Chapter 7

Moving-Magnet Disc Inputs

The Vinyl Medium

The vinyl disc as a medium for music delivery in its present form dates back to 1948, when Columbia introduced microgroove 33 rpm LP records. These were followed soon after by microgroove 45 rpm records from RCA Victor. Stereo vinyl did not appear until 1958. The introduction of Varigroove technology, which adjusts groove spacing to suit the amplitude of the groove vibrations, using an extra look-ahead tape head to see what the future holds, allowed increases in groove packing density. This density rarely exceeded 100 grooves per inch in the 78 rpm format, but with Varigroove 180–360 grooves/inch could be used at 33 rpm.

While microgroove technology was unquestionably a considerable improvement on 78 rpm records, any technology that is 50 years old is likely to show definite limitations compared with contemporary standards, and indeed it does. Compared with modern digital formats, vinyl has a restricted dynamic range, poor linearity (especially at the end of a side) and is very vulnerable to permanent and irritating damage in the form of scratches. Even with the greatest care, scratches are likely to be inflicted when the record is removed from its sleeve. This action also generates significant static charges, which attract dust and lint to the record surface. If not carefully removed this dirt builds up on the stylus and not only degrades the reproduction of high-frequency information today, but may also damage it in the future if it provokes mistracking.

Vinyl discs do not shatter under impact like the 78 shellac discs, but they are subject to warping by heat, improper storage, or poor manufacturing quality control. Possibly the worst feature of vinyl is that the stored material is degraded every time the disc is played, as the delicate high-frequency groove modulations are worn away by the stylus. When a good turntable with a properly balanced tone-arm and correctly set up low-mass stylus is used this wear process is relatively slow, but it nevertheless proceeds inexorably.

However, for reasons that have very little to do with logic or common sense, vinyl is still very much alive. Even if it is accepted that as a music-delivery medium it is technically as obsolete as wax cylinders, there remain many sizeable album collections that it is impractical to replace with CDs and would take an interminable time to transfer to the digital domain. I have one of them. Disc inputs must therefore remain part of the audio designer’s repertoire for the foreseeable future, and the design of the specialized electronics to get the best from the vinyl medium is still very relevant.

Spurious Signals

It is not easy to find dependable statistics on the dynamic range of vinyl, but there seems to be general agreement that it is in the range 50–80 dB, the 50 dB coming from the standard quality discs and the 80 dB representing direct-cut discs produced with quality as the prime aim. My own view is that 80 dB is rather optimistic.

The most audible spurious noise coming from vinyl is that in the mid frequencies, stemming from the inescapable fact that the music is read by a stylus sliding along a groove of finite smoothness. There is nothing that the designer of audio electronics can do about this. Scratches create clicks that have a large high-frequency content, and it has been shown that they can easily exceed the level of the audio [1]. It is important that such clicks do not cause slew limiting or other forms of overload, as this makes their subjective impact worse.

The signal from a record deck also includes copious amounts of low-frequency noise, which is often called rumble; it is typically below 30 Hz. This can come from several sources:

  1. Mechanical is noise generated by the motor bearings and picked up by the stylus/arm combination. This tends to be at the upper end of the low-frequency domain, extending up to 30 Hz or thereabouts. This is a matter for the mechanical designer of the turntable, as it clearly cannot be filtered out without removing the lower part of the audio spectrum.

  2. Room vibrations will be picked up if the turntable and arm system is not well isolated from the floor. This is a particular problem in older houses where the wooden floors are not built to modern standards of rigidity, and have a perceptible bounce to them. Mounting the turntable shelf to the wall usually gives a major improvement. Subsonic filtering is effective in removing this.

  3. Low-frequency noise from disc imperfections. This is the worst cause of disturbances. They can extend as low as 0.55 Hz, the frequency at which a 33 rpm disc rotates on the turntable, and are due to large-scale disc warps. Warping can also produce ripples in the surface, generating spurious subsonic signals up to a few hertz at surprisingly high levels. These can be further amplified by a poorly controlled resonance of the cartridge compliance and the pickup arm mass. When woofer speaker cones can be seen wobbling – and bass reflex designs with no cone loading at very low frequencies are the worst for this – disc warps are usually the cause. Subsonic filtering is again effective in removing this.

(As an aside, I have heard it convincingly argued that bass reflex designs have only achieved their current popularity because of the advent of the CD player, with its greater bass signal extension, but lack of subsonic output.)

Some fascinating data on the subsonic output from vinyl was given by Tomlinson Holman [2] and shows that the highest warp signals occur in the 2–4 Hz region, being some 8 dB less at 10 Hz. By matching these signals with a wide variety of cartridge–arm combinations, he concluded that to accommodate the very worst cases, a preamplifier should be able to accept not less than 35 mVrms in the 3–4 Hz region. This is a rather demanding requirement, driven by some truly diabolical cartridge–arm set-ups that accentuated subsonic frequencies by up to 24 dB.

Since the subsonic content generated by room vibrations and disc imperfections tends to cause vertical movements of the stylus, the resulting electrical output will be out of phase in the left and right channels. The use of a central mono subwoofer system that sums the two channels will provide partial cancellation, reducing the amount of rumble that is reproduced. It is, however, still important to ensure that subsonic signals do not reach the left and right speakers.

Maximum Signal Levels on Vinyl

There are some definite limits to the signal level possible on a vinyl disc, and they impose maxima on the signal that a cartridge and its associated electronics will be expected to reproduce. The exact values of these limits may not be precisely defined, but the way they work sets the ways in which maximum levels vary with frequency, and this is of great importance.

There are no variable gain controls on RIAA inputs, because implementing an uneven but very precisely controlled frequency response and a suitably good noise performance is quite hard enough without adding variable gain as a feature. No doubt it could be done, but it would not be easy, and the general consensus is that it is not necessary. The overload margin, or headroom, is therefore of considerable importance, and it is very much a case of the more the merrier when it comes to the numbers game of specmanship. The issue can get a bit involved, as a situation with frequency-dependent vinyl limitations and frequency-dependent gain is often further complicated by a heavy frequency-dependent load in the shape of the feedback network, which can put its own limit on amplifier output at high frequencies. Let us first look at the limits on the signal levels that stylus-in-vinyl technology can deliver. In the diagrams that follow the response curves have been simplified to the straight-line asymptotes.

Figure 7.1(a) shows the physical groove amplitudes that can be put on to a disc. From subsonic up to about 1 kHz, groove amplitude is the constraint. If the sideways excursion is too great, the groove spacing will need to be increased to prevent one groove breaking into another, and playing time will be reduced. Well before actual breakthrough occurs, the cutter can distort the groove it has cut on the previous revolution, leading to ‘pre-echo’ in quiet sections, where a faint version of the music you are about to hear is produced. Time travel may be fine in science fiction but it does not enhance the musical experience. The ultimate limit to groove amplitude is set by mechanical stops in the cutter head.

Figure 7.1: (a) The levels on a vinyl disc. (b) The cartridge response combined with the disc levels. (c) The RIAA curve. (d) The RIAA combined with curve (b). (e) Possible preamplifier output restrictions

There is an extra limitation on groove amplitude; out-of-phase signals cause vertical motion of the cutter, and if this becomes excessive it can cause it to cut either too deeply into the disc medium and dig into the aluminum substrate, or lose contact with the disc altogether. An excessive vertical component can also upset the playback process, especially when low tracking forces are used; in the worst case the stylus can be thrown out of the groove completely. To control this problem the stereo signal is passed through a matrix that isolates the L–R vertical signal, which is then amplitude limited. This potentially reduces the perceived stereo separation at low frequencies, but there appears to be a general consensus that the effect is not audible. The most important factor in controlling out-of-phase signals is the panning of bass instruments (which create the largest cutter amplitudes) to the center of the stereo stage. This approach is still advantageous with digital media as it means that there are two channels of amplification to reproduce the bass information rather than one.

From about 1 kHz up to the ultrasonic regions, the limit is groove velocity rather than amplitude. If the disc cutter head tries to move sideways too quickly compared with its relative forward motion, the back facets of the cutter destroy the groove that has just been cut by its forward edges.

On disc replay, there is a third restriction – that of stylus acceleration, or, to put it another way, groove curvature. This sets a limit on how well a stylus of a given size and shape can track the groove. Allowing for this at cutting time places an extra limitation on signal level, shown by the dashed line in Figure 7.1(a). The severity of this restriction depends on the stylus shape; an old-fashioned spherical type with a tip diameter of 0.7 mil. requires a roll-off of maximum levels from 2 kHz, while a (relatively) modern elliptical type with 0.2 mil. effective diameter postpones the problem to about 8 kHz. The limit, however, still remains.

Thus, disc-cutting and playback technology put at least three limits on the maximum signal level. This is not as bad a problem as it might be, because the distribution of amplitude with frequency for music is not flat with frequency; there is always more energy at LF than HF. This is especially true of the regrettable phenomenon known as rap music. For some reason there seems to be very little literature on the distribution of musical energy versus frequency, but a rough rule is that levels can be expected to be fairly constant up to 1 kHz and then fall by something like 10 dB/octave. The end result is that despite the limits on disc levels at HF, it is still possible to apply a considerable amount of HF boost to reduce surface noise problems. At the same time the LF levels are reduced to keep groove amplitude under control. Both functions are implemented by applying the inverse of the familiar RIAA replay equalization at cutting time.

Equalization and its Discontents

Both moving-magnet (MM) and moving-coil (MC) cartridges operate by the relative motion of conductors and magnetic field, so the voltage produced is proportional to rate of change of flux. The cartridge is therefore sensitive to groove velocity rather than groove amplitude, and so its sensitivity is expressed as X mV per cm/s. This velocity sensitivity gives a frequency response rising steadily at 6 dB/octave across the whole audio band for a groove of constant amplitude. Therefore a maximal signal on the disc, as in Figure 7.1(a), would give a cartridge output like that shown in Figure 7.1(b), which is simply Figure 7.1(a) tilted upwards at 6 dB/ octave. From here on the acceleration limits are omitted for greater clarity.

The RIAA replay equalization curve is shown in Figure 7.1(c). It has three corners in its response curve, with frequencies at 50.05 Hz, 500.5 Hz, and 2.122 kHz, which are set by three time-constants of 3180, 318, and 75 μs. The RIAA curve was of USA origin but was adopted internationally with surprising speed, probably because everyone concerned was heartily sick of the ragbag of equalization curves that existed previously. It became part of the IEC 98 standard, first published in 1964, and is now enshrined in IEC 60098, ‘Analogue Audio Disk Records and Reproducing Equipment’.

Note the flat shelf between 500 Hz and 2 kHz. It may occur to you that a constant upward slope across the audio band would have been simpler, and requiring fewer precision components to accurately replicate. But such a response would require 60 dB more gain at 20 Hz than at 20 kHz, equivalent to 1000 times. The minimum open-loop gain at 20 Hz would have to be 70 dB (3000 times) to allow even a minimal 10 dB of feedback at that frequency, and implementing that with a simple two-transistor preamplifier stage would have been difficult if not impossible (must try it some time). The 500 Hz – 2 kHz shelf in the RIAA curve reduces the 20 Hz – 20 kHz gain difference to only 40 dB, making a two-transistor preamplifier stage practical. One has to conclude that the people who established the RIAA curve knew what they were doing.

Figure 7.1(c) shows in dotted lines an extra response corner at 20.02 Hz, corresponding to a time-constant of 7950 μs. This extra roll-off is called the ‘IEC Amendment’ and it was added to what was then IEC 98 in 1976. Its apparent intention was to reduce the subsonic output from the preamplifier, but its introduction is something of a mystery. It was certainly not asked for by either manufacturers or their customers, and it was unpopular with both, with some manufacturers simply refusing to implement it. It still attracts negative comments today. On one hand, it was pointed out that as an anti-rumble measure it was ineffective, as its slow first-order roll-off meant that the extra attenuation at 13 Hz, a typical cartridge–arm resonance frequency, was a feeble −5.3 dB; however, at 4 Hz, a typical disc warp frequency, it did give a somewhat more useful −14.2 dB, reducing the unwanted frequencies to a quarter of their original amplitude. On the other hand, there were loud complaints that the extra unwanted replay time-constant caused significant frequency response errors at the low end of the audio band, namely −3.0 dB at 20 Hz and −1.0 dB at 40 Hz. Some of the more sophisticated equipment allows the Amendment to be switched in or out; a current example is the Audiolab 8000PPA phono preamplifier.

Another possible problem with the IEC Amendment is that it was almost certainly intended to be implemented by restricting the value of the capacitor at the bottom of a series feedback arm, i.e. C0 in Figure 7.2. While electrolytic capacitors nowadays (2009) have relatively tight tolerances of ±20%, in 1976 you would have been more likely to encounter −20% + 50%, the asymmetry reflecting the assumption that electrolytics would be used for non-critical coupling or decoupling purposes where too little capacitance might cause a problem, but more than expected would be fine. This meant that there could be significant errors in the LF response.

Figure 7.2: Series-feedback RIAA equalization, with the IEC Amendment implemented by C0. Component values for a gain at 1 kHz of 35.0 dB. RIAA accuracy is within ±0.1 dB from 20 Hz to 20 kHz

And another problem… is the non-linearity of electrolytic capacitors when they are asked to form part of a time-constant. This is described in detail in Chapter 2. Since the MM preamps of the 1970s tended to have poor linearity at LF anyway, because the need for bass boost meant a reduction in the LF negative-feedback factor, introducing another potential source of distortion was not exactly an inspired move. There is little doubt that even a simple second-order filter subsonic, switchable in and out, would be a better approach to controlling subsonic disturbances. If a Butterworth (maximally flat) alignment was used, with a −3 dB point at 20 Hz, this would only attenuate by 0.3 dB at 40 Hz, but would give a more useful −8.2 dB at 13 Hz and a thoroughly effective −28 dB at 4 Hz. It should be said, however, that not all commentators are convinced that the more rapid LF phase changes that result are wholly inaudible. Subsonic filters are examined more closely at the end of this chapter.

To get back to the levels coming from the cartridge, when the RIAA equalization of Figure 7.1(c) is applied to the cartridge output of Figure 7.1(b), the result looks like Figure 7.1(d), with the maximum amplitudes occurring around 1–2 kHz. This in agreement with Holman’s data [2].

Figure 7.1(e) shows some possible output level restrictions that might affect Figure 7.1(d). If the IEC Amendment is implemented after the first stage, there is a possibility of overload at low frequencies that does not exist if the Amendment is implemented in the feedback loop by restricting C0. At the high end, the output may be limited by problems driving the RIAA feedback network, which falls in impedance as frequency rises. More on this later.

The ‘Neumann Pole’

The RIAA curve is only defined to 20 kHz, but by implication carries on down at 6 dB/octave forever. This implies a recording characteristic rising at 6 dB/octave forever, which could clearly endanger the cutting head if ultrasonic signals were allowed through. From 1995 a belief began to circulate that record lathes incorporated an extra unofficial pole at 3.18 μs (50.0 kHz) to limit HF gain. This would cause a loss of 0.17 dB at 10 kHz and 0.64 dB at 20 kHz, and would require compensation if an accurate replay response was to be obtained. The name of Neumann became attached to this concept simply because they are the best-known manufacturers of record lathes.

The main problem with this story is that it is not true. The most popular cutting amplifier is the Neumann SAL 74B, which has no such pole. For protection against ultrasonics and RF it has instead a rather more effective second-order low-pass filter with a corner frequency of 49.9 kHz and a Q of 0.72 [3], giving a Butterworth (maximally flat) response rolling off at 12 dB/octave. Combined with the RIAA equalization this gives a 6 dB/octave roll-off above 50 kHz. The loss from this filter at 20 kHz is less than −0.1 dB, so there is little point in trying to compensate for it, particularly because other cutting amplifiers are unlikely to have identical filters.

Implementing RIAA Equalization

It can be firmly stated from the start that the best way to implement RIAA equalization is the traditional series-feedback method, as in Figure 7.2. This stage is designed for a gain of 35.0 dB at 1 kHz, and RIAA accuracy is within ±0.1 dB from 20 Hz to 20 kHz. With a nominal 5 mVrms input at 1 kHz the output will be 280 mV. The IEC Amendment is implemented by making C0 a mere 7.96 μF. You will note with apprehension that only one of the components, R0, is a standard value, and that is because it was used as the input to the RIAA design calculations that defined the overall RIAA network impedance. This is always the case for accurate RIAA networks. Here, even if we assume that capacitors of the exact value could be obtained, and we use the nearest E96 resistor values, systematic errors of up to 0.06 dB will be introduced. Not a long way adrift, it’s true, but if we are aiming for an accuracy of ±0.1 dB it’s not a good start. If E24 resistors are the best available the errors grow to a maximum of 0.12 dB, and don’t forget that we have not considered tolerances – we are assuming the values are exact. If we resort to the nearest E12 value (which really shouldn’t be necessary these days), then the errors exceed 0.7 dB at the HF end. And what about those capacitors?

The answer is of course that by using multiple components in parallel or series we can get pretty much what value we like, and it is perhaps surprising that this approach is not adopted more often. The reason is probably cost – a couple of extra resistors are no big deal but multiple capacitors make more of an impact on the costing sheet. More on this important topic later in this chapter.

Be aware that the circuit of Figure 7.2 is not optimized, but attempts to represent a ‘typical’ design. R0 could and should be made lower in value to reduce its Johnson noise contribution.

The only drawback to the series configuration is what might be called the unity-gain problem. While the RIAA equalization curve is not specified above 20 kHz, the implication is clear that it will go on falling indefinitely at 6 dB/octave. A series-feedback stage cannot have a gain of less than unity, so at some point the curve will begin to level out and eventually become flat at unity gain – in other words, there is a zero in the response. Figure 7.3 shows the various poles and zero frequencies of the circuit in Figure 7.2, with their associated time-constants. T3, T4, and T5, are the time-constants that define the basic RIAA curve. T2 is the extra time-constant for the IEC Amendment, and T1 shows where its effect ceases at very low frequencies when the gain is approaching unity. At the high end, the final zero is at frequency f6, with associated time-constant T6, and because the gain was chosen to be +35 dB at 1 kHz it is quite a long way from 20 kHz and has very little effect at this frequency, giving an excess gain of only 0.10 dB. This quite quickly dies away to nothing as frequency falls below 20 kHz.

Figure 7.3: The practical response for series-feedback RIAA equalization, including the IEC Amendment, which gives an extra roll-off at 20.02 Hz

However, if the gain of the stage is set low to maximize the input overload margin, the 6 dB/octave fall tends to level out at unity early enough to cause significant errors in the audio band. Adding an HF correction pole (i.e. low-pass time-constant) just after the input stage makes the simulated and measured frequency response exactly correct. It is not a question of bodging the response to make it roughly right. If the correction pole frequency is correctly chosen the roll-off cancels exactly with the ‘roll-up’ of the final zero.

The technique is demonstrated in Figure 7.4, where several changes have been made. The overall impedance of the RIAA network has been reduced by making R0 220 Ω, to reduce Johnson noise from the resistors; we still end up with some very awkward values. The IEC Amendment is no longer implemented in this stage; if it was then the correct value of C0 would be 36.18 μF, and instead it has been made 220 μF so that its associated −3 dB roll-off does not occur until 3.29 Hz. Even this wide spacing introduces an unwanted 0.1 dB loss at 20 Hz, and perfectionists will want to use 470 μF here, which reduces the error to 0.06 dB. Most importantly, the gain has been reduced to +30 dB at 1 kHz to get more overload margin. With a nominal 5 mVrms input at 1 kHz the output will be 160 mV. The result is that the final zero f6 in Figure 7.3 is now at 66.4 kHz, much closer in, and it introduces an excess gain at 20 kHz of 0.38 dB, which is too much to ignore if you are aiming to make high-class gear. The correcting pole R4, C4 is therefore added, which solves the problem completely. Since there are only two components, and no interaction with other parts of the circuit, we have complete freedom in choosing C3 so we use a standard E3 value and then get the pole frequency exactly right by using two resistors in series – 470 and 68 Ω. Since these components are only doing a little fine-tuning at the top of the frequency range, the tolerance requirements are somewhat relaxed compared with the main RIAA network. The design considerations are (a) that the resistive section R4 should be as low as possible in value to minimize Johnson noise, and on the other hand (b) that the shunt capacitor C4 should not be large enough to load the op-amp output excessively at 20 kHz. At this level of accuracy, the finite gain open-loop gain of even a 5534 at HF has a slight effect, and the frequency of the HF pole has been trimmed to compensate for this. But what about the IEC Amendment? In several of my designs it has been integrated into the subsonic filter that immediately follows the RIAA preamplifier; this gives economy of component use but means that it is not really practicable to make it switchable in and out. To do this a separate 7950 μs time-constant is required after the preamplifier, as shown in Figure 7.4, where R3, C3 give the required −3 dB roll-off at 20.02 Hz. Once again we can use a standard E3 capacitor value and 470 nF has been chosen here, and once again an unhelpful resistor value results – in this case 16.91 kΩ. However, with E24 values, this can be implemented exactly as 16 kΩ + 910 Ω. The switch as shown will not be entirely click-free because of the offset voltage at A1 output, but that is relatively unimportant as it will probably only be operated a few times in the life of the equipment.

Figure 7.4: Series-feedback RIAA equalization reconfigured for 30.0 dB gain at 1 kHz, and a lower impedance RIAA network. The switchable IEC Amendment is implemented by C3, R3. An HF correction pole R4,C4 is added to keep RIAA accuracy within ±0.1 dB, 20 Hz to 20 kHz

The shunt-feedback equivalent of Figures 7.2 and 7.3 is shown in Figure 7.5. It has occasionally been advocated because it avoids the unity-gain problem, but it has the crippling disadvantage that with a real cartridge load, with its substantial inductance, it is about 14 dB noisier than the series RIAA configuration [4]. A great deal of grievous twaddle has been talked about RIAA equalization and transient response, in perverse attempts to render the shunt RIAA configuration acceptable despite its serious noise disadvantage. A series-feedback disc stage cannot make its gain fall below 1, as described above, while the shunt-feedback version can; however, an extra output pole solves that problem completely. Shunt feedback eliminates any possibility of common-mode distortion, but then at the signal levels we are dealing with that is not a problem, at least with bipolar input op-amps. A further disadvantage is that a shunt-feedback RIAA stage gives a phase inversion that can be highly inconvenient if you are concerned to preserve absolute phase.

Figure 7.5: Shunt-feedback RIAA configuration. This is 14 dB noisier than the series-feedback version

Passive and Semi-Passive RIAA Equalization

For many years, series-feedback RIAA preamplifiers as described above were virtually universal, it being accepted by all that they gave the best noise and overload performance. However, human nature being what it is, some people will always want to do things the hard way, and this is exemplified by the fashion for passive (actually, semi-passive is more accurate) RIAA equalization. The basic notion is to split the RIAA equalization into separate stages, and I have a dark and abiding suspicion that this approach may be popular simply because it makes the design of accurate RIAA equalization much easier, as all you have to do is calculate simple time-constants instead of grappling with foot-long equations. There is a price, and a heavy one: the overload and/or noise performance is inevitably compromised.

Clearly a completely passive RIAA stage is a daft idea because a lot of gain is required somewhere to get the 5 mV cartridge signal up to a usable amplitude. The nearest you can get is the scheme shown in Figure 7.6 (a), where the amplification and the equalization are wholly separate, with no frequency-dependent feedback used at all. R2, R3 and C1 implement T3 and T4, while C2 implements T5. There is no inconvenient T6 because the response carries on falling indefinitely with frequency. This network clearly gives its maximum gain at 20 Hz, and at 1 kHz it attenuates by about 20 dB. Therefore, if we want the modest +30 dB gain at 1 kHz used in the previous example, the A1 stage must have a gain of no less than 50 dB. A 5 mVrms, 1 kHz input would therefore result in 1.58 V at the output of A1. This is only 16 dB below clipping, assuming we are using the usual sort of op-amps, and an overload margin of 16 dB is much too small to be usable. It is obviously impossible to drive anything like a volume-control or tone-control stage from the passive network, so the buffer stage A2 is shown to emphasize that extra electronics are required with this approach.

Figure 7.6: Passive and semi-passive RIAA configurations

The only answer is to split the gain so that the A1 stage has perhaps 30 dB, while A2 after the passive RIAA network makes up the loss with 20 dB more gain. Sadly, this second stage of amplification must introduce extra noise, and there is always the point that you now have to put the signal through two amplifiers instead of one, so there is the potential for increased distortion.

The most popular architecture that separates the high and low RIAA sections is seen in Figure 7.6 (b). Here there is an active LF RIAA stage using feedback to implement T3 and T4 with R1, C1, R2, followed by R3, C2, which give a passive HF cut for T5. This is what can only be called a semi-passive configuration. The values shown give an RIAA curve correct to within 0.04 dB from 20 Hz to 20 kHz. Note that because of the lack of time-constant interaction, we can choose standard values for both capacitors, but we are still left with awkward resistor values.

As before, amplification followed by attenuation means a headroom bottleneck, and this passive HF roll-off is no exception. Signals direct from disc have their highest amplitudes at high frequencies, so both these configurations give poor HF headroom, overload occurring at A1 output before passive HF cut can reduce the level. Figure 7.7 shows how the level at A1 output (trace B) is higher at HF than the output signal (trace A). The difference is trace C, the headroom loss; from 1 dB at 1 kHz this rises to 14 dB at 10 kHz and continues to increase in the ultrasonic region. The passive circuit was driven from an inverse RIAA network, so a totally accurate disc stage would give a straight line just below the +30 dB mark.

Figure 7.7: Headroom loss with passive RIAA equalization. The signal level at A1 (trace B) is greater than at A2 (trace A) so clipping occurs there first. Trace C shows the headroom loss, which reaches 18 dB at 20 kHz

A related problem in this semi-passive configuration is that the op-amp A1 must handle a signal with much more HF content than the op-amp in the series-feedback stage, worsening any difficulties with slew limiting and HF distortion. It uses two amplifier stages rather than one, and more precision components, because of the extra resistor. Another difficulty is that A1 is more likely to run out of open-loop gain at HF, as the response plateaus above 1 kHz rather than being steadily reduced by increasing negative feedback. Once again a buffer stage A2 is required to isolate the final time-constant from loading.

Another method of equalization is a variation on the circuit just described, shown in Figure 7.6 (c), where the T5 roll-off is done by feedback via R4, C2 rather than by passive attenuation. This is not really passive in any way, as the equalization is done in two active stages, but it does share the feature of splitting up the time-constants for easier design. As with the previous circuit, A1 is running under unfavorable conditions as it has to handle a larger HF content than in the series-feedback version, and there is now an inconvenient phase reversal. The values shown give the same gain and RIAA accuracy as the previous circuit, though in this case the value of R3 can be scaled if more gain is required.

There are many other alternative arrangements that can be used for passive or semi-passive equalization. There may be a flat input stage followed by a passive HF cut and then another stage to give the LF boost, as in Figure 7.6 (d), which has even more headroom problems and uses yet more bits. In contrast, the ‘all-in-one-go’ series-feedback configuration avoids unnecessary headroom restrictions and has the minimum number of stages. Passive RIAA is not an attractive option.

Calculating the RIAA Equalization Components

Calculating the values required for the series-feedback configuration is not straightforward. You absolutely cannot take Figure 7.2 and calculate the time-constants of R2, C2 and R3, C3 as if they were independent of each other; the answers will be wrong. Empirical approaches (cut-and-try) are possible if no great accuracy is required, but attempting to reach even ±0.2 dB by this route is tedious, frustrating, and generally bad for your mental health.

The definitive paper on this subject is by Stanley Lipshitz [5]. This heroic work covers both series and shunt configurations and much more besides, including the effects of inadequate open-loop gain. It is relatively straightforward to build a spreadsheet using the Lipshitz equations that allows extremely accurate RIAA networks to be designed in a second or two; the greatest difficulty is that some of the equations are long and complicated – we’re talking real turn-the-paper-sideways algebra here – and some very careful typing is required.

My spreadsheet model takes the desired gain at 1 kHz and the value of R1, which sets the overall impedance levels of the RIAA network. In my preamplifier designs the IEC Amendment is definitely not implemented by restricting the value of C0; this component is made large enough to have no significant effect in the audio band, and the Amendment roll-off is realized in the next stage.

RIAA Equalization Component Tolerances

Having calculated the component values we want, we must face the fact that real components have tolerances on their value, and we need to assess what RIAA accuracy is possible without spending a fortune on precision parts; ±1% is the best tolerance readily available for resistors and capacitors, so at first it appears anything better than ±0.1 dB accuracy is out of the question. This is not so; let us consider the circuit in Figure 7.4. The component–sensitivity plots in Figures 7.8 and 7.9 show the effect of 1% deviations in the value of R1 and R2; the response errors never exceed ±0.05 dB, as there are always at least two components contributing to the RIAA response. The rapid fall-off at low frequencies is due to the subsonic filter, which was included in the simulations as it implements the IEC Amendment.

Figure 7.8: The effect on RIAA accuracy of a ±1% change in R1. The worst error is 0.05 dB, and the variation is negligible above 100 Hz

Figure 7.9: The effect on RIAA accuracy of a ±1% change in R2. The worst error is ±0.05 dB around 1 kHz

The RIAA capacitor sensitivity is shown in Figures 7.10 and 7.11, and it can be seen that tighter tolerances are needed for C1 and C2 than for R1 and R2 to achieve the same ±0.05 dB accuracy, as the capacitors have more effect on the response than the resistors.

Figure 7.10: The effect on RIAA accuracy of a ±0.44% change in C1. The effect is less than ±0.05 dB at low frequencies. There is a small and diminishing effect above 1 kHz

Figure 7.11: The effect on RIAA accuracy of a ±0.58% change in C2. The error is less than ±0.05 dB in the top four octaves, and the variation is negligible below 1 kHz

Finding affordable close-tolerance capacitors is not easy; the best solution still seems to be axial polystyrene types, which are freely available at a 1% tolerance. These capacitor ranges usually only go up to 10 nF, so some paralleling is required, and in fact is highly desirable. This is because the sum of multiple capacitors is more accurate than a single component of the same tolerance.

This statistical trick works because the variance of equal summed components is the sum of the individual variances. Thus, if we have four 10 nF capacitors in parallel, the standard deviation (the square root of the variance) increases only by the square root of 4, that is 2, while the total capacitance has increased 4 times; therefore we have inexpensively made an otherwise costly 0.5% close-tolerance 40 nF capacitor. There is a happy analog here with the use of multiple amplifiers to reduce electrical noise. Here we are using essentially the same method to reduce ‘statistical noise’. The application of this principle can be seen in Figure 7.17 below. The same technique could be used with resistors if necessary.

Figure 7.17: A simplified schematic of the Technics SU9600 RIAA stage, with its +136 V supply rail

The figures given above assume that the component values vary with a normal (Gaussian) distribution. If the 1% components were obtained by selection from a population that varied over say 5%, the distribution would be much flatter than Gaussian and the accuracy improvement much reduced. Having said that, this appears not to be the way that polystyrene capacitors at least are manufactured, and the expected improvement really is attained in practice. A very good discussion of managing statistical variation to enhance accuracy is given in Ref. [6]. This deals with the addition of tolerances in optical instruments, but the principles are the same.

Simulating Inverse RIAA Equalization

SPICE simulation is well suited to the task of checking that the RIAA component values chosen are accurate. The best way to do this is to build an inverse RIAA simulation that can be used to feed the RIAA preamplifier under test. This is much, much simpler than designing the preamplifier RIAA network because the time-constants can be completely decoupled from each other by using unity-gain buffers with zero-impedance outputs. The required response can be implemented in many ways, but my version is shown in Figure 7.12. The component values have nothing to do with practical circuitry and are chosen simply for ease of calculation. The first network implements the 7960 μs time-constant of the notorious IEC Amendment, and can be omitted if so desired. Since this first network is the inverse of a bass roll-off, its output must continue to rise indefinitely at 6 dB/octave as frequency falls, and it is therefore implemented with a current source, so that as the impedance of C1 rises the output voltage at node 20 rises indefinitely. The apparently odd value of 1.011 A for the current source is in fact cunningly chosen to give a final output of 0 dBV at 1 kHz, which simplifies SPICE output plotting. The 10 GΩ resistor Rdummy is required as SPICE otherwise considers node 20 to be at an undefined DC level, and objects strongly.

Figure 7.12: Inverse RIAA network for SPICE simulation

The voltage at node 20 controls the output of the voltage-controlled voltage source (VCVS) E1, which has its gain set to unity. It has zero output impedance and so acts as a mathematically perfect buffer. E is the conventional designator for a VCVS in SPICE.

E1 then drives the network R2, C3, R3, which implements the 3180 and 318 μs time-constants. E2 acts as another perfect buffer for the voltage at node 23, and drives R4, C3, R5, which implement the 75 μs time-constant. The very low value for R5 allows the output to go on rising at 6 dB/octave to well beyond 20 kHz; the response does not level out until the T6 zero at 2.12 MHz is reached.

Physical Inverse RIAA Equalization

Building a sufficiently accurate inverse RIAA network for precision measurements is not to be entered upon lightly or unadvisedly. The component values will need to have an accuracy a good deal better than 1%, and this makes sourcing components difficult and expensive. A much better alternative is to use a test system such as those by Audio Precision that allow an equalization file to modify the generator output level during a frequency sweep.

Overload Margins

Since RIAA preamplifiers rarely have variable gain, it is important that they have an adequate margin against overload. The nominal output of a magnetic cartridge is 5 mVrms at 1 kHz, but much greater levels are experienced with high output cartridges that can track a high-level recording. In a classic paper, Tomlinson Holman [7] gives 95 mVrms at 1 kHz as an absolute worst case. If we assume op-amp preamplifier technology with an output limit of 10 Vrms, this means that the gain at 1 kHz cannot exceed 40 dB if this input level is to be accepted without clipping. The nominal output is then 500 mVrms at 1 kHz. The 1 kHz overload margin is thus 95/5 = 25.6 dB, which is not generally considered to be better than mediocre. Reducing the gain of the preamplifier stage increases the overload margin at the cost of reducing the nominal output, so that the 35 dB gain (1 kHz) stage of Figure 7.2 gives an output of 280 mV, an overload level of 178 mVrms, and a margin of 31 dB, which might be called good. The 30 dB (1 kHz) stage in Figure 7.4 gives a nominal 158 mVout and has an overload level of 316 mVrms and a margin of 36 dB, which is definitely excellent, giving 10 dB more headroom. There is a question as to whether 160 mV is adequate for driving external equipment; if not, then another stage with variable gain may be required. Some gains and overload levels are shown in Table 7.1.

Table 7.1   Preamplifier gains, and output levels and overload margins for a nominal 5 mVrms input and maximum output of 10 Vrms (all at 1 kHz)

A complication is that the preamp stage output capability may vary with frequency. In the discrete preamplifier stages described below, most have an emitter-follower output, which is much better at sourcing current than sinking it. An RIAA feedback network, particularly one designed with a relatively low impedance to reduce noise, presents a heavy load at high frequencies because of the capacitors, and this heavy loading was often a major cause of distortion and headroom limitation in discrete RIAA stages that had emitter-follower outputs with highly asymmetrical drive capabilities, with the 20 kHz output capability, and thus overload margin, often reduced by 6 dB or more. Replacing the emitter resistor in the emitter-follower with a current source much reduces the problem, and the slight extra complication of using a push–pull Class-A output can bring it down to negligible proportions (for more details see Chapter 3 on discrete design). Op-amps such as the TL072 also struggled to drive RIAA networks at HF, and it was the advent of the 5532 op-amp, with its excellent load-driving capabilities, that finally solved that problem.

Further headroom restrictions may occur when not all of the RIAA equalization is implemented in one feedback loop. Putting the IEC Amendment roll-off after the preamplifier stage, as in Figure 7.4 above, means that very low frequencies are amplified by 3 dB more at 20 Hz than they otherwise would be, and this is then undone by the later roll-off. This sort of audio impropriety always carries a penalty in headroom as the signal will clip before it is attenuated, and the overload margin at 20 Hz is reduced by 3.0 dB. The effect reduces quickly as frequency increases, being 1.6 dB at 30 Hz and only 1.0 dB at 40 Hz. Whether this loss of overload margin is more important than providing an accurate IEC Amendment response is a judgement call, but in my experience it creates no trace of any problem in a stage with a gain of 30 dB (1 kHz). Passive-equalization input architectures that put flat amplification before an RIAA stage suffer much more severely from this kind of headroom restriction, and it is quite common to encounter preamplifiers that claim to be high-end, with a high-end price tag, but only have an overload margin of 20–22 dB. Bad show, chaps.

Earlier in this chapter we saw that Tomlinson Holman concluded that to accommodate the worst of the worst cases, a preamplifier should be able to accept not less than 35 mVrms in the 3–4 Hz region [2]. If the IEC Amendment is after the preamplifier stage, and C0 is made very large so it has no effect, the RIAA gain in the 2–5 Hz region has flattened out at 19.9 dB, implying that the equivalent overload level at 1 kHz will need to be 346 mVrms. The 30 dB (1 kHz) gain stage of Figure 7.4 has a 1 kHz overload level of 316 mVrms, which is only 0.78 dB below this rather extreme criterion. We are good to go.

At the other end of the spectrum, adding a correction pole after the preamplifier, to correct the RIAA response, also introduces a compromise in the overload margin, though generally a much smaller one. The 30 dB (1 kHz) stage in Figure 7.4 has a mid-band overload margin of 36 dB, which falls to +33 dB at 20 kHz. Only 0.4 dB of this is due to the amplify-then- attenuate action of the correction pole, the rest being due to the heavy capacitative loading on A1 of both the main RIAA feedback path and the pole-correcting RC network. This slight compromise could be eliminated by using an op-amp structure with greater load-driving capabilities, so long as it retains the low noise of a 5534A.

An attempt has been made to show these extra preamp limitations on output level in Figure 7.1(e) above, and comparing Figure 7.1(d), it appears they are almost irrelevant because of the fall-off in possible input levels at each end of the audio band.

To put all this into some sort of perspective, here are the 1 kHz overload margins for a few of my public designs. My first preamplifier, the ‘Advanced Preamplifier’ [8] achieved +39 dB in 1976, partly by using all-discrete design and ±24 V supply rails. A later discrete design in 1979 [9] gave a tour-de-force +47 dB, accepting over 1.1 Vrms at 1 kHz, but I must confess this was showing off a bit and involved some quite complicated discrete circuitry, including the push–pull Class-A output stages mentioned above. Later designs such as the Precision Preamplifier [10] and its descendant the Precision Preamplifier ’96 [11] accepted the limitations of op-amp output voltage in exchange for much greater convenience in most other directions, and still have an excellent overload margin of 36 dB.

Cartridge Impedances

The impedance of the cartridge strongly influences the noise performance of an MM RIAA stage. Manufacturers do not always supply this data, and so I have had to make the best of what is available. Some of the cartridges listed in Table 7.2 are vintage, some more up to date, the collection covering from about 1972 to the present day (2009).

Table 7.2   Some moving-magnet cartridge impedances, both current and historical

Type

Resistance (Ω)

Inductance (mH)

Goldring 1006

Not stated

570

Goldring 1042

Not stated

570

Goldring 2044

Not stated

720

Goldring 2100

Not stated

550

Goldring 2200

Not stated

680

Shure ME75-ED Type 2

610

470

Shure ME95-ED

1500

650

Shure V15V MR

815

330

Shure V15V IV

1380

500

Shure V15V III

1350

500

Shure M44G

650

650

Stanton 5000 AL-II

535

400

CS1 ‘Carl Cox’

430

400

The Shure V15V values are known to be accurate, having been checked by Burkhard Vogel [12]. Resistance ranges from 430 to 1500 Ω, and inductance from 330 to 720 mH.

Cartridge Loading

The standard loading for a moving-magnet cartridge is 47 kΩ in parallel with a certain amount of capacitance, the latter usually being specified by the manufacturer. It is normally in the range 50–200 pF. The capacitance is often the subject of experimentation by enthusiasts, and so switchable capacitors are often provided at the input of high-end preamplifiers, which allow several values to be set up by combinations of switch positions. The exact effect of altering the capacitance depends on the inductance and resistance of the cartridge, but a typical result is shown in Figure 7.13, where increasing the load capacitance lowers the resonance peak frequency and makes it more prominent and less damped. It is important to remember that it is the total capacitance, including that of the connecting leads, which counts.

Figure 7.13: The typical effect of changing the loading capacitance on an MM cartridge

Because of the high inductance of an MM cartridge, adjusting the load resistance can also have significant effects on the frequency response, and some preamplifiers allow this too to be adjusted. The only way to assess the effects of these modifications is to measure the output when a special (and expensive) test disc is played.

Cartridge–Preamplifier Interaction

One often hears that there can be problems due to interaction between the impedance of the cartridge and the negative-feedback network. Most commentators are extremely vague as to what this actually means, but according to the work of Tomlinson Holman [7], the factual basis is that it used to be all too easy to design an RIAA stage, if you are using only two or three discrete transistors, in which the NFB factor is falling significantly with frequency in the upper reaches of the audio band, perhaps as a result of using excessive dominant-pole compensation to achieve HF stability. Assuming a series-feedback configuration is being used, this means that the input impedance will fall with frequency, which is equivalent to having a capacitative input impedance. This interacts with the cartridge inductance and can cause a resonant peak in the frequency response, in the same way that cable capacitance or a deliberately added load capacitance can do.

For this reason a flat-response buffer stage between the cartridge and the first stage performing RIAA equalization is sometimes advocated. One design including this feature was the Cambridge Audio P50, which used a Darlington emitter-follower as a buffer; with this approach there is an obvious danger of compromising the noise performance because there is no gain to lift the signal level above the noise floor of the next stage.

Cartridge DC and AC Coupling

Some uninformed commentators have said that there should be no DC-blocking capacitor between the cartridge and the preamplifier. This is insane. The signal currents are tiny (for MM cartridges 5 mV in 47 kΩ = 106 nA, while for MC ones 245 μV in 100 Ω = 2.45 μA – a good deal higher) and even a small DC bias current could interfere with linearity. I am not aware of any published work on how cartridge distortion is affected by DC bias currents, but I think it pretty clear they will not improve things and may make them very much worse. Large currents might upset the magnet strength. Keep DC out of your cartridge.

Discrete MM Disc Input Stages

Discrete moving-magnet (MM) input amplifiers were almost universal until the early 1970s. For a long time op-amps had, quite deservedly, a poor reputation for noise when used in this application.

When the first bipolar transistor MM inputs were designed, active components were still expensive, and adding another transistor to a circuit was something not to be done lightly. The circuitry from that era therefore looks to us very much cut to the bone, and before disrespecting it we need to remember that it was designed under very different economic constraints.

Figure 7.14 shows a typical two-transistor MM input amplifier from the late 1960s. The configuration is generally considered to have been introduced by Dinsdale [13] in 1965, in a classic preamplifier design that was one of the first to deal effectively with the new equalization requirements for microgroove records. It is a two-stage series-feedback amplifier composed of two common-emitter stages. R3 and C1 make up an RF filter; note the high value of the series resistance. This is considerably greater than the DC resistance of most MM cartridges and looks like a mistake. The RIAA network is R6, R7, C4, C5, and has a high impedance to reduce loading on the stage output. Since R6 has the high value of 1M8, the RIAA network cannot be used for the DC feedback that is required to set the quiescent conditions. There is a separate DC feedback network comprising R1, R4, R5, R10, and C3, which establishes the appropriate voltage across R10.

Figure 7.14: A typical two-transistor MM amplifier as commonly used in the 1960s and early 1970s. Gain is +38.9 dB at 1 kHz with values shown

Because of its simplicity, this stage contains compromises. R11 needs to be high in value to maximize open-loop gain, but low to adequately drive the RIAA network and any external loading. RIAA accuracy is poor, with the errors reaching +1.6 dB at 20 Hz and +0.7 dB at 20 kHz; the IEC amendment is not implemented. THD is 0.011% for 1 Vrms out at 1kHz, and about 3 times greater at 40 Hz for 1 Vrms 1 kHz. Linearity can be much improved by raising the supply rail to +24 V and adjusting the DC biasing to suit.

A moving-magnet input has to deliver a maximum boost of nearly 20 dB at low frequencies, and this is on top of the gain required to get the desired output level at 1 kHz. If the cartridge output is taken as 5 mVrms at 1 kHz, and the amplifier output as 150 mVrms (which is about as low as you could hope to get away with if you are sending this signal to the outside world), then a total closed-loop gain of 20 + 34 = 54 dB is required at low frequencies. The open-loop gain obviously needs to be considerably higher than this, for a decent feedback factor is required not only to reduce distortion, but also to ensure that the RIAA equalization is accurately rendered by the feedback network. By 1970 it had become clear that the two-transistor configuration was really not up to the job, and more sophisticated circuits were developed. It must now be regarded as of purely historical interest.

Adding an extra transistor improves the possible performance remarkably. The most common three-transistor configuration was introduced by Arthur Bailey in 1966 [14], though his version had a rather awkward level-shift between the first and second transistors. Figure 7.15 shows an improved version from the early 1970s, designed by H.P. Walker [4] and later enhanced by me [8]. It consists of two voltage-amplifier stages as before, but an emitter-follower Q3 is added to buffer the collector of the second transistor from the load of the RIAA network and any external load. This means that the second transistor collector can be operated at a much higher impedance, generating more open-loop gain. The original Walker design had a simple 22 kΩ resistor as a collector load for Q2; when I was using this configuration in Ref. [8] I split this into two 12 kΩ resistors, bootstrapping their central point from the emitter of Q3 as shown in Figure 7.14; this further increased the open-loop gain and reduced the stage distortion by a factor of 3. Dominant-pole compensation is applied to Q2 by C1. Once again, the RIAA network has a high impedance and a separate path for DC feedback must be provided by R1; there is in fact no DC feedback at all through the RIAA network as it is connected to the outside of C9. R3 and C8 are the input RF filter. Note the heavy filtering of the supply to the first stage by R8 and C5.

Figure 7.15: A typical three-transistor MM amplifier as commonly used in the 1960s and early 1970s. The original design was by H.P. Walker, the bootstrapping was added by me

The added emitter-follower, running at a much higher collector current than Q2 (5 mA versus 500 μA), has a greater drive capability, which increases the output voltage swing into a load and so improves the input overload margin. However, a simple emitter-follower output stage has an asymmetrical output current capability, so this is less effective at high frequencies, where the impedance of the RIAA network is falling. Low-gain versions of this circuit may have the overload margin compromised by several decibels at 20 kHz. This can be overcome by making the output stage more sophisticated – replacing the emitter resistor R4 with a current source greatly improves matters, and using a push–pull Class-A output doubles the output current capability again. See Chapter 3 on discrete design for more details.

While this stage is a great improvement on the two-transistor configuration, it is not well adapted to dual supply rails, which would allow a still greater output swing without recourse to discrete supply regulators.

Figure 7.16 demonstrates another way to use three transistors in an RIAA amplifier; this configuration consists of a voltage-amplifier stage, an emitter-follower, and then another voltage-amplifier stage. The first transistor is now a PNP type, its collector bootstrapped for increased gain by connecting the lower end of collector load R6 to the emitter-follower Q2. R2 and C8 make up an RF filter, and once more there is a high value of series resistance. Q2 drives the output transistor Q3, which is now a common-emitter voltage amplifier with collector load R8. Note the asymmetrical supply voltages; the positive rail is 8 V greater than the negative rail, and this is almost certainly intended to increase the positive-swing capabilities of the Q3 stage. Once again note the high impedances in the RIAA network to reduce the loading on the output, and the consequent need for a separate DC feedback path via R10, the AC content being filtered out by C7. Versions of this configuration were used by Pioneer and Sonab.

Figure 7.16: Another three-transistor MM amplifier configuration with a quite different structure. This is a simplified version of a circuit used by both Pioneer and Sonab

The record for the highest supply voltage to an RIAA stage was set in 1974 by the Technics SU9600, which employed ±24 V rails and a third rail at a staggering +136 V (this gives a whole new meaning to the phrase ‘third-rail electrification’). To the best of my knowledge this record still stands. The general configuration is shown in Figure 7.17; seven transistors are used, in three cascaded differential voltage amplifiers, followed by an emitter-follower output buffer. The final voltage amplifier and the output stage work on asymmetrical supplies, running between +136 and −24 V. The output sits at +56 V to allow a symmetrical output swing, which accounts for the DC-blocking capacitor C11. Note that several resistors around the output stage are high-wattage types. RV1 allowed gain adjustment, while RV2 and RV3 were for setting the DC conditions. My information is that the maximum input was 900 mV (frequency unstated, but presumably at 1 kHz), the THD was 0.08% (frequency and level unstated), the signal-to-noise ratio was 73 dB with reference to 2 mV, and the RIAA accuracy was ±0.3 dB.

The output stage dissipation is of course enormous for a preamp stage, and the use of a constant-current source or a push–pull Class-A output stage would have allowed this to be much reduced; one can only speculate as to why those techniques were not used. There would have been some fearsome transients at the output on switch-on, and it is notable that an output muting relay was required, probably not so much for reducing audible noise as to give the later stages in the preamplifier a chance of survival.

In the original circuit small capacitors were freely sprinkled over the diagram, leading me to suspect that HF stability was a serious issue during development.

Op-Amp MM Disc Input Stages

The previous section should have convinced you that discrete MM preamplifier circuitry is not that straightforward to design, and there is a lot to be said for using a good op-amp, which if well chosen will have more than enough open-loop gain to implement the RIAA bass boost without introducing detectable distortion. The 5534/5532 op-amps have input noise parameters that are well suited to moving-magnet (MM) cartridges.

Having digested this chapter so far, we are in a position to summarize the requirements for a good RIAA preamplifier. These are as follows:

  1. Provide series feedback, as shunt feedback is approximately 14 dB noisier.

  2. Ensure correct gain at 1 kHz. This sounds elementary, but setting up the Lipshitz equations is not a negligible task.

  3. Ensure accuracy. My 1983 preamplifier was designed for ±0.2 dB accuracy from 20 Hz to 20 kHz, the limit of the test-gear I had at the time. This was tightened to ±0.05 dB without using rare parts in my 1996 preamplifier.

  4. Use obtainable components. Resistors will be from the E24 series and capacitors probably E6, so intermediate values must be made by series or parallel combinations.

  5. R0 must be as low as possible as its Johnson noise is effectively in series with the input signal. This is particularly important when the MM preamplifier is fed from a low impedance, which typically occurs when it is providing RIAA equalization for the output of an MC preamplifier, rather than accepting input direct from an MM cartridge with its high inductance.

  6. The feedback RIAA network impedance to be driven must not be so low as to increase distortion or limit output swing, especially at HF.

  7. The resistive path through the feedback arm should ideally have the same DC resistance as input bias resistor R18, to minimize offsets at A1 output. This is a bit of a minor point as the offset would have to be quite large to significantly affect the output voltage swing.

The circuit shown in Figure 7.18 meets all these requirements, and is a practical development of the circuit shown in Figure 7.4. Note that a 5534A is used at the input stage to get the best possible noise performance. A 5534A without external compensation has a minimum stable closed-loop gain of about 3 times; that is close to the gain at 20 kHz here, so a touch of extra compensation is likely be required for reliable stability. The capacitor shown is 4.7 pF.

Figure 7.18: An op-amp RIAA preamplifier using a 5534 to get the best possible noise performance

The resistors can be made more than exact enough by combining two E24 values in series. The RIAA network capacitances are made up of multiple 1% polystyrene capacitors for improved accuracy. Thus, for the five 10 nF capacitors that make up C1, the standard deviation (square root of variance) increases by the square root of 5, while total capacitance has increased five times, and we have inexpensively built an otherwise costly 0.44% close-tolerance 50 nF capacitor. Similarly, C2 is essentially composed of three 4n7 components and its tolerance is improved by root-three, to 0.58%. An HF correction pole R8, R9, C13 is fitted, and here the resultant loss of HF headroom is only 0.5 dB at 20 kHz, which I think I can live with.

Immediately after the RIAA stage is the subsonic filter. It is shown here because it also implements the IEC Amendment; there is more on such filtering later in this chapter. The filter is a third-order Butterworth HP filter, with its response modified to give a slow initial roll-off that implements the Amendment. This is done by reducing the value of R11 + R12 below that for maximal flatness. The stage also buffers the HF correction pole, and gives the capability to drive a 600 Ω load, if you can find one. A version of this design, using appropriate precision components, is manufactured by the Signal Transfer Company in bare PCB, kit, and fully built and tested formats [15].

Noise in RIAA Preamplifiers

The first priority is to find out the physical limits that set how low the noise can be. With a purely resistive source it is easy to calculate the Johnson noise from the input source resistance. With a noiseless amplifier this would be the equivalent input noise (EIN), but real amplifiers have their own noise, and the amount by which the source/amplifier combination is noisier is the noise figure (NF). I often wonder why NFs are used so little in audio; perhaps they are a bit too revealing. Certainly manufacturers seem to have no interest at all in quoting MM noise specs in a way that would allow easy comparison.

The subject of noise in MM RIAA preamplifiers is considerably complicated by the facts that the signal source is a complex impedance, and the equalization curve is a long way from flat. The best possible EIN for purely resistive sources, such as 200 Ω microphones, is easily calculated to be 129.6 dBu at the usual temperature and bandwidth, but the same calculation for a moving-magnet input is much harder. A highly inductive source – that has no standard value of inductance, or indeed resistance – is combined with the complications of RIAA equalization [5].

The basic noise situation for a series-feedback RIAA stage like that in Figure 7.18 is shown in Figure 7.19. The cartridge is modeled as a resistance Rgen in series with a large inductance Lgen, and is loaded by the standard 47 kΩ resistor Rin. The amplifier block A1 is treated as noiseless, its voltage noise being represented by the voltage generator Vnoise and its current noise being represented by the current generator Inoise. It does not matter which side of Vnoise, Inoise is connected because Vnoise has no internal resistance. Here Vnoise represents not only the noise of whatever input device is used in the amplifier, but also noise generated by resistors in the feedback network. The RIAA equalization is done by the noiseless black box after the amplifier.

Figure 7.19: A moving-magnet input simplified for noise calculations, with typical cartridge parameter values (Shure ME75-ED2)

Figure 7.19 illustrates several important effects that are due to the rising impedance of the cartridge inductance Lgen with frequency:

  1. The impedance seen by the current source Inoise increases with frequency. Here it increases from 640 Ω at 75 Hz to 24.9 kΩ at 18 kHz; the increase at the top end is moderated by the shunting effect of Rin. This increase has a major effect on the noise behavior. For the lowest noise you must design for a higher impedance than you might think, and van de Gevel [16] quotes 12 kΩ as a suitable value for noise optimization; this assumes A-weighting, inclusion of the IEC amendment, and cartridge parameters of 1000 Ω and 494 mH.

  2. The Johnson noise generated by Rin is shunted away from the amplifier input by an amount that decreases with frequency. Here the fraction reaching the amplifier rises from 0.014 to 0.53 from 75 Hz to 18 kHz.

  3. The proportion of noise from Rgen that reaches the amplifier input falls with frequency as the impedance of Zgen increases. Here the fraction reaching the amplifier falls from 0.99 to 0.47 from 75 Hz to 18 kHz.

  4. A complication that is not visible in the diagram is that the effective value of Rgen is not simply the resistance of the coils. It increases in value with frequency (while still remaining resistive – we are not talking about inductance here) as a consequence of hysteresis and eddy current magnetic losses in the iron on which the coils are wound (see Hallgren [17]). According to van de Gevel [16], this has little effect on noise issues.

On top of this complicated frequency-dependent behavior is overlaid the effect of the RIAA equalization.

Clearly this simple model has some quite complex behavior. It could be analyzed mathematically, using a package such as MathCAD, or it could be simulated by SPICE. The solution I chose is a spreadsheet mathematical model of the cartridge input that I call MAGNOISE, the basic method being as described by Sherwin [18], which divides the audio spectrum into a number of bands so RIAA equalization factors can be applied, and Vnoise, Inoise, and Rgen can be varied with frequency if desired. I extended the Sherwin method somewhat by using nine octave bands to cover from 50 Hz to 22 kHz, and my latest version adds switchable A-weighting. An advantage of the spreadsheet method is that it is very simple to turn off various noise contributions so you can experiment with noiseless amplifiers or other flights from physical reality. For example, the noise generated by 47 kΩ resistor Rin is modeled separately from its loading effects so it can be switched off independently. It is also possible to switch off the bottom three octave bands to make the results comparable with real cartridge measurements that require a steep 400 Hz high-pass filter to control the hum. The results match well with my 5532 and TL072 measurements, and experience shows the model is a usable tool. While it is no substitute for careful measurements, it gives a good physical insight and allows noise comparisons at the LF end, where hum is very difficult to exclude completely.

Table 7.3 shows some interesting cases; output noise is calculated for gain of +29.55 dB at 1 kHz, and signal-to-noise ratio for a 5 mVrms input at 1 kHz. The cartridge parameters were set to 610 Ω + 470 mH, the measured values for the Shure M75ED 2. No weighting is used in this table.

Table 7.3   RIAA noise results from the MAGNOISE spreadsheet model under differing conditions, in order of merit (Cases 1–3 assume a noiseless amplifier and are purely theoretical)

Firstly let us see what the target is; how quiet would the circuit of Figure 7.19 be if we had miraculously noise-free electronics?

Case 1. To begin with we will completely ignore the cartridge loading requirements and set Rin to 1000 MΩ, at which value it has no effect. The minimum EIN with these particular cartridge parameters is then −130.0 dBu (Case 1a). This is a very low value, and it is the quietest possible condition. All of this noise comes from Rgen, the resistive component of the cartridge impedance. The only way to improve on this would be to select a cartridge with a lower Rgen but the same sensitivity, or start pumping liquid nitrogen down the tone-arm. (As an aside, if you did cool your cartridge with liquid nitrogen at −196ºC, the Johnson noise from Rgen would only be reduced by 5.8 dB, and if you are using a 5534A in the preamplifier, as in Case 5 below, the overall improvement would only be 0.7 dB. Also of course, the compliant materials would go solid and the cartridge wouldn’t work at all. Hold the cryostats).

With lower, but still high, values of Rin the noise increases; with Rin set to 10 MΩ (Case 1b) the EIN is −129.9 dBu, a bare 0.1 dB worse. With Rin set to 1 MΩ (Case 1c) the EIN is now −129.2 dBu, 0.8 dB worse than the best possible condition.

Case 2. It is, however, a fact of life that MM cartridges need to be properly loaded, and when we set Rin to its correct value of 47 kΩ things deteriorate sharply, the EIN rising by 4.3 dB to −125.7 dBu. We are still assuming a noiseless amplifier, and this appears to be the appropriate noise reference for amplifier design, so the NF is 0 dB.

Case 3. We leave the amplifier noise switched off, but add in the Johnson noise from the resistor in the bottom leg of the NFB network (R0 in Figure 7.4, R7 in Figure 7.18) to see if its value of 220 Ω is appropriate. The noise only worsens by 0.7 dB, so it looks like that resistor is not the first thing to worry about. The noise from this resistor is included in all the cases that follow. The NF is now 0.7 dB.

Case 5. We will now take a deep breath and switch on the amplifier noise. We jump to Case 5, where we have a 5534A as the amplifying element, and using the typical 1 kHz specs for the A-suffix part, we get an EIN of −122.7 dBu and an NF of 3.0 dB. This could be regarded as both good and bad. Using thoroughly standard technology, we are within a few decibels of perfection; on the other hand, the opportunities for showing off some virtuoso circuit design appear limited.

Case 6. It is well known that the single 5534 has somewhat better noise specs than the dual 5532, with both en and in being significantly lower, but does this translate into a significant noise advantage in the RIAA application? Case 6 shows that on plugging in a 5532A the noise output increases by 2.0 dB, the EIN increasing to −120.7 dBu. The NF is now 5.0 dB, which looks a bit less satisfactory. If you want good performance then the inconvenience of a single package and an external compensation capacitor are well worth putting up with – there is no easier way to obtain a 2.0 dB noise improvement than by moving from the 5532A to the 5534A. The rightmost column of Table 7.3 shows how other op-amps compare with the 5534A. If your circuit design ends up with an odd number of half-5532s, a single 5534A can be placed in the MM stage, where its lower noise is best used.

Case 7. Ho-hum, I hear you murmur, it all comes back to the 5534, doesn’t it? What about all the other op-amps on the market? If for Case 7 we take the FET-input OPA2134, which is a very fine op-amp when DC accuracy and low bias currents are required, we find the en is much higher at image but in is much lower at image. It looks like we might be in with a chance, but the greater voltage noise does more harm than the lower current noise does good and the EIN goes up to −119.8 dBu. The OPA2134 is therefore 2.9 dB noisier than the 5534A and 0.9 dB noisier than the 5532A, and it is not cheap. The NF is now 5.9 dB.

Case 8. The LM4562 BJT-input op-amp is the new chip on the block, and a very fine op-amp it is too, giving significant noise improvements over the 5534/5532 when used in low-impedance circuitry. This is because its en is lower at image. However, the impedances we are dealing with here are not low, and the in, at image, is four times that of the 5534A, leading us to think it will not do well here. We are sadly correct, with EIN deteriorating to −117.8 dBu and the NF an unimpressive 7.9 dB. The LM4562 is almost 5 dB noisier than the 5534A and at the time of writing is a lot more expensive. I have done some measurements that confirm this, and since I had not at the time done the noise calculations, it came as a rather unwelcome surprise.

Case 9. The TL072 with its FET input has high voltage noise but low current noise. The voltage noise is in fact very high at image, and we can expect a poor performance. We duly get it, with EIN rising to −116.5 dBu and a rather poor NF of 11.8 dB. The TL072 is 8.8 dB noisier than a 5534A, and 6.8 dB noisier than a 5532A. The latter figure is confirmed| (within experimental error anyway) by the data listed in the section below on noise measurements.

You may be wondering what has happened to other well-known op-amps, particularly the LT1028 and the OP27. Both are sometimes recommended for audio use, because of their low en values, but this ignores some serious problems.

The LT1028 gives a poor performance in MM applications because while it has an appealing low en of image its in is high at image as a result of running big input BJTs at high currents. The EIN is by calculation −120.9 dBu, making it a shade quieter than the 5532A. However, read to the end of this section before ordering any.

The OP27 has a low en and in, and in fact gives a calculated EIN of −123.0 dB, which beats the 5532A noise, but when you measure it in real life it is actually several decibels noisier; I have confirmed this several times. This is due to extra noise generated by bias-current cancellation circuitry. Correlated noise currents are fed into both inputs, and will only cancel if both inputs see the same impedance. In this RIAA application the impedances are wildly different, and the result is greatly increased noise. This problem with the OP27 was originally pointed out to me by Marcel van de Gevel [19].

The LT1028 also has bias-current cancellation circuitry, and the data sheet explicitly states: ‘The cancellation circuitry injects two correlated current noise components into the two inputs.’ In real life it will be several decibels noisier than the 5532A.

We will now go back to Case 4 (I have done it this way to keep the rows in Table 7.3 in order of noise performance), which is that of a single discrete bipolar transistor used as an input device, either as part of a fully discrete RIAA stage or as the front end to an op-amp. If we turn a blind eye to supply difficulties and use the remarkable 2SB737 transistor (with Rb only 2 Ω typical) then some interesting results are possible. We now can decide what collector current to run the device at, so we can to some extent trade off voltage noise against current noise. We know that current noise is important with an MM input, and so we will start with Ic = 200 μA, which gives Case 4c in Table 7.3. By coincidence, this gives exactly the same noise results as for the 5534A. Undiscouraged, we drop Ic to 100 μA (Case 4b) and voltage noise increases but current noise decreases, the net result being that things are now 0.8 dB quieter. If we reduce Ic again to 70 μA (Case 4a) we gain another 0.2 dB, and we have an EIN of −123.7 and an NF of only 2.0 dB. Voltage noise is now increasing fast and there is virtually nothing to be gained by reducing the collector current further.

We therefore must conclude that even an exceptionally good discrete transistor with appropriate support circuitry will only gain us a 1.0 dB noise advantage over the 5534A, and it is questionable if the extra complications are worth it.

Noise Results with A-Weighting

Many commentators feel that it is appropriate to use psychoacoustic weighting when studying RIAA preamplifier noise, because the frequency-dependent nature of the noise behavior interacts with the non-flat human hearing response at normal listening levels. A-weighting is universally used. You may therefore be wondering how the unweighted results described above will be affected by the application of A-weighting. Will the order of merit in Table 7.3 be upset?

Table 7.4 compares the results of no weighting and A-weighting on the results in Table 7.3. The apparent performance improves because the low-frequency noise, which has been emphasized by the RIAA equalization, is now discriminated against by the LF roll-off of the A-weighting curve. Note that the reference case for determining the NF is also subjected to A-weighting.

Table 7.4   The effect of A-weighting on the calculated noise performances of various amplifiers

It is clear that A-weighting does not introduce any revolutionary changes into the order of merit of the various devices. The 2SB737 at 200 μA now comes out as 0.3 dB noisier than the 5534A, but this is the only alteration.

RIAA Noise Measurements

In the past, many people who should have known better recommended that MM input noise should be measured with a 1 kΩ load, presumably thinking that this emulates the resistance Rgen, which is the only parameter in the cartridge actually generating noise – the inductance is of course noiseless. This overlooks the massive effect that the inductance has in making the impedance seen at the preamp input rise very strongly with frequency, so that at higher frequencies most of the input noise actually comes from the 47 kΩ loading resistance. I am grateful to Marcel van de Gevel for drawing my attention to some of the deeper implications of this point [19].

The importance of using a real cartridge load is demonstrated in Table 7.5, where the noise performances of a TL072 and a 5532 are compared. The TL072 result is 0.8 dB too low, and the 5532 result 4.9 dB too low – a hefty error. In general, results with the 1 kΩ resistor will always be too low, by a variable amount. In this case you still get the right overall answer – i.e. you should use a 5532 for least noise – but the decibel difference between the two has been exaggerated by almost a factor of 2.

Table 7.5   Measured noise performance of 5532 and TL072 with two different source impedances

These tests were done with an amplifier gain of +29.55 dB at 1 kHz. Bandwidth was 400 Hz–22 kHz to remove hum, rms sensing, no weighting, cartridge parameters were 610 Ω + 470 mH.

The 1 kΩ recommendation was perhaps made because the obvious measurement method of loading the input with an MM cartridge has serious difficulties with hum from the ambient magnetic fields. To get useful results it is essential to enclose the cartridge completely in a mu-metal can – I use one from a redundant microphone transformer and it works very well. I suppose the ideal load would be a toroidal inductor, but it would be an expensive custom part. It is desirable to use complete electrostatic screening of the amplifier itself. If it has a 22 μF input coupling capacitor and the input is short-circuited, the impedance downstream of the capacitor is 145 Ω at 50 Hz, which is enough to make it susceptible to electrostatic hum pickup.

Electronic Cartridge Loading for Lower Noise

Going back to Table 7.3, you will recall that when we were examining the situation with the amplifier and feedback network noise switched off, adding in the Johnson noise from the 47 kΩ loading resistor Rin caused the output noise to rise by 4.3 dB. In real conditions with amplifier noise included the effect is obviously less dramatic, but it is still significant. In the 5534A case the removal of the noise from Rin (but not the loading effect of Rin) reduces the noise output by 1.7 dB. Table 7.6 summarizes the results for various amplifier options in columns 4–6; the amplifier noise is unaffected, so the noisier the technology used, the less the improvement. Note that the no-Rin NFs in the fifth column look worse because the reference case also has the noise from Rin removed, and is now −130.0 dBu. All noise results in this section are unweighted.

Table 7.6   The noise advantages gained by removing or reducing the noise from the 47 kΩ loading resistor Rin (from MAGNOISE)

This may appear to be utterly academic, because the cartridge must be loaded with 47 kΩ to get the correct response. This is true, but it does not have to be loaded with a physical 47 kΩ resistor. An electronic circuit that has the V/I characteristics of a 47 kΩ resistor but lower noise, will do the job very well. Such a circuit may seem like a tall order – it will after all be connected at the very input where noise is critical, but unusually the task is not as difficult as it seems.

Figure 7.20 (a) shows the basic principle. The 47 kΩ Rin is replaced with a 1 MΩ resistor whose bottom end is driven with a voltage that is phase-inverted and 20.27 times that at the top end. If we conceptually split the 1 MΩ resistor into two parts of 47 and 953 kΩ, a little light mathematics shows that with −20.27 times Vin at the output of A2, the voltage at the 47 kΩ– 953 kΩ junction A is zero, and so far as the cartridge is concerned it is looking at a 47 kΩ resistance to ground. However, the physical component is 1 MΩ, and the Johnson current noise it produces is less than that from a 47 kΩ (Johnson current noise is just the usual Johnson voltage noise applied through the resistance in question). The point here is that the physical resistor value has increased by 21.27 times, but the Johnson noise has only increased by 4.61 times, because of the square root in the Johnson equation; thus, the current noise injected by Rin is also reduced by 4.61 times. The improvement gained using a practical 1 MΩ resistor rather than a less practical infinite one is a bit less, as shown in columns 7–9, but with a 5534A amplifier the difference is only 0.2 dB.

Figure 7.20: Electronic load synthesis. (a) The basic principle. (b) The van de Gevel circuit.

Other combinations of resistor value and amplifier gain could be used – the higher the resistance, the more the noise advantage, as described earlier in this chapter. The problem is that this also requires higher inverted drive voltage at the bottom of the resistor, and you will soon run into a situation where the inverting amplifier clips before the main path amplifier, restricting headroom.

The implementation made known by van de Gevel [16, 19] is shown in Figure 7.20 (b). This ingenious circuit uses the current flowing through the feedback resistor R0 to drive a shunt-feedback stage around A2. With suitable scaling of R3 (note that here it has an E96 value), the output voltage of A2 is at the right level and correctly phase-inverted. When I first saw this circuit I had reservations about connecting R0 to a virtual ground rather than a real one, and thought that extra noise from A2 might find its way back up R0 into the main path. (I hasten to add that these fears may be quite unjustified, and I have not found time so far to put them to a practical test.) The inverting signal given by this circuit is amplified by 20.5 times rather than 20.27, but this has a negligible effect on the amount of noise reduction.

Because of these reservations, I tried out my version of load synthesis as shown in Figure 7.21. This more closely resembles the circuit of Figure 7.20 (a); it is important that the inverting stage A3 does not load the input with its 1 kΩ input resistor R4, so a unity-gain buffer A2 is added. The inverting signal is amplified by 20 times, not 20.27, but once again this has negligible effect on the noise reduction.

Figure 7.21: Electronic load synthesis: the Self circuit

In practical measurements with a 5534A as amplifier A1, I found that the noise improvement with the real cartridge load (Shure M75ED 2, cartridge parameters 610 Ω + 470 mH) was indeed 1.5 dB, just as predicted, which is as nice a matching of theory and reality as you are likely to encounter in this world. There were no HF stability problems. Whether the 1.5 dB is worth the extra electronics is a good question; I say it’s worth having.

This technique has been called ‘electronic cooling’, presumably because it could be regarded as analogous to dipping the loading resistance in liquid nitrogen or whatever to reduce Johnson noise. I must admit I don’t like the term as it could be understood to mean that thermoelectric elements have been used to cool down the input stage, a technique I do not think has been used in hi-fi yet. I prefer to call it ‘electronic loading’, ‘active input impedance’ or ‘load synthesis’, the latter being perhaps the most explicit.

Subsonic Filters

In the earlier parts of this chapter we have seen that the worst subsonic disturbances occur in the 2–4 Hz region, due to disc warps, and are about 8 dB less at 10 Hz. We have also seen that the IEC Amendment gives only 14 dB of attenuation at 4 Hz, and in any case is often omitted by the manufacturer or switched out by the user. It is therefore important to provide authoritative subsonic filtering. What needs to be settled is what order filter to use, because some people at least will be concerned about the audibility of LF phase-shifts, and how far into the audio band the filter should intrude. There is nothing approaching a consensus on either point, so it can be a wise move to configure the subsonic filter so it can be switched out.

The third-order filter already described in this chapter also did the job of implementing the IEC Amendment, so it did not have one of the classic filter characteristics. All the filters described here do just the filtering job and it is assumed that the IEC Amendment is implemented elsewhere, if at all.

High-pass filters used for RIAA subsonic are typically of the second-order or third-order Butterworth (maximally flat) configuration, rolling off at rates of 12 and 18 dB/octave respectively, as shown in Figure 7.22. Fourth-order 24 dB/octave filters are much less common, presumably due to worries about the possible audibility of rapid phase changes at the very bottom of the audio spectrum. The Butterworth response is but one of many possible filter alignments; the Bessel response gives a slower roll-off, but aims for linear phase, i.e. a constant delay versus frequency, and so reproduces the shape of transients better. Other filter alignments such as Chebyshev give faster initial roll-offs than the Butterworth, but they do so at the expense of ripples in the passband or stopband gain, which is not helpful if you are aiming for a ruler-flat response after RIAA equalization.

Figure 7.22: Subsonic filters: second-order (a) and third-order (b) Butterworth high-pass filters, both 3 dB down at 20 Hz

A very handy filter configuration is the well-known Sallen-and-Key type; it has drawbacks when used as a low-pass handling high frequencies (the response comes back up due to the non-zero op-amp output impedance) but works very well for our purposes here. A second-order Sallen-and-Key filter is simple to design; the two series capacitors C1 and C2 are made equal and R2 is made twice the value of R1. Such a filter with a −3 dB point at 20 Hz is shown in Figure 7.21(a). Other roll-off frequencies can be obtained simply by scaling the component values while keeping C1 equal to C2 and R2 twice R1. The response is 24.0 dB down at 5 Hz, by which time the 12 dB/octave slope is well established, and we are well protected against disc warps. It is, however, only 12.3 dB down at 10 Hz, which gives little protection against arm-resonance problems. Above the −3 dB roll-off point the response is still −0.78 dB down at 30 Hz, which is intruding a little into the sort of frequencies we want to keep. We have to conclude that a second-order filter really does not bifurcate the condiment, and the faster roll-off of a third-order filter is preferable.

Third-order filters are a little more complex. Some versions are made up of a second-order filter cascaded with a first-order roll-off, using two op-amp sections. It can, however, be done with just one, as in Figure 21.1b, which is a third-order Butterworth filter also with a −3 dB point at 20 Hz. The resistor value ratios are now a less friendly 2.53:1.00:17.55, and the circuit shown uses the nearest E24 values to this – which by happy chance come out as E12 values. The frequency response is shown in Figure 7.23, where it can be seen to be 18.6 dB down at 10 Hz, which should keep out any arm-resonance frequencies. It is 36.0 dB down at 5 Hz so disc warp spurii won’t have a chance. The 30 Hz response is now only down by an insignificant −0.37 dB, which demonstrates that a third-order filter is much better than a second-order filter for this application. As before, other roll-off frequencies can be had by scaling the component values while keeping the resistor ratios the same.

Figure 7.23: Frequency response of a third-order Butterworth subsonic filter, 3 dB down at 20 Hz

When dealing with frequency-dependent networks like filters you need to keep an eye on the input impedance, because it can drop to unexpectedly low values, putting excessive loading on the stage upstream and degrading its linearity. In a high-pass Sallen-and-Key filter, the input impedance is high at low frequencies but falls with increasing frequency. In the third-order version, it tends to the value of R1 in parallel with R3, which here is 10.6 kΩ. This should not worry the previous stage.

Because of the large capacitances, the noise generated by the passive elements in a high-pass filter of this sort is usually well below the op-amp noise. The capacitances do not, of course, generate any noise themselves. With the values used here, SPICE simulation shows that the resistors produce −125.0 dBu of noise at the output (22 kHz bandwidth, 25ºC).

Capacitor distortion in electrolytics is (or should be) by now a well-known phenomenon. It is perhaps less well known that non-electrolytics can also generate distortion in filters like these. This has nothing to do with subjectivist musicality, but is all too real and measurable. Details of the problem are given in Chapter 2, where it is concluded that only NP0 ceramic, polystyrene, and polypropylene capacitors can be regarded as free of this effect. The capacitor sizes needed for subsonic filters are large, if impedances and hence noise are to be kept low, which means polypropylene has to be used. Anything larger than 470 nF gets to be big and expensive, so that is the value used here; 220 nF polypropylene is substantially smaller and about half the price. There is more information on this, and on highpass filters in general, in Chapter 5 on filters.

Combining Subsonic and Ultrasonic Filters

Scratches and groove debris create clicks that have a large high-frequency content, some of it ultrasonic and liable to cause slew-rate and intermodulation problems further down the audio chain. It is often considered desirable to filter this out as soon as possible (though of course some people are only satisfied with radio-transmitter frequency responses), but an obstacle to this is the extra cost and power consumption of another filter stage. This difficulty can be resolved by combining an ultrasonic filter with a subsonic filter in the same stage. Combined filters also have the advantage that the signal now passes through one op-amp rather than two, and can be extremely useful if you only have one op-amp section left.

This cunning plan is workable only because the high-pass and low-pass turnover frequencies are widely different. Figure 7.24 shows the third-order Butterworth subsonic filter combined with a second-order 50 kHz Butterworth low-pass filter; the response of the combination is exactly the same as expected for each separately. The low-pass filter is cautiously designed to prevent significant loss in the audio band, and has a −3 dB point at 50 kHz, giving very close to 0.0 dB at 20 kHz. The response is −12.6 dB down at 100 kHz and −24.9 dB at 200 kHz. C4 is made up of two 2n2 capacitors in parallel.

Figure 7.24: A third-order Butterworth subsonic filter combined with a second-order ultrasonic filter

Note that the mid-band gain of the combined filter is −0.15 dB rather than exactly unity. The loss occurs because the series combination of C1, C2, and C3, together with C5, form a capacitative potential divider with this attenuation, and this is one reason why the turnover frequencies need to be widely separated for filter combining to work. If they were closer together then C1, C2, C3 would be smaller, C5 would be bigger, and the capacitative divider loss would be greater.

References

[1]  M. Jones, Designing valve preamps: Part 1, Electronics World (March 1996) p. 192.

[2]  T. Holman, Dynamic range requirements of phonographic preamplifiers, Audio (July 1977) p. 74.

[3]  K. Howard, Cut and thrust: RIAA LP equalisation, Stereophile (March 2009).

[4]  H.P. Walker, Low-noise audio amplifiers, Electronics World (May 1972) p. 233.

[5]  S.P. Lipshitz, On RIAA equalisation networks, J. Audio Eng. Soc. (June 1979) p. 458. onwards.

[6]  W.J. Smith, Modern Optical Engineering, McGraw-Hill, 1990, p. 484.

[7]  T. Holman, New factors in phonograph preamplifier design, J. Audio Eng. Soc. (May 1975) p. 263.

[8]  D. Self, An advanced preamplifier design, Wireless World (November 1976).

[9]  D. Self, High performance preamplifier, Wireless World (February 1979).

[10]  D. Self, A precision preamplifier, Wireless World (October 1983).

[11]  D. Self, Precision preamplifier 96, Electronics World (July/August and September 1996).

[12]  B. Vogel, Adventure: noise (calculating RIAA noise), Electronics World (May 2005) p. 28.

[13]  J. Dinsdale, Transistor high-quality audio amplifier, Wireless World (January 1965) p. 2.

[14]  A.R. Bailey, High performance transistor amplifier, Wireless World (December 1966) p. 598.

[15]  http://www.signaltransfer.freeuk.com/

[16]  M. van de Gevel, Noise and moving-magnet cartridges, Electronics World (October 2003) p. 38.

[17]  B. Hallgren, On the noise performance of a magnetic phonograph pickup, J. Audio Eng. Soc. (September 1975) p. 546.

[18]  J. Sherwin, Noise specs confusing? National Semiconductor Application Note AN-104, Linear Applications Handbook (1991).

[19]  M. van de Gevel, Private communication (February 1996).

Small Signal Audio Design; ISBN: 9780240521770

Copyright © 2010 Elsevier Ltd; All rights of reproduction, in any form, reserved.

..................Content has been hidden....................

You can't read the all page of ebook, please click here login for view all page.
Reset
34.204.99.254