0 Members and 1 Guest are viewing this topic. Read 59736 times.
Folks,I have studied amps quite intensively for about eight years now. I have no particularly preference for tubes or SS, and like to build hybrids. I sell two kitset amps called the AKSA, one of 55W and the other of 100W, a new preamp called the GK-1, and a small tweakable two way speaker.I am not an engineer, and resolutely develop my circuits by ear, using only a meter and CRO ocassionally. The 100W AKSA gives 0.045% into 8R resistive at 20KHz and 10 watts, sounds pretty good, and so I offer here my philosophy for designing amps. ****************************************************· Prevent Interstage Crosstalk – Decouple the supply rail for the low current stages.A hard working output stage creates heavy, periodic voltage sag on the supply rail, and this has adverse impact on the voltage amplifier and to a lesser extent on the differential input stage. Interstage crosstalk can be almost eliminated by fitting a diode and small resistor in series from the positive supply rail and a decoupling electrolytic capacitor to ground. Interestingly, sonic testing reveals that a similar network on the negative rail has no beneficial effect. This simple decoupling network ensures that supply rail disturbances created by a hard-working output stage never interfere with the operation of the voltage amplifier and the input differential pair. By ensuring that complex interactions cannot occur between the output and input stages, a potential source of serious intermodulation between the two input stages and the output stage of the amplifier is eradicated. Listening tests reveal a cleaner sound, particularly on heavy musical passages, with superior resolution, clarity and a notable lack of intermodulation.
· Foster High Linearity – Operate the voltage amplifier at constant currentRemembering the sterile sound and high frequency complications of the constant current source leads to an old concept called bootstrapping. The technique uses the amplifier’s low impedance output to dynamically elevate the voltage amplifier power supply in step with the output signal, permitting the voltage amplifier to run at close to constant current. Constant current operation for any amplifying device improves linearity, minimises distortion and greatly ‘speeds up’ the amplifier by removing loading on the voltage amplifier. The bootstrap also has implications for source impedance. The alternative, the constant current source, has been tried many times but always leads to a dry, unexciting sound – and it confers an extreme high frequency bandwidth which needs to be viciously tamed to ensure amplifier stability. The taming process costs musical vitality. To its discredit, the constant current source does confer an asymmetric clip, slicing the negative half cycle long before the positive. However, although a bootstrap does suffer small current variations, it yields good (though deliberately not outstanding) voltage amplifier linearity and intrinsically confers rapid and convenient high frequency rolloff. Early rolloff helps to make the amplifier stable without resorting elsewhere to draconian bandwidth limitation thus bringing some immunity to digital high frequency artefacts and radio frequency interference, allowing the designer to reduce the size of the output inductor and lag compensation. Again, listening tests reveal that a current source powering the voltage amplifier creates a ‘dry’ sound. The bootstrap gives a subjective impression of speed, attack, pace and liquidity – all of which add impressively to the subjective musical experience and take the listener closer to the passion of the original recording.
· Minimise amplitude/phase Intermodulations – Split DC offset and AC feedback control.This technique effectively changes the proportions of resistance and capacitance at the feedback node. This reduces adverse signal damping effects arising from the RC series chain in the network whilst preserving offset control. It also creates two AC paths to the blocking capacitor – usually an electrolytic – and this sets up a secondary charge path which masks the poor sonics of such capacitors by minimizing electrolytic memory effects. The result is a superior, longer lasting decay, particularly on fast percussive material such as cymbals, a considerable reduction in the size of the shunt capacitor and an improvement in sonic ambience.
· Eliminate Switching Transients – Implement charge suckout on the output stage drivers.The output stage introduces switching non-linearities on the output as one output device hands over control of the speaker voice coil to the other. Base junction charge suckout, identified by Self in his impressive 1993 series of articles, involves placing a small resistor and capacitor in parallel between the two driver emitters in a conventional darlington output stage. This parallel RC network ensures that the drivers (and with them, the output devices) turn off more cleanly when speaker current flow passes to their opposite numbers at the crossover of the signal. This greatly reduces switching distortion, the bane of all Class AB amplifiers, by removing the usual spray of high order switching spikes generated by the handover event. It smoothes the transfer of the baton in the musical relay. It is this phenomenon which is partly responsible for the splashy, grainy and often harsh top end of most solid state amplifiers, since the ear readily registers switching disjunctions at high frequency and the feedback loop lacks the necessary speed to correct it.It is clear that many modern amplifiers deliver poor performance because of short term stability problems which manifest only in musical passages but rarely in steady tone testing favoured by contemporary testing regimes.
· Choose Semiconductors with care.The final step in the re-design concerned the choice of semiconductors. To achieve good slew characteristics, the speed of all active devices is important, particularly the voltage amplifier. Also important is the output stage current versus voltage linearity (sometimes referred to as transconductance), since this ensures a high feedback factor under conditions of high output at high frequencies. Note that the speed of the output devices is NOT of the importance one might expect. Of course, cost and availability are also important factors.It is important to maintain a consistent current gain in the output stage as the speaker current demand increases. This parameter relates to large signal linearity; a vital component of the amplifier’s overall transfer function. The choice finally settled on one transistor complementary pair which achieves constant gain (hfe or beta) from 100mA to 7A. This was the 2SC5200 (npn) and the 2SA1943 (pnp). These devices were developed by Toshiba specifically for push pull audio amplifiers, and are rated at a fast 30MHz. These are impressive figures for a 12 amp, 230V, 150 watt transistor with excellent SOAR ratings.
The drivers must also exhibit high current capacity, with good linearity and matching speed. A high current rating ensures they are not destroyed if the output devices are subjected to a momentary short, and a consistent current gain across a wide range fosters linear, stable performance into difficult loads. The devices initially chosen were the ON Semiconductor (formerly Motorola) MJE15030 and MJE15031, and these are still used in the 100W AKSA. These devices are commonly used as drivers in the very high power amplifiers of professional audio, and are virtually unburstable in a small, low power audio amplifier. They can easily withstand a short term current burst of 8 amps, almost 100 times the working requirement in the AKSA circuits. This rating guarantees a long, reliable life – very important in this application, and especially in a kitset where robustness is mandatory. Recently new drivers have become available from Toshiba specifically for the chosen output devices, and if anything, their performance is superior, particularly in terms of linearity. These devices, now fitted to the 55W AKSA, are 2SC4793 (npn) and 2SA1837 (pnp) and their speed is almost triple that of the Motorola devices.The voltage amplifier is particularly important, since the amplifier’s entire voltage gain is incorporated in this common emitter stage. Only this transistor configuration gives both current and voltage gain, and this is heavy going for a transistor owing to Miller capacitance, so it must be very fast and linear with high gain. It must be able to withstand more than twice the rail voltage of the amplifier at 10mA without a heatsink. Such a transistor with very low Miller capacitance is critical to good performance in a quality SS amplifier.
The input differential pair transistors are perhaps the easiest to choose. Since they are small TO-92 devices, their die is small with low parasitics and they are fast. Virtually any PNP device with a rating of at least 20mA and 100MHz is adequate. However, it is fairly important that they be closely matched for gain, as this confers precise differential pair current balance. Current mismatches affect the sonics and the DC offset of the amplifier quite significantly. Finally, the 150V 2N5401 and BF491 were chosen from ON Semi and Philips, as these are inexpensive, quiet, very fast and available in batches with consistent beta, making matching relatively easy.********************************************************This article delineates what I have found to be important. I claim no originality; just careful, painstaking assembly of a variety of simple designs and meticulous listening. The interpretations are mine; some can be disputed on math grounds, and I don't pretend to follow some of the more obscure modes and happily eschew PSpice, which tells me almost nothing about the sonics or even the stability margin. Some of this material amounts to audio heresy, and I make no apology for that, but neither will I strenuously defend my points because I haven't the time or energy. The proof of it all lies in a good listen to my amps, nothing less.I hope it's of some use, and have to say this is a very IMPRESSIVE forum, and my sincere thanks to JohnR, the Borg who set it all up! Cheers,Hugh
...BTW, I have seen mention of saturation of mains power transformers in power amps in various threads. Can anyone explain the mechanism by which this happens?
Hugh, thanks for your contribution and insights. I think you supported your viewpoints quite nicely, and based on what I've read from current owners, they are more than pleased with the finished product. But, taking on the actual assembly of a kit a still a bit too intimidating for someone like me. That's one reason why I've been pressing ('er encouraging) my son to take up the hobby with me. I'll say it again: this discourse is wonderful. Please keep it up guys. DVV, thank you again.Question for future discussion. Assembly techniques such as solder quality, soldering technique, internal wiring, length of wire runs - what potential impact do they have on the sonics: marginal, signifcant, or does it depend?
Are we talking magnetic saturation of the transfomer core here?If so, the flux is determined by the number of turns and the voltage applied i.e. Faraday's Law. I don't see how either of these can increase due to excessive loading.Have I missed something?
I guess my amp really is good...700Va transformer in each monoblock, 4 50000 uf caps and the semis mentioned, the 5200s and 1493s... Only thing that bums me a little after looking at it are a few spots of what looks like flux residue. Nothing a little alcohol won't take care of but with the rep SL has, I didn't expect to see it. It looks surprisingly simple although I don't understand why the 1493s are on one side of the board and the 5200s on the other. The caps have the SL brand name on them so maybe they're custom...smaller caps are mostly Phillips and there are some resistors that look precision there next to them. The case looks like it was designed to be a stereo but I can't imagine how they would fit all the extra stuff into one box. What would I have to be doing to saturate that transformer?cheers,Dick
Very true, I don't think I've made it past eleven on my volume knob. In fact, I think my landlord's bought their grandson a drumkit due to my testing. He needs a little practice. I'm using the BelCantos, also made by SL as speakers and I really wish I could make a smith chart of them, although I like the sound and am not really worried about how they fill out a spec sheet. I guess what amazes me is how simple it looks when built when I know a lot of work went into putting it together. How much time does it take to match components? What criteria do you use to say "yes, that's matched."?cheers,Dick
Hi Dejan,I should correct your comment about purchasing the 5200/1943 Toshiba outputs in Singapore matched. No, this is not possible. I buy large quantities, they come from the same batch at the factory, and then I match them.You'd be interested in how this is done.In a power amplifier, there is usually one driver for multiple outputs. Thus the bases of the outputs are all connected on each side of the amplifier, and clearly for identical bias currents, the voltage drop across the base/emitter AND the emitter resistors must be equal.Thus the principle matching parameter should be Vbe at the lowest current the device will see, namely the quiescent. If we use larger emitter resistors (I use 0R47), then this will insulate us to some extent from beta variations, and at two amperes we are dropping no less than a volt across the emitter resistor, and this hefty negative feedback ensures euqal current sharing is forced at high currents anyway.
So the secondary match parameter is beta. In any event I use 3%, as you identified in your previous post.I designed a matching circuit which sets up a constant current of 50mA through the device under test (DUT), and a constant voltage across the collector emitter. This is a simple circuit consisting of a current source with LED reference, and a differential pair to match base voltages. It is powerered from a 12V gelcel.First I measure Vbe for constant current and voltage, then grade them all within 1mV. There is a surprising spread, and you must be quick, as a heating device chip quickly changes the parameters.Now, within each graded pile, say of 607mV, you can now match for beta. This is done, again under constant current and voltage, by measuring the voltage drop across a base stopper on the device. A higher voltage gives a lower beta; the relationship is reciprocal. The relationship is: beta = 100 x collector current/V(mV) across base stopper. Once again, each device is beta matched within a millivolt using this technique. Using this approach, it is possible to pair off about 60 devices from a sample of 100 in about an hour. I have laid out the circuit on a pcb, attached it to a piece of timber, and as jigs go this is one of the most effective and accurate in my workshop.
I say all this because I wish to illustrate that the traditional, expensive approach of transistor curve tracers and elaborate test gear is neither necessary nor economically viable. I am a low cost outfit, and it is crucial that I achieve high standards at minimal cost. After all, I am Australian.....
Using my matched devices, it is normally possible to get quiescent balance on multiple outputs down to +/-7% on every amp constructed. Good balance has profound impact on sonics.
One of the advantages of sliding bias is that output transistor matching becomes far less crucial. Where it does become important is power share between the output transistors. Over the years I have observed that the PNP transistors always run hotter than the NPN transistors. According to one engineer I used to work with this due to some of the inherent problems related to semiconductor physics. What have you guys found?
I have a recommendation for both Hugh and DVV. Years ago I stopped using the mica insulators with the thermal grease. I use mainly Thermalloy 3 insulators. These have a better thermal coefficent and do not require thermal grease. Having a better thermal coefficent helps to put the heat on the heat sink where it belongs. You guys might find some better performance in certain areas by switching over.
To date I have not been satisified with Japanese discrete semiconductor. I have always used Motorola which is now ON Semiconductor. I wouldn't want to make the LNPA 150 with anything else. For discrete bipolar I think they are the best.
Hi Dejan, Dan,Thank you both for your comments. We are learning from each other; the primary benefit of forums like this. Dejan, I am astonished that you too are using the same matching procedure! It seems I'm in good company.......
