Fullscreen HTML5Med Res (??MB)HTML4

Ultra-LD Mk.3 200W Amplifier Module Upgraded design has even lower distortion! The Ultra-LD Mk.2 (August-September 2008) was the lowest distortion class-AB amplifier board design ever published. But we have not rested on our laurels and have found ways to improve it significantly. The Mk.3 version has less than half as much distortion at frequencies of 2kHz and above. It also boasts much improved thermal stability, is slightly quieter and has a flatter frequency response. By NICHOLAS VINEN T HE NEW AND UPDATED Ultra-LD Mk.3 is by far the best class-AB amplifier module design published anywhere. It has an astonishingly low total harmonic distortion plus noise (THD+N) figure of 0.004% at 20kHz for 100W into 8Ω (20Hz-80kHz measurement bandwidth) and less than 0.0006% THD+N at 1kHz and below. The signal-to-noise ratio has also been slightly improved on the previous version (by 1dB) to -123dB with respect to 135W into an 8Ω load. The power output figures are unchanged with regards to the Mk.2 module. All power measurements were made with a mains voltage of 230VAC, which is now common in Australia (although by no means universal). In locations with a higher mains voltage, slightly more output power is available. For example, if your mains voltage is normally 240VAC, you can expect about 8% more power output, eg, 145W into 8Ω. The quiescent current accuracy, stability and thermal compensation have been dramatically improved compared to the Mk.2 and in fact are superior to any class-AB amplifier that we have tested. The new module has a trimpot so that the quiescent current can be set Specifications & Performance Output Power (230VAC mains).................................200 watts RMS into 4Ω; 135 watts RMS into 8Ω Frequency response.................................+0, -0.3dB (8Ω); +0, -1.0dB (4Ω) – 10Hz-20kHz (see Fig.5) Input sensitivity...................................... 1.26V RMS for 135W into 8Ω; 1.08V RMS for 200W into 4Ω Input Impedance.............................................................................................................................. 12kΩ Rated Harmonic Distortion (8Ω)............... <0.004% 20Hz-20kHz, typically 0.0006% (see Figs.1 & 3) Rated Harmonic Distortion (4Ω)............... <0.007% 20Hz-20kHz, typically 0.0006% (see Figs.2 & 4) Signal-to-Noise Ratio....................123dB unweighted with respect to 135W into 8Ω (22Hz to 22kHz) Damping Factor.....................................................................~180 with respect to 8Ω at 1kHz & below Stability......................................................unconditionally stable with any nominal speaker load ≥ 4Ω 30  Silicon Chip to the optimum (the Mk.2 was a bit hit and miss in this regard). The new thermal compensation arrangement keeps the quiescent current well under control, even during and after sudden changes in dissipation. This contributes to the low distortion as it means that the output stage is always correctly biased. Rationale Making these improvements to an amplifier that already had outstanding performance may seem like gilding the lily. But there are two important reasons why we decided to improve on the Ultra-LD Mk.2. First, we felt that we could produce a design that was even closer to that holy grail of amplifier design: a highpower class-AB module with the low distortion of a Class-A amplifier. In fact, the new design is tantalisingly close to the benchmark SILICON CHIP Class-A amplifier (May-Sept. 2007). Astute readers may have noticed that while the Ultra-LD Mk.2 was clearly superior to the original UltraLD amplifier (SILICON CHIP, March & May 2000), it actually had higher distortion for frequencies above 6kHz. This is because the original Ultrasiliconchip.com.au The Ultra-LD Mk.3 Audio Amplifier module features pluggable connectors, improved thermal stability and extremely low noise and distortion figures. It’s built on a double-sided PCB and is attached to a large finned heatsink which carries the driver and output transistors and a central VBE multiplier transistor. LD featured a more linear output stage, consisting of two complementary compound transistor pairs. By contrast, the Ultra-LD Mk.2 used a standard complementary Darlington emitter-follower output stage, for better current sharing between the output transistors (allowing it to reliably drive 4Ω loads). Since then, we have tweaked the emitter-follower output stage to improve its linearity at high frequencies (more on this later). The end result is that the Mk.3 has distortion lower than or equal to both the original Ultra-LD and the Ultra-LD Mk.2 at all frequencies. siliconchip.com.au It may seem that the distortion products of very high frequencies (10kHz & above) are irrelevant, since they will all be above the audible range. The second harmonic of a 10kHz signal is 20kHz and the third is 30kHz and these are not audible so why are we trying to minimise their level? The answer is intermodulation. While lower order harmonic distortion may be relatively benign, the associated and inevitable intermodulation distortion is definitely not benign; it is audibly unpleasant. To demonstrate, let’s say we have an audio signal consisting of a 10kHz sinewave mixed with an 11kHz sine- wave. Their second harmonics are at 20kHz and 22kHz respectively and are not audible, but the difference products of 1kHz, 2kHz & 12kHz certainly are audible and are musically unrelated. So by minimising harmonic distortion at high frequencies, we are also minimising intermodulation – a far more unpleasant distortion product. Quiescent current Second, we just weren’t satisfied with the quiescent current and thermal compensation arrangement of the Ultra-LD Mk.2. That was our first design using the On Semiconductor July 2011  31 THD+N vs Frequency, 8, 100W, 20Hz-80kHz BW THD+N vs Frequency, 4, 100W, 20Hz-80kHz BW 05/20/11 12:27:35 Ultra-LD Mk.2 Ultra-LD Mk.3 Ultra-LD Mk.2 Ultra-LD Mk.3 0.01 Total Harmonic Distortion + Noise (%) Total Harmonic Distortion + Noise (%) 0.01 0.005 0.002 0.001 0.0005 0.005 0.002 0.001 0.0005 0.0002 0.0002 0.0001 20 05/20/11 12:27:35 0.02 0.02 50 100 200 500 1k Frequency (Hertz) 2k 5k 10k Fig.1: total harmonic distortion plus noise across the audible frequency range for an 8Ω load driven at 100W. This is an “apples-to-apples” comparison between the old and new amplifier modules with an identical power supply and test set-up. The Mk.3 is superior at all frequencies but especially above 1kHz. “ThermalTrak” transistors, which have integral diodes. The literature for these devices claims that they eliminate the need for quiescent current adjustment as well as providing much better thermal tracking than a traditional VBE multiplier circuit. Our initial prototypes seemed to confirm both points. But as more people built modules based on that design, it became apparent that the ThermalTrak transistors vary somewhat from batch to batch and therefore we do in fact need a method to trim the quiescent current. Also, for reasons we shall explain later, many of the Ultra-LD Mk.2 modules built do not have good thermal tracking. That is to say, their quiescent current can vary considerably depending on the output device temperature, which can vary rapidly depending on the program material being played. Once we found out about these problems we took a closer look at the ThermalTrak transistor data sheets. It turns out that the ThermalTrak diode temperature coefficient doesn’t necessarily match that of the accompanying transistor and so using the diodes alone for thermal compensation is not satisfactory. In some cases, the diode temperature coefficient is so much lower than the transistors’ that the result can be thermal runaway – as the transistors get hotter, the quiescent current increases, making them hotter again 32  Silicon Chip 20k 0.0001 20 50 100 200 500 1k Frequency (Hertz) 2k 5k 10k 20k Fig.2: the total harmonic distortion plus noise across the audible frequency range for a 4Ω load driven at 100W. The performance improvement for the Mk.3 module is even larger with a 4Ω load, with less than half the distortion of the Mk.2 version across a large portion of the audio frequency range. until eventually they blow; definitely not a good state of affairs! Back to the drawing board Actually, building a class-AB amplifier with accurate thermal compensation that responds quickly to changes in dissipation is a very difficult task. The basic problem is that to get good performance, the standing current through the push-pull output transistors must be kept within a relatively small range (in this case, about 70140mA per pair). If the quiescent current is too low, the result is significant crossover distortion. As the output voltage passes through zero, the load current is “handed over” from one of the output transistors to the other. Without sufficient bias, one transistor turns off faster than the other turns on, resulting in a discontinuity in the output stage transconductance (ie, the ratio of its input voltage to output current). This makes the amplifier as a whole less linear and so increases its distortion. The opposite problem occurs if the quiescent current is too high. In this case there is actually a sudden increase in the transconductance in a voltage band around 0V. This is called “transconductance doubling” and again reduces linearity. When the quiescent current is in the correct range, these two effects tend to balance out and so the transconduct- ance curve for the output stage is as flat as possible, maximising linearity and thus minimising distortion. So we want to set it within that range and keep it there. High quiescent current also causes excessive dissipation in the output devices – we don’t have to explain why that’s undesirable. Thermal tracking If the transistors were all kept at a constant temperature, correct biasing could easily be arranged by simply placing an adjustable floating voltage source between the base of the two driver transistors and then trimming it with an eye on the current through the output stage. This bias voltage sets the VBE across the driver and output transistors, resulting in a constant standing current through the output stage. Unfortunately, the required VBE for constant current through a transistor depends on its junction temperature. Since the output transistors heat up and cool down during use in an unpredictable way (depending on the program material, load impedance, ambient temperature, airflow, etc), we must come up with a way for the bias voltage to vary with driver and output transistor temperature, to keep the quiescent current as stable as possible. In the Mk.2 amplifier, the bias was developed by passing a constant current through the four ThermalTrak diodes contained within the output siliconchip.com.au Total Harmonic Distortion + Noise (%) 0.05 THD+N vs Power, 8, 1kHz, 20Hz-20kHz BW 05/20/11 14:59:08 0.1 Ultra-LD Mk.2 Ultra-LD Mk.3 0.02 0.01 0.005 0.002 0.001 0.0005 0.0002 05/20/11 14:57:55 Ultra-LD Mk.2 Ultra-LD Mk.3 0.02 0.01 0.005 0.002 0.001 0.0005 0.0002 0.06 0.1 0.2 0.5 1 2 5 Power(W) 10 20 50 100 Fig.3: total harmonic distortion plus noise against power level for 1kHz into 8Ω. The slightly lower noise figure makes the Mk.3 marginally superior at low powers, with it pulling further ahead above 4W due to its lower harmonic distortion. Note that the maximum power available has hardly changed from the earlier design; the small variation is mainly due to the test procedure. transistor packages. As the output transistors heated up, the required VBE for a given current dropped and so did the forward voltage of the associated diodes. If the two thermal coefficients matched, then theoretically the diodes would correctly compensate for the changing transistor properties with temperature. Since that clearly wasn’t happening, we decided to ignore what the application literature said about these transistors and instead analyse the circuit from first principles. We are not the only people to notice this problem. Douglas Self experienced similar difficulties using this type of transistor, which he documents in his Audio Power Amplifier Design Handbook (Fifth Edition). In that book, he points out that if the ThermalTrak transistor data sheet is correct, the diode forward voltage temperature coefficient is -1.7mV/°C but the transistor VBE temperature coefficient is -2.14mV/°C. Clearly then, we cannot use a single ThermalTrak diode to compensate for a single ThermalTrak transistor without risking thermal runaway (or at least a wildly varying quiescent current). But that wasn’t the only problem. The four ThermalTrak diodes compensated for four transistor VBE drops but only two of those drops are from the base-emitter junctions of the power transistors that the diodes thermally siliconchip.com.au 0.0001 200 0.06 0.1 0.2 track. The other two are the driver transistors (Q10 and Q11, MJE15030/ MJE15031). So even if the diode thermal coefficients matched those of the output transistors, they wouldn’t necessarily correctly compensate the driver transistors. Also, there is significant thermal lag between the output transistors and the driver transistors, since during periods of high output power, the power transistors can get significantly hotter than the heatsink. It takes a while for the heatsink temperature to heat up in response to the increased dissipation and then there is a further thermal lag from the heatsink back to the driver transistors. Fig.5: the frequency response for the Ultra-LD Mk.3 module. Note that there is less roll-off at both the lowfrequency and highfrequency ends for the Mk.3 compared to the Mk.2. The high-frequency rolloff is greater for 4Ω loads (about -1dB at 20kHz). This can be slightly improved (to -0.7dB) by changing the inductor – see Pt.2 next month. 0.5 1 2 5 Power(W) 10 20 50 100 200 Fig.4: total harmonic distortion plus noise versus power for 1kHz into 4Ω at a range of power levels. Here the Mk.3 module really shines, providing significantly lower distortion across the entire range. The Mk.3 can easily produce the rated power of 200W into 4Ω. Note that the measurement bandwidth (20Hz-20kHz) is smaller than in Figs.1 & 2, so the figures are better. +1.0 We had to find a better solution. As a result, we came up with several ideas for circuits that would provide a bias voltage with more accurate and reliable thermal tracking, then ran them through circuit simulations before building a prototype incorporating the most promising. New design Our new solution harks back to that tried and true bias compensation scheme, the good old VBE multiplier. But we have also incorporated the ThermalTrak diodes as they are critical in allowing us to provide compensation for rapidly varying output device temperature. Frequency Response, 4 & 8, 1W 05/20/11 12:43:21 Ultra-LD Mk.2 (8) Ultra-LD Mk.3 (8) Ultra-LD Mk.3 (4) +0.5 0 Amplitude Variation (dBr) 0.0001 THD+N vs Power, 4, 1kHz, 20Hz-20kHz BW 0.05 Total Harmonic Distortion + Noise (%) 0.1 -0.5 -1.0 -1.5 -2.0 -2.5 -3.0 10 20 50 100 200 500 1k 2k Frequency (Hertz) 5k 10k 20k 50k July 2011  33 Fig.6: load lines for the Ultra-LD Mk.3 amplifier. The red line is the 1-second Safe Operating Area (SOA), outside of which transistor second breakdown becomes likely. The mauve and green lines represent realistic speaker operating areas for 8Ω and 4Ω units respectively, taking into account their reactance. Ultra-LD Mk.3 Load Lines (4 Output Transistors) 10 2 x ThermalTrak 1 second SOA, 90% Sharing 8 Resistive Load 8 Reactive Load, 135W (5.6+5.6j) 8  Resistive Load Collector Current (Amps)  Reactive Load, 200W (2.83+2.83j) 6 4 2 0 0 20 40 60 80 Collector-Emitter Potential (Volts) We are now using two ThermalTrak diodes to compensate for the two power transistor VBE drops, in series with a VBE multiplier to compensate for the driver transistor VBE drops. The VBE multiplier transistor is mounted on the heatsink, between the driver transistors, to best track their temperature. Now that we have a VBE multiplier, this allows us to easily provide an adjustment by placing a trimpot in the multiplier network. This means that the quiescent current can be configured correctly regardless of variations in the output transistors. The adjustment will however slightly degrade the thermal tracking, since in changing the absolute voltage contribution of the VBE multiplier (by changing the multiplication factor) we also change its thermal coefficient. But our testing shows that this is a relatively minor factor and the tracking is still more than good enough. Actually, because the temperature coefficient of the ThermalTrak diodes is lower than that of the associated transistors, in order to achieve correct compensation, the VBE multiplier must slightly over-compensate for changes in temperature. We found that if we used a BD139 for the VBE multiplier, we achieved the required over-compensation. Simulation shows that the resulting quiescent current variation with temperature is virtually flat. The prototype Ultra-LD Mk.3 modules were built from two different batches of ThermalTrak transistors and they bear this out. As a happy coincidence, it turns out that the 34  Silicon Chip 100 120 best current to use for the new bias generating arrangement is the current that we originally chose for the UltraLD Mk.2 (9.5mA) to provide the best performance. Parallel diodes While we stated earlier that we are only using two of the ThermalTrak diodes, we have actually wired up all four on the PCB, in two parallel pairs which are then connected in series. This makes it possible to build the amplifier with only two output transistors (the outer pair), for applications where less power is required. The supply voltage is also reduced in this case, to reduce overall power dissipation. Lower distortion As stated, the Ultra-LD Mk.3 has less than half the distortion of the Mk.2 at frequencies of 2kHz and above (see Figs.1-4). It also has lower distortion at low frequencies but there is so little to measure that it tends to be lost in the noise floor (not that there is much of that either). There are two main changes which reduce the distortion and these are the new frequency compensation arrangement and the new driver transistor emitter resistor configuration (ie, for Q10 & Q11). Of these, the latter is the most important but they both contribute to the excellent performance. With the Ultra-LD Mk.2, the driver emitters were connected to the output via 100Ω resistors. For the new circuit, the emitters are instead connected to each other via a 220Ω resistor which is bypassed with a 470nF capacitor. This allows the driver transistors to reverse-bias one pair of the output transistors to switch them off quickly, when the slew rate is high (ie, at high frequencies). This was not possible with the old arrangement. Reverse-biasing the output transistor base-emitter junction rapidly removes the charge carriers from it, preventing conduction which would otherwise occur for some period after the normal base drive was removed. The 470nF bypass capacitor assists in the switch-off process. The bottom line is lower distortion at high frequencies. Two-pole compensation The new compensation scheme also helps to lower the distortion. Instead of a single 100pF, 100V ceramic capacitor between the base of Q8 and the collector of Q9, we now have two 180pF 100V polypropylene (plastic dielectric) capacitors and a 2.2kΩ resistor. This dramatically increases the open-loop gain within the 20Hz20kHz frequency range without affecting stability. For more details on why and how this works, see the separate feature article titled “Amplifier Stability and Compensation” in this issue. We found that polypropylene capacitors gave measurably less distortion compared to C0G/NP0 ceramic capacitors of the same value, presumably due to their higher linearity. Ceramic capacitors can be used but the distortion at 20kHz will increase from around 0.0048% to about 0.0055% (with proportionally similar increases at lower frequencies, down to about 1kHz). Feedback network changes During the course of testing the prototypes, we ran into a problem with the capacitor in the feedback network (above and to the left of Q8). The purpose of this capacitor is to reduce the amplifier’s DC offset at the output, by reducing the DC gain to one. The original capacitor was specified as 220µF but we found that if the capacitor value was on the low side and/or the capacitor used had particularly bad non-linearity (as is sometimes the case), the result could be a significant rise in distortion below 50Hz. By changing this capacitor to 1000µF, we eliminated that possibility. This also improves the signal-to-noise siliconchip.com.au Fig.7: an oscilloscope screen grab illustrating the shape of the distortion residual waveform for a 20kHz sinewave at 100W into 8Ω. It is primarily second harmonic, with some third harmonic (how much depends on how well-matched the output transistors are in terms of beta). We have to demonstrate the distortion at a high frequency and power level otherwise it’s hard to see! ratio, by about 1dB, because it lowers the source impedance seen by the inverting input (the base of transistor Q2) at low frequencies. In addition, it flattens the low frequency response, as can be seen in Fig.5. Input filter changes We have increased the value of the RF filter capacitor at the input, from 820pF to 4.7nF. This allows it to better reject supersonic components of the input signal (eg, digital-to-analog converter switching artefacts). This value suits signals sources with low output impedance (0-220Ω). Virtually all CD/DVD/SACD/BluRay players, preamplifiers, computer sound cards and DACs should be within this range. If a volume control potentiometer is to be installed immediately before the power amplifier, with no buffering between the two, or if the signal source(s) will have an output impedance above 220Ω, reduce this capacitor value to 1nF. Otherwise, the high frequency response of the amplifier will suffer. PCB improvements As well as updating the board to include the circuit changes, we have made further tweaks to the PCB pattern itself. The most important is that we completely removed the three top layer tracks which connected Q12, Q13 and Q14 to their supply rails, which were on the bottom side of the board. siliconchip.com.au Fig.8: by contrast with Fig.7, this scope grab shows the extremely low distortion when delivering 100W into 8Ω at 1kHz. Note that the distortion is virtually buried in the noise (blue trace). Averaging the distortion product signal shows it to be mainly second harmonic at a very low level. This low-level harmonic distortion is virtually the same whether at 50mW or 100W. That current is now routed entirely through bottom layer tracks, eliminating 30 current-carrying vias, six wire feed-throughs and one signal via (a via makes an electrical connection between tracks on different layers of the PCB). We have also “tented” all the vias on the board, except for those which require wire feed-throughs to be installed (for robustness under fault conditions). This means that the solder mask layer goes over the vias, exposing as little copper as possible and thus reducing the chance of short circuits when probing around the board. Some vias have also been moved under components, further ensuring that you can’t accidentally make contact with them. For boards without plated through-holes or solder masks, feed-throughs can still be installed in these locations since the components they are under (the 5W resistors) are mounted proud of the PCB anyway. We also rearranged some components to take account of the range of sizes available. This includes the 220nF 400V capacitor at the output, the 470µF 63V bypass capacitor for the negative rail and the 47µF bipolar input capacitor. There should now be enough space for just about any components with these ratings. Note that the PCB retains the most important aspect of the previous design: the layout of the current-carrying tracks results in the induced magnetic fields being almost perfectly cancelled, keeping the distortion low even with a high output power. The updated output filter also improves the outputcurrent magnetic field cancellation, reducing high-frequency distortion by around 20%. Better connectors For the Mk.3 design, we have also changed the connector arrangement. All connectors are now pluggable, making it easier to install and remove the module and simplifying testing and repair. Making reliable connections to a terminal block can be awkward with the module inside a case. More than once we thought we’d made a solid connection but then found that we could easily pull the wire out. The new connectors eliminate that problem. We have replaced the signal input terminal block with a right-angle RCA socket. For the power input and speaker/headphone outputs, we are now using Molex “Mini-Fit Jr” plastic locking connectors (in horizontal or vertical format). The power connector has three keyed pins and the speaker/ headphone connector has four keyed pins, so that they can’t be swapped around or connected backwards. The Mini-Fit Jr connectors are rated at 9A per pin, which is sufficient for this application. In addition, the final version of the PCB (not shown here) can also accept July 2011  35 210mV Q3 BC546 B A K D1, D2: 1N4148 4.7nF† 100 E C B E C E C 210mV K 180pF 100V D1 470 F 63V D2 A Q4 BC546 A K B C E 180pF 100V 2.2k B B B E K A K A C E K A DQ14 K A 120 VR1 1k B 330 DQ12 Q7 BF470 Q9 BF469 C C E 68 2SA970, BC639 2.2k E Q8 BC639 22k C 12k 100nF B 2.2k 56.3V 1000 F 16V 510 6.2k 6.2k B Q6 BC556 * Q16 IN THERMAL CONTACT WITH HEATSINK NEAR Q10 & Q11 100nF 68 B Q1, Q2: 2SA970 68 C E 100 2.2k 47 F 35V ULTRA-LD MK.3 200W AMPLIFIER MODULE † USE 1nF IF Z source > 220  10 1M 12k 47 F NP 6.8k 1W B 47 F 35V E E C C C E Q14 NJL1302D B C B E BD139, BF469, BF470 B B C C 100nF E MJE15030, MJE15031 Q15 NJL1302D FUSE2 6.5A C E 0.1  5W 0.1  7-10 5W mV 7-10 mV E E C 0.1  5W B 100nF 0.1  7-10 5W mV E C Q13 NJL3281D FUSE1 6.5A 7-10 mV B Q12 E NJL3281D C Q11 MJE15031 B 10  1W BC546, BC556 B B Q10 MJE15030 470nF MKT 2.2V 56V 56V 100 DQ15 Q16* BD139 DQ13 100 100nF C B E 390  1W –57V (NOM.) 0V 0V SPEAKER OUT PHONES OUT CON3 CA K NJL3281D, NJL1302D 1000 F 63V 220nF 400V 6.8  1W L1 10 H 1000 F 63V +57V (NOM.) CON2 Fig.9: the complete circuit diagram for the Ultra-LD Mk.3 amplifier. Changes from the Mk.2 circuit are highlighted with yellow boxes. We have improved the output stage bias circuit and the compensation network, while a new driver emitter resistor configuration speeds output transistor switch-off, reducing distortion. A larger feedback capacitor (1000μF) lowers noise and extends the bass response. In addition, L1 has been increased from 6.8μH to 10μH which partially cancels the magnetic field produced by the output current, reducing high-frequency distortion. 2011 SC  CON1 SIGNAL IN C E 100 45V Q5 BC556 100 220 36  Silicon Chip siliconchip.com.au vertical connectors in two locations (ie, a vertical RCA socket for the signal input and a vertical 3-way Mini-Fit Jr connector for the power input). These let you build a stereo amplifier, with the two amplifier modules mounted on either side of the case. The new, slimmer power supply board (described next month) can then fit between them. Heatsinking As stated, the additional transistor for the VBE multiplier is located on the heatsink between the two driver transistors (Q10 & Q11). To make room, the output transistor pairs are now closer together. This allows us to position the mounting holes for all transistors so that they fall in between heatsink fins, with the board centred on the heatsink. It is therefore no longer necessary to blind-tap the mounting holes or to offset the board from the centre of the heatsink if the transistor machine screws are fastened with nuts, as was the case with the Mk.2 design. Note that if you plan to run the amplifier at continuous high power levels (100W or more) into a 4Ω load then it will probably be necessary to use a larger heatsink (with lower thermal resistance to the air) and/or fanforced cooling. If driven at full power (200W) into a 4Ω load continuously, the heatsink becomes too hot to touch even in free air (70°+) and this will be even worse if it is mounted in a chassis with limited ventilation. For continuous high power levels into 8Ω, a larger heatsink is also a good idea although it may not be strictly necessary if the ventilation is good. Note that in either case (4Ω or 8Ω), for music program material, if the amplifier is not driven into clipping then heatsinking should not be an issue. This is because even heavily compressed pop music typically has a dynamic range of at least 10dB, so even if the peak power is close to maximum, the average power will be significantly less. Load lines When we described the Ultra-LD Mk.2, we did not publish any load line curves. Such graphs show the range of transistor currents and dissipations that can occur with speaker loads (resistive and reactive) and the Safe Operating Area (SOA) of the transistors in the amplifier. siliconchip.com.au The relevant load lines and the SOA curve for the Mk.3 are shown in Fig.6. By comparing the SOA curve for a pair of ThermalTrak transistors to reactive load lines for typical 4Ω and 8Ω loudspeakers, we can determine whether the transistors are likely to exceed their ratings during periods of high power output. If they can be driven beyond the safe operating area, the output transistors may be destroyed by second breakdown. Second breakdown is a phenomenon which can occur in bipolar transistors, where high temperature and dissipation lead to thermal run­ away in a small area on the silicon die, ultimately resulting in the silicon melting. We need to ensure that this is not possible under normal conditions. As you can see from Fig.6, the load lines for 4Ω and 8Ω resistive and reactive loads are within the safe operating area. This curve is computed based on the ThermalTrak transistor data sheets and assuming that no single output transistor is required to carry more than 55.6% of the total load current. It is specified for signal durations of one second. Since the reactive load curves are within the SOA then the amplifier should be quite robust. Unless the load impedance is dramatically less than we are assuming (eg, due to a short circuit at the output), the power transistors should be safe from destruction. All in all, plotting the load lines gives us a reasonable idea of how close to the limits we are pushing the power transistors. Circuit description Let’s now look at how the circuit works in more detail – see Fig.9. As shown, the input signal is applied to CON1 and is coupled to the WARNING! High DC voltages (ie, ±55V) are present on this amplifier module when power is applied. In particular, note that there is 110V DC between the two supply rails. Do not touch the supply wiring (including the fuseholders) when the amplifier is operating, otherwise you could get a lethal shock. base of PNP transistor Q1 by a 47µF non-polarised capacitor. The intervening RC filter (100Ω/4.7nF) attenuates any supersonic signals present, eg, switching artefacts from a DAC. The 12kΩ resistor provides the bias current for Q1’s base. PNP transistors Q1 and Q2 are the differential input pair, with Q1’s base being the non-inverting input of the amplifier and Q2’s base being the inverting input. These are configured as a “long tail pair”, fed with current by PNP transistor Q5, which is configured as a current source. The 100Ω resistor at its emitter sets the current through this stage to around 6.5mA (0.65V/100Ω). Some of this current flows through Q1’s collector-emitter junction and the rest flows through Q2’s. How the current is split depends on the difference in voltage between the two bases. Most of this current then flows through NPN transistors Q3 and Q4, which are configured as a current mirror. This current mirror keeps the current through Q3’s collector-emitter junction equal to the current through Q4’s, so any difference in the current through Q1 and Q2 must then flow to the base of Q8. Thus the current to Q8 is proportional to the difference in voltage between the bases of Q1 and Q2, ie, the two amplifier inputs. The 100Ω resistors at the emitters of You Must Use Good-Quality Transistors To ensure published performance, the 2SA970 low-noise transistors must be from Toshiba. Be wary of counterfeit parts. We recommend that all other transistors be from reputable manufacturers, such as NXP Semiconductors, On Semiconductor, ST Microelectronics and Toshiba. This applies particularly to the MJE15030 & MJE15031 output driver transistors. During the course of our testing, we came across some BC556 transistors which, when used in the amplifier, resulted in excessive distortion. Despite this, their hFE figure tested as normal. Replacing them with a different batch returned the distortion to normal. Use good-quality transistors throughout to guarantee good performance. July 2011  37 Parts List 1 double-sided PCB, code 01107111, 135 x 115mm 1 black anodised aluminium heatsink, 200 x 75 x 45mm (L x H x D) 4 M205 PCB-mount fuse clips 2 6.5A M205 fast-blow fuses (F1,F2) 1 10µH air-cored inductor (L1) (or 1 20mm OD x 10mm ID x 8mm bobbin and 2m of 1mm diameter enamelled copper wire, plus one length of 10 x 20mm diameter heatshrink tubing) 1 1kΩ multi-turn vertical trimpot (VR1) 2 TO-220 mini flag heatsinks, 19 x 19 x 9.5mm 5 TO-220 silicone insulating washers 4 TO-264 or TOP-3 silicone insulating washers 2 transistor insulating bushes Screws, nuts, spacers & washers 4 M3 x 9mm tapped spacers 7 M3 x 20mm machine screws 2 M3 x 10mm machine screws 8 M3 x 6mm machine screws 9 M3 nuts 9 M3 flat washers Connectors 1 black PCB-mount switched RCA Q1 and Q2 are “emitter degeneration resistors” which provide some local negative feedback, increasing their linearity at the cost of reduced gain (which in turn reduces the overall open loop gain of the amplifier). The 6.8kΩ resistor simply reduces the dissipation in Q5. The 68Ω emitter resistors for Q3 and Q4 improve the accuracy of the current mirror. Voltage amplification stage The circuitry described above comprises the first stage of the amplifier and as explained, it converts the differential input voltage into a proportional current. This current is then converted back to a single-ended voltage, relative to the negative rail, by the following stage (the “voltage amplification stage” or VAS). This consists primarily of NPN transistors Q8 and Q9 as well as PNP transistor Q7. 38  Silicon Chip connector, or one vertical PCBmount RCA connector (CON1) 1 Molex Mini-fit Jr 3-pin rightangle PCB-mount male socket (Element14 order code 9963545); OR one vertical PCB-mount Mini-fit Jr male socket (Element14 order code 9963570) (CON2) 1 Molex Mini-fit Jr 4-pin rightangle PCB-mount male socket (CON3, Element14 order code 9963553) 1 Molex Mini-fit Jr 3-pin female line plug (CON2, Element14 order code 9963480) 1 Molex Mini-fit Jr 4-pin female line plug (CON3, Element14 order code 9963499) 7 Molex Mini-fit Jr female pins (for CON2 & CON3, Element14 order code 9732675) 1 MJE15031 PNP transistor (Q11) 2 NJL3281D NPN ThermalTrak transistors (Q12,Q13) 2 NJL1302D PNP ThermalTrak transistors (Q14,Q15) 1 BD139 NPN transistor (Q16) 2 1N4148 signal diodes (D1,D2) Semiconductors 2 2SA970 low-noise PNP transistors (Q1,Q2) 2 BC546 NPN transistors (Q3,Q4) 2 BC556 PNP transistors (Q5,Q6) 1 BC639 NPN transistor (Q8) 1 BF470 or 2SA1837 PNP transistor (Q7) 1 BF469 or 2SC4793 NPN transistor (Q9) 1 MJE15030 NPN transistor (Q10) Resistors (0.25W, 1%) 1 1MΩ 1 220Ω 1 22kΩ 1 120Ω 2 12kΩ 6 100Ω 1 6.8kΩ 1W 3 68Ω 2 6.2kΩ 1 10Ω 1W 4 2.2kΩ 1 10Ω 0.25W 1 510Ω 1 6.8Ω 1W 1 390Ω 1W 4 0.1Ω 5W 1 330Ω 2 0Ω 2 68Ω 5W (for testing) NPN transistor Q8 amplifies the current from the previous stage and feeds it to NPN transistor Q9. Together they form a compound transistor similar to a Darlington, which is set up as a common-emitter amplifier. PNP transistor Q7 is the current source load for this amplifier and the standing current is set to around 9.5mA by the 68Ω resistor (0.65V/68Ω). This current flows from Q7, through the output stage bias network (DQ12-DQ15 and transistor Q16) and thence to Q9. This common-emitter amplifier converts the current delivered to the base of Q8 into a voltage, at Q9’s collector. This voltage is then proportional to the input voltage differential at the base of Q1 and Q2. Transistor Q6 provides the negative feedback for both current source transistors (Q5 and Q7), regulating the current through them. The two 6.2kΩ Capacitors 2 1000µF 63V electrolytic 1 1000µF 16V electrolytic 1 470µF 63V electrolytic 2 47µF 35V electrolytic 1 47µF non-polarised (NP) electrolytic 1 470nF 63V MKT 1 220nF 400V MKT 5 100nF 63V MKT 1 4.7nF 63V MKT 2 180pF 100V polypropylene (Rockby Stock No 36350) resistors, in combination with the 47µF capacitor, form a bootstrapped current sink which turns on both Q5 and Q7. Once the right amount of current is flowing through each, Q6 turns on and reduces the base current to both in order to maintain it at that level. The two 180pF capacitors and the 2.2kΩ resistor between Q8’s base and Q9’s collector are the compensation network described earlier, which takes the place of the traditional Miller capacitor. This reduces open loop gain at high frequencies by reducing the gain in this stage, as well as linearising the operation of Q8 and Q9. The negative supply rail for this stage and for the previous (input) stage is filtered using a 10Ω resistor and a 470µF capacitor. This low-pass filter prevents 100Hz power-supply ripple from coupling into the signal path, especially when the output power is siliconchip.com.au Another view of the completed Ultra-LD Mk.3 amplifier module. The full constructional details will be published next month. frequencies where the load’s reactance may cause the amplifier to oscillate. The parallel 6.8Ω resistor acts as a “snubber”, preventing the output filter from oscillating in response to signal pulses from the amplifier. The inductor also prevents any RF signals picked up by the speaker leads from being coupled to the base of Q2 where they may be rectified to an audio frequency signal. The 220nF capacitor, in combination with the inductor and resistor, presents the amplifier with a constant load to high frequencies and keeps the output impedance low at high frequencies, even if no speaker is connected. This ensures that oscillation can not occur. Any output cable capacitance will be swamped by the 220nF capacitor across the output. This filter arrangement was developed by A.N.Thiele (Load Circuit Stabilising Network for Audio Amplifiers, Proceedings of the IREE 299, September 1975). Power for the output stage is filtered by 1000µF and 100nF bypass capacitors across each rail. If an output transistor fails, one or both of the 6.5A fuses will blow, protecting the power supply. Loudspeaker protection high. It’s the negative rail that requires filtering most of all because the output voltage of the VAS common-emitter amplifier is relative to this. Output stage The output stage is a current buffer/ unity gain voltage follower formed from two complementary emitterfollowers. These are arranged in Darlington configuration, with a single driver transistor for each half (Q10 and Q11) providing current to the bases of two output transistor pairs (Q12/Q13 and Q14/Q15 respectively). There is a 100Ω resistor in series with the base of each driver transistor, to limit current in the event of a (brief) output short circuit. The voltages at the bases of the two driver transistors are controlled by the common-emitter amplifier in the previous stage. The DC voltage between them is set by the bias generator described earlier. The 100Ω resistors also function as RF “stoppers” which reduce the possibility of parasitic oscillation in the emitter-follower output stages. The class-A amplifier (VAS) current passes through the bias generator and siliconchip.com.au the voltage across it is determined by the forward voltage of the two ThermalTrak diodes which have the lowest forward voltage within each pair, plus the voltage across the VBE multiplier. The voltage across the VBE multiplier is roughly Q16’s base-emitter voltage multiplied by a factor set by VR1. This bias voltage varies with the junction temperatures of Q10-Q15. The 0.1Ω emitter resistors for Q12Q15 force each pair (Q12-Q13 and Q14-Q15) to share the load current, as well as providing a small amount of current feedback. RLC filter After the output stage is the RLC filter consisting of a 10µH air-cored inductor, 6.8Ω resistor and 220nF capacitor. This isolates the amplifier circuitry from any load reactance (capacitance or inductance) caused by the cabling or loudspeakers. The 10µH inductor presents a low impedance to audio-frequency signals but a high impedance at supersonic frequencies, at which the amplifier might oscillate. Therefore it isolates the amplifier from the load at critical Note that it is necessary to connect a loudspeaker protection module between the amplifier and speaker terminals, so that the load is disconnected in the event of an amplifier failure. Failures usually cause the output to be connected directly to one or other of the ±57V supply rails and unless a protection module is present to immediately disconnect the loudspeaker, it may be damaged and quite possibly catch fire due to the resulting high current flow through the voice coil. More to come That’s all we have space for this month. Next month, we will describe how to assemble, test and adjust the amplifier module and also present an updated version of the power supply board. That article will also include some suggestions for putting it all together in a case, as a mono or stereo power amplifier. Finally, for those who have already built an Ultra-LD Mk.2 module, don’t despair. We plan to present a small adaptor board which will allow you to fully upgrade its performance to the SC Mk.3 standard. July 2011  39