Fully Differential Amplifier

    PCBs are available for this project, and replace P87A and P87B.  Click on the PCB Image for the pricelist.

Projects 87A and 87B have been available for some time now, and although their usefulness is in no way diminished by this version, some applications really do demand the highest performance possible.  The basic circuit for this project is shown in the article Balanced Inputs & Outputs - The Things No-One Tells You, but has been adapted as a project.  The circuit has also been built and tested, and even with 1% resistors out of the box (i.e. not matched more closely as needed for best performance), I still measured an input common mode rejection of 63dB at 1kHz.  The PCBs for P87A and P87B are no longer available - both are replaced by this PCB.

Photo of Completed Board

A fully differential amplifier (FDA) can be used to convert balanced inputs to an unbalanced output, an unbalanced input to balanced outputs, or balanced inputs to balanced outputs.  The latter isn't particularly useful, because if you already have a balanced signal, passing it through another stage isn't going to add anything.  However (and within impedance constraints), it can be used to reduce the output impedance of a floating signal source.  The inputs can be driven from buffers (e.g. unity gain FET input opamps) if the signal source is very high impedance.  When used as balanced to unbalanced (or vice versa) the circuit can also be used in inverting or non-inverting mode, and it has as much flexibility as you are likely to need.

Common-mode rejection ratio (CMRR) depends on the tolerance of the resistors, but is also affected by the gain rolloff of the opamps at high frequencies.  This causes the CMRR to fall at frequencies above around 2kHz (opamp dependent), but this is inevitable unless you have access to 'ideal' opamps.  Since these exist only in theory and simulators, it's simply a fact of life that you have to accept.  Dedicated FDAs are no better, but some may 'hide' the reduction of CMRR at increasing frequencies by simply not providing a graph showing CMRR vs. frequency.

While the functions of an FDA are easily fulfilled with a dedicated IC such as the OPA1632, these are only available in a SMD (surface mount) package, and they are not inexpensive.  You still need to use precision resistors (although the number is reduced), and the recommended values in the datasheet are somewhat lower than ideal for most circuits.  There are many other FDA ICs, but most are SMD, with only a few DIP versions offered.  Some are designed for a maximum of 5V supplies (or ±2.5V) which might be alright in conjunction with a DAC or ADC, but isn't much use as a line driver for professional quality audio.

Many so-called 'differential amplifier' ICs are not fully balanced - they simply use the basic single opamp differential circuitry seen in countless circuits, and offer only a balanced input.  Others provide only a balanced output.  If an IC does not offer both a balanced input and a balanced output in the same chip, it's an instrumentation amplifier, a differential amplifier or a differential line driver.  A true FDA offers both balanced input and balanced output in the same package.

Despite claims to the contrary that you may see, using balanced circuitry does not improve 'sound quality' unless the sound is affected by noise picked up by input/ output cables.  This is a common myth, and it's simply not true.  The use of balanced connecting cables is only necessary to prevent noise, and it's not necessary in the vast majority of home systems.

The circuit diagram is shown below.  The selection of resistor values depends on the signal level and the desired input impedance.  Lower value resistors contribute less thermal noise, but reduce the input impedance such that it may be too low for some sources.  With high signal levels you can get away with (relatively) high value resistors, but it's suggested that you don't exceed 33k, with 22k being my personal maximum.  Where possible, lower values mean lower noise and are preferred.  Only one channel is shown - when (if) the PCB is made available it will be dual channel.  Note that supply bypass capacitors are not shown in the schematic, but they are essential or the circuit will oscillate (usually at RF).

All input and gain resistors should never have more than 1% tolerance, with 0.1% used if possible.  You can select the values from 1% types with an ohm meter to get close tolerance, which is cheaper than the 'real thing'.  For example, if you use 5.6k resistors, 1% tolerance means that they can be 56 ohms above or below the claimed value.  0.1% tolerance limits that to ±5.6 ohms, which can be measured with most decent multimeters.  The worst case common mode rejection ratio (CMRR) for both input and output is 40dB with 1% resistors, or 60dB with 0.1% types.  Closer resistor tolerance means better CMRR for input and output.  The exact value of matched resistors isn't important, only the accuracy of matching.  If they are all (for example) 5.3k (±5.3 ohms for 0.1%) that's all you need.  The resistors must be metal film, high stability types so they are not affected by thermal drift.

Figure 1 - Fully Differential Amplifier (One Channel Shown)

With a 1V balanced input, each output provides 1V, so when used as a balanced to unbalanced converter, the gain is unity.  For unbalanced inputs, if the unused input is left floating (not recommended as it will pick up some noise), the gain is 0.67 (-3.52dB) at each output, an overall gain of 1.33 (2.5dB).  When supplied with a 1V unbalanced input (with the unused input grounded), the gain is unity, but to both outputs in reverse phase.  This is again an overall gain of two at the balanced output.  C1 is optional, and its function is described in the 'Input/ Output Filters' section below.

The input impedance (at each input) is 3.73k unbalanced, or 7.46k for a balanced input.  The output impedance is set by R109 and R110 and is 100 ohms (at low frequencies).  These resistors can be reduced, but that may cause instability if the load is highly reactive (such as a long shielded cable).  Unlike most ESP designs, I've elected not to include coupling caps for the inputs and outputs, so the circuit is DC coupled.  Normally, any DC offset will be minimal, but the source must have no DC offset or it will be passed straight through the FDA.  It's worth pointing out that Projects 87A/ B (balanced receiver /driver respectively) also don't include coupling caps, and no-one has ever reported this as an issue.

Circuit layout is fairly critical, and it should be as symmetrical as possible to ensure that stray capacitance affects both opamp circuits equally.  An asymmetrical layout may seriously compromise CMRR.

Note that the input and output levels depend on the source and load impedances, so may be different from the ideal cases described above when in use.  While it is possible to change the default gain of the circuit, it's not usually required because any gain needed is provided by the circuitry of whatever is being adapted to/ from balanced connections.  If you wanted a gain of two (close enough), you simply increase the values of R2, R4, R6 and R8 to 11k.  Increasing the gain will reduce CMRR.

The circuit shown works very well, and I've run tests on a Veroboard version in my workshop.  The distortion was measured at around 0.0007% (THD + noise), and response extends to well over 100kHz.  I tested it with a 30kHz squarewave, and could not fault it in any way.  Ultimately, my test equipment was the limiting factor, not the circuit.  Input CMRR is better than 60dB up to 100kHz, and although not measured, output CMRR will be similar.

Power connections are not shown in the circuit.  Pin 8 for most dual opamps is positive (+5 to +15V) and pin 4 is negative (-5 to -15V).  The IC supply must be bypassed with a 100nF multilayer ceramic as close to the supply pins as possible, with board-level bypassing using not less than 10µF electrolytic caps from each supply rail to earth/ ground.  Poor bypassing can lead to unexpectedly high distortion and increased noise.

Figure 2 - Fully Differential Amplifier Connections

Connections to/ from the FDA are as shown above.  The connection for 'unbalanced in, balanced out (inverting)' isn't shown, but can be obtained either by grounding the non-inverting input or swapping the outputs.  Unused outputs must not be grounded - they are left floating, and are shown as 'N.C.' - not connected.  This one circuit can replace individual balanced-to-unbalanced and unbalanced-to-balanced converters, and offers performance that's only limited by the opamp you choose and resistor tolerances.

If you have to ensure that RF or other high frequency noise does not cause problems, you may need to add an input and/ or output filter.  This becomes critical, because the capacitors have to be carefully matched to ensure that CMRR isn't affected at high frequencies.  Ideally, any caps used should be matched to the same tolerance as the resistors, which is likely to be an arduous task.  The caps must be high stability types, so don't use 'high-k' ceramic caps, which have very poor stability with both voltage and temperature.  Polyester is alright, polypropylene is better, or if you can get them, use polystyrene.  You can also use G0G (aka NP0) ceramics, but you'll have to match them yourself.  I'll leave this to the constructor, but an example is shown in the article referenced ^[ 1 ].

Some basic input filtering that doesn't affect the CMRR can be provided by adding a small cap (C101) between the non-inverting inputs of each opamp.  A value of 100pF to 220pF will help get rid of most high frequency noise above 70kHz or so.  Adding ferrite beads to the inputs and/ or outputs (right at the point of entry into the chassis) can also be helpful, but most of the time the circuit will be used as-is.  For truly intractable noise issues, use a good quality transformer with an electrostatic shield in front of the balanced input.

Adding an output capacitor will also help to maintain CMRR at high frequencies, and may help to prevent RF interference from getting back into the circuit.  With the value of 2.2nF shown in Figure 1, the response is less than 0.3dB down at 50kHz (that includes a 100pF cap for C101).  The effect within the audio band is negligible (0.07dB at 20kHz).  Note that the output cap must be placed after the output resistors, or opamp instability will result.

An output transformer can also be used, but you'll have to ensure that there's no DC offset, and the values of R9 and R10 will need to be reduced to provide a low impedance source to the transformer.  Around 22 ohms is the minimum I'd normally suggest, but it might be possible to use less.  This has to be tested with the transformer you intend to use.  Driving a transformer with a balanced driver may seem like overkill, but if you need particularly high performance it's a simple (if expensive) solution.

In some cases, constructors may want (or need) a fully balance preamp.  This isn't a common requirement because most sources are unbalanced, but a few 'high end' manufacturers seem to have convinced some people that balanced systems sound 'better'.  This is simply untrue, but "never let the truth ruin a good story" seems to be a common theme in the high-end market.  It's unrealistic and completely un-necessary to make a preamp balanced throughout, and getting a 4-gang pot with tracking to within 1% or better is necessary to preserve the balance while not degrading the CMRR.  Such pots are unobtainable in conventional rotary styles, so the only option would be to use 'digital pots' with a micro-controller to handle changing the volume.  It's far easier (and a great deal cheaper) to use an FDA at the input and output of a more conventional preamp.

Figure 3 - Balanced-In, Balanced-Out Preamp Example (One Channel Shown)

The drawing shows how this may be done.  An input FDA converts the inputs from balanced to unbalanced, the preamp is completely 'normal' using (for example) P88 or P97 preamp boards, followed by another FDA at the output to generate the balanced output signals.  Electrically, the system appears fully balanced from input to output, but avoids the need for unobtainable or hard to use parts.  This is the general scheme of all analogue mixing consoles for example.  The inputs and outputs are balanced, but nearly all internal circuitry is unbalanced.

To do otherwise would make even the most cost-effective mixing console so complex (and costly to build) that no-one would be able to afford it.  There's no reason (sensible or otherwise) to try to maintain the signal as balanced throughout.  Doing so will almost always result in reduced performance overall, as well as vastly more expensive pots and double the number of parts.

The performance of fully integrated versions of the circuit may (at least in theory) be marginally better, but remember that you can choose the opamps you use, and there's no reason not to use the very best available if you need the highest performance.  Unlike a dedicated IC, the opamps can be expected to be around for a long time, because they are used in their millions in innumerable pieces of equipment, from home hi-fi to audio mixing consoles and everything in between.

The circuit shown is certainly not the only way to create a fully differential amplifier, but of those I've looked at it's by far the easiest to implement, and its performance is better than any of the other variations I've looked at.  While complete integrated circuit versions may require less PCB space, their cost is greater and most are only available in SMD packages, with many being limited to only a 5V supply.  Availability (or lack thereof) in some countries is another factor, but the circuit presented can use opamps that are available anywhere.  While not all can offer the performance you can get with the NE5532, LM4562 or equivalent, in many cases even a lowly TL072 may be all that's needed for the purpose.

It may not be apparent, but with exact value resistors the CMRR when used with a balanced input and balanced output is theoretically infinite, limited only by the CMRR of the source itself.  Real world limits on resistor tolerance will obviously affect this, but you can expect the overall result to be extraordinarily good.  However, balanced-in to balanced-out is not a common requirement, so you may never get to experience just how good the CMRR can be.  In general, you should be able to achieve a CMRR of around 80dB with careful resistor matching.  Any stray capacitance will affect the result - as little as 10pF will degrade performance above 10kHz.

This is a versatile topology that uses just one dual opamp, and as noted will generally have more than acceptable performance even if you don't bother to select the resistors for 0.1% tolerance.  It has a reasonable input impedance, which can be increased if necessary.  However, you'll have to accept the additional thermal noise from high value resistors, but up to 10k will add little extra.  If you really do need high or very high input impedance, use FET input opamps as buffers in front of the FDA.  This will allow the input impedance to be as high as you like, but it's critical that the opamps used have reasonable precision, or the input balance will be disturbed and CMRR will be reduced.  A single opamp can be used as a buffer if the input source is unbalanced.

1 - Balanced Inputs & Outputs - The Things No-One Tells You - ESP
2 - OPA1632 FDA Datasheet

Main Index

Projects Index