Balanced vs. Unbalanced Analog Interfaces

If you look at the back of any Benchmark product, you will find balanced XLR analog-audio connectors. As a convenience, we also provide unbalanced RCA connectors on many of our products. In all cases, the balanced interfaces will provide better performance.

We build our unbalanced interfaces to the same high standards as our balanced interfaces, but the laws of physics dictate that the balanced interfaces will provide better noise performance.

This paper explains the advantages of balanced interfaces as compared to unbalanced.

Noise Sources

There are several sources of noise that can contaminate the audio when two audio products are connected together with a cable. These noise sources include ground loops, radio interference, magnetic interference, thermal noise and noise from the input and output buffers.

Signal to Noise Ratio (SNR)

The signal to noise ratio (SNR) is the ratio between the maximum signal voltage and the idle-channel noise voltage. This ratio is usually expressed in dB. The higher the ratio, the better the performance. To keep noise completely inaudible, the A-weighted SNR in dB should exceed the peak sound pressure level (SPL), at the listening position, expressed in dB SPL. When this condition is satisfied, the noise will be below the 0 dB SPL threshold of normal hearing.

The peak sound pressure should not be confused with the average levels read by an SPL meter. The peak level will be substantially higher but can be calculated from the amplifier output voltage and the speaker efficiency. This calculation is covered in some of my other application notes.

Most audio systems produce some audible noise. In other words the noise level emitted from the loudspeakers exceeds 0 dB SPL. Many listeners have accepted this distraction as a necessary part of audio electronics, but the technology exists to produce entire systems that emit no audible noise. Balanced interfaces are one important feature of these state-of-the-art systems.

SNR of Audio Interfaces

In this paper we will be focusing on the capabilities of the interfaces that connect audio components together to form a system. In many systems, the interfaces limit the noise performance of the overall system. To avoid this limitation, the interfaces should provide better performance than the poorest performing component in the signal chain.

Noise Figure

If we want an interface to impose no more than a 1 dB reduction on the system SNR, it will need to have an SNR that is at least 6 dB better than the lowest performing device in the signal chain. If a noise source causes a 1 dB reduction in noise performance, we call this a 1 dB "noise figure".

If our interface SNR is 12 dB better than the worst component in our system, then the interface will impose a 0.27 dB noise figure on the system. 12 dB is a 4:1 ratio. At this 4:1 ratio of component noise to interface noise, the interface still causes a slight reduction in the system SNR. This demonstrates the need for interfaces that are much quieter than the devices they connect.

Example - Connecting a DAC to a Power Amplifier

If we connect a Benchmark DAC3 (128 dB A-weighted SNR) directly to a Benchmark AHB2 monoblock amplifier (135 dB A-weighted SNR) we should be able to achieve a system SNR of about 127 dB. The SNR of the AHB2 is 7 dB better than that of the DAC3 so it will impose a noise figure of just less than one dB on the DAC3.

To avoid more than an additional dB increase in noise, the interface between the two boxes will need an SNR that is at least 6 dB better than the combined 127 dB SNR. This means that the interface needs to have an SNR of at least 133 dB. If we attempt to use the 2 volt (8.2 dBu) RCA output on the DAC3, we will not achieve the desired result. If you do the math (8.2 dBu - 133 dB) , the unbalanced outputs and inputs would need to achieve a noise level of -124.8 dBu. This is nearly impossible because this is equivalent to the thermal noise produced by a 600-Ohm resistor. The unbalanced input and output stages would need to use very low impedances and we would need to hope that we didn't pick up any additional noise from ground-loop interference.

In our example, the upstream unbalanced output on the DAC3 would be the limiting factor in terms of noise. The DAC3 has high-quality low-noise unbalanced outputs, but these cannot support the full 128 dB SNR of the DAC3. Under ideal lab conditions, we would suffer about a 10 dB reduction in system SNR when connecting the DAC3 to the AHB2 using the unbalanced outputs on the DAC3. Outside of the lab, ground-loops and electromagnetic interference (EMI) could add additional noise to this unbalanced interface.

If we use the balanced outputs on the DAC3, this interfacing task is easy and we will achieve our system SNR goals while adding immunity to interference.

Why are there no Unbalanced Inputs on the AHB2 Power Amplifier?

If you look at the back of the AHB2 you will see that there are no unbalanced inputs. Unbalanced inputs have no business being on a device that delivers a 135 dB SNR. Unbalanced interfaces operate at 2 Vrms which is about 8.2 dBu. If you do the math (8.2 dBu - 135 dB - 6 dB), you will find that the noise on the interface would need to be -132.8 dBu to achieve a 1 dB noise figure. To put this in perspective, a single 91-Ohm resistor produces a thermal noise level of -132.8 dBu, and the best microphone preamplifiers have an equivalent input noise (EIN) of about -130 dB. If we want to use 2 Vrms input levels, we would need an input amplifier that is better than any microphone preamplifier ever built! Due to the laws of physics, this requirement is impossible to achieve. 

The only solution is to use higher signal levels. These are only available from balanced outputs.

For compatibility with consumer equipment, the AHB2 has three input gain settings: The high-gain mode supports 2 Vrms inputs, the mid-gain mode supports 4 Vrms inputs, and the low-gain mode supports professional +24 dBu (12.28 Vrms) inputs. To achieve maximum performance, it is essential to feed the amplifier with professional studio-level inputs while using the low-gain setting.

Increase the Signal Levels

One of the easiest ways to improve the SNR of an interface is to increase the signal level. For example, if we double the signal voltage, we are increasing the level by 6 dB and this increases the SNR of the interface by 6 dB.

Studio-Grade Balanced Interfaces

Professional balanced interfaces generally operate with much higher signal levels than unbalanced consumer interfaces. This voltage difference gives these balanced interfaces a significant SNR advantage over unbalanced interfaces. Most RCA interfaces operate at a maximum signal level of +8.2 dBu which is 2 Vrms. In contrast, professional balanced interfaces usually operate at a maximum signal level of +24 dBu which is 12.28 Vrms. If you do the math (24 dBu - 8.2 dBu) you can see that the signal level is 15.8 dB higher on the professional-grade balanced interface. If the noise is the same on both interfaces, the balanced interface will provide almost a  16 dB improvement in the interface SNR.

But, balanced interfaces always require dual output buffers and dual input receivers. These additional active devices contribute some noise and this tends to reduce the SNR improvement by about 3 dB. Taking this into consideration, the interface SNR of a professional balanced interface is still about 13 dB better than that of an unbalanced consumer interface.

In addition, balanced interfaces provide rejection of many types of interference. This immunity to interference can provide a 50 to 100 dB reduction in these unwanted noises. This immunity to interference is usually more than enough to keep the interference inaudible.

Professional interfaces are more expensive to build. The high signal levels generally require the use of +/- 18 volt power supplies within the audio product. To save costs and reduce the power consumption, consumer products usually use much lower supply voltages. As a result, it is rare to find consumer audio products with balanced interfaces that can support professional signal levels. These products have dumbed-down balanced interfaces that operate at much lower voltages.

Consumer-Grade Balanced Interfaces

Many high-end consumer products have balanced interfaces, but they operate at a maximum level of 4 Vrms which is + 14.2 dBu. This is 10 dB lower than the level used in professional interfaces. This means that the 13 dB advantage provided by a +24 dBu balanced interface is reduced to just 3 dB when operating at a maximum level of +14.2 dBu. Consumer-grade balanced interfaces are definitely a step better than unbalanced interfaces, but the signal levels are too low for use in very high perfomance systems.

Benchmark D/A converters are equipped with professional-grade +24 dBu outputs. These outputs have 10 dB passive pads that can be engaged in order to drive consumer-grade 4 V balanced inputs. If you find you need these pads, it is a good indication that the downstream device is limiting the SNR performance of your system. Likewise, the AHB2 has a gain setting that supports inputs from 4 Vrms consumer-grade balanced outputs. Again, these consumer-grade devices will be the weak link in the system.

Check the specifications and look for balanced interfaces that support professional signal levels. Consumer-grade balanced interfaces may look like professional interfaces, but they do not provide the same level of performance.

Common-Mode Noise Rejection

The largest benefit of balanced interfaces comes from their ability to reject common-mode noise signals. Balanced input receivers have active differential amplifiers or passive transformers that respond to the difference between the + and - input pins. The ground pin on the XLR connector is only used for shielding. The ground connection is not part of the audio signal and it is  ignored by the input amplifier or input transformer.

Ground currents, and differences in ground voltage, can create noise signals that are identical on both the + and - pins. This common-mode noise will cancel and be rejected by a transformer or a well-trimmed differential amplifier.

CMRR

The common-mode rejection ratio (CMRR) is a measure of how well a balanced receiver rejects common-mode noise. A high CMRR is an indication that the balanced receiver will reject most of the noise voltage caused by ground loops and other sources of common-mode interference. Transformers generally provide a very high CMRR but may add distortion and frequency response problems. Active differential amplifiers can provide better transparency than transformers and can provide a very high CMRR if they are well trimmed.

Good differential amplifiers provide at least 50 dB of rejection at AC line-related frequencies. This is usually enough to reduce AC ground loop interference to inaudible levels. Well-trimmed differential amplifiers, such as those used in Benchmark products, may provide a CMRR of 70 to 100 dB at AC line-related frequencies.

Differential Amplifiers - The Key Components in Balanced Receivers

There are some consumer products that have "balanced" inputs with a CMRR of 0 dB. Let's just call these "fake" balanced inputs. The input is equipped with an XLR connector, but the - pin (pin 3) is ignored. These are just unbalanced inputs wired to an XLR connector. They offer no advantage over unbalanced RCA inputs. In most cases, pin 2 of the XLR is tied directly to the center contact of a nearby RCA input. This makes the XLR jack nothing more than an adapter.

There are other devices that use the + and the - pins, but they fail to remove common mode noise before the signal leaves through a balanced output. These devices simply pass the common mode noise on to the next component in hopes that the next component will remove the common mode noise. If you can't find a specification for CMRR, the balanced input may not have a differential amplifier. If this is the case, the input is just a pair of unbalanced inputs feeding a pair of unbalanced outputs.

Differential Amplifiers in D/A Converters

High-quality D/A converter chips use balanced outputs. The purpose of these balanced outputs is to increase the signal level by a factor of two (6 dB) while providing a means for removing the common-mode distortion produced by the converter chip. This common-mode error tends to be odd-harmonic distortion and is not musically pleasing. This low-level distortion can change the character of musical voices and detract from the music. This common-mode distortion is only removed if the D/A converter feeds a well-trimmed differential amplifier.

I have seen many D/A converters that completely omit the differential amplifier. This omission is actually common practice in most "high-end" D/A converters. Unfortunately this leads to very unpredictable performance. The D/A converter will still measure well when it is connected to the balanced input on an audio analyzer, but may not perform nearly as well in a typical system. A good analyzer will always have an excellent CMRR and this will reject the common-mode distortion produced by the D/A converter chip. In a real system, the D/A converter could be driving a balanced input with a poor CMRR and the common-mode distortion would not be rejected. Furthermore, If this D/A converter also has unbalanced outputs, these outputs will be contaminated with distortion that could have been removed by a differential amplifier. Once this distortion reaches an unbalanced output, it cannot be removed by a downstream device.

To avoid these problems, Benchmark D/A converters include well-trimmed differential amplifiers to remove the converter common-mode distortion before it reaches any of the outputs. From a distortion standpoint, the balanced and unbalanced outputs on a Benchmark D/A converter will have identical THD performance. But, from a noise standpoint, the balanced outputs provide better performance.

Solving Ground Loop Problems

In a balanced system, ground loops generate common-mode noise, and balanced receivers can provide near-perfect (50 dB to 100 dB) rejection of this troublesome noise. For this reason, balanced interfaces are considered absolutely essential in professional environments.

Balanced interfaces are also the solution to the many ground loop problems that occur in a complex home hi-fi system. Cable boxes, TV antennas, computer USB ports and AC ground pins may all be at different ground potentials. By code, cable TV and antenna cables must be grounded where they enter the building. These ground points are rarely the same as the ground delivered on the ground pin of an AC outlet. USB shields may be grounded to a relatively noisy computer mother board and these can also cause additional ground loops.

Galvanically isolating a USB cable is a band-aid fix to ground loop problems. This band-aid fix may partially reduce audio interference but it can increase the RF emissions produced by the USB cable. Ideally, the USB shield should be tied to a chassis ground at each end of the cable. When this is done, any ground-loop interference is easily removed through the use of balanced analog audio interconnects.

AC currents flow between various grounds when audio components are connected together. In most cases these currents flow through the ground shield on the outer layer of the audio cables.

Unbalanced cables use the shield to form one of the two audio conductors. This dual use of the shield makes unbalanced interfaces very sensitive to ground voltage differences between the two ends of the cable.

In contrast, balanced cables use a dedicated shield. This outer shield is not one of the audio conductors. Noise on the shield is well isolated from the audio circuit.

RF Rejection

The braided copper or foil shields on balanced and unbalanced cables are intended to protect against radio-frequency (RF) interference. This "Faraday shield" isolates the internal wires from radio interference.

In an unbalanced cable, this shield is also used to carry the audio ground. When the shield serves a dual purpose, some of the RF energy can contaminate the audio. This double use of the shield can make unbalanced interfaces somewhat more sensitive to RF interference. This may lead to audible interference from nearby radio stations and cellular phones. In other cases, the RF interference will increase the distortion produced by the audio system.

In a balanced cable, the shield has a dedicated purpose. It is only a shield. It is not used to carry the audio signal. For this reason, balanced cables can provide slightly better shielding against RF interference. In most cases, this shielding will be sufficient to prevent any audible interference from radio signals.

Magnetic Rejection

The braid or foil shield on the outside of a cable cannot provide any shielding against magnetic fields. This can be demonstrated with a magnet and a copper penny. The force of the magnet will pass through the copper penny without much change. Likewise, magnetic fields can pass through multiple layers of copper and foil shielding without being attenuated.

Power supplies in audio devices, computers, and chargers produce AC magnetic fields that can cause interference in an audio cable. When an audio cable passes near an AC magnetic field, the audio conductors act like the secondary winding in a transformer and picks up a magnetically-induced voltage. This voltage may be AC line hum, AC line-related buzz, or a variety of other unwanted and ugly sounds.

Balanced interfaces can only reject magnetic interference when both the + and - conductors receive exactly the same common-mode interference. If one conductor receives more magnetic interference than the other, the rejection is greatly reduced. In practice, one internal wire will be closer to the magnetic interference and it will see a stronger magnetic field. This imbalance reduces the rejection.

If four-conductor star-quad cable is used, the rejection of magnetic interference can be improved by about 20 to 50 dB compared to standard 2-conductor balanced cable. Star-quad cable uses two conductors for the + audio and two for the - audio. The precise geometric configuration of these conductors causes an equal common-mode pick-up of magnetic interference on both of the + and - conductor pairs. This magnetically-induced common-mode voltage will be rejected if the balanced receiver has a good CMRR.

In practice, star-quad cable is rarely needed for short line-level balanced interfaces, but it is almost always beneficial on microphone feeds. Nevertheless, star-quad cable provides an extra margin against magnetic inference when it is used with line-level balanced interfaces. For this reason, Benchmark recommends star-quad cable for all balanced interconnections. Star-quad cables are good insurance against unexpected sources of magnetic interference.

Thermal Noise (Johnson Noise)

Every electrical component produces a certain amount of thermal noise (known as Johnson noise). This includes passive components such as resistors. Yes, passive components produce electrical noise! This noise is caused by the thermally-induced random motion of electrons.

For example, a 10 k Ohm resistor produces a noise level of -112 dBu over the audio band at room temperature. If you want to achieve a 130 dB SNR through this 10 k resistor, the signal level will need to be 130 dB higher than -112 dBu which is 18 dBu. If you are using consumer-level 2 Vrms (8.24 dBu) unbalanced signals, you will be about 10 dB short of achieving a 130 dB SNR after the signal passes through a 10 k resistor. If you want to achieve a higher SNR, you have two choices; reduce the value of the resistor, or increase the signal level.

Any practical audio circuit contains many resistors. The noise contributions of each resistor are cumulative. For this reason, drive impedances must be kept low and signal levels must be kept high. 2 Vrms consumer unbalanced interfaces are woefully inadequate. Unbalanced interfaces will usually limit the SNR to about 100 dB to 110 dB in a very well engineered product. Consumer-grade products often deliver an SNR of just 80 to 100 dB over unbalanced interfaces.

The Myth of "Unbalanced" Headphones

This discussion would not be complete without mentioning that there is no such thing as an unbalanced headphone transducer (with the possible exception of DC-biased electrostatic headphones such as those manufactured by Stax).

Headphone transducers respond to the voltage difference between the two wires that feed them. They have perfect rejection of common-mode interference because there is no path to ground or to any other conductor. In other words, there is no path for ground loops.

Headphone transducers are electrically isolated from everything other than the two wires that feed them. It doesn't matter if both conductors are driven differentially or if only one conductor is driven. The headphone transducer will reject common-mode noise.

There are some advantages to using separate wires to feed the left and right transducers, and there are some advantages to using XLR connectors instead of TRS headphone jacks, but none of these have anything to do with balanced vs. unbalanced drive. Balanced drive can provide twice the audio voltage for a given power supply voltage. XLR connectors often provide better electrical connections than TRS jacks. A separate return pin for the left and right channels can reduce crosstalk but channel separation is not really a concern with headphone listening.

In the context of this paper it is important to understand that headphone transducers always behave like perfect balanced inputs. It doesn't matter how they are driven. Headphone transducers provide perfect rejection of common-mode noise because they only have wires. The current flowing through these wires will be equal and opposite because there is no other path for the electrons to flow. 

The Myth of "Balanced" AES Digital Interfaces

In the early days of digital audio, the Audio Engineering Society (AES) decided that it would be handy to use existing analog XLR cabling to carry digital audio. In my opinion, this was a really bad idea!

Balanced connections provide no advantage with digital audio signals. Digital signals provide substantial immunity to noise. The data format used to carry digital audio was designed so that it would have no spectral content at audio frequencies. This feature allows the use of a simple high-pass filter to remove AC line-frequency interference.

Digital pulses produce substantial energy at RF frequencies. The shape of these pulses is only preserved when the cable has a controlled impedance and is terminated with resistive loads that match the cable impedance. Existing analog XLR cables had a variety of impedances and these impedances were not well controlled. Analog cables proved to be completely unreliable for digital signals and special digital XLR cables had to be created. So much for using existing cables! We now have analog and digital audio cables that look nearly identical. Digital cables are acceptable for analog audio, but analog cables cannot be used for digital audio.

The AES initially gave us a standard (AES3) for digital audio using XLR connectors and special 110-Ohm cable. But, it has been shown that coaxial cables provide better signal integrity over long transmission distances. Coaxial cables support cable runs as long as 1000 m while the 110-Ohm cable is limited to about 100 m. The video industry created a separate standard for digital audio over 75-ohm coax. As a result, the AES3 standard was updated to include digital audio over coaxial cable.

Given a choice, we would strongly recommend using unbalanced coaxial digital connections instead of balanced XLR digital connections when making long cable runs (over about 50 m). Some professional products use BNC coaxial connectors instead of RCA connectors. Consumer and professional digital audio formats are designed to talk to each other. Simple adapters can be used to connect RCA and BNC connectors. Transformers are required when adapting between balanced and unbalanced digital audio connectors.

Typical Performance of Audio Interfaces

Based on what I have seen while testing audio products in the lab, I have attempted to put together some typical performance numbers for audio interfaces. These are only approximate numbers, and it is possible to do better with careful engineering. Nevertheless, I believe these are fairly typical performance numbers.

• Unbalanced 2 Vrms interfaces are rarely capable of delivering an SNR better than 100 dB (the SNR equivalent of a 17-bit digital system).
• Consumer-grade 4 Vrms balanced interfaces may deliver an SNR up to about 125 dB (the SNR equivalent of a 21-bit digital system).
• Professional-grade 24 dBu balanced interfaces may deliver an SNR up to about 135 dB (the SNR equivalent of a 23-bit digital system).

As a general rule, professional-grade balanced interfaces are the only interfaces that can deliver performance that matches that of today's best converters and amplifiers. In contrast, unbalanced interfaces tend to limit a system to CD-quality performance.

Summary

Professional-grade balanced analog audio interfaces can provide a 12 to 16 dB SNR advantage over unbalanced interfaces due to the high +24 dBu signal levels used on balanced interfaces. Consumer-grade balanced interfaces can only provide 3 to 6 dB SNR advantage due to the relatively low +14 dBu (4 Vrms) signal levels.

In addition, the differential amplifier or transformer in a balanced input can provide an incredible 50 to 100 dB rejection of ground-loop interference. This is usually sufficient to reduce ground-loop interference to completely inaudible levels. 

In an unbalanced interface, the shared use of the shield places ground-loop currents in the audio path. Unbalanced interfaces are very sensitive to ground currents flowing between audio components. This is not a problem with balanced interfaces due to the use of dedicated audio conductors.

Copper braid and foil layers provide shielding against RF interference. In a balanced cable, the shield does not carry the audio signal. The audio conductors are fully surrounded by the shield but are electrically isolated from the shield. In an unbalanced system, the RF shield also serves as the audio ground. This dual use of the RF shield, in an unbalanced system, causes a slight increase in susceptibility to RF interference.

Copper braid and foil shields do not provide any protection against magnetic interference. Magnetic fields easily pass through copper and foil. If star-quad cables are used in a balanced system, magnetic interference can be rejected by the CMRR of the balanced input receiver. In a balanced system, 4-conductor star-quad cables can reduce magnetic interference by 20 to 50 dB when compared to standard two-conductor balanced cables.

These numbers should be hard to ignore, but the hi-fi industry has been slow to change. Many high-end audio products still are not equipped with balanced interfaces. Others have consumer-grade 4 Vrms balanced interfaces. These are a partial step in the right direction.

The facts show that it is virtually impossible to achieve state-of-the-art audio performance using unbalanced interfaces. We see this in the lab when we measure balanced and unbalanced interfaces under ideal well-controlled conditions. Outside, in the real world, the advantages of balanced interfaces are larger than a set of balanced vs. unbalanced specifications would indicate on a product data sheet. The differences can be extremely large when ground loops, RF interference, and magnetic interference are encountered in a typical audio system.

Our recommendation? Avoid unbalanced (RCA) analog interfaces whenever possible! Look for professional-grade balanced interfaces when buying audio products. Look for CMRR specifications on balanced inputs. Consider replacing audio devices that do not support balanced interconnects. These unbalanced-only devices are probably a weak link in your audio chain.

Additional Reading

"What's Special About Star-Quad Cable?" - SOS, November 2009