Phono Pre-amp Design Theory
There are plenty of perfectly good moving-magnet phono RIAA pre-amp circuits published, and I was originally planning to just copy a simple op-amp based version, but I decided to be a bit more adventurous. There are well known but rarely used methods both for noise reduction and rumble 'filtering' I want to experiment with.
Some older records were recorded using different equalisation time-constants, so although I only want an RIAA version for my own use I have chosen a circuit with easily calculated and modified time-constants so that they can easily be changed. Determining what equalisation was used on old recordings is not always easy or even possible, so my own choice would be to add bypassable tone controls or a graphic equaliser and just stick to RIAA. I have a few old 78 rpm discs in my collection, but my turntable, a Pro-Ject 1 Xpression, has no 78rpm speed. It does have a low voltage ac motor, so a variable speed control may be a future project.
Part 1. Noise
The minimum noise achievable, even if we had noiseless amplifying devices, is limited by the circuit configuration used. For a moving-magnet cartridge pre-amp requiring a standard 47k input resistance there are two well known circuits, shown next as A and B. The first of these is an inverting 'virtual earth' circuit which has a 47k resistor in series with the cartridge (shown for simplicity as a 600mH inductance). This unfortunately adds the thermal noise voltage of the resistor to the cartridge output voltage. At very high frequencies the cartridge impedance is high and reduces the gain of the circuit, so noise starts to fall above some frequency, typically 12kHz. Adding the RIAA frequency compensation however causes the pre-amp gain to rise at low frequencies, so reducing low frequency input noise is of some importance, and version B is far better in this respect because here the 47k resistor is in parallel with the cartridge impedance, which is low at low frequencies and therefore attenuates the resistor noise.
The majority of published circuits are of type B.
The circuits are shown without the RIAA frequency compensation, which can be done in several different ways, and adds a few difficulties, and will be covered later. For now we only consider the input noise.
The other two circuits, C and D, use feedback to control the input impedance. This is done by increasing the resistor value to 470k but increase the voltage across the resistor to 10 times the cartridge output voltage by applying an amplified and inverted signal voltage to the other end of the resistor. D is perhaps easier to understand than C. If the cartridge output voltage is Vi then this is amplified by the first stage by a factor of 9, and then inverted by the second stage, giving an output -9Vi. With Vi at one end and -9Vi at the other end of the 470k the resistor has 10Vi applied, so its current is identical to a 47k resistor with Vi applied, so the current taken from the cartridge is exactly the same as for a simple 47k resistor as in B, and so the cartridge sees the required 47k load. The great advantage of this approach is that the thermal noise of the higher 470k is attenuated more by the cartridge impedance. The thermal noise is proportional to the square root of the resistance, so the 470k has 10dB higher noise than the 47k, but the 10 times higher resistance gives about 20dB more attenuation with the cartridge impedance included, and so there is a net reduction of around 10dB in noise level. A greater reduction can be achieved by using a higher resistor and higher amplifier gain, but increasing the gain too far will adversely affect overload margins, and other sources of noise will limit the ultimate signal to noise ratio.
This 'active input impedance' idea has been known for many years, I think I saw it first in Practical Electronics or Wireless World around 1972, but the general principle was known long before (See references). I have found a discussion on DiyAudio from 2004 about the 'Pro-ject Phono Box' which uses this method, which says that it can only improve noise by 3dB, so maybe I am missing something obvious. More reason to build and test. The idea seems to have been rediscovered and renamed a few times, being sometimes known as 'actidamping' or 'electronic cooling'.
Shown next is the equivalent input noise voltage (in units of nV per sqrt Hz) for the above circuits. As mentioned B has much lower noise compared to A at low frequencies, but above about 12kHz the noise of A falls below that of B because the high cartridge impedance above this frequency reduces the gain of A but becomes increasingly ineffective in attenuating the resistor noise in B. The improvement of about 10dB for versions C and D can be seen to extend through most of the audio frequency range, reducing at higher frequencies. The usual inclusion of a capacitance in parallel with the input will change the high frequency levels, as of course will the inclusion of the RIAA compensation. There are a few other sources of noise such as internal cartridge resistance not included here, but the results are adequate for comparing the different versions.
One advantage of A compared to B is that there is only a very small voltage at the amplifier input. The non-inverting input is earthed, and for a high amplifier feedback loop gain the voltage at the inverting input may be just a small fraction of a mV. In contrast B has practically the full cartridge voltage at both inputs, and careful design is needed to prevent significant common-mode distortion.
Version C also has one earthed amplifier input and can have low voltage at the other input, and so also avoids common-mode distortion. The one possibly serious drawback is that neither side of the cartridge is earthed. For a pre-amp built into a turntable this may not matter too much, but it may be inconvenient for general purpose use, requiring the cables and connectors fitted to the turntable to have two screened conductors. That may be a good idea anyway, but I prefer to avoid the problem.
There are well known and effective techniques for reducing common-mode distortion, so my choice for this project is based on variation D. The addition of RIAA compensation will change the feedback element used to set the input impedance, and this will no longer be just a 470k resistor, a slightly more complex arrangement of resistors and capacitors is needed, and I am as yet not entirely certain how this will affect the noise reduction. Rather than calculate or simulate I hope to measure the actual improvement achieved compared to a standard type B circuit to determine whether it is worthwhile. In practice disc noise will almost certainly be well above circuit noise so this 'improvement' is optional.
Update Aug 2013. As suggested the disc surface noise was found to be by far the greatest noise source, and the circuit noise entirely unobtrusive, even with my low output cartridge, so it never seemed worth the effort to try to improve this further.
An article. 'Low-noise Audio Amplifiers' by H.P.Walker, in Wireless World, May 1972, pp 233-237, analysed circuit types A and B including the effect of RIAA equalisation. He found that ignoring amplifier noise the signal to noise ratio for a 50Hz to 20kHz bandwidth relative to 2mV at 1kHz was 58.5dB for circuit A and 72dB for circuit B. The difference was confirmed by measurements of practical examples. He used a 600mH cartridge for his calculations, which is why I also chose this value so that I can more easily compare my results to his.
Part 2. Rumble Filter.
Although a modification to the RIAA equalisation was introduced to reduce gain below 20Hz, with the intention of improving rumble rejection, this appears to be unpopular, partly because no corresponding boost is added to the discs to maintain a flat response, and partly because a simple -6dB per octave reduction is of limited effectiveness. A second or third order high-pass filter would be better, to give less attenuation at 20Hz, where there may still be signal components, and more attenuation below 10Hz where it may really be needed.
Much of the low frequency rumble we want to reject is, in effect, vertical modulation of the record groove, which will produce pickup cartridge output voltages of opposite polarity from left and right channels. If these signals were to be amplified and applied to the speakers unaltered then they would cancel fairly well acoustically, reducing audible annoyance. There still remains a problem because any high amplitude sub-sonic signal reaching the speakers may cause large cone excursions in the bass drivers, particularly in bass-reflex enclosures and other types where cone movement is limited primarily by the driver compliance. The large amplitude cone movements can cause intermodulation and Doppler effects which may be at an audible level, so rumble filters can certainly serve a useful purpose.
Many years ago I saw a circuit idea in one of the electronics magazines which suggested using the opposite polarity of the unwanted sub-sonic signals to cancel its effect, simply by adding the left and right signals at low frequencies to convert them to mono. Very low frequency components contain little directional information, which is why central sub-woofers are generally found to be acceptable, so the loss of channel separation at low frequencies may be far less noticeable than the effects of sub-sonic rumble. The circuit can of course be made switchable, but my own experience of this method was that it could be very effective, and I never noticed any significant problems. It was particularly effective when using headphones, where of course there is no acoustic cancellation of opposite polarity sub-bass. Here is a very simple example of this idea, three other versions are included in the references section at the foot of the page.
In a practical circuit the output op-amps will need large value resistors from the inputs to 0V to set dc levels. Assuming the output op-amps have high input impedance, if the signals are equal and of the same polarity (L = R) there will be no current through the 10k resistor and no attenuation of the signals. If L = -R then there will be current through the 10k and therefore a signal voltage drop across the 1u capacitors, and both signals will be attenuated. An easy way to make this circuit switchable is to have two equal resistors in series instead of the single 10k and switch the junction between the two resistors to earth so that there is then no interaction between the channels, and instead we have a simple first order high-pass filter. Alternatively a switch in series with the 10k can eliminate its effect.
Next the circuit gain is plotted as a function of frequency for both L = R and L = -R.
Equal signals are passed with no attenuation, but low frequency opposite polarity signals are attenuated as required. This is fine provided the low frequency signals are of equal amplitude in both channels, but suppose there is a wanted bass signal in only one channel. It can be argued that allowing a recording to have bass on only one channel is bad practice, but this is no guarantee that it never happens. Certainly simple recordings from a pair of directional microphones should never have this problem.
The next diagram shows what happens with an input on only one channel.
The channel with the input is shown in red, and there is now attenuation, with a response -3dB around 20Hz, eventually levelling off at -6dB at lower frequencies. The other channel, in green, now has bass output, and at very low frequencies both channels are at the same level. Adding the levels from the two channels, taking relative phase angles into account, we again find a flat response, so all is well, maybe.
Update: Since I wrote this section some alternative circuits have been analysed by Douglas Self and found to have problems. The only other one I simulated myself is the Oldfield circuit linked to in my references, Stereo Rumble Filter, and for similar attenuation of out of phase rumble at 100Hz the attenuation at 10Hz is about 30dB compared to only 10dB for my simple first order circuit. (I increased the capacitors in the Oldfield circuit to 68n to more closely match the frequency range of my first order filters). The response is flat for in-phase signals, and for an input on only one channel the total of the two channels is again flat. All is not perfect however, for input only on one channel the output from the other channel rises at low frequencies and rises to a peak about 3dB above the channel with the input, which has a compensating dip in the response in the same range.
I suggested on my 'Latest News' page that two or more of my first-order circuits could be used in series to give higher attenuation at 10Hz, which seemed obvious enough without needing a simulation, but having seen the problem with the higher order Oldfield circuit I decided to check what happens with a signal at only one input, and unfortunately it turns out to have the same sort of problem, the result shown next is for two first-order filters in series:
The signal originally entirely on the right channel now has a higher level on the left channel at frequencies up to 30Hz. This is not necessarily fatal, it may not have any serious audible effect, and signals below 30Hz on just one channel are highly unlikely.
Initial simulations suggest that adjusting resistors to reduce the channel reversal effect in the Oldfield circuit also reduces the 10Hz attenuation, so it may be little or no improvement on my in-series first-order filters. Another idea I checked was that if two of my circuits in series produce this reversal effect maybe more stages will reverse it back again, but it doesn't work that way, adding more stages increases the levels of peaks and dips and also moves them higher up the frequency scale, so using more than two stages is not recommended. The results for just a right channel input using 1,2,3 and 4 first-order filters in series are shown next:
Part 3. RIAA Equalisation (single network).
About fifteen years ago I checked the RIAA equalisation in 7 published phono pre-amps, and found that only 2 had anywhere near the correct component values. Hopefully this was just bad luck and was an unrepresentative sample, but I have seen at least one expensive pre-amp measured by Stereophile with surprising levels of frequency response error. The point of mentioning this is to suggest that just copying a published design is no guarantee of accuracy, and it is better to calculate components starting from the required RIAA time-constants. The calculation gets more difficult the more time-constants we try to include in a single network.
The example shown next uses an inverting 'virtual earth' configuration, and for this the network has 3 time-constants, which is not too difficult to calculate. (The non-inverting version with 4 time-constants has now been added, see the link at the end of this section). The noise penalty mentioned earlier for this circuit can be overcome by using a buffer stage at the input, which allows the inverting stage input resistor to be reduced, e.g. to under 1k, with consequently lower noise contribution.
Note that there is another variation on this equalisation network, with R1 connected in parallel with C1 instead of across both capacitors. This will lead to different component values. I have always used the version shown here, but if correctly calculated they should be interchangeable.
The gain of this circuit is proportional to the impedance of the feedback network, assuming the feedback loop gain is high enough to avoid significant errors from that source. I will eventually check the requirements for loop gain, but my first guess is that anything in excess of 40dB loop gain from 20Hz to 20kHz should be enough. Starting by calculating the network impedance we can then find relationships between the component values needed to give the time constants, T1 = 3180us, T2 = 318us, and T3 = 75us.
The calculation is rather long, so I have put it on a separate page, RIAA calculation.
The result is:
C1R1 = 2937
C1R2 = 236.79
( R in Ohms, C in uF)
For anyone who doesn't trust my calculation I refer them to an article 'A Design in Retrospect' by J Dinsdale, Wireless World, Nov 1969, page 506, which gives only the first part of the calculation, with the useful phrase 'It may be shown that:' followed by essentially the same figures I arrived at, but rounded to 3 significant figures. A simulation in Part 6 also shows that component values calculated from these equations give the correct response very accurately.There is no real need for such precise figures, but it does no harm to be excessively accurate up to the point of choosing component values.
One possible starting point is to work out the ratio of the capacitors,
C1 / C2 = 236.79 / 81.205 = 2.91595.
(This is how I checked the published designs mentioned earlier, the ratio of these two capacitors varied from 2.6 to 4.5, only two were close to 2.9. Note however that this ratio will be different for the alternative network mentioned earlier with a resistor in parallel with each capacitor, I have not calculated this myself, but it appears the capacitor ratio is then typically 3.5.)
Suppose we then choose C1 = 4n7, then C2 = 1.61182nF
We can now calculate R2 = 50k3809
Then R1 = 624k894
The gain at 1kHz may be a more useful starting point, and a good approximation is R1 / 10 Ri, so for the previous example if Ri = 1k then the gain will be about 62. A 5mV peak input at 1kHz will then produce a peak output of 310mV, and if we want at least 20dB overload margin at 1kHz the supply voltage must be chosen to enable in excess of 3V output, which is no problem with typical op-amp supplies of + /- 12V to 18V. Overload margin is covered in more detail in Part 7.
Note that for non-inverting circuit type B above there is some interaction with the value of the resistor from op-amp inverting input to earth, shown as 100R. The calculation includes 4 time-constants and gets more difficult, and initially I wanted to avoid it, but later decided to add an example anyway, and this is on a separate page as Op-amp based Phono Pre-amp, together with a rather untidy hand-written calculation which involved solving a quadratic equation, but nothing worse. With a single op-amp the finite gain has some inevitable effect, and some adjustment of component values can compensate, but there is no point calculating this, just a little trial and error using a simulation is quicker, and a result within +/- 0.01dB was found to be possible. In practice 1% component tolerances can be a greater source of error.
Actually I am not certain whether there is an exact solution with zero error with op-amp gain taken into account, my guess is not, but I don't know any good reason why I should want to find out.
These calculations can easily be avoided by simply splitting the equalisation into two stages so that each stage involves no more than two time-constants and the calculation is then easier, as in the next example.
Part 4. RIAA Equalisation (two stage).
The circuit is now split into two parts. For now a simple 47k resistor is used to define the input resistance.
To reduce the effect of the second stage noise we need sufficient gain from the input stage so that input stage noise dominates. Resistor R1 contributes to input noise and so needs to be small, but not so small that the current required via the feedback network is too high for the first stage op-amp to provide without difficulty, and for this I have chosen 100R. Any high signal voltages will be at fairly high frequencies where the RIAA pre-emphasis has greatest effect, so it may be a good idea to apply the high frequency attenuation part of the RIAA equalisation in the input section, and leave the bass boost part for the second stage. We therefore require R2 and C1 to provide the 75us time-constant.
I have seen suggestions that the 75us high frequency attenuation should be in the second stage to reduce noise from both stages, but this assumes that noise in later stages is a problem, which it need not be provided the input stage has enough gain and circuit impedances are chosen with care. A more serious concern may be the high frequency pre-emphasis of the cartridge output signal. Feedback amplifiers are sometimes criticised on the grounds that square wave or step functions can have a high percentage overshoot in the input stages unless the open-loop bandwidth is high, but for a RIAA pre-emphasised square or step there will already be a high overshoot before the signal even reaches the input stage, and this will continue to exist right through the amplifier stages up to the point where the high frequency equalisation takes effect, which suggests that this needs to be applied as early in the circuit as possible. One solution is to use feedback to the emitter or source of the first amplifying device, as is done here, to apply the 75us equalisation, and as is done in the most common single amplifier designs with a single overall feedback network. Almost any other approach, including a flat gain input stage followed by passive equalisation, will allow the step function overshoots to potentially cause transient intermodulation effects in any circuitry prior to the 75us equalisation. Fortunately normal recordings are unlikely to include such difficult signals, and all that really matters is to ensure that each stage remains adequately linear with the maximum signal amplitudes it will be required to handle in normal use.
The non-inverting configuration has a gain falling to unity at some high frequency rather than continuing indefinitely to fall at -6dB per octave as required for the official RIAA response. It is sometimes suggested that some record cutting lathes have an extra time-constant at 3.18us to limit high frequency pre-emphasis beyond 50kHz, and that consequently an equalisation response should compensate for this effect. Personally I prefer not to boost output beyond 50kHz, but having the unity gain problem anyway with this circuit it seems reasonable to arrange for the resulting additional time-constant to be at 3.18us. If this effect is not wanted then adding another capacitor in the second stage can replace the missing high frequency attenuation.
Again I have put the calculation on a separate page, RIAA calculation.
The result is:
C1R1 = 3.321
C1R2 = 75
C4R5 = 2862
(R in Ohms, C in uF)
Part 5. Inverse RIAA Simulation
Checking the frequency response using simulations is useful for revealing errors, and also for establishing component tolerances and levels of amplifier loop gain needed for keeping errors within some limit. Equalisation accuracy between 20Hz and 20kHz is typically specified as something between +/- 1.0 dB and +/- 0.1dB. The effects of component tolerances can be calculated, but simulations are a lot quicker and easier, so this approach will be adopted. The question of how accurate we need to be is of course debatable, and it may be that even 0.1dB variations can be detected by some listeners in direct comparison, but given typical speaker-room variations and recording variations it seems unlikely that achieving such accuracy needs to be a high priority, and few listeners will be switching between pre-amps to listen for differences. My approach is to aim for the best accuracy achievable using readily available and affordable component tolerances, i.e. +/-1% resistors and capacitors. 1% polystyrene and polypropylene capacitors are available from Farnell (UK) at reasonable prices, e.g. 4n7 Vishay polyprop at 44p each. Even 0.1% tolerance resistors are not really expensive compared to the rest of the components, so may be a reasonable choice.
The easiest way to check for errors is to start with an accurate RIAA pre-emphasis simulation (sometimes called an 'inverse RIAA' network). Then after passing through our pre-amp simulation we should get a flat frequency response, and any deviation from a flat line will be easily seen and specified.
I found an example from an old magazine article, and checked the calculation, but found an error. After correcting this I used it for a Spice simulation. Just to be certain I worked out another version using feedback loops where each time-constant is separated and each has just single R and C values, which makes errors virtually impossible. With 1k for all resistors the capacitor values in nF are just the time constant values in us, i.e 3180n, 318n, 75n and 3.18n. Comparing the result with the corrected magazine version they matched perfectly, so I am sure they are both now very accurate. I also checked the phase responses and these also matched perfectly. The point of this is that the magazine version used passive networks with just a unity gain buffer stage, but my own version used two feedback loops with high loop gain. Correctly done there is no significant difference between gain or phase of passive and active equalisation.
My active version simulation shown next uses amplifier stages with gain 1,000,000 for good accuracy, but it need not be so high. At 10,000 there was still no significant difference compared to the passive version, but at 1,000 small differences could be seen at high frequencies, reaching about 0.5dB at 100kHz. This may not be ideal as a practical circuit, but for Spice simulations we can use extremely high loop gain without worrying about stability.
(Although I have made E1 and E2 have negative gain the result appears to be exactly the same with positive gain, which may look wrong, but the difference between positive and negative feedback has little to do with the phase shift or polarity of the feedback, unless the loop gain is low, which it never is in this simulation. In real circuits of course the gain falls at high frequencies so the polarity becomes important.)
Inverse RIAA Vin 1 0 ac 0.1 R1 1 2 1k R2 2 3 1k R3 3 4 1k R4 4 5 1k C1 1 2 3180n C2 2 3 318n C3 3 4 75n C4 4 5 3.18n E1 3 0 2 0 -1000000 E2 5 0 4 0 -1000000
Part 6. Pre-amp Simulation
Single Stage Design
The first circuit checked for errors is the single stage inverting circuit from Part 3 using the component values worked out there. The 3.18us time-constant was not included there so C4 from the inverse RIAA is omitted. The result shows the +/- 1dB levels, and it can be seen that there is no significant deviation from flat overall gain. In a real circuit with lower loop gain and +/- 1% tolerance components there will be greater error, but this test is just to check that the derived component values are correct, I have no plan to use this circuit in a practical version. The phase was also checked, and again looks perfectly flat.SINGLE STAGE RIAA Vin 1 0 ac 0.1 R1 1 2 1k R2 2 3 1k R3 3 4 1k R4 4 5 1k R5 5 6 47k R6 6 8 624.8936k R7 7 8 50.38085k C1 1 2 3180n C2 2 3 318n C3 3 4 75n C5 6 7 4.7n C6 7 8 1.6118227n E1 3 0 2 0 -1000000 E2 5 0 4 0 -1000000 E3 8 0 6 0 -1000000
Two Stage Design
The final circuit checked is the two stage circuit from Part 4. Now the 3.18us time-constant is reinstated in the inverse RIAA. The gain and phase plots were again perfectly flat, confirming that the earlier analysis was correct. This will be used in a practical design eventually. The gain if the circuit can be changed by adjustment of R7, shown as 560R. For low output cartridges this could be reduced to 100R without danger of overloading the first amplifier, giving both higher gain and lower noise contribution. For high output cartridges 1k or more could be used to maintain a good overload margin in the output stage, and then the noise will be less important.
Two Stage Pre-amp Vin 1 0 ac 0.1 R1 1 2 1k R2 2 3 1k R3 3 4 1k R4 4 5 1k R5 6 0 100.6303 R6 6 7 2272.73 R7 7 8 560 R8 8 10 42088.235 R9 9 10 4676.47 C1 1 2 3180n C2 2 3 318n C3 3 4 75n C4 4 5 3.18n C5 6 7 33n C6 8 9 68n E1 3 0 2 0 -1000000 E2 5 0 4 0 -1000000 E3 7 0 5 6 1000000 E4 10 0 8 0 -1000000
33n and 68n are available from Farnell with 1% tolerance. Resistors with 0.1% to 1% tolerance are also available, but parallel combinations are needed to get close to the calculated values:
R5 = 110R // 1k2
R6 = 2k4 // 43k
R8 = 47k // 402k
R9 = 4k7 // 1M
To eliminate the 3.18us time constant if this is not wanted an additional capacitor is needed. To minimise interaction with the existing network it should be connected in parallel with the 42k088, R8. The whole network should really be recalculated, but I first tried adjusting the additional capacitor in the Spice simulation, arriving at a value of 760pF. The result is then flat apart from less than 0.1dB boost below 60Hz. The standard value 750pF is near enough, there is then also a slight boost, under 0.1dB, above 25kHz. The output op-amp now needs to be stable at unity closed-loop gain, and if not it may be necessary to add a capacitor, e.g. 3n3, between the op-amp inputs.
In the above Spice simulation substitute C4 8 10 750p for the C4 4 5 3.18n line.
Recalculating the network and adjusting the resistor values gave a perfectly flat result, but keeping the original values makes little difference and makes adding or removing the 3.18us time-constant a simple option.
If 0.1% tolerance resistors are available and affordable for 110R, 2k4, 47k, 4k7, then the larger parallel values, 1k2, 43k, 402k, 1M can be 1% because their errors have less effect. Such accuracy is not really necessary, using all 1% values should be more than good enough in practice. For example a 1% error in R5 gives a small error, probably no more than 0.1dB at 100kHz.
Part 7. Overload Margins.
A graph of peak recorded velocities as a function of frequency was published by Shure some years ago. There are occasional claims to have measured higher levels, but these may be primarily the result of scratches rather than recorded music signals, or using moving-coil pickups with additional pre-amps or step-up transformers. (Moving coil pickups can have frequency response extending beyond 50kHz and with big resonances, which should not be a problem with normal music signals but can make sensitivity to scratches a problem. I tried to find an example, but reviews don't now seem to measure frequency response, maybe for good reasons.) A maximum of 75cm/sec around 4kHz suggests that a typical cartridge with output 1mV/cm/sec will rarely exceed 75mV peak output. An overload margin far above this may not therefore be essential, but it does no harm to add a wide safety margin. Some cartridges have output in excess of 2mV/cm/sec, so anyone wanting to use such a high output type needs an extra margin.
There are three important ways in which the two stage equalised pre-amp can be overloaded, the first being the current required through the feedback network in the input stage. This is easily analysed because the current required is given simply by the peak cartridge output voltage divided by the 100R resistance in the feedback network. The peak voltage appears across this 100R, and the resulting current is provided via the feedback resistor and capacitor. With 75mV across 100R the current is 0.75mA, and so this is the maximum current level needed from the output of the first stage. (Plus the current needed to drive the following stage). We could easily use an amplifier capable of driving 7.5mA into the feedback network, then inputs up to 750mV could be handled, giving a 20dB safety margin.
The second place to look for problems is the available voltage swing at the first stage output. With the components used in the two stage design in Part 6 there is a voltage gain of about 24 at low frequencies, falling by 3dB around 2kHz and falling to unity beyond 50kHz. Applying the gain curve to the peak recorded levels the peak output should be around 1V at 2kHz, and again we can easily exceed this by a 20dB safety margin using a stage capable of 10V peak output.
We do eventually encounter problems when we reach the second stage. If we want to maintain the same margin all the way to the output then with a 10v input at 2kHz the output stage could only have a little over unity gain at that frequency, and also at 1kHz. The entire amplifier then has a gain of about 30 at 1kHz, and the nominal 5mV input will produce only 150mV output. With a power amplifier gain of 20 that will give 3V peak at the speakers, which for average sensitivity speakers is still fairly loud, and may be enough for some listeners. Heavily modulated recordings producing 75mV will still produce 45V peak at the speakers, well in excess of 100W, so maybe this is not so bad, and anyway most 100W amplifiers will have gain more than 20.
The problem is that many recordings are well below this peak level, so the required gain may be higher.
One solution to this problem is to add another volume control between the stages, or switched gain levels. Another solution for the DIY enthusiasts is simply to make the input resistor of the output stage easily accessable, soldered to terminal posts, so that the value can be chosen by experiment to match the listening level requirements. Starting with 4k7 we can maintain the previous safety margin, or if the resulting gain is insufficient even at maximum volume setting a lower value can be used, down to 470R at which a 75mV cartridge input will produce 10V output, and a typical op-amp will be only a few dB away from its clipping level.
The big problem with this is that changing the output stage gain means we also need to change all the components in the active input impedance circuit. There is, however, an alternative way to do this, which adds a little complexity but should have little effect on performance. This is shown in the next part.
An alternative approach to achieve high overload margins, used in some commercial designs, is to make the output stage capable of producing very high voltages, maybe 40V or more. Provided there are no more active stages prior to the volume control this approach can have enough gain for very low output cartridges while still maintaining sufficient overload margin to allow the use of very high output types. Personally I would choose switched gain levels as an easier and possibly more effective option, avoiding the problem of needing very low volume control settings for high outputs. Typical dual log potentiometers can have very poor channel matching at low settings.
Possibly the best way to specify overload margin is as a graph of peak input voltage for 1% distortion levels at the output as a function of frequency. For switched gain types there would then be several curves for the different gain settings. Anyway, that is how I plan to specify my next design. Including curves similar to the Shure peak recorded levels for extreme high output cartridges will then show how much safety margin we can claim to have.
Part 8. Active Input Impedance With Variable Gain
Circuit D in Part 1 used a 470k resistor with an amplified and inverted input signal applied to make the effective input impedance the required 47k. With RIAA equalisation applied something more complex is needed instead of the single 470k, and in theory a network with three resistors and two capacitors is needed. As mentioned in Part 7 this then becomes a great inconvenience if we want to adjust the output amplifier gain for cartridges of different sensitivity or for different output level requirements. There is however a simple way to avoid this difficulty, which is to use a third amplifier as an inverting stage for the active impedance. In my planned design the input stage will be a jfet input discrete amplifier, and the other two stages can be a dual op-amp. The added complexity is then minimal. An additional advantage is that the 3.18us time-constant can also be added or removed without disturbing the input impedance. An example is shown next:
Two of the resistor values are slightly different to those shown in Part 6 for the two stage version. The values were recalculated for the above diagram to be correct with the optional 760p included. It can be seen that the differences are very small, and using the closest parallel pair of standard value resistors plus even 0.1% tolerance could add greater errors, so this is not something to be concerned about. I plan to include the 760p in my future design, which is why I checked the calculation including this.
'Synthesis of Driving-Point Impedances with Active RC Networks' By I.W.Sandberg, The Bell System Technical Journal, July 1960.
Thanks to Dimitri Danyuk for sending me this reference, I had searched with no luck for an early mention of this active impedance method. My suggestion that it was known over 35 years ago can be revised to at least 51 years, maybe much more.
The Phonobox Story.
A commercial implementation of the active input impedance technique is the Pro-ject Phono box. One of these was tested and then taken apart and 'reverse engineered' by Dimitri Danyuk. This revealed a few problems, with incorrect RIAA equalisation and other circuit errors.
Phono Preamp Using active RIAA Equalisation by AndyC presents an investigation of the effects of component tolerance, DC amplifier gain, and gain-bandwidth product.
On Reference RIAA Networks. This includes a passive compensation network which I mentioned, which I recall was published in a magazine some years ago. I mentioned that I used this to check my simulated inverse RIAA circuit, and initially found an error, which is in Fig.4 where R4 is given the value 353.3. This is the correct value for R2, actually 353.3333 but R4 should be 332.0802. It makes very little difference, but with the correct value there is accurate agreement with my active version.
On RIAA Equalisation Networks by Stanley Lipshitz (1979) is a detailed analysis of RIAA networks. Impressive, but not really essential reading, the networks with 3 or 4 time-constants can be calculated in a page or two, and the accuracy of the result confirmed by simulation, as I did. Once the correct results are known there is no need to recalculate everything for a new design.
'A Design in Retrospect' by J Dinsdale, Wireless World, Nov 1969, page 506 is an earlier example of a correct calculation for a 3 time-constant network.
Stereo Rumble Filter. A Circuit idea by M. L. Oldfield, Wireless World, Oct. 1975, p 474.
I found this example of the rumble filter idea I described in Part 2. This is not exactly the same circuit, it uses a second-order filter, but essentially the same idea. This looks very familiar, and may be where I originally saw this technique. The version I showed may just be my own 're-invented' version based on an incorrect memory, so I am no longer certain I ever saw that exact circuit in print.
Differential rumble filter by J.P.Macaulay (Wireless World Sept 1979 page 75), published a few years later looks even less like my 'remembered' version.
Another Second Order Rumble Filter. This one is from Dimitri Danyuk, and uses the same rumble filter idea to cancel vertical rumble, but in addition there is a high-pass filter for horizontal rumble reduction. This filter can be adjusted with a single variable resistor, R2, to give the responses shown in the second diagram. I assumed most rumble to be vertical modulation, but bearing noise and also off-centre pressings can contribute horizontal low frequency effects also.
The earliest version of the rumble filter I found so far is a passive circuit published in Wireless World, Feb 1969, page 81, in a page of reader's 'Circuit Ideas', this idea being contributed by David Ralph.