Design of moving-coil head amplifiers
For many years moving-coil head amps were exotic devices of highly specialised design. They were usually all-discrete, to minimise noise, and because of this, and the very low feedback impedances to be driven, linearity was poor, and only acceptable because the signal levels were so low.
This design is a hybrid discrete/op-amp configuration, which gives very low distortion even at signal levels monstrously above the normal operating conditions. It sidesteps several problems by having much more gain than is normally required; this would normally be a very bad move, severely curtailing headroom, but here it works as it can be assumed that the MC amp is always followed by a moving-magnet disc stage with substantial gain, so clipping will occur there first. When the article appeared, the 2SB737 transistor, with its magnificently low Rb, was expensive and not that easy to obtain, but in a year or two this situation changed and nobody now would consider devices like the 2 N4403 for this application.
In recent years, moving-coil cartridges have increased greatly in popularity. This is not the place to try and determine if their extra cost is justified by an audible improved performance; suffice it to say that a preamplifier now needs a capable moving-coil cartridge input if it is to be considered complete. The head-amplifier design presented here as an example was originally intended to be retrofitted to the precision preamplifier previously published in Wireless World,1 feeding the existing moving-magnet disc input. However, it is adaptable to almost any preamplifier and cartridge as the gain range available is very wide; it should therefore be of interest to any engineer working in this field. Hereafter ‘moving coil’ is abbreviated to m.c., and ‘moving-magnet’ to m.m.
Traditionally, moving-coil cartridges were matched to moving-magnet inputs by special transformers, which give ‘free gain’ – in a sense – and are capable of a good noise performance if the windings are carefully designed for very low series resistance. However, the inescapable problems of low-frequency distortion, high-frequency transient overshoots and the need for obsessive screening to avoid 50 Hz mains pickup render them unattractive and expensive.
The requirements for a high-quality m.c. head-amplifier are as follows. The overwhelming need is for a good noise performance, as the signals generated by m.c. cartridges are, in general, very low. However, this sensitivity is also much more variable than that of m.m. cartridges, where one can take a nominal output of 5 mV r.m.s. for 5 cm/s at 1 kHz as being virtually standard. In contrast, a survey of the available m.c. cartridges gave a range from 2.35 mV (Dynavector DV10X IV) to 0.03 mV (Audionote 102 vdH), though these are both exceptional and the great majority fell between 0.2 mV and 0.4 mV. Figure 1 shows the output levels of a number of current m.c. cartridges plotted on a scale of dBu (i.e. referred to 775 mV) and m.m. cartridges are included on the right for comparison. It is notable how these bunch together in a range of less than 7 dB.
A representative m.c. cartridge used both as a basis for design, and for testing, is the Ortofon MC10 Super, which has an output of 0.3 mV for 5 cm/s, and an internal resistance of 3 Ω. There is general agreement that this is a good-sounding component.
As detailed above, there is a need for easily variable gain over a wide range. This can be quite adequately provided in switched steps, avoiding the problems of uncertain stereo balance on dual potentiometers. From the above output figures, a gain range of 6 dB to 46 dB appears necessary to cater for all possible cartridges. It would seem, at the low-gain end, that the amplifier is virtually redundant, and so a minimum gain of 20 dB was chosen.
Moving-coil cartridges are very tolerant of the loading they see at an amplifier’s input, as a result of their own very low internal impedance. For example, Ortofon, who might be reckoned to know a thing or two about m.c. cartridges, simply state that the recommended load for most of their wide range of cartridges is ‘greater than 10 Ω’. Nonetheless, since experimenting with cartridge loading is a harmless enough pastime, provision for changing the input loading resistor over a wide range has been made in this design.
The preamplifier should have the ability to drive a normal m.m. cartridge input at sufficient level to ensure that the head amplifier does not limit the disc headroom. Any figure here over about 300 mV r.m.s. should be satisfactory. A less obvious point is that the input impedance, apart from the nominal 47 k resistive component, usually includes a fair amount of capacitance, either to adjust cartridge frequency response or to exclude r.f. This can cause head-amplifier instability unless it is dealt with.
Finally, a head amplifier should meet the usual requirements for frequency response, crosstalk, and linearity. Capacitive crosstalk is usually not a problem, due to the very low impedances involved, but for the same reason, linearity can present problems despite the low signal levels.
Design problems
The theoretical noise characteristics of amplifiers have been dealt with very competently in other articles,2 and there is no need to repeat the various mathematical derivations here. The designer’s options are usually limited to choosing a suitable input device, operating it at roughly the right current, (not usually critical due to the flat bottoms of the noise curves) and then making sure that the surrounding circuitry doesn’t mess things up too much. M.c. head amplifiers are almost always built around discrete devices, with or without the addition of an accompanying op-amp (for an exception see Ref. 3). Figure 2 shows the reason why: when source resistances are low (say below 1 k) even advanced op-amps are easily out-performed by discrete devices, due to the inevitable compromises in i.c. fabrication. The values of equivalent input noise (e.i.n.) in Figure 2 were taken from five samples of each device, using a source resistance of 3R3, and the general circuit configuration in Figure 3. The rather non-standard measurement bandwidth is due to the use of the internal filters on a Sound Technology measuring system; adding a third-order 20 kHz active filter at the ST input would be very difficult, as the levels of noise being measured are so low. To convert to 20 kHz upper bandwidth limit, subtract 1.5 dB. One of the prerequisites for good performance in this role is a low value for Rb, and this has led to a fine miscellany of devices being applied to a job they were never intended for: medium power devices, print-hammer drivers, (a lot of transistors seem to have been designed as print-hammer drivers) and so on.
Apart from careful device selection, the other classical way of reducing noise with low source impedances is to use multiple devices. The assumption here is that m.c. amplifier noise will swamp the miniscule Johnson noise inherent in the source (this is usually all too true) and therefore, if two input devices have their outputs summed, the signals will simply add, giving a 6 dB gain, while the two uncorrelated device noise contributions will partially cancel, giving only 3 dB.
Thus, there is a theoretical gain of 3 dB in noise performance every time the number of input devices is doubled. There are, of course, clear economic limits to the amount of doubling you can go in for; eight parallel devices is the most that I have seen. It also seems difficult in practice to get the full theoretical benefit.
M.c. head-amplifiers in use today can be roughly divided into three common topologies, as shown in Figure 4. That shown in 4(a) relies on a single device with low Rb, and the combination of limited open-loop gain and the heavy loading of the low-impedance of the feedback network on the final transistor means that both linearity and maximum output level tend to be uninspiring. Given the technical resources that electronics can deploy, there seems no need to ask the paying customers to put up with any measurable distortion at all. An amplifier of this type is analysed in Ref. 4.
Figure 4(b) shows the classic multiple-parallel-transistor configuration; the amplifier block A is traditionally one or two discrete devices, that usually have difficulty in driving the low-impedance feedback network. Effort is usually expended in ensuring proper current-sharing between the input devices.
This can be done by adding small emitter resistors to swamp Vbe variations, but these will effectively appear in series with the source resistance, and compromise the noise performance unless they are individually decoupled with a row of very large electrolytics. Alternatively, each transistor can be given its own d.c. feedback loop to set up its collector current, but this tends to be even more prodigal of components. Having said this, experiment proved that the problem of current-sharing was not as serious as conventional wisdom holds; this is explained below. For examples of circuitry see Ref. 5.
Figure 4(c) shows the series-pair scheme. This simple arrangement allows two input devices to give the normal 3 dB noise improvement without current-sharing problems as substantially the same collector current goes through each device. The collector signal currents are summed in Rc, which must be reasonably low in value to absorb-any current imbalance. This configuration has its adherents but it also has its difficulties, such as indifferent linearity.
It was therefore originally decided to base the design presented here on a single well-chosen device, with the spadework of providing open-loop gain and output drive capability left to an op-amp. This leads to the configuration in Figure 4(d), which gives excellent linearity, and less than 0.002% THD at full output may be confidently expected. The first problem to be dealt with is the very low value of Rf2; this must be as low as possible (say 10 Ω) as it is effectively in series with the input source resistance and will degrade the noise performance accordingly. This means that Cf must be very large, of the order of 2200 μF, to preserve the l.f. response. A 3R3 resistor in the Rf2 position demands 4700 μF to give − 3 dB at 10 Hz; this is not elegant. The capacitance Cf cannot be dispensed with, since there is a d.c. level of + 0.6 V on the emitter of the input device, leading to a wholly impossible offset at the output of the op-amp.
One solution to this is the use of a differential pair, as in Figure 4(e). This cancels out the Vbe of the input transistor Tr1, at the cost of some degradation in the noise performance of the circuit, and hopefully the d.c. offset is so much smaller that, if Cf is omitted and the offset is amplified by the full a.c. gain, it will not seriously reduce the output voltage swing. In effect, the second transistor Tr2 is an emitter follower transferring the feedback signal to the emitter of Tr1, and such a circuit element introduces a small but inescapable amount of extra noise. In this case, with the component values shown, the degradation is about 2.8 dB.
A possibly more serious objection to this circuit is that the offset at the output is non-negligible, about 1 V, much of which is due to the base bias current flowing through Rn. A d.c.-blocking capacitor on the output is essential, and if it is an electrolytic there may be some doubt as to which way round to put it, as the exact level of input pair balance is unpredictable.
After practical trials, it was decided that a 3 dB noise penalty was too great, and that a way had to be found to use a single-ended input.
A new approach
The new method evolved is shown in the block diagram Figure 5. There is no Cf in the feedback loop, and indeed no overall d.c. feedback at all. The two halves of the circuit, the input transistor and the op-amp, each have their own d.c., feedback systems. The transistor relies on simple shunt negative feedback via d.c. loop 1, while the op-amp has its output held precisely to a d.c. level of 0 V by the integrator A2. This senses the mean output level, and sets up a voltage on the non-inverting input of A1 that is very close to the level set on Tr1 collector, such that the output stays firmly at zero; its time-constant is made large enough to ensure that an ample amount of open-loop gain exists at the lowest audio frequencies. Failure to do this results in a rapid rise of distortion as the frequency is lowered. Any changes in the direct voltage on Tr1 collector are completely uncoupled from the output. However, a.c. feedback passes through Rf1 as usual and ensures that the linearity of the compound arrangement is near-perfect, as is often the case with transistor op-amp hybrid circuits. Due to the high open-loop gain of A the a.c. level on Tr1 collector is very small and so a.c. feedback through d.c. loop I does not significantly affect the input impedance of the amplifier, which is about 8 kΩ.
The device chosen for the input transistor was the 2 N4403, a type that has been acknowledged as superior for low-noise applications for some years. The Rb is quoted as about 40 Ω.5,6 More modern purpose-designed devices such as the 2SB737 will improve the noise performance by up to 1 dB, but the extra cost is significant.
A single device used in the circuit of Figure 6 gives an e.i.n. of − 138 dB with a 4 mA collector current, which is certainly not bad, but it was consistently found that putting devices in parallel without any current-sharing precautions whatever always resulted in a significant improvement in noise performance. On average, adding a second transistor reduced noise by 1.2 dB, and adding a third reduces it by another 0.5 dB. Beyond this the law of diminishing returns sets in and, since further multiplication was judged unprofitable, a triple-device input was settled on. The currentsharing under these conditions was checked by measuring the voltage across 100 Ω resistors inserted in the collector paths. Using 3.4 mA as the total current for the array it was found after much device-swapping that the worst case of imbalance was 0.97 mA in one transistor and 1.26 mA in another. No attempt was made to ensure that all the devices came from the same batch. It therefore appears that, for this device at least, matching is good enough to make simple paralleling worthwhile, and it was therefore decided to use three devices in parallel in the final circuit.
There now remains the problem of setting the gain. Usually it would be simple enough to alter Rf1 or Rf2, but here it is not quite so simple. The resistance Rf2 is not amenable to alteration, as it must be kept to the lowest practicable value of 3.3 Ω, and Rf1 must be kept up to a reasonable value so that it can be driven to a full voltage swing by an op-amp output. This means a minimum of 500 Ω if the op-amp is to be of an easily obtainable type such as the 5534. (It is paradoxical that amplifiers whose output is measured in millivolts are required to chuck around so much current.)
These two values fix a minimum closed-loop gain of about 44 dB, which is far too high for all but the most insensitive cartridges. The only solution is to use a ladder output attenuator to reduce the overall gain; this would be anathema in a conventional signal path, because of the loss of headroom involved, but since an output of 300 mV r.m.s. would be enough to overload virtually all m.m. inputs, we can afford to be prodigal with it. If the gain of the head amplifier is set to be a convenient 200×(+ 46 dB) then attenuation to reduce overall gain to a more useful + 20 dB still allows a maximum output of 480 mV r.m.s.; this comfortably exceeds the input capability of the intended host preamplifier, though one previous design would accept it all and come back for more.7 Smaller degrees of attenuation to provide intermediate gains allow greater outputs, and these are summarized in the specification. The Ortofon MC10 was used with + 26 dB of gain, to give similar output levels to m.m. cartridges driving the precision preamplifier RIAA stage direct.
The last constraint is the need to provide a low output impedance to the succeeding m.m. input stage, so that it can give a good noise performance; it is likely to have been optimized to give of its best with a source impedance of 500 Ω or less. This implies that the ladder attenuator will need low resistor values, imposing yet more loading on the unfortunate op-amp, so this problem has been side-stepped by making the ladder an integral part of the a.c. feedback loop, as shown in Figure 6. This is only practicable because it is known that the load resistance presented by the next stage will be too high at 47 kΩ to cause any significant gain variations.
The final circuit
This is shown in Figure 6, and most closely follows the configuration of Figure 4(d), with the exception that the input devices have suddenly multiplied themselves by three. Capacitor C1 is soldered on the back of the m.c. input phono sockets and is intended for r.f. filtering rather than modification of the cartridge response. If the need for more capacitive or resistive loading is felt, then extra components may be freely connected in parallel with R1. If R1 is raised in value, then load resistances of up to 5 kΩ are possible, as the impedance looking into C2 is about 8 kΩ. Capacitor C2 is large to give the input devices the full benefit of the low source impedance, and its value should not be altered. Resistors R2, R3 make up d.c. loop 1, setting the d.c. operating conditions of Tr1,2,3, while R4 is the collector load, decoupled from the supply rail by C9 and R5, which are shared between the two channels. Opamp IC1 provides the main a.c. open-loop gain, and is stabilized at h.f. by C4: R6 has no real effect on normal operation, but is included to give IC1 a modicum of negative feedback and hence tidy behaviour at power-up, when this would otherwise be lacking due to the charging time of C2 the other op-amp, IC2, is the integrator that makes up d.c. loop 2, its time-constant carefully chosen to provide plenty of open-loop gain from IC1 at low frequencies, and to avoid a peaking in the l.f. response that can occur due to the second time-constant of C2.
The ladder resistors R8 – R12 make up the combined feedback-network and output-divider, overall gain being selected by a push-on link in the prototype. A rotary switch could be used instead, but this will produce loud clicks when moved with the volume up, since the emitter current of Tr1 – Tr3 flows through R7, and a small current therefore flows down the divider chain. The output resistor R15 ensures stability when driving long screened cables, and C5 is included to eliminate any trace of d.c. offset from the output because the stage might find itself driving a horribly vulnerable ‘esoteric’ input stage with direct coupling and possibly substantial gain at d.c. Anything is possible these days.
Comparing performance parameters
These are given in the specification, and I think there will be few opportunities to quibble. On the vital question of noise it would be instructive to compare it with other preamplifiers – not easy because the noise performance of m.c. head amplifiers is specified in so many different ways it is virtually impossible to reduce them all to a similar form, particularly without knowing the spectral distribution of the noise. Noise performances are specified with and without CCIR-ARM weighting, over different band-widths, and with different source impedances. This article has dealt throughout with unweighted noise referred to the input, over a 400 Hz–30 kHz bandwidth, and with RIAA equalisation not taken into account. Without getting bogged down in invidious comparisons, I can only say that it is my belief that the design given here is quieter than most current designs, being within 6 dB of the theoretical minimum.
When using this design with the precision preamplifier, it was noted with some surprise that it was so quiet that the m.m. RIAA stage actually caused the noise performance to deteriorate by about 3 dB. Since the RIAA stage is itself very quiet (s/n ratio − 81 dB referred to 5 mV r.m.s. input) it is considered that the design goals were met.
Practice
P.c.b. layouts of require some care if the full performance is to be realised. First, the grounding should be carefully planned, as it must be realised that with such low impedances as R7 (3R3) playing a vital role, the resistance of tracks can be significant. It is suggested that a single star ground point be chosen on the p.c.b., and critical paths (input ground, R1, R7) all connected to this, to prevent signal currents causing voltage drops where they are least wanted. It is vital to avoid making loops in the input path that will pick up 50 Hz magnetic fields.
It is essential to place the decoupling capacitor C8 next to IC1 to prevent insidious h.f. oscillation which makes its presence known only by severely impaired linearity. When interfacing the head amplifier to an existing design, note that about 8 mA flows down the ground connection.