Fullscreen HTML5Med Res (??MB)HTML4

20W Class-A Amplifier Module PT.1: By LEO SIMPSON & PETER SMITH This new 20W class-A power amplifier module is a refinement of the very popular 15W class-A module published in SILICON CHIP in July & August 1998. It features ultra-low distortion levels, very low noise levels and a greatly simplified power supply which improves overall efficiency. Since it runs in pure class-A mode, there is no crossover distortion at all. 34  Silicon Chip siliconchip.com.au The MJL21193 and MJL21194 output transistors are spaced well apart and bolted to a large heatsink. The heatsink may look big but it has to be that size to safely dissipate around 50W continuously. This view shows the lefthand power amplifier module. The righthand module is laid out almost as a mirror image. In the result, we have made quite a few minor improvements to the original amplifier module. Together, these added up to an overall major improvement which enabled us to dispense with the regulated power supply. This makes the overall circuit more efficient and means that the amplifier can now use some of the power previously wasted in the regulated supply. That also reduces component cost and actually helps reduce distortion in an already exceptional design. Some of the changes in the design are based on ideas and circuits published by the noted audio designer Douglas Self and outlined in a number of his books (available from the SILICON CHIP Bookshop). All in the same case The 15W/Channel Stereo Class-A amplifier presented in August 1998 also featured a separate power supply box because hum radiation from the power transformer was quite high. This new design will feature a shielded toroidal transformer which means that there is no need for a separate box. We will talk more about this aspect in a future article. Redesigned PC board T HIS UPGRADED CLASS-A amplifier has been a long time coming. Virtually since the original circuit was published in July 1998, readers have been hankering for more power. Until recently, we have resisted because we knew that increasing the power output would bring a proportional increase in overall power consumption which was already quite high. This is the great drawback of any class-A design. While they are beautifully distortion-free, they dissipate the same high power whether they are delivering a milliwatt, one watt or full siliconchip.com.au power. And the total power consumption, and therefore heat dissipation, of the previous 15W/Channel Class-A Stereo Amplifier was 100 watts. That’s quite a lot of power dissipation for not very much audio output. So how could we increase the power output while staying within the original parameters – ie, the original large single-sided heatsinks and the 160VA toroidal power transformer? The answer was not simple but essentially involved analysing the weaknesses of the original design to see if we could make worthwhile improvements. We have completely re-designed the PC board so that the two power output transistors are spread much further apart. Instead of concentrating the heat in the centre of the heatsink, it spreads the heat over a wider area and makes more efficient use of the available heatsink area. In fact, while the new amplifier module can deliver up to 25W (instead of the original 15W), the heatsink temperature remains about the same as the original design; ie, about 30°C above ambient. By the way, we must stipulate that even though the amplifier can deliver up to 25W at the onset of clipping, it only provides pure class-A operation up to 20W. Beyond this, it is operating class AB – still with very low distortion but not genuine class-A. We made this compromise to reduce the temperature rise on the heatsinks. With sufficient quiescent current to ensure class-A operation up to 25W, the heatsinks simply became too hot. In fact, the new circuit is actually slightly more “voltage-efficient” than the old one, so that the available output voltage from the balanced supply rails is greater than before. We will see just May 2007  35 Fig.1: this graph plots the total harmonic distortion (THD) at 1kHz from 100mW to just over 25W. Fig.2: the distortion versus frequency at 10W & 20W into an 8-ohm load (measurement bandwidth 22Hz to 80kHz). Fig.3: distortion vs frequency at 10W from 20Hz to 20kHz (measurement bandwidth 22Hz to 22kHz). Fig.4: the frequency response is ruler flat over the audible frequency range, with -3dB points at 1.5Hz and 190kHz. how these improvements have come about as we go through the circuit description. Performance Since many readers will not be familiar with the original design published in July & August 1998, we will present the complete circuit description and mention the differences with the older design where appropriate. But first, let’s talk about performance. The distortion of this new design is actually lower than the original, amazing though that may seem. For those who have the original articles and who want to make direct comparisons, we have produced equivalent distortion plots. If you don’t have the original articles, you will just have to take our word for it that the distortion is lower. 36  Silicon Chip Fig.1 shows the total harmonic distortion at 1kHz for power levels from 100mW up to clipping which occurs in excess of 25W. Note that the distortion for power levels between say 5W and 20W is far below .001% and is typically less than .0006% at around 10W. Similarly, Fig.2 shows the distortion versus frequency for power levels of 10W and 20W into an 8-ohm load, using a measurement bandwidth of 22Hz to 80kHz. This is a far more stringent test as the distortion for any amplifier, even quite good designs, usually rises quite markedly at high powers for frequencies above 5kHz. But for this design, at 10W, the distortion at 20kHz is only marginally above that at 1kHz and is considerably better across the whole spectrum than the older design. At 20W, the new design has about half the distortion of the original design at 15W and that is right across the spectrum, not just at one frequency! Fig.3 is included largely as a matter of academic interest and is taken for a power output of 10W for frequencies from 20Hz to 20kHz but with a bandwidth of 22Hz to 22kHz. Note that this means that harmonics above 22kHz will be ignored and therefore the distortion for signal frequencies above 10kHz will be artificially attenuated. Having said that, the distortion levels shown on Fig.3 are less than half that for the equivalent distortion plot (also Fig.3) in the July 1998 article. Frequency response is ruler flat, as shown in Fig.4. It is -1dB at 90Hz and -3dB at 1.5Hz and 190kHz. This is a much wider frequency response than the original design and comes about because we have used much gentler siliconchip.com.au Parts List 1 PC board coded 01105071 (“left”) or 01105072 (“right”), 146mm x 80mm 2 Micro-U TO-220 heatsinks (Altronics H-0630, Jaycar HH8502) 3 TO-126 heatsink pads (Altronics H-7230) 2 TO-3P heatsink pads (Farnell 936-753 recommended, see text in Pt.2) 1 diecast heatsink, 300 x 75 x 49mm (W x H x D) (Altronics H-0545) 1 PC-mount RCA socket 2 M3 x 10mm tapped spacers 2 M3 x 6mm pan head screws 2 M3 x 10mm pan head screw 2 M3 x 20mm pan head screws 6 M3 flat washers 4 M3 nuts 5 M4 x 10mm screws 5 M4 flat washers 5 M4 shakeproof washers 5 M4 nuts 5 6.3mm single-ended chassismount spade lugs (Jaycar PT-4910) 4 M205 fuse clips (F1 & F2) 2 3A M205 slow-blow fuses 1 11.8mm or 13.8mm ID bobbin (Altronics L-5305) 1 2-metre length of 1mm-diameter enamelled copper wire 0.7mm diameter tinned copper wire for links 1 1kW 25-turn trimpot (Altronics R-2376A, Jaycar RT-4644) Semiconductors 2 2SA970 low-noise PNP transistors (Q1 & Q2) (avail-able from www.futurlec.com) 4 BC546 NPN transistors (Q3, Q4, Q8 & Q9) 3 BC556 PNP transistors (Q5- Q7) filtering at the input of the amplifier. We will describe the reasoning behind this later in the article. Residual noise measurements have also improved. Unweighted signal-tonoise ratio with respect to 20W into 8W is -115dB while the A-weighted figure is -118dB. Even though those noise figures are highly creditable, they are not low enough to enable us to accurately siliconchip.com.au 2 BD139 NPN transistors (Q10 & Q11) (Farnell 955-6052) 1 BD140 PNP transistor (Q13) (Farnell 955-6060) 1 MJL21193 PNP transistor (Q12) (Jaycar ZT-2227, Farnell 955-5781) 1 MJL21194 NPN transistor (Q14) (Jaycar ZT-2228, Farnell 955-5790) 2 1N4148 diodes (D1, D2) Capacitors 1 1000mF 35V PC electrolytic 2 470mF 35V PC electrolytic 4 47mF 25V PC electrolytic 1 220mF 25V PC electrolytic 1 820pF 50V ceramic disc 1 100pF 50V NPO ceramic disc (Jaycar RC-5324) 4 100nF metallised polyester (MKT) 1 150nF 250VAC metallised polyester or polypropylene (Farnell 121-5452) Resistors (0.25W, 1%) 1 1MW 1 510W 4 10kW 1 270W 3 2.2kW 8 100W 1 1kW 3 68W 1 680W 1 16W 1 6.8W 1W 5% 1 10W 1W 5% 2 0.1W 5W 5% wirewound 2 1.5W 5W 5% wirewound (for testing) Power Supply 1 PC board coded 01105073, 134mm x 63mm 1 16V+16V 160VA magnetically shielded toroidal transformer (see text in Pt.2). 4 M3 x 10mm tapped spacers 4 M3 x 6mm pan head screws 6 M4 x 10mm pan head screws measure the distortion at low power (ie, below 5W). This is because the residual noise becomes a significant part of the measurement and largely masks the actual distortion. We discussed this in some detail in the July 1998 article and published some noise-averaged scope plots of the distortion products to demonstrate this mechanism. We hope to feature some equivalent scope plots next month. 6 M4 flat washers 6 M4 shakeproof washers 6 M4 nuts 3 6.3mm single-ended chassismount spade lugs (Jaycar PT-4910) 3 6.3mm double-ended 45° or 90° chassis-mount spade lugs (Jaycar PT-4905, Altronics H-2261) Extra heavy-duty hook-up wire and spade crimp lugs for lowvoltage wiring Mains connection hardware to suit installation Semiconductors 1 KBPC3504 400V 35A bridge rectifier (Altronics Z-0091) 2 3mm red LEDs Capacitors 6 10,000mF 35V or 50V snap-in PC-mount electrolytics (max. 30mm diameter) (Altronics R-5601, Farnell 945-2869) 2 100nF metallised polyester (MKT) Resistors 2 2.2kW 1W 5% Transistor Quality To ensure published performance, the MJL21193 & MJL21194 power transistors must be On Semiconductor branded parts, while the 2SA970 low-noise devices must be from Toshiba. Be particularly wary of counterfeit parts. We recommend that all other transistors be from reputable manufacturers, such as Philips (NXP Semiconductors), On Semiconductor and ST Microelectronics. This applies particularly to the BD139 & BD140 output drivers. For the moment, we can unequivocally state that this new class-A amplifier module is one of the lowest distortion designs ever produced, anywhere! Circuit description Fig.5 shows the full circuit of the new amplifier. While the general configuration is similar to that used in our July 1998 design, very few component May 2007  37 Performance: Class-A Amplifier Module Output power: 20W into 8W (pure class-A); see text Frequency response: 0dB down at 20Hz; ~0.2dB down at 20kHz; -3dB <at> 1.5Hz and 190kHz (Fig.4) Input sensitivity: 625mV RMS (for full power into 8W) Input impedance: ~10kW Rated harmonic distortion: <.002% from 20Hz – 20kHz, typically .0006% (Fig.2) Signal-to-noise ratio: -115dB unweighted, -118dB A-weighted (with respect to 20W into 8W, 22Hz-22kHz bandwidth) Damping factor: 180 at 1kHz Stability: unconditional values are the same. Some of the transistors have been changed, the cascode stage has been omitted, the biasing arrangements for the constant current sources (Q5, Q6 & Q7) have been significantly changed and the impedance of the input and feedback networks has been substantially reduced. These changes were made to improve the residual noise, the power supply rejection ratio (PSRR) and the voltage efficiency of the amplifier. In fact, the only stages which are largely unchanged are the Vbe amplifier (Q10) and the complementaryfeedback pair (CFP) power output stage. So let’s go through the circuit. The input signal is coupled via a 47mF 25V electrolytic capacitor and 100W resistor (R2) to the base of transistor Q1, one of an input differential pair (ie, Q1 & Q2) using Toshiba 2SA970 PNP low-noise transistors. The 100W input resistor and 820pF capacitor (C1) constitute a low pass filter with a -6dB/octave rolloff above 190kHz. This is a much lower impedance network than the previous design, in order to provide the lowest impedance for the signal source. In fact, a simple 20kW volume control, as used in the previous design, will also degrade the amplifier’s noise performance and for that reason we will be presenting an active volume control circuit in a future issue. Both the bias resistor for Q1 and the series feedback resistor to the base of Q2 are set at 10kW (instead of 18kW in the original design), again to minimise source impedance and thereby, Johnson noise. The gain of the amplifier is set by the 38  Silicon Chip ratio of the 10kW and 510W feedback resistors to a value of 20.6, while the low-frequency rolloff (-3dB) of the gain is set by the 220mF capacitor to 1.4Hz. Readers may wonder why we used such large electrolytic capacitors in the input and feedback networks. The answer is that we are acting to eliminate any effects of capacitor distortion in the audio pass-band. Readers might also wonder why we have not used non-polarised (NP) electrolytics for these functions since they are normally preferable where the capacitor operating voltage is extremely low. The answer is that NP electrolytics could have been used except for their greater bulk and we wanted to minimise any extraneous signal pickup by physically larger capacitors. That is one of the unwanted sideeffects of a much wider frequency response – the amplifier is more prone to EMI and in the extreme case, to supersonic oscillation if the wiring details are not duplicated exactly. D1 & D2 are included across the 220mF capacitor as insurance against possible damage if the amplifier suffers a fault which pegs the output to the -22V rail. In this circumstance, the loudspeakers would be protected against damage by a loudspeaker protection module (to be published in a coming month) but the 220mF capacitor would be left to suffer the reverse current. We have used two diodes here instead of one, to ensure that there is no distortion due to the non-linear effects of a single diode junction at the maximum feedback signal level of about 1V peak. Most of the voltage gain of the ampli- fier is provided by Q9 which is fed via emitter follower Q8 from the collector of Q1. The emitter follower is used to buffer the collector of Q1 to minimise non-linearity. Q9 is operated without an emitter resistor to maximise gain and output voltage swing. The collector loads for Q1 & Q2 are provided by current mirror transistors Q3 & Q4. Similarly, the collector load for Q9 is provided by a constant current load comprising transistors Q6 & Q7. Interestingly, the base bias voltage for constant current source Q5 is also set by Q6. Q5 is the constant current “tail” for the input differential pair and it sets the collector current through these transistors. Power supply rejection ratio The reason for the rather complicated bias network for Q5, Q6 and Q7 is to produce a major improvement in the power supply rejection ratio (PSRR) of the amplifier. Similarly, the PSRR is improved by the bypass filter network consisting of the 10W resistor and 1000mF 35V capacitor in the negative supply rail. Why is PSRR so important? Because this amplifier runs in class-A, it pulls a constant current in excess of 1A (actually 1.12A) from the positive and negative supply rails. This is a great deal higher than the typical quiescent current of a class-B amplifier which is typically around 20-30mA. The result of this is that the 100Hz ripple superimposed on the supply lines is about 500mV peak-peak, when two modules are connected. Hence we need a PSRR that is much higher than for a typical class-B amplifier. That is why we employed a regulated power supply for the previous classA design. The output signal from voltage amplifier stage Q9 is coupled to driver transistors Q11 and Q13 via 100W resistors. These protect Q7 and Q9 in the event of a short circuit to the amplifier output which could possibly blow these transistors before the fuses blow. The 100W resistors also have a secondary function in acting as “stopper” resistors to help prevent parasitic oscillation in the output stage. As already mentioned, the output stage actually uses complementary feedback pairs, based on Q11 & Q12 and Q13 & Q14. These give a more linear performance than the more usual Darlington transistor pairs used siliconchip.com.au siliconchip.com.au May 2007  39 Fig.5: the circuit is a conventional direct-coupled feedback amplifier with complementary feedback pairs (Q11 & Q12 and Q13 & Q14) in the output stage. The Vbe multiplier (Q10) is adjusted to give a quiescent current of 1.12A. Here’s a preview of the power supply module. It’s driven from a bridge rectifier and carries six 10,000mF 35V filter capacitors plus two LED circuits to discharge the capacitors after switch-off. in many push-pull amplifiers. In effect, they are connected as feedback pairs with 100% current feedback from the collector of Q12 to the emitter of Q11 by virtue of a 0.1W “emitter” resistor. To make the CFP concept easier to understand, consider Q11 as a standard common emitter amplifier with a 100W collector load resistor. Q12’s base emitter junction is connected across this 100W resistor and so it becomes a current amplifier stage and its collector load is the common 0.1W resistor which provides the current feedback to the emitter of Q11. Because there is 100% local feedback, these output pairs have unity gain and a very high degree of linearity. We should also mention the output transistors specified for this amplifier. They are the MJL21193 and MJL21194 plastic encapsulated transistors which have been featured in quite a few of our higher-powered amplifiers over the years. They are rated at 250V, 16A (30A peak) and 200W, and are clearly far more rugged than they need to be for an amplifier of this rating. We use them here because they are among the best complementary power transistors for linearity made by any manufacturer in the world (originally made by Motorola and now sourced by On Semiconductor). Another circuit change in this new module is that we have used a BD139 and a BD140 as the driver transistors in the complementary feedback pairs in40  Silicon Chip stead of using the lower power BC337 & BC327. This was necessary because of the higher power dissipation in the driver transistors. Vbe multiplier stage Q10 is the Vbe multiplier and it has exactly the same arrangement as in any class-B amplifier. A “Vbe multiplier” is a temperature-compensated floating voltage source and in this case it provides about 1.6V between the bases of Q11 & Q13. Q10 multiplies the voltage between its base and emitter, by the ratio of the total resistance between its collector and emitter to the resistance between its base and emitter. In practice, VR1 is not adjusted to produce a particular voltage across Q10 but to produce the specified quiescent current of 1.12A in the output stage. This requires a voltage of 112mV across each 0.1W emitter resistor. In practice too, the emitter resistors have a 5% tolerance so we average the voltage across each of these resistors at 112mV. Note that you will need a digital multimeter for this adjustment (more on this next month). An interesting point about Q10 is that we have specified a BD139 for this task instead of a much-lower rated BC547 or similar transistor which would certainly be adequate from the point of power dissipation. The reason for using the BD139 is that its package and junction does a much better job of tracking the junction temperature of the driver and output transistors and thereby gives much better bias stability. In fact, Q10 is bolted to the same heatsink as driver transistor Q11 to improve tracking. Also included to improve temperature compensation is the 16W resistor in the collector circuit of Q10; a small point but still worthwhile. Output RLC filter The remaining circuit feature to be discussed is the output RLC filter, comprising a 6.8mH air-cored choke, a 6.8W resistor and 150nF capacitor. This output filter was originally produced by Neville Thiele and is still the most effective output filter for isolating the amplifier from any large capacitive reactances in the load, thereby ensuring unconditional stability. It also helps attenuate any RF signals picked up by the loudspeaker leads and stops them being fed back to the early stages of the amplifier where they could cause RF breakthrough. Finally, as with any high-quality amplifier design, the PC board itself is a very critical part of the circuit and is major factor in the overall performance. Even small deviations in PC layout can have major deleterious effects on the distortion performance. That’s all for now. In Pt.2, we’ll show you how to build the matching left and right amplifier modules and describe SC the power supply assembly. siliconchip.com.au