Elliott Sound Products | Project 243 |
Main Index Projects Index |
It seems that there's a yearning from some for 'retro' sound reproduction. This can be seen in the resurgence of vinyl, and even cassette tapes seem to be making something of a (low-level) comeback. An undeniable part of this is that listeners have a physical 'thing' they can interact with, as opposed to a digital file that is nothing more than ones and zeros stored on a personal device or memory stick. These physical interactions don't make the music sound 'better', but they certainly affect the person's feelings. When it come to the audio chain, many people just don't like the idea of Class-D, having their music 'chopped up' into tiny pieces then filtered to extract the audio component. We'll ignore the fact that a CD does just that anyway, and it can be argued that tape and FM radio do something similar (the high-frequency bias used when recording to tape, and the RF carrier used to transport FM broadcasts). This is actually a false equivalence, as they do no such thing, as both are completely analogue processes.
A sensible approach is to take the best bits of retro designs (in particular the power amplifier), and augment them with modern circuitry for the preamp, tone controls and phono preamp (if required). We gain the benefits of almost immeasurable distortion with modern opamps, low noise and very predictable performance. The original transistor stages are shown, but opamp equivalents are also provided to keep noise and distortion to the minimum possible (based on 'traditional' circuits rather than anything especially exotic).
Many retro amplifiers were surprisingly good - not just for 'their time', but by modern standards as well. Not all circuitry is applicable of course, as early systems generally used very simple circuits, often using single-transistor gain stages everywhere other than the phono equaliser. There is some appeal, because getting good performance from very simple stages is satisfying, albeit difficult. The circuitry described is modified from the Sansui AU-555A, with the power amp being the least modified - apart from the transistors which have all been replaced with readily available equivalents. Even if the original Japanese transistors could be obtained new, they would almost certainly be different from the 1970s era devices. Fabrication techniques have evolved, and the old methods aren't used any more. Some suppliers even warn the user that the 'modern' version is not an exact equivalent.
This project is based on the Sansui design, but there are significant changes, particularly for the preamp, tone controls and phono preamp (if used). This isn't something I normally do - most ESP circuits are 'new' designs (that will almost always be based on common design principles). The difference here is that the circuits have been improved (albeit marginally in some cases), but are fairly true to the original circuits. Re-designing a 'retro' hi-fi from scratch is an oxymoron, as it probably shares almost nothing with any circuitry of old, because the original transistors are difficult or impossible to obtain. The amp is rated for 25W into 8Ω and 33W into 4Ω. There should be close to double the power with a 4Ω load, and the reduced figure indicates that the power transformer was too small.
We need to define (or perhaps re-define) 'good performance', because the vanishingly low distortion and low noise that we can get with a modern opamp simply cannot be achieved with one transistor. We are also stuck with inverting stages, but provided we use an even number of them, the output will be 'in-phase' with the input (i.e. not inverted). Not that this actually matters. No-one ever knows what processing and/ or polarity inversions have occurred during the recording, mixing and mastering processes, but it's generally true that a positive pressure on a mic diaphragm will cause a positive pressure in front of the reproducer (the speaker).
There's nothing that demands no polarity reversal, but it's generally considered to be desirable that the 'absolute polarity' is retained. It is audible with some tones and a few instruments, but can only be heard when the signal is switched. If you leave the room and someone else makes the switch, you won't hear it. There is no polarity reversal if all stages shown here are used.
Provided we don't expect a vast amount of gain from a single stage, it's not especially hard to keep distortion below 0.1%. That's pretty woeful compared to a good opamp, but it's within the scope of hi-fi, and is generally better than you'll get with most valve (vacuum tube) stages. One thing that simple transistor circuits are not very good at is being a unity-gain buffer. The closest is an emitter follower, but these show a loss of signal level (up to 1dB) and they aren't linear when the peak input voltage is greater than 25% of the supply voltage. Mostly, that's not a limiting factor with preamp circuitry. Simple transistor stages are really bad at providing a unity gain inverting buffer.
Because resistor values with a 'solid-state' circuit are much lower, thermal noise is reduced compared to a valve stage, and the gain can be set a little more predictably. A supply voltage of around +24V is ideal, as that allows for reasonable headroom before clipping.
We are interested in handling signals of around 150 to 250mV input (some sources are much higher of course), with an output voltage of perhaps 1V (RMS) into a load of 10k or more. If we aim for a total gain of around 10dB (preferably easily varied to suit the application) that gets us close to where we need to be for the preamp stage. The power amp has an input sensitivity of ~770mV RMS for full output.
While most early designs used single transistors, adding an emitter follower means just one more transistor and resistor. At the time, transistors were fairly expensive and economy was the key. Now we can use as many as we like at almost no extra cost. However, in keeping with the 'retro' idea, we'll still keep the transistor count as low as possible.
One thing that I have not included is the original phono preamp - modified or otherwise. Like everything else, the circuit was fairly simple, but would (probably) do a passable job. However, it lacks any finesse, and is the basic arrangement that was used by countless manufacturers of the era, but was (IMO) rather poorly designed. It's a two transistor feedback pair, with a conventional EQ network than manages to work quite badly in terms of accurate EQ. Also, some of the resistor values are much too high, so I'd expect it to be comparatively noisy (especially if compared to my Project 06 design). If there is any interest, I'll rework the design and include it, but the opamp version shown is vastly superior. Of course, you can get all of the original Sansui circuits on-line and take it from there if you prefer. The version shown uses an opamp, and when the input is derived from an inverse RIAA filter, it's commendably flat across the range from 20Hz to 20kHz. The maximum deviation is ±0.1dB from 27Hz to 20kHz, with a rolloff of under 0.2dB at 20Hz.
The input stage is basically just switching unless you add the phono preamp. The original pot was 250k, with a tap to allow 'loudness' compensation. A 50k pot is more appropriate for a modern system. That sets the input impedance, and with few people using valve (vacuum tube) peripheral equipment any more, this is a more sensible input impedance. I don't like the idea of the volume control being right at the input, because that means that all gain/ EQ stages operate at full gain all the time. This increase the likelihood of noise, especially at low levels. As a result, it's been moved to the output of the tone circuit.
The opamp phono preamp is a far better proposition than any 2-transistor gain stage. Its input impedance is higher, output impedance is lower, and the distortion will be below measurement system limits (depending on the opamp used of course). Although 'pedestrian', a 4558 or TL072 will outperform the original by a wide margin, and the RIAA EQ section is far more accurate than the one used when the amp was made. The Sansui circuit has some rather bizarre low-frequency behaviour depending on the impedance of the phono cartridge. My first impression when I simulated the response was "WTF !" - it was pretty odd, but when supplied via a reverse RIAA network is wasn't quite as bad as it looked at first. However, there was a very pronounced 'bump' at 23Hz and a very unwelcome peak at 330kHz (over 15dB). The EQ circuit shown in Fig. 1 is based on a design by John Linsley-Hood (see Project 25 - Phono Preamps For All). The original claimed it was correct to within 0.3dB, but it's actually better than that. Note that R8 really is 8k - there's a small error if you use 8.2k, but it's still (just) within specifications. 8kΩ is preferred though - use 1.2k and 6.8k in series.
The preamp section is basic, but it is still capable of acceptable performance. It's not something I'd bother with if I were to build any of the circuits described. My choice would be based on opamps, although this can be limiting because the input gain stage must be inverting to preserve 'absolute' polarity if a tone control stage is included (they are almost always inverting). Does this really matter? Most blind tests say "no", but it's traditional to ensure that there are no polarity inversions throughout the system. The volume control was originally straight after the input switching, but that means the circuit has full gain at any volume setting, with an increased noise floor. The volume control has been moved to the output of the tone section, and the first gain stage can be preset for the gain you need.
A better arrangement would be to use a 4-gang pot, with one of each pair used at the input, and the other at the output of the preamp. Unfortunately, 4-gang pots are very hard to get, so this option is not available to most constructors.
A single transistor stage with a gain of 10dB can have less than 0.1% distortion quite easily, with a well-designed circuit approaching 0.01% at an output level of less than 1V peak. Frequency response can extend from a few hertz up to 100kHz, within a small fraction of 1dB. Interestingly, it's actually easier to design a circuit with high gain (around x10) than it is to get a gain of x3 (10dB close enough). The reason for this will become clear when we look at the design.
The source impedance is also an important factor. Because we are dealing with an inherently nonlinear amplifying device (and a transistor is better in this respect than a JFET or a valve), any significant source impedance allows the transistor to have greater non-linearity than if the source impedance is low. This is one reason that an amplifying transistor may be preceded by an emitter-follower. However, that's not ideal, as the follower will contribute noise, but no amplification.
As transistor circuits became more advanced due to falling prices and a better understanding of their limitations, it became possible to design circuits with highly predictable gain, and much lower distortion than the simple 1-transistor circuits could manage. The ESP article Discrete Opamp Alternatives shows a number of circuits that became very common. The two-transistor feedback pair was used in countless early preamplifiers, in some cases well past the time when 'decent' opamps became available.
In all cases, the designs shown here are optimised for a 24V DC supply. A lower voltage can be used, but the levels available are very limiting. Single transistor stages are sensitive to the load impedance. The load is effectively in parallel with the collector resistor, reducing the stage gain. The first circuit shown is based on the gain stage of the Sansui AU-555A, and it includes a bootstrap arrangement to increase the input impedance. This was important in the early days of transistor amplifiers, as many customers would have valve-based AM/ FM tuners and/ or tape recorders. The AU-555A was released in the mid 1970s, at a time when many people were still using valve equipment.
The original gain was 5.2 (14dB) when loaded by the tone control circuit or a 10k resistor. That's a bit too high, so the circuit has been re-configured to make it variable, using VR1 (Gain Preset). This lets you set the gain to suit your signal sources. In general, you'll need a gain of about three (10dB close enough). The maximum input level is limited to 2V peak (1.4V RMS), as the circuit will clip with anything above that. With the maximum input level, distortion is over 1% (1.3% as simulated), but it falls as the input is reduced. With an input of 500mV peak (350mV RMS), the distortion is 0.053%. At this level, harmonics extend to the 4th, at -100dBV (10μV). There is a smooth decay of harmonics, and as expected, the 2nd is predominant. It's at -62dBV, with the 3rd harmonic at -76dBV. While there are harmonics beyond the 4th, their level is insignificant.
This is expected with all simple designs. The price paid for no high-order harmonics is higher overall distortion, but at 'sensible' levels it's fairly benign. The frequency response of the Fig. 1 circuit is from 10Hz to 500kHz (-0.1dB). The AU-555A used a bootstrapped input that caused slightly unpredictable response, and the bootstrap circuit has been removed. Input impedance is 100k across the audio range.
The gain of the stage is altered by changing the setting of VR1, which maintains the total resistance of 2kΩ. If the setting of VR1 is reduced, C2 has to be increased in proportion, but 100μF is suitable for any typical setting of VR1. Unfortunately, when the collector of Q1 is at its optimum value (~13V DC), the emitter is at ~3V, and at high levels the difference between the two signals approaches zero (no voltage across Q1). This sets the clipping levels to about 20V and 5V at the collector and emitter respectively. The quiescent collector current is about 1.9mA. A negative supply solves problems such as this, but almost all early amps used a single supply.
Normally, the emitter resistor would be a much lower value, but that increases the gain. The gain is determined by the ratio of the collector and emitter resistors (excluding bypassed resistors - part of VR1 as shown). The load impedance is in parallel with R3, so for flat response the load impedance needs to be constant (something not provided by the tone control circuit shown below). The emitter-follower stage is an addition that improves the tone control performance. It's not an absolute requirement, but the extra cost is well under $1 so it would be silly to omit it. The output impedance of the emitter-follower is under 10Ω.
The relatively high source impedance for the preamp (due to a 250k volume control) meant that distortion was much worse than it should have been. This was probably unavoidable given the constraints that existed at the time. Now, we'd use a 50k pot that keeps the distortion below 0.1% at any setting. Note that the 10k load is for testing - it's not used once the tone stage is connected (this has an approximate impedance of 10k). The maximum output required is about 700mV (full-power sensitivity for the power amp).
The distortion figures are much worse than even a common-or-garden opamp can achieve (e.g. 4558 or TL072), and are way behind some of the newer devices such as the LM4562 or OPA1642. Distortion of these is close to immeasurable without sophisticated equipment. Opamps can have any gain you want (including unity) and drive low impedance loads (> 2kΩ for 'basic' types) with an output of over 6V RMS with a 24V supply. No single transistor circuit can come close, even when we add an emitter-follower.
The opamp gain stage uses a non-inverting amp for the first stage (which provides the gain), followed by a unity-gain inverter to maintain the correct polarity after the tone control stage (which is also inverting). This approach allows the inverter to use low value resistors for lowest noise. The input could be applied to an inverting stage, but the resistances involved will be too high, and inverting amplifiers are also noisier than non-inverting amps. It uses more parts than the discrete version, but has much better performance.
The gain is adjusted to suit your system with VR1, and can be set from 2.6 (8.3dB) up to 8.3 (18dB). The box marked 'V/2' is a half-voltage reference, that's used for the other channel and the tone control circuits. It's well filtered to minimise hum problems.
The next circuit is a tone control. The Sansui AU-555A included a midrange control that required an 800mH inductor, but that has been eliminated to minimise cost (decent 'audio-grade' inductors are relatively expensive). To say it was of marginal use is high praise, and a physical inductor increases the chances of external hum fields causing problems. This problem was (apparently) solved, but the midrange control was a gimmick at best, with limited range. The circuit included a 'loudness' control, and used a 250k volume pot at the input. The loudness control supposedly compensates for our hearing response, but no-one ever got it to work properly. The control boosted bass by 8dB at 50Hz and treble by 3dB at 10kHz, but only at low volume settings. The loudness control requires a tapped pot - almost impossible to get now (and not used here).
The tone controls are the 'classic' Baxandall feedback type, and there's nothing remarkable about them. The distortion performance with the controls 'flat' is pretty good, and it actually (partially) cancels some of the even-order harmonic distortion from the input stage. However, the distortion increases when boost is applied, because there is less feedback.
The original tone controls were somewhat convoluted, and they've been simplified. The change also makes the input impedance flatter. As built by Sansui, the tone control circuit added almost 1dB of boost at 100Hz, with the boost starting at 750Hz (+0.2dB). This could not be disabled, as the tone network had an input impedance of 27k at 50Hz and 8.6k at 3kHz. Because of the relatively high output impedance of the first (Fig. 1, but without the emitter follower) stage, this caused the problem. The modification doesn't fix this, but the boost is reduced to 0.2dB at 63Hz.
Had the input preamplifier used an emitter follower (which I have added), this issue wouldn't exist, but it does increase distortion a little. The circuit also had switchable high and low-pass filters. With -3dB frequencies of ~120Hz and 6kHz, they would be very intrusive (telephone bandwidth) and their inclusion cannot be recommended.
Boost (and cut) are set for just under 12dB. More is possible, but is rarely useful. The bass and treble boost are centred on 1kHz (which is conventional, but I prefer ~400-500Hz, roughly an octave lower. The values for C1 and C2 (in brackets) are preferred (IMO). Note that R2 is 18k rather than 22k as you'd expect. This was done to prevent a bass 'shelf' that was about 0.5dB below where it should be. You can use 22k if preferred, and simply give a tiny bit of bass boost to get a flat output.
The power supply for any single-supply circuitry has to be very quiet (especially hum), as PSRR was marginal at best. The input bias resistors also connect to the supply, making low noise an absolute requirement. This is easily done now with an IC regulator, but it was harder before they became popular (and low-cost). The AU-555A used a simple 2-stage resistor/ capacitor filter, which is not ideal, but seems to work well enough. The original supply voltage was 26V, but that would vary depending on the mains voltage. A regulator is (IMO) essential, but the common 7824 types are a bit noisy, so an R/C filter after the regulator is worthwhile.
This is completely different from the original design, and utilises the low impedance of an opamp to reduce all resistance values to minimise noise. It can provide up to 15dB of boost or cut, with the 'centre frequency at about 600Hz. The response isn't shown - it's similar to that in Fig. 4, but not quite the same.
The combined response of the input stage and tone controls is shown above. The bass and treble controls are shown at 100%, 75%, 50% (flat), 25% and 0%. The preamp gain is set for 10dB, which is usually sufficient. The overall distortion of the two stages is level dependent, but (as simulated) it's about 0.015% with an output level of 700mV RMS (full power from the power amp). The volume control follows the tone circuits, ensuring minimum circuit noise at low settings.
There is a downside to using opamps with a single supply. Unlike simple transistor stages, opamps can't easily be provided with a slow voltage rise after power-on (especially the inverting stages). That means that there may be an audible 'thump' when power is applied. This can be minimised with a muting circuit if it's found to be too loud, or you just don't like the idea. Many early amps had this problem to some degree, and no-one worried about it.
The power amp is the least modified of the circuits, but I've made a few changes because this really is the 'heart-and-soul' of any hi-fi system. The transistor types are changed to commonly available devices, and as the amp is no 'powerhouse' TIP3055 (or TO-3 2N3055) are adequate. Personally, I'd use TIP35C transistors - much more rugged and they have lower case to heatsink thermal resistance. The original output devices were 2SC1030, rated for only 50W. You can also use MJL21194 (or MJL3281) if you wish. These are better than the other options, but will be more expensive. However, their greater gain linearity will give better performance than 'lesser' transistors.
The input stage got its power from a very basic 'capacitance multiplier' circuit, but that's been modified and moved to the power supply. The one 'multiplier' can handle both channels, and provides a +55V supply. It's possible to delete it altogether, but there will be some ripple breakthrough. Note that all component designators are mine, and they do not match the Sansui schematic.
The feedback (AC and DC) is applied to the emitter of Q1, and this is a current feedback design rather than the now-common voltage feedback topology. There are three feedback paths, with DC feedback applied via R7 and R8. The AC component is bypassed by C3. AC feedback is split in two, with half provided via R12 and C4, and the other half taken from the output via R11. This helps to reduce any distortion from the output cap (C9, 1,500μF/ 2,200μF), and also ensures that the response doesn't roll off prematurely. Without the secondary feedback, the response would be down by 3dB at 22Hz, but that's improved to 11.5Hz by the extra feedback.
The DC level is set by VR1, and should be a dynamic setting - VR1 is adjusted for symmetrical clipping with an 8Ω load at just beyond rated output. If you were to set VR1 to get exactly half the supply voltage with no signal, that doesn't compensate for supply voltage sag when the amp is drawing current. VR1 will typically be set for about 29V DC at the positive end of C9 with no output. The actual voltage will depend on the DC supply and mains voltage.
Q3 is the bias servo transistor, and it must be mounted to the heatsink so it senses the output transistor temperature. A hole can be drilled into the heatsink, and Q3 inserted along with heatsink 'grease'. The leads (along with R14, R15, R16 and VR2) can be attached to a small piece of Veroboard that's attached to the heatsink with a screw. If you use this method, place some insulation under the Veroboard so that a 'stray' lead can't short to the heatsink. You can also use a small piece of blank PCB material, with 'mechanical etching' (i.e. using a rotary tool to make separate 'pads'. Check that are separate with a multimeter - even a tiny whisker of remnant copper will short the pads together.
The voltage amplifier stage (VAS) uses bootstrapping, with C5 providing the bootstrap to the junction of R10 and R13. The Class-A (VAS) and driver transistors should not require a heatsink as their dissipation will be less than 250mW under all 'normal' operating conditions. The bias is set my measuring the voltage across R22. For the suggested 30mA quiescent current, you should measure 14mV across R22. The circuit uses the current feedback topology, which is normally stable without the need for a 'dominant pole' capacitor, although it has one anyway (C6, 47pF). While this is nice in theory, you may need to increase the value of C6 if there's any trace of instability. Anything above 100pF is unlikely to be needed.
The original circuit used 10k for R1, which was an 'uninspiring' choice. It's been reduced to 100Ω, which reduces both noise and distortion (albeit ever-so-slightly). The circuit simulates with a distortion of 0.07% at 26W into 8Ω. This falls to 0.05% when the level is reduced by 10dB, and falls further to 0.03% with another 10dB level reduction. The optimum quiescent current is around 30mA, and at that value crossover distortion is all but completely eliminated.
As you can see, I retained the quasi-symmetry output stage. It's not particularly well-known (and is probably counter-intuitive), but this topology usually has slightly lower distortion than symmetrical Darlington pairs. Sziklai pairs (aka CFP or compound feedback pairs) are better again, but the lower half can be prone to oscillation. Interestingly, the lower pair in this design is a CFP, but when used like this they never oscillate. The reason for this is not known. If the output stage is re-configured to use Sziklai pairs, the -10dB THD falls to 0.003%, 0.00012% at -20dB and at 26W output (8Ω) it's 0.04%. This is significant, but it's very unlikely that you'll hear the difference. Of course, these are simulated results, but I know from experience that SIMetrix is surprisingly close to reality in most simulations of this type.
The power supply is fairly basic, and requires a 40V transformer. The power needed isn't high, but anything less than 100VA would be unwise. At full output, each channel will draw an average of about 800mA. Both channels will therefore pull ~1.6A, which is fine for a 100VA transformer. You can use a bigger transformer which will improve regulation and allow more power into 4Ω loads, but of course it will be larger and more expensive. The original only used a 2,200μF main filter cap, but that should be at least doubled (4,700μF is the suggested minimum). The original DC filter for the preamp is very effective, but it requires three resistors (all 1kΩ) and three fairly large caps (2 x 470μF and 1 x 1,000μF). A regulator can also use a 'post-filter' if you are paranoid about minimising hum, but it's rarely necessary.
Q1 is a 'capacitance multiplier' (which is actually a basic active filter - there is no 'multiplication'). The original was only single-pole (one base resistor, one capacitor) but a 2-pole filter is dramatically more effective for the same total resistance and capacitance. The output tracks the input, but attenuates any hum component by at least 40dB, usually more. R3 was omitted in the Sansui circuit, which reduces the effectiveness of the filter.
A MOSFET pre-regulator is used to reduce the input voltage to the regulator. The output from the MOSFET will be about 31V, so it's within the allowable range for the LM317. The maximum allowable input voltage for the LM317 is 40V. The zener current (ZD1) is about 5.75mA, enough to ensure good voltage stability. You can use a 7824 if you wish, but the pre-regulator is still necessary. The LM317 adjustable regulator is preferred because it is quieter than the 78-series, but the difference is not great in real terms. You can use an LM317HV (60V maximum input) if you don't wish to use the pre-regulator. The voltage-setting resistors for an LM317/HV are 100Ω and 1.8k (23.75V output) as shown. Make sure that D5 is included - the IC can be damaged if it's left out. The input to the pre-regulator is from the main supply, which works because Q2 already reduces hum by using the filtered supply (via Q1) to provide zener current. Ripple to the input of the LM317 is reduced by at least 40dB.
While I can't think of a good reason to build a retro hi-fi, I'm sure that there are people who like the idea, but are put off by the relatively high cost of an original amplifier. I've seen the AU-555A advertised for anything from AU$450 to over AU$1,000 on-line. That's rather a lot for an old 20W/ channel amplifier! The condition is unknown of course, and while your purchase might work, it will almost certainly need some work done. Capacitor replacement is not especially expensive, but the value and ESR need to be checked. There are people who claim that you must 're-cap' an old amplifier, but that's only true if they have degraded (low capacitance, high ESR or both). The myth that caps suffer from 'degraded sound' even when they test fine is just that - a myth.
As noted, I strongly recommend that the preamp and tone control circuits are run from the regulated supply. The 56V supply requires that the voltage be reduced before the regulator. Failure to keep the regulator's input voltage below 40V (for a 7824 regulator) will result in failure. This is shown in the suggested power supply circuit. It may not be as 'retro' as the simple RC filter used, but it will perform better, with less chance of hum.
Overall, the most interesting part of the circuit is the power amp. The current feedback topology is (in many respects) superior to the long-tailed pair and voltage feedback, but it does have less gain so there's also less feedback. To many people, this is a good thing. The capacitor-coupled output is unfortunate, but current feedback amplifiers don't have the same low offset that you get with a long-tailed pair. That means that a dual-supply version is harder to design for low DC offset. The method of choice is to use a DC servo, but that has to be very carefully designed to ensure that it doesn't introduce other problems (particularly very low frequency instability).
Note that the copyright for the AU555-A amplifier remains the property of Sansui, and the circuits described are for reader information only. No PCBs will be produced for this project.
Main Index Projects Index |