circuit because of the wide variation of fet characteristics and the resulting need to either select devices or add a preset adjustment, but I still like this circuit and may try it sometime. The current source is slightly unusual, but is an old idea, for example it is shown in a 1970 National Semiconductor application note AN-32 'FET circuit applications'.
I have many old circuits I never tried, and probably never will, and I have now added a page Untried and Abandoned Versions, and may add more to it later. Some of these could work well, others not at all, and I have added a few notes about some of the possible problems.
I have tried one experimental version of the 8 transistor MJR8, which I have called the MJR8-Mk2, just to show that the feedforward arrangement I had suggested really does work. The 2nd harmonic of 1kHz is almost unchanged, but I could trim the feedforward to get all the other harmonics down under -130dB. I may try the LM234Z idea with this circuit, that should help with the 2nd harmonic. A simplified simulation suggests that distortion components up to 20kHz could be reduced to under -120dB, but that seems rather over-optimistic.
I have added a possible final circuit using a two transistor feedforward buffer to allow more correction current for difficult loads. I hesitate to call this my final 'no compromise' mosfet amplifier, but at least regarding distortion this circuit is probably the best I can do with a single pair of lateral mosfets. From a sound quality point of view it may be entirely pointless, but it is interesting to see how far these simple circuits can be taken.
Design and testing of the MJR7-Mk4 is just about finished, and I have added a board layout page for anyone who wants to try this version. The Mk3 had lower distortion and lower switch-on thump, so the only real advantage of the Mk4 is simpler layout and more easily repeatable results.
It is not clear why the distortion should be lower for the Mk3 version. There are a few possible reasons which I have yet to investigate, but whether the second harmonic of 20kHz is at 0.001% or 0.003% is really of no great importance, either figure is very good, and anyway the Mk3 figure was only an estimate. The Mk4 20kHz 2nd harmonic can be reduced to 0.0008% as shown in the test results, but that is just demonstrating a cancellation trick to get a better but not entirely honest specification.
The MJR7 distortion figures could certainly be improved by using parallel pairs or the dual-die type lateral mosfets, or switching to higher gm vertical mosfets (which would need added temperature compensation), or adding feedforward error nulling as I suggested on the MJR-8 page, but that is getting further away from the original point of these designs, which was to achieve low distortion from a simple low cost circuit with no need for component matching or accurate adjustment, and easily predictable stability. (I still managed to get it wrong in earlier versions, see update from 24-Oct-07 on this page concerning low capacitance loads).
Stability near clipping with capacitive loads is rarely shown in amplifier tests, but is, I believe, a good way to reveal whether feedback has been either badly implemented or pushed too far in the quest for better distortion figures. A typical MJR7 result is shown near the end of the Mk3 page, but without anything to compare this to it is difficult to be sure how good or unusual this is.
Testing with low capacitance loads can also be revealing, there being the possibility of series resonance with the output inductor near the feedback loop unity gain frequency. Capacitances of a few nF can then be a problem, and some speaker cables are in this range.
I have now tried the Exicon mosfets, which are sometimes said to be better matched than the Renesas types, but the difference in distortion I measured is very small. If anything the Exicons are slightly worse, but without trying many samples this may just be random variation rather than a repeatable difference. The Renesas types have internal gate protection zeners, so I still prefer to use these. I have included the Exicon distortion spectrum at 1kHz on the MJR7-Mk4 distortion measurement page.
One remaining problem is the switch-on thump, which is not excessive with my own speakers, but may be a problem in some applications. Increasing the current source bias smoothing capacitor from 10uF to 1500uF the output pulse at switch-on was only 1V, so I thought this was a solution. Then I found that the capacitor needs to be discharged before switching on otherwise the output pulse could be 6V or more. Checking the capacitor voltage showed that after switching off it falls quickly from 1.7V to about 0.6V, but then almost stops. This is a more complex problem than I had expected, and for now I only recommend the Mk4 for speakers not likely to be damaged by a low frequency 6V pulse. Most speakers should have no problem with this, but if there is any doubt it would be better to use the Mk3 version with its anti-thump circuit, or add this circuit to the Mk4. There is enough empty space on the board for this addition. For now I have left the current source smoothing capacitor as 220uF.
The distortion measurement page now shows two methods I have tried for measuring the Mk4 distortion. The first is a circuit for my distortion extraction circuit, and the second is a more direct approach with a filter added to a conventional signal generator and the amplifier output reduced by 20dB to reduce the risk of damaging the soundcard. I have suggested this to some people, so I wanted to try it myself. I have included initial distortion results, but need to try different transistors. I know some of those I used are not genuine, as can be seen in the photo further down this page.
The second harmonic of 20kHz is at least partly a supply current effect, and adjusting supply cable position and separation the figure can be brought down to 0.0008%. This sort of cancellation effect is not a satisfactory method of achieving an impressive distortion figure, and it is more honest to adjust the cables for best supply rejection when driving the other channel and accept that most of what remains is amplifier nonlinearity. A figure of 0.003% is probably more realistic.
I have done the Mk4 supply rejection and crosstalk testing, and the results are added on a new page Supply Rejection and Crosstalk. The first diagram on that page shows the wide range of frequency components present on the 60V supply line, and the second diagram shows how much of that finds its way to the output of the amplifier, i.e. practically nothing at a detectable level.
The first MJR7-Mk4 is built and working. I will start testing soon, but have already added a page with initial circuit diagram and a few photos.
I have also added a MJR7 Constructor's Page which will show test results and photos from anyone who has built the MJR7 or any of my other designs. There is an email address at the end of that page for anyone who would like to send their results. There are a few of those who bought boards or made their own who have said they will do their own tests, and the first set of results received have now been included.
My computer is back online following problems mentioned earlier with leaking electrolytics. Instead of replacing 19 faulty capacitors on an outdated board I decided on a new motherboard, the Asrock K7S41. This board at least uses a few Rubycon electrolytics, so hopefully this manufacturer has avoided the bad-caps problem which still seems to continue in some products.
I have a final board design for the two channel mosfet amplifier. I had to simplify the circuit a little to fit a 6" x 4" board, so I have called it the MJR7-Mk4. It is probably not an improvement compared to the Mk3, but the simplified layout may have some advantages, making predictable and repeatable results more certain. I will have the first prototype built and ready for testing in a week or two, and when I am happy that there are no problems I will add a new page with more information. (I mentioned a 'final' version back in May 2009, but I made a few improvements since then.)
An update regarding transistor availability, here is a photo of a few transistors I bought in the past year.
The 4 on the left were all described as Sanyo. All are different, and the 3rd is marked "isc", which is a large Chinese manufacturer, Inchange Semiconductor Company Ltd., so I am at least sure this is a genuine isc device, I think there is little chance anyone is yet making fake versions of Chinese transistors. A quick check of current gain shows that all the isc types in my collection have lower gain than any of the others, but still within the Sanyo specification. For genuine Sanyo devices Farnell may be the safest bet. I have had 2SC2911 and 2SA1209 from Nikko and both these are "isc", so they are now supplying these as substitutes for both Sanyo types.
The transistor characteristics are not highly critical in this application, apart from the noise level of the 2SC2547E, and here also are two examples on the right, which are not even the same size. Here the current gains are more of a worry, one is 590, the other 180. The specification is 400 to 800.
A genuine 2SC2547E is far better than it needs to be for this application, it has very low rbb' (base-spreading resistance) which is not important here with a 10k resistance in series with the input signal. What matters is the noise figure with a 10k source resistance and a 0.5mA collector current. A high current gain helps, and the 2SC2240BL should work well. The Magnatec versions available from RS may be worth trying, but they list current gain as 200 to 700, which is not just the 350-700 BL range, it also includes the lower 200-400 GR gain range. There are plenty of easily available 'low noise' transistors such as the BC550C, but the noise figure at 10k and 0.5mA is not so good (9dB at 120Hz, and that appears to be typical rather than worst case). The MPSA18 has a better wideband figure, has a very high current gain, and is also very cheap, so I may test a few.
One explanation sometimes given for components being discontinued is that they are not compliant with the European RoHS regulations on hazardous substances, introduced in the UK in 2006. The Magnatec 2SC2547E is listed as non-compliant, and the 2SC2240BL as compliant, so this may have a better long-term availability. Another explanation is the increasing use of surface-mount devices (SMD), so demand for non-SMD types is reduced.
Anyway, I think my next project needs to be a transistor tester!
I bought an EMU1820M soundcard on eBay so that I can do some more accurate distortion testing. Soon when I have the new two channel board finished that will be the first to be tested. I will make a new distortion extraction circuit to include input and output network effects instead of the easier method measuring the input stage signal, which is relatively inaccurate.
Also I will add a page of pictures and test results from those who bought boards and did their own measurements. So far I have only one set of results, but I hope more will follow. I have a few comments from listening tests, and so far they have been entirely complimentary, so that is most encouraging.
For now I have to replace a few capacitors in both the EMU and my computer. Both use the infamous G-Luxon LZ electrolytics which it has been reported were made with a missing ingredient, and consequently they generate hydrogen gas. This is revealed by the bulging tops, and in extreme cases leaking electrolyte. The EMU has just 2 of the bad caps, but my computer motherboard has 19. Some other makes of low ESR electrolytics are affected, and there is a list of suspect makes here. There are a few brands widely agreed to be reliable, the ones I have most often seen mentioned are Panasonic and Rubycon. I am ordering some Panasonic FM and Rubycon ZL, both types are available from Farnell.
Regarding the MJR7 distortion specification, the 'under 0.001%' I sometimes mention applies only to distortion components up to 20kHz. I have now reworded the home page description to make this clearer. Using my usual measurement method of nulling the test signal and analysing the remaining output with a computer soundcard I can only measure components a little beyond 20kHz. My Santa Cruz soundcard is long overdue for an upgrade, but until then I can only give a very approximate estimate of 0.001% THD for a 20kHz input based on oscilloscope traces of a poorly nulled test signal. Some of those who bought boards have said they will do their own measurements, so I hope for some independant confirmation. Distortion components at 40kHz and beyond are inaudible to normal humans, but THD-20 figures are useful for comparing different designs.
The reason why such low distortion was achieved is that over 60dB feedback is used at 20kHz, and with open loop distortion around one or two percent a closed loop figure of 0.001% is quite possible. If it is a little higher it could probably be brought down to this figure just by increasing the quiescent current. Such a high level of feedback requires a rate of attenuation greater than -6dB/octave to keep the unity gain frequency low enough, and the consequent phase shift with a capacitive load can then exceed 180deg at some frequencies, which need not be a problem provided it is reduced before the unity gain frequency and also reduces near clipping. A 180deg excess phase shift does not necessarily create positive feedback as is sometimes stated; if the loop gain is still high the feedback is still negative with all the normal distortion reducing advantages. Feedback is positive if it increases gain, negative if it reduces gain. The phase shift becomes important only near the unity gain frequency. At other frequencies where the loop gain is high the phase shift primarily affects the phase of the closed-loop distortion. See my article here. The unity gain frequency is unfortunately variable, particularly near clipping, and it helps if the high frequency compensation is applied in such a way that its phase shift reduces near clipping.
There will eventually be a two-channel PCB as mentioned earlier, and I have no plans to make more of the single boards. To encourage constructors to make their own boards for the single channel version I have rewritten the 'Part 1. Adding The Components' page listed further down this page under the heading 'HOW TO BUILD THE MJR7'. This now includes a high resolution image showing the PCB track layout and a few words of advice for anyone not already experienced in making boards. There are more complete sources of information, so please don't just depend on my own rather incomplete advice. The two channel board will not include a heatsink mounting bracket, which may be some advantage, it is difficult to get angle bracket in small lengths, my local supplier would only supply 5 metre lengths. The disadvantage is that fixing the mosfets direct to a heatsink may be more difficult, and also the PCB needs mounting holes and spacers to fix it in a case.
My article about Slew Rate and TID showed, as Fig.1, the overshoot at an amplifier input produced by a filtered step function input. From this I concluded that extending the open-loop -3dB frequency to 20kHz by adding a resistor in parallel with the compensation capacitor would increase input stage distortion for both transient and steady state conditions, and so was an entirely bad idea. A similar analysis using a step function was used in an article by Daugherty and Greiner entitled 'Some Design Objectives for Audio Power Amplifiers' (March 1966, IEEE Transactions on Audio and Electroacoustics.) Subsequent articles have also tended to consider only a step input, and I had never thought this was a problem. Recently I decided to check my previous result using a Spice simulation. My original version was worked out around 1979 using a TI59 programable calculator. To my relief the Spice result for a step function was about the same.
In addition to the updated Spice result there was one small surprise, and this also is included in a new article TID - Part 2.
I am making a list of alternative transistors to help anyone having difficulty finding those specified. For the Hitachi 2SC2547E and 2SA1085E alternatives are Hitachi 2SC1775E and 2SA872E, and also the Toshiba 2SC2240BL and 2SA970BL. The noise figure of the NPN device is important, it needs to be low at 10k source impedance and Ic 0.5mA. The types listed are mostly around 0.5dB. The 2SA872 was the input transistor specified about 30 years ago in the original Hitachi power mosfet application note, so it is rather ancient, as are some of the others, so I will keep a lookout for more modern alternatives. The 2SC2547 is available from RS Components in the UK, but they list the manufacturer as Magnatec.
The Sanyo 2SA1209 (pnp) and 2SC2911 (npn) are not so critical. The BF470 (pnp) and BF469 (npn) could be tried if the Sanyo types are difficult to get, but I have had one report of higher than expected distortion when using the BF types. I have used them myself in the past with good results, so they may not be to blame. Obviously I am unable to test every possible transistor which could be substituted, so I must strongly recommend using the same types I specified if at all possible to give predictable and repeatable results.
I have a final board design for the two-channel version of the MJR7, which has two channels on a single 6" x 4" board. To fit the reduced space I am using single 4700uf output capacitors instead of the parallel pair of 2200uF, and have left out the anti-thump circuit and the on-board mosfet mounting bracket. I will add the final circuit diagram when it has been tested, and may make the boards available.
I have another project I am working on, which is a cut down version of the MJR7. The idea is to see how far I can reduce the total number of components so that two channels will fit on one standard size 4x6 inches board without reducing track width and spacing too far. Using only 7 transistors is less of an achievement if the number of other components is excessive. The performance should be almost as good, but I will leave the MJR7-MK3 as a less convenient 'top-end' design and add the MJR7x2 as a slightly cheaper and easier alternative. One feature I regret needing to abandon is the heatsink mounting bracket, which was once a common feature on DIY designs and has some advantages, but I will conform to the more common practice of putting the power devices near the side of the board for direct mounting on a heatsink. The anti-thump circuit is also left out, so it may not be suitable for some high sensitivity speakers or active crossover applications. On the plus side the external wiring is minimised and there are no links needed on the board, and the star earth is also on the board, so the two channel earthing can be improved and simplified.
An address for enquiries concerning the MJR7 design is email@example.com but remove the two letters x and z (this is to prevent automatic spam emails, I already get about 100 per week at this address). Use AMPLIFIERS as the subject, otherwise it may get missed and deleted with the spam. I am sometimes slow to reply so please be patient.
HOW TO BUILD THE MJR7.
Part 1. Adding the components.
Part 2. Parts List.
Part 3. Setup and testing.
Part 4. Photos of stages of construction.
11-Feb-2009.One small problem with the MJR7 is that setting the output stage operating voltage to half the supply voltage is difficult if this is done with no load connected. This is because there is then only a 1k load and the output capacitor time constant affects the overall feedback in such a way that adjusting the preset control has only a very slow effect on the voltage. I was turning the control from minimum to maximum and finding the voltage at the mosfet sources changed very little, and thought something serious was wrong, but being more patient and waiting a minute or two after each adjustment revealed that it was behaving exactly as expected. Using a small resistor load, e.g. 22R, during adjustment makes setup far quicker, as does connecting a speaker, but until everything is set up and confirmed to be working correctly it is safer to use the resistor.
01-Feb-2009. Testing the clipping performance of the MJR7 with a 2uF load revealed only a very small 'glitch' coming out of clipping. It also revealed that when clipping occurs the LED used to bias the current source flickers. This does not appear to be a problem, but it occurs to me that this LED could easily be mounted on the amplifier front panel and used as a clipping indicator. I have added square wave test and clipping test results to the amplifier page.
I have rewritten a few pages and moved some. The Introduction page is now listed at the top of the Designing Audio Power Amplifiers section, and includes a footnote about the use of high global feedback. The MJR6 page still includes some useful information, including extracted distortion results using music and speakers, but the home page link has been removed.
04-Jan-2009. I have a final MJR7-Mk3 board layout finished, but part of the problem remaining is to decide which components to use. For example a 2u2 capacitor can be anything from a small electrolytic with lead spacing 2mm up to a huge polypropylene over 4cm long. Having adopted a theory plus measurement based design approach it seems consistent to ignore the many differing claims about component sound, but in most cases it should do no harm to choose types reputed to have lower distortion, so for example I have used polypropylene capacitors for most of the low values, but chose others on purely technical grounds where these are important. The 2u2 input capacitor needs to be small to avoid interference pickup at this sensitive location, and low leakage to avoid changes to dc levels. This eliminates both polypropylene and electrolytic, and a small polyester is specified. Similarly the output capacitors are chosen primarily for current rating, which is of some importance in this application. Some time ago I wrote a piece about capacitor distortion which may be interesting to anyone who remains undecided about 'capacitor sound'. It includes a link to an article giving a different view, which some may find more convincing than I did. I have also tried to identify components widely available, or for which substitutes of similar size can easily be found. For this reason some component values have been changed, but the original values work just as well if they are obtainable and will fit the board layout.
I need to improve the input earthing arrangement. When I built a complete stereo version in a case there were small pulses at the amplifier output as shown here. I traced the problem to the toroidal transformer I used, and found that rotating it a few degrees round its fixing bolt would reduce interference pickup below the noise. Another approach is to reduce the field from the transformer. The interference pulses seem to start with a small pulse followed by a much bigger spike, suggesting that one happens when the rectifiers start to conduct and the other greater effect is when they switch off. In addition to taking steps to reduce this field we can reduce the pickup at the amplifier input, which I found was achieved by connecting the two earth inputs together, which causes an earth loop, but reduces a loop in the input signal circuit. This confirms my suspicion that earth loops are not the worst problem, as I suggested on one of the amplifier design pages: Earth Loops. (I was never entirely happy with this article and it is now in the archive section). Such a loop can have a current induced in it by a varying magnetic field through the loop, but not necessarily any potential differences between different points round the loop as we might expect. To avoid both earth and signal loops one idea is to take the input section earths and input source earths to a separate star earth near the input and connected to the main star earth via a single heavy gauge wire.
29-Oct-2008. I have another project which may be worth supplying as a kit, and this is a class-A discrete component headphone amplifier. I have seen the sort of prices being asked for some fairly ordinary designs, and it should be possible to do better for much less. The current favourite version uses a jfet input stage with optional direct coupled input, and an active volume control to achieve something sufficiently close to a log response from a dual linear control with the advantage of better tracking compared to a typical log control. Having the PCBs made could be expensive, typical charges for small quantities are over £20 each, but I could make these myself, and I may have more spare time soon after Christmas. My designs have the advantage that only a few of the passive components could have any significant effect on performance, so anyone who believes component quality is a real problem can easily change these few parts to upgrade the designs.
The Physics section seems increasingly out of place on this primarily audio site, so I have removed the link on the home page. Anyone who is interested can still find the index page here: PHYSICS ARTICLES. INDEX. The articles about cables and conductors may be of some interest to audio enthusiasts. The 'Reflections on a Transmission Line' article is a much expanded version of an unpublished 'Letter to the Editor' submitted to Wireless World many years ago in response to one of their articles. It deals with the relationship between the electromagnetic field and the conduction electrons in a cable, and although somewhat flawed I still think it explains this better than most other treatments of the subject I have come across, so maybe it deserves a rewrite someday. 'Cable effect' enthusiasts are unlikely to find it either interesting or encouraging.
01-Oct-2008 It has become apparent that the MJR-7 is a very good circuit design, but although the original layout is found to work well there is room for improvement. To keep construction easy I want to do this without abandoning the single-sided board, for example by using external wire links for the high nonlinear current paths so that these wires can be twisted, or at least kept close together, to minimise current loops. One disadvantage is that the result may look rather untidy, but my layouts have never had neatness as a high priority, so nothing new here.
23-July-2008 I never really finished my MJR-7 mosfet design, I left the inductor and its damping resistor as just the original values I started with. My earlier simulation results, some of which are shown here suggested that a lower damping resistor could be a good idea, but were inconclusive. Then, as mentioned further down this page, I learned that quite small values of load capacitance could be a problem, and although a change to the internal compensation RC solved this there could be some variation if different transistor types or different layout is used, so I suggested a 100nF across the output for added safety to avoid smaller capacitance loads being a problem. After further thought and simulations I concluded that a 1R resistor in series with the 100nF is needed to damp any resonances with the load. Without this an inductive load could in theory resonate with the capacitor, giving the same impedance as the originally troublesome capacitor load near the unity gain frequency. Reducing the damping resistor across the inductor from 7R5 to 2R2 also helps.
I checked the older MJR-6 circuit, and the same output network is good for this design also, with the 100n plus 1R across the output, but here reducing the inductor damping resistor to 1R is better according to my simulations. Smaller damping resistors improve phase margin with high capacitance loads but higher resistance helps with smaller capacitance loads. Another benefit of including the 100n output capacitor is that by reducing the range of possible load capacitance this makes choosing the damping resistor easier.
A final version, the MJR-7-Mk3, will be added here eventually, there are just a few details to finalise. There will be only a few small differences compared to the MJR-7 and the Mk2, and if anyone has built either of these just adding the 100n plus 1R across the output and reducing the 7R5 inductor damping resistor to 2R2 are the only changes I would suggest.
26-Oct-2007. I have now tried my MJR-7 with low capacitance loads, and found that with the original circuit it became unstable with capacitive loads from 1.1nF to 3.6nF. With 100nF added across the output terminals there was no problem, but I also tried changing the resistor in series with the 220pF compensation capacitor, and found that values less than 169R or greater than 285R lead to instability for some load capacitor values. A 220R resistor, about half way between these values, is now recommended, and I have updated the circuit diagrams accordingly. My test results are shown here. The 220R value is the optimum value for my own amplifier, but there may be some variation using different samples of the specified transistors, or with different layouts, so the 100n at the output is still a good idea for added safety.
I have also rechecked the clipping performance with a 2uF load, which I mentioned earlier is cause for concern because the phase shift at 100kHz is worryingly high at lower signal levels, but hopefully drops to safer levels near clipping where it otherwise could be a problem. Using a 10kHz sinewave with the 2uF load there are a few very small ripples when coming out of clipping, but not enough to worry about. Real speakers are never likely to be a pure capacitance, so this is even less of a problem in practice.
24-Oct-2007. I never tried the original MJR-7-Mk2 myself, but I have new information from Peter Schoellhorn, who has tried this 'improved version'. Serious stability problems were found, suggesting that I was over-optimistic about the possible increase in loop gain, so I have changed the circuit diagram to return to the original MJR-7 compensation. Even that may have problems with low capacitance loads, and the simplest way to avoid that problem is to add a capacitor across the load, initial simulations suggest 100n is a good choice. With low capacitance loads the series resonance with the output inductor may be above the loop unity gain frequency, and this adds extra phase lag at this frequency, and I believe this is the cause of the problem. I have also changed the Mk2 version in other ways, but again for now this will remain as untried 'work in progress'. Whether the MJR-6 also would benefit from a capacitor added across the output is not yet clear, but for now I suggest including this. This is not a complete solution to the MJR-7 stability problem, instead of being unstable with a range of capacitor load values there will now be a range of inductance values at the unity gain frequency which may be a problem, and simulations with 10n across the output are not so good because of this. Another change which should help is to increase the 100R in series with the 220p compensation capacitor. The 100R pulls back the phase lag by 45deg only at 7MHz, but the input stage already rolls off around 5MHz giving an extra 45deg, so to better compensate for this the resistor should be increased, e.g to 200R. My most recent simulation shows instability for loads from 500pF to 9nF with the 100R, but no instability with 200R. Many thanks to Peter Schoellhorn for letting me know his test results, I had previously assumed that low value capacitances would be no problem, and only tested with higher values.
19-Sept-2007. The more accurately I calculate feedback loop phase shifts for the MJR-6 mosfet amplifier the more likely it seems that shifts in excess of 180deg are possible with high capacitive loads in the 100 kHz region. This is generally regarded as a bad thing because if the loop gain falls near clipping it is possible to have positive phase feedback combined with unity gain at some frequency and consequent oscillation. There is no sign of such a problem with the amplifiers built, and the reason appears to be that close to clipping there is also a fall in phase shift so that long before the loop gain falls to unity the phase shift has reduced to a harmless value. This will not be the case for all amplifier circuits, for example using the more common Miller compensation capacitor method there can be feedforward through the capacitor near clipping, which will make the phase shift worse. The question of whether the driver or output stage clips first may also be relevant.
I was initially surprised by the level of feedback I could use in the MJR-6 circuit while still maintaining excellent stability. I had just finished reading the article by Baxandall which warned about the Miller capacitor effect, so my idea was to use a rather more predictable compensation method, which in the MJR-6 uses the mosfet capacitances, and in the MJR-7 improves on this with a more linear component. The use of a 10p capacitor in parallel with the feedback resistor plus a 390p capacitor from input base to earth is a useful circuit trick to accurately define feedback network gain and reduce the impedance at the input at high frequencies, and I believe this is part of the reason why such a high feedback level is possible.
20-Aug-2007. The 'Square Wave Testing' article mentioned that I find it more useful to check for problems near clipping, and now I have added oscilloscope traces showing examples of latch-up and instability.
20-Jun-2007. Updated further on 17-Aug-2007. The 'Output Inductor Problems.' article explains some of the problems involved when choosing the output inductor and its damping resistor. A negative resistance component at the input of the output stage when driving a capacitive load was also explained.
18-May-2007. I have probably written far too much about amplifiers, and for a change I have started a Speaker Design Section. I have more limited expertise regarding speaker design, and my current attempts are intended only as a simple low cost 'fun project'. I may eventually try one or two new projects when I have more spare time, but for now there are just two pages.
10-Feb-2007. The inverting amplifier configuration I used in all my recent designs avoids the problem of common-mode input distortion, but has a few problems of its own, and in a link on the MJR-7-Mk2 page I have illustrated these problems and my chosen solution, together with feedback network phase response plots for the latest version of the mosfet amplifier.