Classic Audio by Michael Fidler

Practical Line Input Circuits


This article is the first part of a series on practical line level pre-amplifier design, following from the beginning of the signal path at the line input, to the end of it at the line output. Tedious analysis of various circuit topologies that fall short of the mark, or simply don’t deliver any practical benefit will be avoided and only brief reference will be made to them if any. The circuitry described, with the exception of some description of topologies used in the past will feature the use of commonly available op-amps such as the NE5532 and TL072, in tandem with split supplies of ±15V unless stated otherwise. This approach allows the great deal of flexibility and simplicity needed to realise the optimised circuits to be revealed throughout the course of articles. In addition to this, some additional auxiliary circuitry will be shown to demonstrate the correct application of some of the topologies detailed in this article, to give the reader a flavour of how to make the most of their relative merits in regards to the bigger picture. As the reader will discover through the rather high word count of this article and the plethora of complementary circuits of this nature, it is a big picture indeed to say the least.

Analogue line inputs are quite essential features that appear on almost all audio equipment, from power amplifiers to digital audio recorders, that needs to accept some sort of external analogue audio signal in the region of 100mV RMS to over 2V RMS in the case of many digital audio sources. Although nothing could appear easier than simply connecting an external signal to a circuit node of choice, things are invariably a little more complicated in the inconveniently imperfect reality that is the world of analogue audio design. As a result of this, this article is considerably longer than a description of such a direct connection.

Line Level Connectors

Figure 1. Common RCA, TS, and TRS connectors

The most commonly seen line inputs in consumer audio take the form of RCA connectors, as can be seen in Figure 1. These unbalanced connectors are quite inexpensive, and were originally designed by the Radio Corporation of America in the 1940s as a low cost and easily mass produced means of connecting external phonographs with crystal cartridges up to radio sets of the day. They were designed to replace banana plugs which, being unshielded, were quite unsuitable for the high source impedance of the newer crystal cartridges, resulting in excessive electrostatic hum injection. Originally not a significant problem for the earlier heavy magnetic pick-ups with had output levels hundreds of times greater than the ones we see today, featuring many thousands of turns windings in the pick-up coils. Due to their low production cost, they became quite the favourite with manufacturers of consumer audio equipment, and have been the de-facto unbalanced connector ever since.

Traditionally nickel plated, modern trends that have very little to do with reason have endowed these connectors with a very thin layer of gold plating to keep up appearances. This is quite unnecessary, and in order to prevent the brass connector material from 'spotting' through the gold layer, a layer of nickel must be plated between the two. Nickel has excellent corrosion resistance, and unless the line connection is going to be situated outdoors, close to the sea, the gold plating performs no function except that of an anodyne to the ever more tortured mind of the consumer who has been bombarded with highly dubious claims of the 'musicality' of connector plating material from the usual suspects. In most cases the gold plating wears off after a handful of mating cycles to expose the lustrous nickel underneath; slightly visible on the black banded RCA connector in Figure 1.

Having reluctantly dismounted from a favourite hobby-horse, it is conducive to effect a cursory glance in the direction of professional audio connectors, which are usually balanced. Cable runs will often greatly exceed the meter or so which is acceptable for an unbalanced line before various forms of interference make such a connection infeasible. The TRS connector is the most common by a small margin over the XLR connector, mainly due to its ease of connection, especially as a patch cable. RCA connectors, which are not anywhere near as durable, are often nowhere to be seen and if an unbalanced connection is required, then a TS connector like the one in Figure 1 is used; note the nickel plating.

Line level cable should be shielded, preferably with a foil type shield to prevent electrostatic pick-up of any stray electrical fields that may conspire to impose themselves upon the audio signal on the centre conductor, or conductors in the case of a balanced cable. The ground conductor should also be of sufficient gauge to have a low enough resistance to shunt these capacitively coupled sources of interference to ground without developing a large enough voltage to cause similar problems. The centre conductor does not need to be of a particularly heavy gauge due to the relatively high impedance of the signal it carries, although it should not be so thin as to be fragile. Aside from this, and despite the claims of subjectivists, there is little else required to fulfil the objective criteria for a high quality audio cable. It is always a good idea to keep line level cabling as short as possible; especially true for unbalanced connections.

The need for line inputs

Before embarking on the creditable mission of designing a high quality line input, it is a good idea to understand exactly why line input circuitry is necessary. If the line were to be connected directly to the sort of circuitry found in a noise-optimised high quality pre-amplifier, compressor, or active crossover, then what's the worst that could happen? For some very simple circuitry, such as passive attenuators that masquerade as pre-amplifiers in some DIY systems, not a great deal. For many power amplifiers with simple unbalanced inputs that go straight to the input stage of the power amplifier, the same answer is given, although distortion performance may begin to be severely compromised if the line impedance exceeds 1kΩ as a result of the non-linear input currents drawn by conventional power amplifier input stages.

In order to realise excellent noise performance with active circuitry, special care must be taken to keep circuit impedances as low as possible so as to minimise the effects of thermal noise which increases in proportion to the square root of circuit impedance. In almost all cases the depths to which this low impedance can sink are determined by the load driving ability of the active devices in the circuitry, which is one of the main reasons why op-amps such as the NE5532 are so greatly, and worthily, favoured in audio design. Typically these impedances will be optimised to be on the order of 750Ω at a minimum with the ubiquitous sort of NE5532 based circuitry to be described below. Many line outputs will be unable to drive this sort of impedance without significant insertion loss, current limiting, increased distortion or all of the above. It is therefore very important that there is some line input circuitry to present a reasonable impedance to the line that can be then buffered onto much lower internal impedances to allow for proper noise optimisation.

There is also the question of affording some degree of protection and isolation from whatever might be lurking on the other end of the line. Large DC offsets, RF interference, and potentially damaging transients are only an RCA connector away from potentially doing a great deal of damage to what would be some very serviceable analogue circuitry.

For balanced audio, the necessity is immediately apparent. With the exception of 'variable mu' valve compressors, and for a very brief point in the signal path of differential ADCs and DACs, the signal path within the unit will almost certainly be single ended. It is sensibly assumed that in any competently designed monolithic piece of audio equipment there will be no common mode noise or interference in want of rejection. There is almost no benefit to be had with differential internal circuitry, for the price of almost doubling the component count and power dissipation. It is prerequisite to implement some input circuitry that only passes a differential signal while affording some of the preventative measures mentioned in the previous paragraph - quite a precise definition of a balanced line input.

Line level in the wild

Before setting any kind of specification to design a prospective line input to, it is important to first consider what sort of signal levels and impedances that said line input will be made the subject of. Line level is always defined as a voltage and this can be confirmed with a quick look around the internet. Searching for line level will immediately return information to the effect that consumer line level is -10dBV or about 300mV RMS, and professional line level is +4dBu or 1.228V RMS. The former figure is rather misleading as it seems to imply that there is some standard in effect that in reality is not. Consumer line level can be significantly lower and higher than the nominal -10dBV by almost an order of magnitude in either direction. The world of consumer audio is very much a non-standard environment as far as line level signals go, mainly due to a lack of consumer objectivity and cost considerations. This is especially true with older equipment featuring simple discrete circuitry where distortion would increase as as square of signal level; simply reducing the level by a factor of two could improve the distortion performance by a factor of four and could cover up various other flaws in a cost cutting design such as inferior drive ability inherent to lower quality op-amps and discrete circuitry built with a minimum transistor count.

As far as line impedance in consumer audio is concerned, there is likewise a very wide variation with higher line impedances being more often found in older audio equipment. In most cases the line impedance does not exceed 1kΩ, but exceptions exist with some low cost discrete designs such as tape recorders and tuners from the 1970s which can have output impedances of greater than 5kΩ. Cheap valve equipment that uses capacitively coupled line outputs can exhibit a line impedance of greater than 30kΩ in some instances, usually when the output is taken off the anode of a high gain double triode such as the 12AX7. The misguidedly beloved and often DIY'd RCA phonostage based around the same double triode exhibits an output impedance of 37kΩ, which is the worst that the author has come across in the real world.

Professional level is not referenced directly to voltage in the same fashion as consumer line level is. This harks all the way back to the very early days of electronic audio and is based upon the work done to extend the telephone system over long distances in the early 1920s, where long line lengths required careful impedance matching. As a standard, these lines would present a 600Ω impedance that would be matched at both ends to permit a useable frequency response to be obtained over distance. When a truly revolutionary method of electrical recording was developed in 1925 by telephone engineers working for Western Electric, the use of such lines naturally carried over into the very new field of professional electronic sound recording, especially as live recordings were often made through the very same telephone lines from the concert hall to the studio, or a recording van situated outside. The reference level for the system was not expressed in terms of voltage, but power with dBm referring to a power of 1mW. For some reason, possibly due to optimisation of signal-to-noise ratio against overload and non-linearity, the reference level was set at +4dBm early on in the 1930s or possibly even sooner with, or possibly before, the standardisation of the VU meter.

To transfer 1mW of power into a 600Ω load of the era required a voltage level of 0.775V RMS and this implication gradually led to this voltage being assumed by default for audio use. In professional audio, the 600Ω loading has gradually become less and less common with the introduction of electronic as opposed to transformer balancing, with the load at the other end typically being on the order of 10kΩ or greater, so now the term dBu, decibels unloaded, is used to refer to the 0.775V level previously needed to induce a power of 1mW into the other end of the line regardless as to whether the power was delivered in full due to a higher load impedance. From this point onwards levels will be referred to in dBV as a priority with dBu taking a back seat, as it is the opinion of the author that unless professional audio is being discussed as-is then this rather inelegant and archaic unit should be avoided, although it is very easy indeed to flit between the two by adding 2.21 to dBV to convert to dBu.

The professional reference level of +4dBu equates to a voltage, of 1.228V RMS with no load placed upon the line. This almost always comes from a drive impedance of less than 600Ω, which implies impedance bridging is to be used with a much higher impedance on the end of the line than the driving impedance. In reality the received voltage will always be slightly less as the load at the end of the line will pull the level down slightly but this is considered to be negligible. The move away from impedance matching to bridging is not an issue here as although line connections in professional audio can be very long indeed, reaching up to a hundred meters of cable quite easily in some professional sound reinforcement systems, this truly pales in comparison to the 1,200km wavelength of 20kHz in a typical transmission line. Most of the time a line impedance of not more than a couple of hundred ohms can be assumed with modern equipment.

Professional line level is almost, if not always 1.228V RMS as standard. Meanwhile, consumer line level can vary considerably, as levels are most often not monitored and differences in levels between sources are usually rectified with the volume control. This variation is usually as a result of practicality of making do with what is available and keeping costs to a minimum. To get a flavour for the very wide degree of variation, and where it occurs the table below lists some line levels commonly seen out in the wild in plain old volts RMS, volts peak, dBV, and finally dBu.

Line source Volts RMS Volts peak dBV RMS dBu
Technics SA-200 receiver 150mV 212mV -16.48dBV -14.27dBu
Phonostage 5mV at 32dB 200mV 283mV -14dBV -11.79dBu
Nominal consumer line level 316mV 447mV -10dBV -7.79dBu
Phonostage 4mV at 40dB 400mV 566mV -7.96dBV -5.75dBu
Technics ST-7300 tuner 500mV 707mV -6.02dBV -3.81dBu
Professional line - 0VU 1.23V 1.74V 1.78dBV 4dBu
ESI U24XL USB audio interface 1.72V 2.43V 4.68dBV 6.90dBu
Technics SL-PG570A CD player 2V 2.83V 6.02dBV 8.23dBu
Chord Qutest DAC 3V 4.24V 9.54dBV 11.71dBu

Real world line output levels compared

Starting from the top of the list is a rather low maximum output of the Technics SA-200 receiver, the lowest line level output in the authors possession. This was fairly typical for equipment of this era and as already alluded to, the lower level helped to improve the distortion spec while also reducing the drive current requirements; something that reduced the cost of the power supply considerably in some cases. Most amplifiers from this era were very sensitive indeed and with the noise bottleneck of the analogue formats available at the time, the lower signal-to-noise ratio was tolerated quite amicably by even the more concious consumers. The greater sensitivity was often effected to proscribe similar non-linear gremlins in the amplifier's cheaper resistively loaded input stage from generating excessive distortion, so sensitivity to only 100mV or so was already indelibly embossed onto the menu.

Many phono pre-amplifiers of this era had comparatively low gains, and it was not uncommon to find gains of 30dB or even less. Noise performance was less of an issue here also, with many phono pre-amplifiers only managing an SNR of 70dB against a 5mV source. As amplifiers were already quite sensitive, it did not make sense to have a particularly high gain which would bring with it a decrease in linearity due to the lower feedback factor and increased level, along with the very real possibility of overload on a single supply rail as low as 15V in many cases. It is still common to find phono pre-amplifiers with low gain and a fair few out there today exhibit a gain of 40 times, or 32dB, hence the inclusion into the table. The justification for this is usually to boast a very high overload margin for the spec sheet that simply isn't possible at higher gains with op-amp circuitry due to supply rail limitation.

On the other side of the nominal consumer line level, sits a more modern phonostage, the Classic Audio Spartan 10, with a gain of 40dB. The cartridge level is now 4mV, which has been brought down from the nominal 5mV simply to demonstrate the comparatively small variations in output that are seen in moving magnet cartridges. With an output level of 400mV mated to such a cartridge, the level is high enough not to have a pronounced discontinuity in level when switching between the phono pre-amplifer and modern digital sources, assuming that the digital source is not mastered with an objectionable amount of dynamic range compression and corresponding make-up gain - unfortunately the case for far too much source material out there today. Most modern phonostages tend towards higher gain for just this reasons, along with the reduced chance of overload with sensible split-supply rail op-amp circuitry, in contrast to the rather limited single supply discrete circuitry that preceded it.

Just below the 40dB phonostage, and above it in relative level is the Technics ST-7300 tuner, selected for this short-list to demonstrate the differences in output level from units of the same era. Manufactured in the late 1970s, like the SA-200, the ST-7300 was a more upmarket product that had considerably more invested in it in the way of electronics. This manifests itself in an increased output level that would have propelled it more into the realm of its professional contemporaries. A better output stage is used that has far better driving ability at the cost of a few more transistors, which at the time would have been almost an order of magnitude more expensive than they are today. The higher output level would have resulted in an overall lower signal-to-noise ratio when used in conjunction with the more sophisticated electronics sold at its price range and generally increased its versatility of use.

Up next, and with an output level well in excess of the professional reference level is a far more modern piece of equipment in the form of a common-or-garden USB audio interface. The one listed is unremarkable and shows typical output level characteristics of such devices. The RMS voltage output seems like a rather odd and un-rounded figure at a first inspection, but it is quickly put into context by the peak level of 2.43V, which is half of the 5V USB supply that the interface uses to generate its audio output. This level is very common indeed with USB powered equipment, or equipment running off of a single 5V rail. For most applications, this presents a very significant headroom disadvantage, but in the context of digital audio with its well defined limits, an SNR of 100dB is quite easily realised.

Following the USB audio interface, another typical digital source is brought in - this time a CD player. Here the output level is a little stronger and is set at a total of 2V RMS, which is considerably higher than professional level. Most CD players specify this output level as it is strong enough to rise well above the sort of noise floors found in all but the very worst analogue pre-amplifiers while not being so high as to make the possibility of overload very high. Most DACs contained in such devices naturally give this output level, so all that is required is a little buffering to drive the line. It was often the case in the 1980s, continuing through even into the 1990s for pre-amplifiers to contain an attenuator on the CD line input to either make the discrepancy in level with weaker sources already described a little more tolerable when switching between them, or to reduce the likelihood of overload, particularly as the tone controls were manipulated to provide boost. Very often, the attenuator brought the relative signal to noise ratio below that of the CD itself, resulting in a rather unhappy compromise that went unchallenged for quite some time. Fortunately, with properly designed circuitry equipped with the sort of ample headroom that can be realised with the ubiquitous NE5532 op-amps, there is no need whatsoever to engage in this deleterious practice any longer. In the second age of HiFi, it seems that analogue sources are stepping their game up in terms of output level to match that of digital audio, as opposed to simply attenuating and amplifying; never a good idea if the best signal-to-noise ratio is a design objective.

As a last example at the bottom of the table is the Chord Qutest DAC, a unit that features a considerably higher 3V RMS. The author does not have this unit in his possession, but is showcasing its specification to show how some digital audio sources can have an output level well over twice that of professional reference level. It is a well known phenomenon that when compared side-by-side, of two otherwise identical sources with slightly different levels, the higher level source will be perceived to sound better, even if the level change is imperceptible as a level change to the listener. With this in mind, many manufacturers of CD players and DACs have contrived to build their products with an output level slightly higher than the once-standard 2V RMS so as to give prospective buyers the impression of better sound in the sort of side-by-side subjective comparison that many HiFi retailers use to demonstrate their products. Meridian Audio were among the first to do this quite early on with CD player output levels of 2.4V RMS, about 1.6dB above the standard 2V RMS; just enough to give the impression of better sound without the obvious perception of a level change.

There are some line sources that exceed even 3V RMS as a nominal level, for instance the output levels of some valve phonostages that have hundreds of volts of not-very-linear headroom at their disposal. Likewise, there are also some DACs that are quite capable of heaping scorching hot levels onto whatever sort of line input might be waiting for a scalding downstream of the signal chain. These DACs often have some form of a variable attenuator included either on the back panel, or the front where it might be intended to function as a sort of very simple volume control for the sort of system that aficionados of the very weird-and-wonderful esoteric high end might apply to the inputs of valve power amplifiers built with 1930s directly heated triodes and magical capacitors made of platinum and silver. It is the opinion of the author, that these sources can be discounted from reasonable line input design as including their very unfriendly levels would mean making a compromise elsewhere to the detriment of the vastly more common sensible line level voltages.

Line input requirements

Having examined the basic necessities of an effective line input in conjunction with the sort of levels that we are likely to encounter, it is prudent at this point bring these musings and observations to bear to the purpose of defining exactly what we want out of the line input circuitry currently strewn across the drawing board. The essential requirements are listed below:

The above is a list of what the author considers to be the essential characteristics of a high quality line input, but there are a few more desirable attributes that are worth pursuing to cater for some of the most undesirable scenarios that might be dredged up in the pursuit of using the line input with either poorly designed, poorly made, or very low cost equipment that might combine either of these two factors.

Basic line input

Figure 2. Simple high impedance line input buffer

Figure 2 exhibits a typical line input that fulfils the essential criteria of the first set of bullet points, and the first bullet point of the latter list. It is a very simple, yet effective, unbalanced line input that consists solely of an input RC network and an NE5532 acting as a buffer to drive whatever low impedance might be antecedent. R1 drains away any standing DC that may reasonably be expect to sit on the line; C1 and R2 perform high impedance DC decoupling between the line and the op-amp input bias currents; R3 and C2 form an RF filter, while stopper resistor R4 gives a little extra RF immunity; U1 then buffers the recovered line signal.

As soon as the line comes in, the shell of the RCA connector is connected to chassis ground, which will minimise the sinister machinations of possible ground loop currents inside and outside of the unit. The effective parallel resistance of resistors R1 and R2 then determines the input impedance of the line input, in the circuit above just over 220kΩ; a veritably feather light loading that will be tolerated by almost all audio gear in existence. Although it seems tempting to increase the values further, a compromise is inevitably at play here, and the other side of the equation must be taken into account. Firstly, R1 must not be so high in value as to be ineffective in its role to shunt minute, but present nonetheless, DC currents to ground if there is no similar drain resistor present on the other end of the line; often the case for older equipment where component count was at a premium. Secondly increasing the value of resistor R2 will increase the effects of op-amp input bias currents on shifting the voltage offset generated across it as a direct consequence of its resistance. At 330kΩ, the offset voltage induced is a total of 66mV with the typical input bias currents specified in the NE5532's data-sheet, but could rise all the way up to 264mV worst-case, which would result in a corresponding loss of headroom 528mV in the positive direction. Putting these two inevitable ills aside, an overly high biasing resistor will also generate enough noise to cause consternation when nothing is connected to the line input, potentially deleterious to a products esteem.

In conjunction with R2, C1 decouples any DC that may remain on the line due to either direct-coupling of the line-driver's output, or any leakage current flowing through ageing capacitors. The turnover frequency, at 1Hz, would seem rather low, and C1 could possibly be decreased by a factor of at least 2 without effecting the response at 20Hz by more than 0.1dB at a not inconsiderable cost benefit. Polyester film capacitors are not the cheapest of components as values start to exceed 100nF; using a 220nF part might return a saving of 40% or so - but there is a hidden demon here. Doing so would increase the impedance at the op-amp input from just over 16kΩ to 30kΩ, which would start to degrade noise performance. Op-amp input current noise disobligingly starts to rise toward the bottom of the audio band, just as the impedance of the DC decoupling network rises also. Like all of its BJT input brethren, the NE5532 has a typical input noise current of 0.7pA/√Hz, but this rises all the way up to 2.7pA/√Hz at 30Hz; perhaps 60nV/√Hz of additional noise voltage would result from this current passing through 220nF at 30Hz, while 470nF would produce just over 30nV/√Hz. Luckily this rise in noise is only present in a very small, and not particularly important, portion of the audio band, where human hearing is insensitive to put it mildly. Things are not as bad as they seem objectively as well; the bandwidth of this rise in noise is only a few tens of Hz and the overall noise contribution is small. For a A-weighted measurements that are insensitive to pretty much anything going on below 200Hz or so, the 220nF capacitor might pass unnoticed, but on a proper analyser using 470nF does make a slight difference that is worth it in the sentiment of the author, for a few more pence.

Last before the buffer is the RF filter, composed of R3, C2, and R4. It is a simple RC filter, combined with a second 'stopper' resistor going into the op-amp input. With the values shown, the cut-off frequency works out at 2.04MHz with a 100Ω source impedance, which may seem quite a high turnover point for an audio amplifier stage. A quick calculation reveals that in order not to affect the response at 20kHz by more than the target 0.1dB, a first order cut-off only 131kHz is necessary. Why not increase the value of C2 by a factor of 10 all the way to 1nF and still be left what appears to be a very comfortable margin of error to spare? The answer lies in the unpredictable output impedances of some line sources, as already mentioned, some line sources can exceed 10kΩ, which would result in a very disagreeable turnover of 14.9kHz, with an equally disagreeable response droop of -4.47dB at 20kHz. Even with a more reasonable 1kΩ source impedance there is still a drop of 0.19dB. It would be nice if it could always be assumed that source impedance would be under 1kΩ, and it is a sign of a good design if it is under 600Ω, but the real world is a cruel place where we always have to be prepared for the worst eventualities. C2 should be a NP0 or C0G dielectric type with a low ESR, as other dielectrics can exhibit non-linearity that will compromise the distortion performance past the limitations of the buffer stage towards the top of the audio band.

Resistor R3 could be increased to lower the cut-off frequency of the RF filter to possibly better effect rejection of lower frequency RF interference without the possibility of pulling the high frequency response down with high source impedances. There are two penalties to pay if this is done. The first is that there will be a greater noise contribution from the resistors, which will be made worse by the input noise current of a BJT op-amp. Increasing the value of R3 by a factor of five from 680Ω to 3.3kΩ, bringing the cut-off down to 482kHz, will degrade the noise performance of the circuit from a laudable -119.9dBV to -117.9dBV. The second penalty is reduced linearity at higher frequencies; distortion almost doubles at 20kHz with the NE5532. Most op-amps experience some degradation in linearity as a significant impedance mis-match develops between the two inputs with a significant common-mode input, and the NE5532 is no exception to this rule, although it is a good deal better than JFET input types which have a very non-linear capacitance of around 20pF or so from the input pins to the supply rail. Interestingly, it has been shown that a mismatch of around 1kΩ actually improves the linearity of the NE5532, which seems to indicate that there is another complementary distortion mechanism at play that is being partially cancelled. The distortion then starts to rise again the mismatch exceeds 2kΩ. A solution to this could be derived by putting an equal resistance in series with the inverting input, but the noise contribution will not be insignificant.

The cut-off of the RF filter, at 2.04MHz, may seem rather high, definitely quite far into the region of what could very reasonably be described as RF. With a slew rate of 9V/μs, the NE5532 configured with a voltage gain of unity or less can tolerate a 2MHz input of over 700mV before slew rate limiting and envelope detection occurs, resulting in audible interference noise in the audio band. This is far more, by at least 2 orders of magnitude, than what the input is likely to see in practice in even the very worst environments; which stays constant above this point as the first order roll-off of the RF filter and the direct relationship between frequency and required slew rate complement each other. The RF that really causes problems is usually well above 10MHz, such as GSM interference at 1800MHz, which has no problem coupling onto impedance bridged lines.

Adding a stopper resistor, R4, gives further inoculation towards excellent UHF RF immunity, providing much reduced envelope detection artefacts than simply adding its series resistance ahead of C2. Using the esteemed CAT S60 smart-phone/UHF signal generator in close proximity mode to an open ended RCA cable with one end connected to the line input, replacing R4 with a shorting jumper and more than doubling the value of R3 to 1.5kΩ resulted in an almost four-fold increase in the amplitude of detection artefacts, while increasing the value of C2 to 220pF did almost nothing the help the situation. Reverting to setting both R3 and R4 at 680R not only significantly reduced detection artefacts, but also almost doubled the the distance from the RF source to the line at which no RF artefacts were apparent. 680Ω also seems to be the optimal value for an effective stopper resistor, most likely as a result of parasitic capacitance starting to dominate as resistance increases any further; usefully a low enough value to be fairly noise free. To be most effective R4 should be connected as near as possible to the op-amp input pin, taking priority of proximity in the PCB layout. Using a double-sided ground plane PCB layout also goes a very long way to further enhancing RF immunity.

Good behaviour during input switching is also a good idea, as even if no provision for internal switching is made, external line switching boxes are not uncommon. The line input must not generate switching transients while different sources are selected. This is taken care of by DC decoupling network R1, C1, and R2. If switching is to be done, it must be performed ahead of this network to avoid input bias currents from generating clicking transients as the bias resistance changes.

Line input selection

Figure 3. A simple two way line input selector

Figure 3 shows a simple line switching circuit commonly found in a lot of audio equipment consisting of a DPDT switch to switch between two stereo channels. As the switched output will temporarily be open-circuit as the contact moves between the two inputs, the line input must also tolerate being disconnected for a short period of time without producing any transients or excessive noise. The second requirement puts limitations on the maximum input impedance that will determine the noise output of the line input with an open input. Input drain resistor R2 of Figure 2 is especially useful here as if the input is switched with a signal present, decoupling capacitor C2 can absorb a small amount of charge and potentially generate a distinct click as the switch makes contact with the second source input; particularly if the switch is operated slowly.

It might also seem a good idea to incorporate DC drain resistors ahead of the line switching so as to further diminish the likelihood of switching transients in cases where there may be DC leakage current on a line input to be switched. Adding these will increase complexity, reduce input impedance and may also keep the user in blissful ignorance of a fault in the equipment upstream that would otherwise possibly induce repair.

Although only two line input poles are shown in Figure 3, it is possible to add many more. In order to do so, a rotary switch is used, as is the case for a great plethora of HiFi amplifiers, most often a 6 or 4 pole device to select from 6 or 4 inputs respectively. It is uncommon to find more than 6 pole switches, and care must be taken with wiring the multiple inputs to the switch to keep parasitic capacitance and hence crosstalk to an acceptably low level. When a line source with a high output impedance is selected against an adjacent source of sufficient level, the adjacent source can be sometimes heard faintly fizzing across the parasitic capacitance that exists across the switch contacts. There is little to be done for this, other than using considerably more complicated switching arrangements that require twice as many expensive switches, or buffering each line input individually and applying the switching afterwards from the low output impedance of the buffers. If high quality low impedance line sources are used, then crosstalk across switch contacts is reduced to negligible levels well below the noise floor. In any case, the problem can be solved by switching the adjacent source off.

In the opinion of the author, the last two problems described in relation to input switching lie very much in the source equipment for having significant DC offset and output impedance. Having put the blame squarely onto the shoulders of these most thoughtless failure modes and design inadequacies, further consideration will not be given to them.

Unbalanced lines and their woes

Unbalanced, or single-ended line connections, despite their susceptibility to ground loops, remain the most popular means of connecting consumer HiFi equipment together. This has more to do with cost than anything else, as the electronics, connectors and cabling are considerably less expensive to implement and mass-produce than their balanced counterparts. As line connections in typical HiFi set-ups do not exceed one meter, with the exception of line level connections to active speakers, and the interconnected equipment is typically situated in close proximity, the risk of ground loop noise and interfering signals coupling onto the line is considerably reduced. With care, it is quite easily possible to bring ground loop noise well below the noise floor, especially if 30cm line cables are used in a stack arrangement.

RCA connectors are most often used for unbalanced line connections in high fidelity audio. With the advent of portable digital media players and PC sound card manufacturers looking to cut down on connector costs, the 3.5mm TRS connector has become the connector of choice for cheap stereo line connections, sometimes sharing a common ground conductor in the cabling as well as the connector sleeve, promising not-very-good channel separation. Many HiFi amplifiers now feature an auxiliary 3.5mm stereo unbalanced TRS input for connecting a smart-phone or MP3 player. 6.35mm TS connectors are sometimes also used for unbalanced connections, although these durable connectors are most often seen in more demanding environments such as connecting electric musical instruments to their amplifiers.

Unbalanced line cabling especially, should be selected to have as low a ground conductor resistance as possible, so as to ensure that ground loop, or electrostatically coupled currents manifest themselves in the smallest voltages possible. The easiest way to do this is to make the cable as short as possible, but it always helps to select cabling for low ground path resistance. For this reason, sometimes changing out one cable for another with half the ground conductor resistance can reduce ground current interference by a factor of two. If the cabling doesn’t need to go through sharp bends, then RF coaxial cable is a good option, as it is quite easy to find this sort of cabling with a desirable foil shield and a nice thick braid that will put the series resistance of most commercial offerings, including the esoteric kind, quite to shame. The only disadvantage with this option is that the cables will have to be cut, put into sleeving if they are stereo, and finally have the connectors soldered onto them by hand; part of the fun if the cable-builder is into DIY.

Ground loops and other ground current interference problems have been causing grief and despair for users of unbalanced connections since what seems to be time immemorial. Most often, the cause of interference in unbalanced connection is due to voltage generated across the resistance of the ground path as a result of these currents, although sometimes interference can couple directly onto the centre conductor. The most common means by which interference is induced, along with their somewhat partial remedies will now be covered.

The classic ground loop where both the source and the line receiver are connected to mains ground forming a loop from the sources connection to mains earth, to the line output ground, through the line's ground conductor, to the line input ground, back to mains earth, possibly through some domestic electrical wiring if the source equipment is connected to a different wall power socket than the receiving equipment. The loop converts changing magnetic fields in its proximity, in the same manner as a transformer winding, into changing electrical currents which will run through the ground path of the line as part of the loop. If a considerable length of domestic wiring is included into the loop, when different wall sockets are used, any currents flowing through them may also flow through the audio path on the other side of the loop exacerbating the problem further. The larger the loop, the more effective it will be at turning changing magnetic fields, which will be in abundance in a domestic setting due to the AC mains and the magnetic leakage of mains transformers.

The solution to the classic ground loop is to keep the loop as small as possible, which means reducing the length of cabling in all parts of the loop. The equipment must be physically close together and the power plugs must be connected to either the same power block or wall plate, the power block option being a better one as it allows the equipment power cable length to be reduced quite considerably without worrying whether it will reach the wall plate or not. It also helps to cable-tie the power leads of the two pieces of equipment together if they are going to be in place for a significant period of time, aside from making things look a good deal neater, this also further reduces the the loop area. In keeping with a recurring theme, the audio line cable should be kept as short as possible, reducing the loop area even more so.

What absolutely should not be done to tackle ground loops, is to remove or break the safety ground path on one of the pieces of equipment, which will result in it becoming possibly lethal if a fault develops in the insulation between the mains and the chassis. Sometimes a 'ground lift' switch is included on some equipment that performs this highly risky function, but it is best avoided as the ground conductor of the line cable which will make the path to mains ground through the receiving equipment will most likely not be rated for the sort current needed to protect the user in case a fault develops. Sometimes a series resistance network is included between the chassis and line input ground and the mains ground is included in the equipment itself to partly fracture the loop, although this also has safety implications and doesn't yield a difference big enough to make any serious amount of ground loop interference tolerable.

Ground current interference can also be induced capacitively, which will be the case for double insulated class 2 equipment that forgoes connecting the chassis to mains safety ground, in favour of an extra layer of insulation between the mains side and the low voltage side connected to the chassis and audio line. With the mains ground disconnected, the chassis and low voltage of the equipment will essentially be 'floating' and without a mains ground connection to shunt away any capacitively coupled currents, these currents will have to flow through the ground conductor of the audio line to get to mains ground on the other end of the line - if the other end of the line is connected to mains earth. This effect can sometimes be demonstrated by turning the gain up on the line receiver and then touching the chassis of the bouble insulated source, coupling a small capacitive current from the body onto the line ground, similar to how touching the centre pin of an RCA connector yields a buzz or hum. In the absence of human interaction, these currents are typically coupled across the power transformer windings of the double insulated equipment from the mains, from mains wiring that might be nearby, or from a switching power supply if the unit in question uses one. The latter case is particularly troublesome as the suppression capacitor these power supplies need to use in order to meet EMC regulations will couple quite a surprisingly high amount of noise current onto the line, its value typically being well over 10 times higher than the leakage in a high quality double insulated mains transformer. Interfering currents can also be coupled onto the ground conductor of the line by adjacent cables running alongside the line. Cables carrying high frequency digital information; the likes of USB cables and ethernet cables, mains cables, as well as those running out of the already admonished SMPS wall-wart supplies, are the worst offenders here.

To combat capacitive ground current interference, audio line cables should be kept well away from mains cables and other cabling carrying any sort of high frequency interfering signal that might like to couple onto the ground conductor. The usual habit of keeping cables as short as possible will also help to minimise the effects of the interfering current. There's little else to be done here, but with class 2 equipment that uses a proper linear power supply, the capacitive coupling of mains current across the transformer windings is small enough to be negligible with short lengths of decent quality cable.

Unbalanced cables are also susceptible to the effects of capacitive interference if the centre conductor is not properly shielded, which is the case for some very cheap throw-away cabling that is included as a token gesture in the bottom of the packaging of new equipment. Interference can also couple magnetically onto the centre conductor if there is adjacent cabling that is carrying appreciable current, another case for keeping unbalanced lines and mains cabling asunder as far as is practical. Magnetic coupling onto the centre conductor is usually the least likely of all the means by which interference can occur to cause any trouble in the experience of the author.

Having heaped an almost insurmountable pile of execration and admonishment onto the shortcomings of unbalanced connections, it is important to note that they do have their advantages. The main advantage, aside from their simplicity and cheaper connectors, is that the receiving circuitry has quite considerably better electronic noise performance, as will be revealed further on. As long as the cabling can be kept to less than an arms length with a few precautions taken, they can realise this better noise performance without it being swamped by interference. Balanced line is starting to be seen as a higher prestige and quality means of connecting HiFi, however, and it may be the case that not before too long the humble unbalanced connection will be seen as a mark of low quality.

Series feedback receiver

Having already considered a simple unbalanced high-impedance unbalanced input, it is useful to consider how the circuit can be elaborated on the other side to yield a little extra function and flexibility. Simply buffering the line is all very well, and at that point the requirements for a line input have already been fulfilled, but it may be a good idea to add the option of increasing the gain or adding some additional level control. The line input might also be able to do double-duty, functioning as a gain control, tone control, or any other function that can be performed in a series feedback loop.

Single stage gain control

Figure 4. High impedance receiver with a series feedback active-passive level control

Figure 4 exhibits the versatility that can be realised with a series feedback high impedance line input. Along with input decoupling and RF filtering networks, an active gain control is realised by making use of the feedback path and the drive capability of U1 to drive a passive variable attenuation network, but also drive a line output. The circuit was originally built by the author for use as a level control to make transcriptions from a variety of line sources to a simple USB audio interface; a job that it does quite well indeed presenting a nice high impedance to the older tape decks and valve equipment it has been connected to on one end, while driving a 10kΩ audio interface load to 5V RMS with no trouble at all. As the potentiometer is rotated clockwise, the wiper, connected to ground, moves towards the bottom of the feedback network, decreasing the resistance at the bottom on the network and causing the gain of op-amp U1 to increase. When the potentiometer is rotated anti-clockwise the bottom half of the attenuation network pulls the gain down reducing the signal on the output.

To avoid doing too much amplifying and attenuating at the same time, R6 and R9 are set to 2.2kΩ, which permits the circuit to amplify and attenuate not more than 3dB simultaneously. Increasing these resistor values will improve the gain law of the circuit, which is a little too flat in the centre to be considered for a proper HiFi volume control, but will decrease headroom and increase output impedance to an unacceptable level. With the values shown the unloaded gain with the potentiometer centred is -0.3dB with an output impedance of 1460Ω, rising to 20.4dB and 1740Ω fully clockwise, which is a little high but fine if the circuit will only be connected to an audio interface as intended. C4 and R10 then DC decouple to the line output. The noise performance is a very impressive -117dBV with the gain control centred. Not too bad at all for a simple circuit!

A disadvantage of the series feedback receiver are that it can be susceptible to damage from over-voltage if input clamping is not used. Applying input clamping means deploying a pair of diodes between the op-amp inverting input and the power supply rails which can mean that any noise on the supply rails can couple through the diode's parasitic capacitance and onto the the input. Furthermore, the parasitic capacitance is rather non-linear as a function of reverse voltage, a characteristic that is exploited in varicap diodes in radio receivers, but certainly not wanted here and will lead to a substantial increase in high frequency distortion that will only get worse as source impedance increases. Fortunately, there are other topologies to come to the rescue...

Shunt feedback receiver

The line input does not always have to be non-inverting, in many instances where there might be another inverting stage down the signal chain, it may be a good idea to use an inverting line input so that the overall phase is preserved. Inverting line inputs are easily accomplished by deploying a shunt feedback amplifier configuration shown below.

Shunt feedback line receiver

Figure 5. Shunt feedback receiver

The basic inverting line input of Figure 5 is a simple circuit that works very well in some of the most unforgiving real-world environments that a line input can rather unreasonably be subjected to. Resistor R1, parallel with R2 and R3 set the input impedance, while R4 determines the gain of the receiver, which quite unlike the series feedback receiver can go below 0dB if required. Capacitors C1 and C2 are connected back-to-back to effect DC decoupling from the line, while C3 performs RF filtering, and C2 ensures high frequency stability, which can be an issue for shunt feedback amplifiers if it is omitted.

The first thing to be noticed is the input impedance is much lower for the shunt feedback receiver than it is for the series feedback one. This is due to the necessity to keep resistor values as low as is feasible in want of keeping the thermal noise they contribute as low as possible, as this time the load resistors are connected in series with the input. R2 and R3 have been chosen so as to give a combined series resistance of 11.2kΩ to the virtual earth on the inverting input of U1, which when combined in parallel with drain resistor R1 gives a pleasingly round 10,072Ω; close enough to the standard loading of 10kΩ that all decent sources can be reasonably expected to drive. Even with the resistor values optimised for the best noise performance, the noise generated by this receiver is a good 10dB worse than the series feedback one, resulting in a noise output of -109.3dBV in conjunction with a 100Ω source impedance. Aside from the higher, and therefore noisier series resistance that the line signal will have to pass through, the noise gain of a shunt feedback amplifier configured for unity gain is will be two relative the the signal, instead of one for a series feedback type; doing nothing to help the problem at all. The noise falls down to -111.9dBV when the input is left open, prompting the noise gain also fall to just over a decibel. Things are not so bad in reality as the sort of small signal circuitry that succeeds the line input in a typical HiFi pre-amplifier will have a very hard time indeed of coming within 6dB of this figure in total. To put things into perspective, a pre-amplifier that generates a commendable -103dBV of noise after the line input, will suffer less than a decibel of noise degradation when used with the shunt feedback receiver of Figure 5, compared to using the series feedback type detailed in Figure 3.

As the input impedance is necessarily lower, the coupling capacitors C1 and C2 are now electrolytic types of a considerably higher value than the film types described earlier for high impedance inputs. If film capacitors were to be used, then they would have be at 4.7μF parts so as not to pull the response down by more than 0.1dB at 20Hz, which are bulky and rather expensive to say the least. Electrolytic capacitors must always be selected to give a time constant of 1s or greater for audio coupling to avoid the low frequency distortion they exhibit if more than a few millivolts of signal voltage appear across them. Two back-to-back electrolytic capacitors are shown, but there is no reason why non-polar electrolytic capacitors can be put into service, although high quality through hole types are not as easy to source for the same value as a pair of high quality polar ones.

Traversing into the realm of benefits, the RF filter looks quite similar to the one put to use in the series feedback line inputs, but it is in fact much more effective in doing its job as a consequence to the values of the filter and stopper resistors, now R2 and R3, being over 8 times as high as their earlier counterparts. In this topology the cut-off frequency has dropped down all the way to 568kHz, which in tandem with the higher value RF stopper resistance of R3, assures that the very best rejection of RF will be delivered, without having to make any compromises further than setting the values of R2 and R3 high enough so as not to overload the line source.

On the other side of the feedback network, resistor R4 sets the gain to 0.6dB. The gain could of course be set to exactly 0dB by using another pair of 5.6kΩ resistors in series, but consider that the lower input impedance of this receiver will induce a slight bridging loss of up to 0.5dB with a source impedance of 600Ω, and 0.6dB to compensate seems a little bit more appealing. The gain of the stage is very easily changed by altering the value of R4, however with a BJT input op-amp it is not advisable to put a potentiometer that will be adjusted in normal use in the place of R4, as op-amp bias currents may result in crackling as it is turned. The voltage put across R4 and the output as a result of input bias currents will typically be a meagre 2.4mV, giving license to increase it from 12kΩ without needing to worry about output swing being compromised. R4 can also be reduced to bring the gain below 0dB if required, and if this is done the input will be able to accept an input greater than the supply rails of U1, although such high levels are quite rare.

A further benefit that the shunt feedback receiver offers, is quite excellent immunity to the most diabolical input over-voltage conditions that can be afforded to it short of connecting the line input to the mains; not recommended. Tucked away inside the NE5532, between the two input pins of each op-amp section, lie two input protection cross clamping diodes that make certain that the inputs do not deviate from each other by more than 0.6V, which would damage the sensitive input transistors. With the non-inverting input connected to ground, these diodes clamp the inverting input to ground within a ±0.6V range. According to the data-sheet, they are able to clamp a maximum of 10mA current, which means that a voltage of over 110V in either direction can be applied to the line input without risk of damage to the op-amp. If the voltage is DC, it must appear briefly enough so that it does not charge C1 or C2 past their rated voltages; 25V capacitors are used here. This excellent over-voltage capability means that the shunt feedback receiver is perfect for use with potentially unruly sources that might be hot-swapped as part of standard use; auxiliary inputs for smart-phones, laptops and various other portable devices not intended for audio use above all things being a good example. If no diodes are used internally, as with JFET input op-amps, then 1N4148 types can be added between the non-inverting input and ground, although C4 might need to be doubled to make certain that a comfortable HF stability margin is retained.

Shunt feedback hifi preamp

Figure 6. Simple shunt feedback pre-amplifier

Figure 6 is offered as a quick example of the use of the shunt feedback amplifier in a typical pre-amplifier, demonstrating how the phase inversion of the line input corrects an otherwise out of phase latter stage so that input and output are both in phase; something that doesn't make any audible difference with music, but is rightly taken as a sign of careful well-thought-out pre-amplifier design. The receiver applies a little more gain to the input, and then feeds it into a simple active volume control built around U2, which also drives the line output. With the potentiometer centred, the gain is a satisfying 0dB and the noise is very low at -108.5dBV, going all the way down to -120.3dBV when the potentiometer is rotated anti-clockwise all the way to silence. Maximum gain is 20.8dB; enough to take even the puniest 100mV line source past the 1V most power amplifiers require for full output.

By increasing R4 to 15kΩ, the line receiver assumes a gain of 2.5dB, mainly as a means to fix the overall gain of the pre-amplifier to a round 0dB when the potentiometer is centred as the volume control exhibits a loss of 2.5dB at centre position. Adding a bit of gain also permits slightly better noise performance to be realised as the signal going into the volume control is relatively stronger, and as there are no interceding stages between them, the risk of overload is much reduced. The line receiver can still quite comfortably accept an input voltage in excess of 10V peak without any overload behaviour occurring. The noise gain of a shunt feedback stage will always be the signal gain plus one, so increasing the signal gain slightly lends a marginal proportional advantage relative to this undesirable head start.

Given that the line input is connected to a virtual-earth node in the receiver, it is possible to sum multiple inputs through different input networks onto the aforementioned virtual-earth in a single stage, without the risk of them contaminating each other on the line itself. Summing line inputs together may be performed for a number of reasons, but the most common reason is for the purposes of converting a stereo input to a mono signal. If two line level inputs are simply connected together, either directly by shorting them together, or through a series resistance of 10kΩ or so, then it is very possible, if not certain, that heavy crosstalk between the two inputs will occur either completely in the first instance, or partially in the second to a very unsatisfying degree as they will be resistively coupled together. Similarly mixing two audio signals together in this way, when they also need to be routed in separation somewhere else will also cause vexation.

Summing input subwoofer crossover

Figure 7. Simple sub-woofer crossover with summing shunt feedback input

Notwithstanding that the situation described in the last paragraph seems a little far-fetched for an every-day application, the schematic detailed in Figure 7, an active sub-woofer crossover, is presented as an example of such a case in a very common home-audio type of setting. Both stereo inputs need to be summed together to produce a mono signal for the crossover and sub-woofer, but in such a way that preserves channel separation on the line so that amplifier driving the small stereo speakers will not suffer from a lack of stereo separation. Two sets of input networks are used for the left and right channels which are then connected to the non-inverting input of U1. As the non-inverting input will always be at, or at least tremendously close to ground potential, there is no risk of cross coupling between right and left back up the line.

The noise performance of the stage is already quite limited by the TL072 op-amp specified, and the input impedance for each input is set at 20kΩ by the input networks; not 10kΩ, to keep the loading light enough so the stereo amplifier can also be connected to the line without loading it too heavily. R7 allows the gain to be trimmed so as to adjust the sub-woofer level against the stereo speakers. A simple second order multiple feedback filter is used to produce a typical crossover point of 116Hz. Because of the low pass action of the crossover filter, the noise output is surprisingly low at less than 108dBV. Both the line input and the crossover filter use shunt feedback stages; a necessity for the TL072 which has a tendency to be rather non-linear when unequal common-mode impedances are presented to it's inputs, and misbehave when common mode conditions that are not confined to a relatively small range compared with other op-amps. Subsequently, if a TL072 or a TL082 is to be used as an unbalanced line receiver, the shunt feedback receiver topology must be used.

Balanced line

Balanced line, using two differential signal conductors instead of just one, has the highly desirable ability to reject to a very high degree all of the ground interference problems that plague unbalanced line; detailed previously. This becomes a necessity as soon as the line length exceeds a metre or so, and even sooner in some instances. Balanced line connections are found on almost all professional audio equipment as a testament to this necessity, where runs of cabling can easily exceed 10 metres. Balanced line is also slowly creeping into the world of high-end consumer audio, where although most cable lengths are combined to under a metre, it is considered to be a more serious approach. One exception to this are high fidelity active speakers, which should always use a balanced connection, as cable runs will most often be over several metres long.

Aside from excellent rejection of ground noise, balanced connections can also reject any interference that is common to the two hot and cold conductors carrying the signal. As the two signal conductors are twisted together, any interference that might want to couple onto them will do so in equal measure on each side, resulting in a common interference signal which is then rejected by the receiver. The line receiver is designed so as to only pass the differential signal between the conductors, and reject any portion which is common to both of them. The ability of the receiver to reject the common portion of the input signal is referred to as the common mode rejection ratio, or CMRR for short. The higher the CMRR, the better the line receiver will be able to reject common mode interference, making it desirable to have as high a CMRR as possible.

Balanced inputs are usually implemented using 3 pin XLR connectors, or TRS connectors. The latter are used when either ease of use, low cost, or if panel space is at a premium. XLR connectors are a little more robust and easier to solder onto the end of cabling, but are not as easy to connect together, requiring rotational alignment of the connector before a connection can be made. XLR connectors also require more space on the rear panel; if 10 or more balanced connections need to be made onto the back of a 1U piece of equipment, they will have to be TRS. XLR connectors do have the advantage that the ground shell of the connector makes first contact as the connection is made, which is much kinder to the line input/output circuitry than running the hot tip of a TRS connector onto the chassis of a double insulated piece of equipment that might be floating at 50V or so before the full connection is made. TRS connectors are wired hot, cold, ground to tip, ring, sleeve respectively, while XLR connectors are wired correspondingly to pins 2, 3, and 1.

In professional audio, balanced line level is defined as the voltage across the hot and cold conductors, so the receiver must have a differential gain of unity so as to receive the line level signal without altering its intended level. In high fidelity audio, the story is a little different. A high fidelity balanced output usually applies the audio signal to the hot pin, then inverts it and applies the inverted version to the cold pin, so that twice the signal level appears across the line. To side-step effecting an overload-inviting level increase of 6dB every time a connection is made, the receiver must have a gain of -6dB, or half; the case for all the balanced receiver circuits to follow. A further advantage of balanced line is now apparent; the differential signal level on the line can be twice as high as an unbalanced one due to the bridging operation of the line driver, bolstering the strength of the desired signal further against potential interference before common mode rejection is even applied.

Balanced line does have its disadvantages, though, the most obvious one being increased complexity and cost for more expensive connectors and cabling with a pair of centre conductors instead of a single one, in addition to twice or more components being used on the electronic side. The worst disadvantage, however, lies on the electronic side where perversely the electronic noise generated by a simple electronics balanced input will be around 10dB higher than an unbalanced non inverting one. This will be examined and remedied further below, but to put it shortly, it is as a consequence of the combined effects of higher noise gain and resistor noise in differential amplifiers.

Transformer balancing

Before the arrival of high quality audio op-amps and close tolerance resistors in the latter half of the 1970s, using a transformer was the only way to make a good quality balanced input with a reasonable CMRR. They are still used in modern reproductions of vintage equipment, especially of the valve variety. Transformers provide total isolation, within the limits of their insulation, between separate windings, so transformer balanced inputs are inherently very robust when it comes to input over-voltage. Floating transformer inputs can also be connected to unbalanced lines in such a manner that takes advantage of the isolation to break ground loops without any safety risk. Transformer balanced inputs are also very tolerant of RF interference, due to their inability to pass high frequency energy outside of the audio band. For these reasons, transformer balanced inputs are still applied to some very harsh environments that require these levels of robustness. Additionally, since no differential amplifier is needed, better noise performance can sometimes be realised by using transformer balancing, although this can be mitigated by employing the low noise design techniques to be discussed later on in this article.

Having extolled the virtues of transformer balanced inputs, it is now a good idea to examine their deficits, and why their popularity has waned significantly since the late 1970s. There are some serious disadvantages to transformer balanced inputs that warrant precluding them from high quality, cost effective audio design as listed below:

Transformer balanced line input

Figure 8. Legacy transformer balanced receiver

A discussion of transformer balanced inputs would be incomplete without at least offering at simple circuit in the form of Figure 8 as an example of a typical application. It may be immediately noticed that Figure 8 does not use the NE5532 op-amp circuitry set out as the standard at the beginning of the article, but uses a simple discrete arrangement on a single supply rail. Due to the severe limitations of audio transformers set out in the bullet points above, the author does not consider them viable for modern high quality audio design. The transformer balanced input is here exhibit as a legacy circuit, complete with an obsolete BC108 NPN transistor voltage follower, to imply that such inputs belong firmly in the past.

From the TRS connector, J1, the cold side of the line feeds the top of the transformer input winding, while the cold side feeds the top part; the transformer will invert the signal on the other side as part of its operation. Q1 is biased to through the transformer via divider network R1 and R2, with C1 as AC decoupling, and follows the transformer's output onto R4 which performs the role of a crude current souce. During the early 1970s, discrete design was still rightly seen as superior to the underwhelming LM741 op-amp which was starting to be used in low cost equipment, and it was acceptable to apply transistor bias through input transformers, as the bias current was low enough so as not to saturate the transformer significantly. R3 and C2 are a necessary impedance equalisation network that the transformer needs to see as a load so the response doesn't fly up more than 6dB as a result of parasitic resonance of leakage capacitance and transformer inductance; generic values are shown but they would have to be optimised for the specific transformer used - another disadvantage. From experience, the author can attest that the distortion performance of this sort of simple circuitry will not be better than 0.01% at 1kHz 1V RMS, although transformer distortion will dominate below a couple of hundred Hz.

Interestingly, in the strange and disorienting world of subjectivist high-end audio, whisperings of the alleged sonic superiority of audio transformers over other forms of coupling and connection are growing louder and louder, setting yet another trend in quite the wrong direction. This is probably either due to the rarity of such transformers in the 21st century, the great expense required to obtain them - which is always attractive to the irrationally inclined, or because transformers are the only way to implement the ever-more esteemed and desirable balanced connectivity in 'audiophile' valve equipment, regardless of whether it is necessary or not.

Classic single stage receiver

Putting transformer balancing aside as an inadequate and overly expensive option for a modern high performance balanced input, it is necessary to consider electronic balancing as the only option left. Electronically balanced inputs have been used for the past 50 years, with the introduction and mass production of the highly popular 741 op-amp that still geriatrically rattles off the production line today. Initially quite inferior, but still much cheaper than the already flawed transformer balanced inputs in terms of CMRR and especially noise, op-amp differential amplifier based balanced inputs gradually came to dominate the scene as better op-amps such as the TL072 and NE5532 became available, along with cheaper 1% tolerance resistors needed to obtain a satisfactory CMRR. These parts are as cheap as chips today, and it is possible to build a very satisfactory electronically balanced line receiver for significantly less than the price of said chips.

Classic balanced receiver circuit

Figure 9. The classic single stage electronically balanced receiver

Figure 9 details the classic single op-amp balanced input, variations of which have been used extensively for over half a century at the time of writing. At its heart, it is a simple differential amplifier of the kind that can be gleaned from the pages of any textbook that describes op-amp circuitry even briefly. All the resistor values, with the exception of input drain resistors R1 and R4 are 6.8kΩ. The top arm of the circuit composed of R2, R3, and R7 determines that the gain for the cold input on the inverting shunt feedback upper arm will be half, while the lower arm attenuates the hot input to a third, which is then amplified through series feedback action one and a half times, yielding an identical gain of half. Any interfering signal common to both inputs will be amplified by one half, inverted, and then summed with half of itself, resolving to zero - complete rejection. The differential gain of the circuit is -6dB, intended for the high fidelity applications where the differential signal on the line will be 6dB stronger due to the inversion on the cold side.

To effect a differential gain of -6dB, the input side of each arm needs twice the series resistance as the feedback/attenuation side. The easiest way to do this is to use two resistors in series, affording a handy spot to place RF filtering capacitors C3 and C2. Although there will be some common mode voltage on the op-amp inputs at high impedance - usually a recipe for increased distortion - the level will be a third of the signal level on the line relative to ground, and the impedances are matched at each op-amp input so no harm is done. Even a humble TL072 can be used here without risk of excessive distortion, although the noise performance will be almost 6dB worse. As the common mode voltage is a third of that on the line, the receiver can accept an input over-voltage of three times the op-amps power supply rails without suffering damage, making it robust enough to forgo any further consideration of input protection.

So as not to present a heavier load than 10kΩ to the line, the load and feedback resistor values for Figure 9 must not be lower than 6.8kΩ which, through their thermal noise, limits the quietness of the stage to -110.5dBV of noise on the output. The noise gain for the amplifier stage is 1.5 times, and the attenuate-then-amplify action of the non-inverting hot side of receiver doesn't help much, but is a necessary evil here. Not bad in the grand scheme of things, but not as low as might be desired for a piece of equipment claiming to have very low noise. As the input impedance is low, electrolytic coupling capacitors have to be used.

The receiver's CMRR depends very strongly on the tolerance of the 6.8kΩ resistors, and parts with a tolerance at least 1% must be used to push the rejection level up high enough to render any common mode interference it might be presented with to an acceptably low level. It used to be the case that the CMRR was adjusted manually by reducing the value of R8 and placing a trimmer resistor in series with it, so as to adjust the gain of the hot input to that of the cold one. For effective rejection to occur, the gains of each input must be closely matched to achieve as high a level of cancellation as possible. As the resistors determine the gain of each input, they must be very close to their specified value. The worst-case scenario occurs when the first two resistor tolerances in the upper arm and the resistor tolerance at the bottom of the lower arm of the receiver deviate as far as they can in one direction, while the resistor tolerances for the third resistor in the upper arm and the first two in the lower arm go as far in the other direction as the tolerances permit. The table below demonstrates the effect of resistor tolerance on CMRR relative to the differential input level.

Resistor Tolerance Typical CMRR Worst-case CMRR
10% 24dB 12dB
5% 30dB 18dB
2% 39dB 27db
1% 46dB 34dB
0.5% 52dB 40dB

Typical and worst-case CMRR for a range of resistor tolerances

The worst-case CMRR figures above look quite intimidating, but in reality these are about as likely to occur as rolling a set of 6 die to a sum of either 36 or 6. The typical figures in the middle column are given as an example of what can be reasonably expected most of the time, but they are heavily dependent on the the distribution of deviation of resistor tolerance. Experience indicates that 12dB greater CMRR than the worst-case figure, or more in some cases, will be realised in practice. The first three rows are mainly there as a testament to how badly things can go wrong if wide tolerance resistors are thoughtlessly thrown into service. The sort of metal film resistors that will need to be used for high quality audio design usually come with a tolerance of 1% or 0.5%, so a worst-case CMRR of more than 33dB is already off the menu. 5% and 10% parts are usually very cheap carbon film types that are not suitable for audio design as they are not perfectly linear and will generate additional noise if current flows through them, although 5% resistors were sometimes used in very cheap early equipment that employed the 741 op-amp.

In the experience of the author, through-hole metal film resistors of a specified 1% tolerance in the same batch can be expected to be within 0.2% of each other, although there may be some skew on average within the 1% tolerance itself. When resistors are manufactured, most of the deviation in the tolerance occurs around a small area within the tolerance which may move around very slowly during a production run of thousands upon thousands of resistors. A balanced receiver can take advantage of this to phenomenon to reliably yield a CMRR 12-14dB greater than the dreaded worst-case scenario that the manufacturers tolerance implies, so long as resistors from the same batch are used. With 0.5% tolerance parts, of which the through hole types seem to be only perhaps 10-25% more expensive than 1% specified ones, a very respectable CMRR of 52dB or more can be reliably obtained. Resistors are very cheap components, so in the opinion of the author, it is very worthwhile to invest less than a penny or so per line receiver buying 0.5% ones to effect twice the CMRR.

The decoupling capacitors C1, C2, C4, and C5 may also affect the CMRR negatively, and this is worth investigating. Simulation reveals that if a tolerance of ±20% is specified and the upper arm of the receiver sees a series capacitance of 75μF, while the lower arm sees 125μF, then the CMRR will degrade to 49dB at 50Hz - all other things being perfect. It will then improve by a first order function as frequency increases from this point. Again, in practice this worst-case is very unlikely if electrolytic capacitors from the same batch are used, especially if they are connected back-to-back as per Figure 9, resulting in further tolerance averaging. The 12dB rule-of-thumb can be applied here as well, with 61dB being reasonably expected. Doubling the capacitor values to 220μF might improve the CMRR degradation by 6dB, but will increase costs significantly. Looking back upon this information, it is a good idea to ensure that the coupling capacitors are placed on the layout in such a manner that one set is not closer to a source of heat than another which might cause them to dry out a little quicker, resulting in a mismatch and loss of low frequency CMRR. Electrolytic capacitors should of course be kept away from hot components to preserve their longevity, but if it is inevitable that they are to get a little warm, then they should all be equally warm in a balanced receiver so as to dry out in harmony with each other.

An additional set of capacitors to take into account when investigating the optimisation of CMRR against component tolerances are RF filter capacitors C3 and C6, which at 100pF set the cut-off at 351kHz; high enough so as not to affect the response at 20kHz by more than 0.1dB with a reasonable source impedance. There is no problem in using ±5% tolerance parts here, as the effect of the capacitance on the gain within the audio band is so small. A worst-case scenario of 70dB CMRR at 20kHz is ascertained through a quick calculation considering each C3 at 95pF and C6 at 105pF; a negligible amount, all other things considered. As with the resistor worst-case scenario, this can be expected to be at least 10dB better in practice if components from the same batch are used.

The op-amp at the centre of the balanced receiver can also reduce CMRR through its real-world limitations, most often in the form of limited open loop gain which further reduces as frequency increases. Op-amps themselves also have only a limited, if not very high, degree of common mode rejection built into them. If a suitable op-amp such as the NE5532 or even TL072 is chosen, the effects of op-amp gain deficits on CMRR are reduced to well under 90dB within most of the audio band, possibly degrading as far as 70dB at 20kHz; still significantly less compromise than that already afforded by the resistor tolerances that cause the most degradation overall. Fortunately, most of the common mode interference that that balanced line was designed to reject, mains hum, is much lower than 20kHz and the effects of op-amp limitations can be considered negligible without much further thought.

A slightly unsettling characteristic of the single stage balanced line receiver is that when driven by differential signal, the input impedances on the hot and cold side is different from each other. Unequal loading on each side may imply that CMRR may be compromised, but this is really not the case. Input impedance conditions relative to ground for Figure 9 are detailed in the table below, and it quickly becomes obvious that the the unequal differential loading will have effect on CMRR.

Input condition Hot impedance Cold impedance
Differential drive 19,806Ω 10,049Ω
Common mode drive 19,806Ω 19,806Ω
One side driven 19,806Ω 13,333Ω

Input impedances for different input conditions

It's evident from the table that under differential drive conditions the hot input impedance is almost twice as high the cold input impedance, which is an acceptable 10kΩ. The cold input is connected through the series resistance of R2 and R3 to a third of the voltage on the hot input, realised on the op-amps inverting input, so if an equal but opposite signal is applied to the hot input, the series resistance of R2 and R3 will appear to be reduced by a quarter. This may seem a little awkward, but it does no harm whatsoever to the CMRR because as far as a common mode interfering signal is concerned, the impedance is identical as the signal voltage will be identical on both op-amp inputs which connect back through to the line via equal series resistances. This is confirmed in the second row of values in the table above. Any common mode signal is therefore equally loaded on both hot and cold sides of the receiver so the effects of potentially unequal loading and incomplete cancellation are avoided. There is a version of the balanced receiver that adds an additional inverting stage driven from the output, connected to the bottom of the hot side input arm that equalises the differential impedance known as the 'superbal' circuit. Its benefits are purely aesthetic though, and it does nothing to improve performance; quite the opposite in fact, it degrades performance by injecting additional noise from the inverting stage into the bottom of the hot side input arm.

Double stage shunt feedback receiver

An electronically balanced line receiver can be split into two stages that each perform an inverting function. This might be useful for situations when a significant common-mode voltage cannot be tolerated on op-amp inputs, or greater flexibility is required that cannot be realised with the simple single stage differential amplifier input.

Shunt feedback balanced input circuit

Figure 10. Double stage shunt feedback balanced receiver

Figure 10 exhibits the general idea of a two stage differential line receiver. The hot input is inverted U1, and then subtracted from the cold input which is also inverted by U2. Two input decoupling and filter networks identical to those of the unbalanced shunt feedback receiver are used to give an input impedance of 10kΩ on each input, which remains constant regardless of whether the input is differential or common mode. While the gain of the first stage is just under unity at -1dB, the second stage has an effective gain of -6db, so as to bring the overall differential gain of the receiver to -6dB.

Like the unbalanced shunt feedback receiver, the noise performance of the balanced two stage shunt feedback receiver is limited by the resistor values that determine the input impedance. As shown, the circuit puts out a modest -110.3dBV of noise on the output, about 0.2dB more than the single stage differential receiver. All the resistor values other than line drain resistors R1 and R4 could be set to 5.6kΩ, but doing so would increase the noise gain of U2, and reduce the signal gain relative to the noise gain of U1 if R6 and R7 were brought down to this value. As it stands, they are 10kΩ which gives U1 a gain of just under -1dB which is then effectively mirrored in the other direction by R8, bringing the noise gain of U2 to just over two.

To ensure stability when the non-inverting inputs of U1 and U2 are connected to the very low ground impedance, capacitors C7 and C8 are necessary. It is therefore a good idea examine the effect that the very slight low pass action of C7 in conjunction with resistor R7 will have on the receiver's overall CMRR, as it is not also applied to the cold input at the top of the circuit. A simple calculation implies that it will not be worse than 68dB at 20kHz, which is low enough to not worry about. The resistor value tolerances and coupling capacitor tolerances have effects on the CMRR of the receiver that are similar enough to that of the classic single stage receiver to forgo any additional analysis.

As already discussed, a shunt feedback input can perform mixing of multiple line sources without risk of cross coupling between those sources. The double shunt feedback receiver can do this with balanced and unbalanced line inputs, meaning that it lends itself well to being put into use as a multi-purpose input for both kinds of line inputs.

Shunt feedback balanced and unbalanced input circuit

Figure 11. Shunt feedback balanced and unbalanced input receiver

A receiver that can cater to both balanced and unbalanced inputs is drawn up in Figure 11 with a balanced input at the top of the circuit and an unbalanced one going into the hot side inverting stage. The resistor values for the balanced part have been increased by a factor of two so that the circuit can still realise a gain of -6dB in balanced mode, while realising unity when used in unbalanced mode. The noise performance will be about 3dB worse due to this. The unbalanced input goes through a two inversions, so unlike the shunt feedback unbalanced input, there is no signal inversion on the output. The input impedance is 10kΩ for the unbalanced input, and 20kΩ for the balanced input. Due to the higher resistor values in the op-amp feedback networks, the effects of CMRR limitations at high frequency as a result of stability capacitors C8 and C9 are made worse by 6dB. The shunt feedback balanced receiver can also be usefully employed to sum stereo balanced inputs into a mono one if two sets of input networks are used, similar to it's unbalanced counterpart in Figure 7.

Circuitry that combines both balanced and unbalanced inputs into a single receiver inevitably results in heavy compromise that limits the performance of either mode of operation by a very significant degree compared to dedicated receiver circuitry. Figure 11 is a good example of the sort of compromise that can be expected, it being the lesser of some greater evils that plague multi-purpose inputs. Although the extra versatility of a multi-purpose input may seem attractive, the cost of connectors is usually higher than that of the circuitry itself and compromising the performance to save less than a pound of electronics, or less for a simple unbalanced input, is simply not worth it in the estimation of the author.

Shunt feedback balanced optical limiter circuit

Figure 12. Fully balanced shunt feedback optical limiter

The feedback side of the second shunt feedback stage can also be put to good use, as the gain can vary from nothing at all, all the way up to the reasonable limits of the op-amp used. Simple equalisation, a gain control potentiometer, or even a variable gain element such as a resistive opto-isolator can be used as in Figure 12; a fully balanced optical limiter utilising only three op-amp stages to effect a balanced input, limiting circuit, and balanced output. The usual input networks are put into place, but within the feedback network of U2 an LDR, coupled to a suitable LED, is connected in parallel with feedback resistor R9 so that when the LED is illuminated, the gain of the balanced input drops below -6dB, all the way down to about -28dB at full illumination. The LED is driven from a full-wave rectified arrangement that takes further advantage of the third inverting stage, amplifier U3, beyond its primary function of inverting the limiter stage’s output to generate a fully differential balanced output.

With the values shown in Figure 12, there is a swift and remarkably linear limiting action that occurs once the peak output voltage reaches the combined threshold voltages of the rectifier diodes, D1 and D2, and the LED. With the diodes shown, and a typical 2V green indicator LED, limiting begins when the peak output voltage reaches 2.3V, which makes it especially useful as a last resort limiter connected to a digital audio interface for live recording, where overload may occur once the level surpasses 2.5V peak. There will still be a little peak short-term clipping as the photo-resistor will take about 10ms or so to respond to the light emitted by the LED, but this is usually inaudible, with the intention of the limiter to protect against the gross and prolonged distortion that would occur from heavy and sustained clipping if it were left out in such an instance. D1 and D2 are somewhat less common small signal Schottky diodes, although these devices can be found, and be expected for some time to be found in the repositories of the major component suppliers; a necessary evil as using silicon devices would raise the threshold to a less useful 2.7V.

The final inverting stage, U3, performs the necessary inversion needed to generate a fully differential output from U2’s output and drives the line in tandem with U2. The differential output is very conveniently used to provide full-wave detection to the LED through current limiting resistors, R13 and R14, along with the rectifier diodes, D1 and D2. All the stages use virtual earth amplifiers, which is quite aesthetically pleasing. As a result of the doubling of differential gain effected by the last inverting stage, along with its own noise, the differential noise across the output is over 6dB higher than the line input itself, at -103.7dBV. This is not too bad at all, all things considered, and is certainly much better than many JFET limiters lurking around in dusty, and hopefully forgotten corners, which have to reduce the signal level across the limiting element to such an extend so as to avoid severe distortion, that noise performance seldom comes within 12dB of the optical limiter of Figure 12. Distortion performance is very good indeed when there is no limiting action, although it does deteriorate to about 0.01% with 6dB of limiting, increasing by a factor of 5 or so at maximum attenuation, relative to the output signal level; not ideal, but certainly better than severe peak clipping and generally acceptable from such a simple circuit.

Double stage series feedback receiver

Although the classic single stage receiver, and the double shunt feedback receiver have their respective advantages, simplicity in the former, and robustness in the latter, they do not have either the best noise performance in the world, or the most friendly load impedance. A purely series feedback would not only be able to yield a very high input impedance that would not tax even the fussiest of equipment - such equipment is very rarely seen in the domain of the more serious balanced line - and reduce impedance bridging losses that might amount to just over half a decibel with a 600Ω source impedance into their necessarily low 10kΩ input resistor networks. Like the unbalanced series feedback receiver, an improved noise performance to the tune of a most euphonic 10dB might also be obtained.

Series feedback balanced input with gain

Figure 13. Double stage series feedback balanced receiver

Figure 13 illustrates a double stage series feedback receiver that is able to realise a high impedance input, in addition to a much better noise performance than the balanced receivers seen so far. High impedance input networks identical to the network used in the unbalanced series feedback receiver are put into use here to yield a 220kΩ input impedance from each side of the line to ground, or 440kΩ differentially. The cold input towards the bottom of the circuit diagram is then amplified by 6dB by amplifier stage U2, with equal value 2.2kΩ resistors in its series feedback loop. U2 then drives the bottom side of the feedback arm of amplifier U1, which via shunt feedback action, through R9 and R10, inverts the output of U2 onto the receivers output. U1 also amplifies the hot input by 6dB via series feedback action into the low impedance output of U2 so that a differential amplifier with a net gain of 6dB is the final product of the two stages.

CMRR is determined mainly by feedback resistors R9, R10, R11, and R12; helpfully all the same value, and a very common value at that of 2.2kΩ, making the sourcing of 0.5% parts very easy. Deviations to the resistor values within their tolerance have an effect are almost identical to those in the classic single stage receiver, so further discussion will be curtailed at this point. What might bear thinking about is the effect of capacitor tolerances, as with the now polyester input capacitors, C2 and C4, the cut-off frequency has been raised by a factor of three all the way up to 1Hz. If 10% parts are to be used, then the worst case effect of tolerance on CMRR at 50Hz will be 73dB, increasing to 79dB if 5% parts are used. Further applying the 12dB rule of what can reasonably be expected to occur in the real world gives a very acceptable 85dB at 10%, and 91dB at 5%; an order of magnitude greater than the limitations imposed by using 0.5% 2.2kΩ resistors. As the cut-off frequency of the RF filter networks is much higher as a result of the lower stopper resistor values, the effect of component tolerance in the RF shunt capacitors, C1 and C3, is improved by over 10dB relative to the previous two receivers.

The noise output of the receiver is similar to that of the previous two receivers that have already been laid out, despite the differential gain being some 12dB higher, it only presents -109.3dBV of noise on its output. Relative to the classic single stage receiver and the double shunt feedback receiver, with differential gain accounted for, this is just over 11dB better in comparison. Placing a theoretically noiseless 12dB attenuator on the output would give a desirable overall differential gain of -6dB, with a noise output of just -121.3dB; quite an unparalleled performance as far as balanced line inputs are concerned. Unfortunately, it is not possible for the receiver to be configured for a differential gain of any less than 6dB without seriously compromising headroom. This receiver topology must therefore only be used in applications where the differential input can be expected never to exceed half of the output voltage capability of the op-amps used, or some stage further down the line experience overload before the receiver itself does.

As an example of input voltage limitations in practice, if NE5532 op-amps running off ±15V supply rails are used to implement the line receiver, then the maximum peak differential input voltage will be 6.2V with the component values shown. This translates to 4.38V RMS, 12.83dBv, or 15.04dBu; there will be headroom margin of 11dB over a professional 4dBu reference level, cutting things a little fine for standard professional use. If a very well defined line source such as a DAC is used, then this is quite acceptable. Many professional audio DACs will allow adjustment to increase the output all the way up to 12dBu in some cases, which still leaves 3dB headroom to play with; a tolerable margin for the precise and absolute output level of any DAC. If a 12dBu input level is used as the maximum reference level, and a suitable attenuator is placed after the line receiver to induce a 12dB level reduction - 1kΩ and 330Ω in an L-pad arrangement, resulting in a suitably low output impedance of 248Ω - then this line receiver topology will outperform all of the others to follow in terms of noise and simplicity, making it an excellent choice for dedicated low noise DAC inputs.

Balanced input active crossover

Figure 14. Simple active crossover with a high impedance balanced input

If the input level cannot be guaranteed not to exceed the maximum input voltage of the receiver, then it is still possible to use it without experiencing any negative consequences if the one of the stages that succeeds it will reach overload before the receiver circuit itself does. Figure 14 demonstrates one such application; a simple second order Linkwitz-Riley active crossover. Active loudspeaker systems most often specify maximum SPL with an input voltage that is most often no greater than 2V RMS, with the power amplifier clipping or going into some form of non-linear limiting past this point. It is therefore quite possible and most appropriate to use the double series feedback balanced line receiver topology for this application as its low noise will be much appreciated by those who might seek to subjectively test such systems by pressing their ears firmly against the loudspeaker, drivers with a mind to discern quality in attempting to detect the hiss of white noise.

In Figure 14, the line receiver to the left not only ensures low noise recovery of the audio on the line, but also brings the level up by a useful 6dB to overcome the noise of the active crossover networks to the right. As the high frequency path needs to be inverted so as to be in phase with the low frequency path at the crossover point, the hot and cold inputs are swapped around from Figure 13 to effect an inversion. Ease of phase inversion is a blessing that is inherent to balanced line, but can also become a curse if it is accidental; with an unbalanced line it is immediately apparent as the signal is will be shorted to ground and will simply not be present. The low frequency path, a multiple feedback filter, incurs a second phase inversion and brings the low frequency path in phase with the line where it makes more sense to preserve absolute phase.

The preset resistors, R15 and R19, bring the level back down so as to set the input sensitivity at a suitable level in conjunction with the input sensitivities of the power amplifiers and loudspeaker drivers that the crossover will be connected to. If power amplifiers with a common-or-garden 1V RMS input sensitivity are put into play with the active crossover of Figure 14, then the preset resistors can be set to attenuate by 6dB so as to make up for the gain of the line receiver and produce an overall sensitivity of 1V RMS. This attenuation is no problem at all, as the op-amps in the line receiver and crossover are more than capable of sustaining a level of 2V RMS. The presets could even be turned down by half again to bring reduce the sensitivity to a requirement of 2V RMS without any risk of overload in the crossover. In this configuration, the noise performance is quite exceptional with an equivalent differential input noise voltage from the other end of the crossover of less than 114dBV; dead quiet.

Buffered classic receiver

So far the only balanced line input topologies discussed have had either noise or headroom limitations that bring their overall performance, either in terms of noise or headroom, below that of the series feedback unbalanced input. If a versatile general purpose balanced input with a noise performance comparable to a high quality unbalanced one is to be effected, then a little more complexity is going to be required to tackle some of the obstacles that lie in the way of a simple balanced receivers ability to be significantly quieter than -110dBV. The major source of noise in the classic single stage receiver all the way up in Figure 9 is the input loading and feedback resistors, which have to be set high enough so as not to put too heavy a load on the equipment driving the line. If these resistors were reduced by a factor of 10 or so, then their noise voltage could be reduced by a very helpful 10dB, although the input impedance of the differential stage would then be a factor of 10 lower, approaching 1kΩ. While a balanced line output cannot be reasonably expected to drive such a low impedance, a suitable op-amp buffer most certainly can; all the way down to 700Ω without any complaining.

Buffered balanced input circuit

Figure 15. The classic receiver buffered for improved performance

A rendition of this concept in practice is provided in Figure 15. Two separate high impedance line input networks are mated with buffers U1 and U2 which then drive the differential amplifier U3. The feedback and attenuator resistor values are much lower, almost 10 times as low as the classic single stage receiver at 750Ω, with the input resistors now combined into a series resistance of 1.5kΩ, setting the differential gain at -6dB. As a result of the buffers, not only is the input impedance substantially raised all the way up to 220kΩ, but the noise output of the receiver is brought down by a substantial 5.4dB over the single stage receiver to -116.1dBV; not too bad at all, but still not quite as good as the series feedback unbalanced input. Although the effects of resistor noise are reduced by almost 10dB, the effects of amplifier noise in U3, as well as the addition of noise from the buffers and their input networks limit the benefit to about half of what might have been expected if ideal noiseless amplifiers were to be used. Most of the noise comes from amplifier U3, operating at a noise gain of 3.5dB.

Like the single stage differential input, the lower arm presents the lowest drive impedance of 1.125kΩ. The resistor values could be reduced further, but this is not a good idea, not only as op-amp noise is dominating, but there is no doubling of values on the E24 resistor scale until the values drop all the way down to 360Ω and 180Ω respectively; far too low for the NE5532. Perhaps 1kΩ resistors could be used for R9 and R11, and then double paralleled to form 500Ω resistances to replaced R10 and R12, but the noise improvement would be a piecemeal fraction of a decibel on a good day. Using 1kΩ resistors would mean the lower arm impedance dropping down to 750Ω, which is as low as is sensible to go with even excellent op-amps like the NE5532. Driving 600Ω loads is of course possible as per the datasheet, but linearity is starting to degrade at this point. It might still be a good idea to use 1kΩ resistors in the manner alluded to as it does exclude an E24 resistor value in favour of a much commoner, and therefore inexpensive one, while also potentially reducing the number of different parts used facilitating slightly more efficient manufacturing; ideal for surface mount technology as an example. If good old-fashioned through hole technology is going to be used, then the author believes the arrangement as shown is best; the receiver will also be less current hungry as the load will be some 33% lighter.

Bearing the lower value feedback resistor values in mind, it is worth contemplating their effect on the receivers overall ability to drive the succeeding stage as these lower feedback resistor values are going to eat significantly into the op-amp's current drive capability. If a fully differential input is applied to the receiver, then U3 must drive a feedback impedance of 1.5kΩ; meaning that so as not to overload with op-amp, the stage following must not load the receiver's output by less than 1.3kΩ to prevent the total load dropping below 700Ω and reducing linearity. If the 1kΩ resistor arrangement laid out in the above paragraph is to be used, the lower limit is raised to 2.3kΩ.

CMRR once again depends mainly on the tolerances between resistors. It may seem that the extra close component to component tolerance of resistors of the same batch is not possible here, as there are two different resistor values used, but due to the fact that each arm has the same combination of resistors, the CMRR thankfully does not suffer and the 12dB better than worst case scenario can be expected. The input networks have the same effect on CMRR as they would in the double stage series feedback receiver of Figure 13; very little. With 0.5% resistors, a CMRR in excess of 50dB can be quite safely assumed.

Paralleled and buffered receiver

After a little exertion to beat the noise down by over 5dB in the previous section, through the addition of two op-amp stages, it is a good idea to have a look at what more can be done to make an electronically balanced receiver even quieter, perhaps as quiet as the series feedback topology described for an unbalanced connection at the beginning of the article. Within the remit of the NE5532 op-amp, three is not a particularly happy number for any discrete and separate input circuitry as two op-amps are contained within the 8 pin DIP package. Combining a single package between two separate channels to share an op-amp on each side is not advisable, as it will almost certainly introduce some small level of capacitive crosstalk as a result of the close proximity. It would not go amiss to put the fourth op-amp into use somehow further improving the noise performance of the balanced receiver. The main culprit for amplifier noise in the buffered classic receiver is the differential amplifier, operating at a noise gain 3 times as high as the buffers, relative to the output of the receiver. If a second differential amplifier stage is put in parallel with the one already there, then the signal voltages on the outputs can be summed, but the noise voltages, being inherently random, will partially cancel out towards a reduction of 3dB; a very useful reduction to the noisiest part of the circuit.

As well as examining the differential amplifier as a source of noise, the buffers themselves can be put under a little scrutiny to see whether their noise performance can also be improved, as they are effectively in series with the line input after all. The RF stopper resistors contribute about the same amount of noise to the input buffers as the follower amplifiers themselves, adding 3dB of extra noise. Balanced line outputs typically do not exhibit an output impedance of more than 600Ω, and cases where the output impedance is much greater than this are far rarer than in unbalanced line technology. Consequently, it is safe to assume that a little more leeway can be given to increasing the RF shunt capacitance and reducing the stopper resistor values to yield better noise performance from the input buffers. By switching to electrolytic input coupling capacitors and reducing the input impedance accordingly, the effects of low frequency current noise in the buffers, which are already quite small, can be reduced further.

Low noise balanced input circuit

Figure 16. Adding a parallel differential stage for low noise

A realisation of the improvements detailed in the two paragraphs above, Figure 16, is shown as the final iteration of what can be done to make a versatile balanced line input with quite superb noise performance. The input decoupling networks have been modified to lower the input impedance to a still-amicable 103kΩ, while the RF stopper resistors have been decreased by a factor of three, with the shunt capacitors increased by a factor of two to remain effective. The buffered input is then fed into a double differential amplifier arrangement, made up of U3 and U4, sharing a handful of resistors, R9, R10, and R11, in the non-inverting input arm so as to save on parts. Each differential amplifier has its own inverting arm separate from the other one, as otherwise minute variations between the two op-amps will be subject to their full open-loop gain, wreaking current-limiting havoc through the very low output impedances of the separate amplifiers if they do not have their own separate negative feedback paths. The two differential amplifier outputs are then summed through 10Ω resistors, R16 and R17, so that they share the burden of output current equally and do not allow tiny DC offsets to pull each amplifier into severe current limiting, rendering the circuit useless for anything other than electrical heating. As a result of all this tweaking and extra complexity, the line receiver's noise output is now at a very impressive -120.6dBV at a gain of -6dB; finally breaking through the 1μV barrier.

With electrolytic coupling capacitors C1, C2, C4, and C5, the biasing resistors have now been reduced from 330kΩ in Figure 16, to 150kΩ, so that the time constant is not so long that DC offsets will take more than a second or two to settle down below thump inducing levels at power-up. Similarly drain resistors, R1 and R5, have been reduced from 680kΩ to 330Ω with the same intention to let things settle at a swifter rate. Further optimisation of the RF filter networks for low noise has resulted in a noise improvement at the buffer outputs of 1.6dB, which helps things along nicely. The reduction of the values of R3, R4, R7, and R8 from 680Ω to 220Ω, is ultimately a compromise between trading off RF immunity for noise performance and it is the opinion of the author that the values shown give the best one. If R3, R4, R7, and R8 were to be eliminated completely and replaced with dead shorts, the extra benefit would only amount to 1.1dB; clearly not worth it if that extra bit of imperceptible silence, amounting to a fraction of a decibel at the other end, is going to be irrevocably polluted by RF hash. Increasing C3 to from 100pF to 220pF goes some way to mitigate the slightly lower rejection of the new RF filter, although the cut-off frequency is now about 50% higher at 3.28MHz. It might be possible to increase the value of C3 even further, but transformer balanced line outputs, which we may reasonably expect to be connected to the receiver typically exhibit an unsettling rise in output impedance as frequency rises, so the effect of the extra loading on frequency response may be enough to cause a significant droop at 20kHz. The RF filter is still quite excellent, however, and offers better rejection than the vast majority of contemporary equipment one can expect to find in a professional and consumer environment.

All the resistor values in the differential amplifiers are optimised to give the lowest noise possible without taxing buffer amplifiers, U1 and U2, into poverty of linearity. The non-inverting arm, consisting of 470Ω resistors R9, R10, and R11, common to both differential amplifiers, with a paralleling of the resistors on the bottom half of the arm, presents a load impedance of 705Ω to U1; the lowest impedance it can be reasonably expected to drive with good linearity. The impedance at the non-inverting inputs of U3 and U4 is just 157Ω, with a very low corresponding noise voltage of -132.9dBV. As the impedance on the non-inverting inputs is so small the effects of their combined input current noises, which will sum at this point are negligible. Resistor values feedback arms of the two differential amplifiers, R12, R13, R14, and R15, are also selected to be as low as possible, calculated to induce a loading of 1500Ω for each differential amplifier input when presented with a fully differential input. Both differential amplifiers are connected in parallel so a total loading of 750Ω results for the cold input buffer amplifier U2 to drive; again as low as is possible. To avoid overly complicating the schematic diagram, resistors R12 and R14 are shown as 2kΩ resistors which means that different batches of resistors have to be used, possibly degrading CMRR as component tolerances between matches will be greater than those within the same batch. It is strongly recommended that they are composed in practice by connecting two 1kΩ in series with each other, not only to do away with an E24 value that may be slightly more expensive, but to bring about the better CMRR that can be effected by using components from the same batch.

Each differential amplifier will be driving an effective feedback load of 1.5kΩ relative to its output with a fully differential input. If a maximum loading of 700Ω is to be put upon both U2 and U4 without fear of loss of linearity, as is the case for the NE5532, then it is possible for them to both work together and drive a load impedance of 656Ω downstream of the receiver, which is considerably better than the single differential amplifier arrangement of Figure 15. Current sharing resistors, R16 and R17, are selected to keep the effects of small differences in gain and offset voltages well at bay, while also not being too high in value so as to increase the combined output impedance of the differential to the point where it might adversely affect the operation of whatever might lie downstream of the receiver. A Baxandall tone control, for instance, may start to unacceptably sag at both ends of its frequency response if the driving impedance exceeds more than 10Ω, even if the control is set to flat. R16 and R17 are therefore set to the universally obtainable value of 10Ω each, with a parallel combination that determines the output impedance of the receiver at 5Ω. Considering the worst-case effects of offset voltages in both U3 and U4, 5mV full-range; the maximum offset voltage induced current drain from R16 into R17 is a very tolerable 500μA. In the real world it will almost always be less than 100μA. This is nowhere near high enough to eat into the output current capacity by any appreciable degree. What R16 and R17 do not do, however, is isolate the receiver from the destabilising effects of capacitive loading that 100Ω resistors in a typical line output do. If a significant capacitance is going to be put across the output of the receiver, then the usual precautions must be taken.

There may be more to be gained by adding even more differential amplifiers in parallel, in addition to also paralleling the input buffers, so as to reduce their noise as well in addition to driving the ever decreasing impedance of more differential amplifiers. At this point, it is a good idea when to consider throwing in the towel and being happy with what has already been accomplished. If we double the number of components and put two of the line inputs of Figure 16 in parallel, then the noise will be reduced by a further 3dB at a doubling of components. It is quite possible to put four in parallel, quadrupling the number of components for an improvement of 6dB, theoretically giving a very quiet -126.6dBV of output noise. Unfortunately, it is unlikely to the point of being beyond consideration, that the circuitry antecedent to the line receiver will be even to come anywhere near this; even the very best ADCs don’t get close, so designing such a line receiver may not make the slightest difference at all. There is also the effect of putting multiple buffer inputs in parallel with each other, summing the noise current which may start to hurt more than it helps if the line driving impedance is unsuitably high.

Balanced line receivers using 8 sections in parallel have been built before and put into commercial products, such as in the Cambridge Audio 840A amplifier, although surface mount technology was used to avoid occupying a very large amount of PCB real estate. The power consumption was, of course, very high for line level audio electronics with a total of 12 NE5532 packages per channel just for the line input, each drawing about 8mA a piece for a total draw of 96mA before a drive signal even appears on the line. This was not an issue for the unit in question, as it had beefy power amplifier supply rails to run off, but definitely a problem for a smaller pre-amplifier running off an external transformer power supply that may only be able to supply 100mA or so.

It is the opinion of the author that the circuitry detailed in Figure 16 is not only the best place to stop when through hole technology is used, but is actually preferable to that of Figure 15 - which is only really more feasible if there is a spare op-amp section going somewhere nearby - as it makes use of two whole packages facilitating a better PCB layout, and is also 4.5dB quieter. The performance is very good indeed, just on the brink of diminishing returns, and in the grand scheme of things chasing it down any lower simply will not yield any result worth the anguish and extra cost when the bottom line is reached. In comparison to the classic single stage receiver, it has a 10.1dB noise advantage for about twice the overall complexity; a very fair trade for some inexpensive components which will only amount to a fraction of the cost of the balanced connector the receiver circuitry will be wired up to. It is still interesting to note that even with this great improvement, Figure 16 is still 0.7dB noisier in terms of equivalent input noise voltage than the much simpler, but less versatile, Figure 13.


By now this article has reached well past the 20,000 word mark, and even so cannot be considered to be a complete discussion of all the line receivers that may be found or effected in the real world. Ultimately, as is always the case with electronic design, it all reduces to making the right compromise and keeping a good view of the bigger picture, particularly where noise is concerned. A more complicated balanced line input that uses 4 op-amp sections may be some 10dB quieter than the single stage version which manages -110dBV, but does it make a difference if the rest of the circuitry barely inches -100dBV and the overall effect of a very low noise input only results in a difference of a fraction of a decibel? There are also power and cost requirements to take into consideration also, as greater complexity usually results in greater power consumption, which may not be possible if a small power supply is going to be used. The extra space required on the PCB must also be taken into account. And so it goes on...

To the best of the author's knowledge, all the most important topologies have been covered, and hopefully there is enough detail around them to allow the reader to fully understand their operation, and the reasons behind the component values selected to effectively put them into functional use with real world line level, op-amps, and as part of a complete unit. Although NE5532 op-amps have been shown, there is no reason why any of the circuitry detailed throughout the article cannot be used with a more state-of-the-art device, such as the myriad of SMT only op-amps with slightly better noise performance, but from bitter experience, a rather shorter longevity of production. As long as there is a 600Ω drive capability, and a reasonable common mode input voltage range, there is no reason why more exotic parts cannot be used. The author will not be able to recommend any such devices, with the exception of possibly the LM4562 op-amp for the very last circuit diagram shown. The latest-and-greatest op-amps have a very unsettling habit of completely disappearing from the market after only a few years of production, costing well over several times what a NE5532 costs, and generally being a little too fast for their own good so as to cause stability issues with typical layouts. By specifying the NE5532 in all the schematic diagrams, which are after all a very mature technology that already works very well without any additional exotic seasoning, it is assumed from a retrospectively justified point of view that they will remain relevant for many years to come.