Design Feature:July 20, 1995
Bonnie Baker,
Jerald Graeme,
Graeme Consulting
When you try to reduce the noise in an analog circuit, you'll eventually hit a wall if you focus only on the parts of the circuit that you can see. What you can't see can hurt: Time-varying electric and magnetic fields couple energy into the signal path by way of parasitic inductance and capacitance. The coupled energy becomes background noise. Reducing the noise may force you to pay attention to layout or shielding when you'd rather spend time on the circuit itself.
Understanding how noise gets into a circuit will help you to keep noise out or to render it harmless. You should begin by separating noise sources from susceptible components. However, you can reduce noise further by taking advantage of op-amp circuits' differential nature. Balancing the impedances often allows you to convert noise sources into common-mode signals that differential amplifiers reject. Finally, minimizing inductive-loop areas reduces the voltage that such loops radiate and pick up.
To limit the effects of external noise sources, pay attention to the circuit location, topology, layout, and shielding. Electric- and magnetic-field noise enter a circuit through parasitic elements--mutual capacitances for electric-field noise and mutual inductances for magnetic noise. In both the electric- and magnetic-field cases, keeping the susceptible element away from the noise source minimizes the coupling. Although shields can divert electric and magnetic fields from susceptible elements, the two types of fields often require different shield materials.
To further reduce magnetic coupling, identify op-amp circuits' potential pickup loops and evaluate the corresponding noise gains. This procedure tells you where you'll find the biggest payback from minimizing the loop area. The same procedure also defines circuit loops that potentially radiate, rather than receive, magnetic signals. By selecting op-amp circuits that have differential inputs, you can often make noise signals appear as common-mode signals, which the differential circuits' common-mode rejection (CMR) then attenuates. Balancing the impedances that drive the differential inputs extends this CMR benefit to a wide variety of circuits.
Electric-field coupling, such as from ac power lines, introduces noise through the capacitance that exists between any pair of objects. The objects serve as the plates of a capacitor; the intervening medium (usually air) acts as the dielectric. Differences in the ac voltage between the two objects cause noise currents to flow between the objects. Ideally, electrostatic shielding intercepts these currents and shunts them to ground. The shield material and grounding determine the effectiveness of the shield. Making the shield with a high-conductivity material ensures that the coupled currents produce little voltage drop across the shield. This low voltage drop, thus, guarantees that little of the original field continues within the shield. To be effective, the shield must connect solidly to earth ground. Earth ground is the only common reference for the separate objects involved in the coupling.
Ungrounded shield introduces noise
You should also connect the shield to the system ground to minimize the effects of the parasitic capacitances that the shield introduces. All shield enclosures form parasitic capacitances with each component they shield. Capacitance between an ungrounded shield and the circuit's larger signals develops a voltage on the shield. The shield then couples this voltage back into sensitive circuit points through other mutual capacitances. (In other words, an ungrounded shield is often worse than no shield.) Returning the shield connection to the system common removes the shield voltage and the associated coupling. In amplifier circuits, grounding the shield also avoids shunting the feedback elements with the capacitance to the shield and, thus, restricting the bandwidth.
You can use op-amp CMR as an alternative to shielding to reduce the effects of electric-field coupling or to remove the residual effects of imperfect shields. Much electric-field noise couples almost equally from the ac power line to all points of a circuit. The equal coupling makes this interference a natural candidate for removal by CMR; op-amp CMR readily removes signals at the power-line frequency. However, the most common op-amp circuit configurations don't take advantage of the amplifier's CMR. An impedance imbalance at an op amp's two inputs can defeat the CMR. Switching to a differential-input connection or balancing the impedances that drive the amplifier inputs allows the CMR to attenuate the coupled signal. An additional benefit of these measures is a reduction in the dc error. This approach doesn't always eliminate the need for shielding, however, because the electric-field coupling to the two op-amp inputs isn't always identical.
Electric fields couple noise current into an op amp's feedback network through mutual capacitance.
Fig 1 models the basic electric-field coupling effect with an inverting op-amp configuration, an electric-field noise source, eE, and mutual capacitance, CM. Source eE represents any ac voltage that creates an electric field near the amplifier. This source couples a noise current, iNE, through CM and into the amplifier's feedback network. That current flows through the feedback resistor, R2, to produce an output noise signal, iNER2. In practice, other mutual capacitances couple noise currents to other points in the circuit. However, the low impedances at these other points minimize the effects of the currents.
A difference amplifier rejects the effects of electric-field coupling by developing a canceling signal and rejecting it with the op amp's CMR.
Switching to a differential-input configuration produces balanced conditions that allow the op amp's CMR to reject the noise signal, iNER2. Fig 2 models this case with the difference amplifier and two mutual capacitances coupling to the op amp's two inputs. If the two inputs are essentially equidistant from the noise source, these capacitances match. In addition, the voltages on the capacitances match because the amplifier's feedback forces the op amp's two inputs to the same voltage. Thus, the two capacitors couple equal iNE noise currents to the circuit's resistor networks.
The current coupled to the op amp's noninverting input produces a noise voltage of iNE(R1||R2). The circuit amplifies this voltage with a gain of 1+(R2/R1) to produce an output noise component of iNE2. The current iNE coupled to the op amp's inverting input flows through the feedback resistor, R2, developing an output noise component of iNER2. Thus, the two iNE noise currents produce canceling output noise voltages. At the two op-amp inputs, the only remaining signal is a common-mode voltage, iNE(R1||R2), which is attenuated by the amplifier's CMR.
Unbalanced impedance ruins CMR
This circuit operation suggests a simple method of noise reduction for more general op-amp configurations. As long as RB=R1||R2, the signal iNE(R1||R2) at the op amp's inputs does not change when a balancing resistor, RB, replaces the lower R1 and R2 resistors. The fact that this change does not affect the circuit's noise cancellation suggests that simple modifications to other op-amp circuits can also reduce electric-field-noise.
Adding impedance-balancing resistors in series with the op-amp's noninverting input extends the noise rejection of the difference amplifier to more general op- amp configurations.
Fig 3 shows this technique for typical inverting and noninverting configurations. Adding balancing resistances in series with the op amps' noninverting inputs cancels noise coupled to those inputs. Further examination reveals the impedance balance that the technique produces. From the figure, the impedance that drives from the op amp's inverting input also equals R1||R2. R1 returns to the low impedance of a source, and R2 returns to the low impedance of the amplifier output. Thus, for impedance-analysis purposes, these two resistors effectively return to ground and appear in parallel. Therefore, the op amp's CMR rejects electric-field coupling, as long as the circuit presents equal impedances to the two op-amp inputs.
The RB resistors resemble those often added to reduce the dc offset produced by amplifier input current. For reducing dc offset, adding a resistance of R1||R2 in series with the op amp's noninverting input cancels offset. Bypassing the resistors with capacitors avoids two side effects of balancing input resistance to cancel offset: increased resistor noise and, for the noninverting configuration, reduced bandwidth. Although resistors that cancel coupled noise also cancel offsets caused by input current, the desired coupled-noise cancellation precludes bypassing. Bypassing would again cause an imbalance in the impedances that drive the amplifier inputs and would prevent coupled-noise reduction. Because RB has no bypass, its noise adds to the circuit's noise response. Also, for noninverting configurations, the combination of RB and the op amp's input capacitance forms a low-pass filter that may restrict the circuit's bandwidth.
The effectiveness of this noise cancellation depends upon precise resistance matching and high op-amp CMR. At lower frequencies, the resistance mismatch between RB and R1||R2 usually dominates and limits the cancellation accuracy. At higher frequencies, the response roll-off of the op amp's CMR increases the error in proportion to 1/CMR. Using high-frequency op amps minimizes the effect of reduced high-frequency CMR. The OPA132 shown provides a CMR of at least 40 dB from dc to 800 kHz.
Magnetic materials absorb magnetic fields
Magnetic-noise coupling and radio-frequency interference (RFI) introduce noise through a common coupling mechanism: mutual inductance. The interference source acts as the primary winding of a transformer, and circuit loops act as secondary windings. Although designers often think of RFI as a separate effect, it actually represents a high-frequency form of parasitic magnetic coupling. When selecting shielding materials, however, you must deal with RFI and magnetic coupling separately. The frequency differences of the two interference sources necessitate the use of different shielding materials. At lower frequencies, only ferromagnetic materials offer the properties you need to construct shields of a practical thickness. At higher frequencies, decreases in both the shield-thickness requirement and the magnetic response of ferromagnetic materials make copper a good alternative; even the copper layer of a ground plane becomes an effective magnetic shield.
A comparison of magnetic- and electric-field coupling and a transformer analogy illustrate the added requirements of magnetic shielding. Electrostatic shielding requires a high-conductivity shield. The high conductivity grounds currents transmitted through mutual capacitances. Magnetic fields couple through mutual inductances, rather than capacitances, and typically produce voltage, rather than current signals, in op-amp circuits. The coupled signals develop in all circuit loops within the field and have magnitudes proportional to the loop areas.
For a magnetic field, a high-conductivity electrostatic shield forces an equipotential condition only at the shield boundaries. Grounding this shield does not terminate the magnetic field and may merely establish a zero-voltage reference. By analogy, grounding the center tap of a transformer's secondary winding establishes a reference voltage but does not terminate the transformer's coupling. Thus, an electrostatic shield distorts a magnetic field but does not necessarily remove the field energy. Some portion of the field's energy penetrates the region the shield encloses.
Fortunately for noise reduction, some of the magnetic-field energy dissipates in eddy currents and ohmic absorption within the shield. At high and low frequencies, the materials you need for optimum shielding are different. To effectively reduce magnetic coupling, a shield must absorb the field energy as the field travels through the shield's walls. A combination of high conductivity and high magnetic permeability produces the greatest field absorption. High electrical conductivity ensures that the induced shield currents produce little voltage drop across the shield, preventing electric-field coupling from continuing the field within the shield. High magnetic permeability ensures that the shield efficiently absorbs the magnetic field moving through the shield.
At the power-line frequency common to many magnetic-coupling effects, ferrous materials reduce magnetic coupling an order of magnitude better than do other shield materials (Ref 4). Ferrous materials can act this way because of their high magnetic permeability. The realignment of magnetic dipoles in the ferrous material dissipates the magnetic field energy as eddy currents. This effect greatly improves magnetic-shield effectiveness, even though the electrical conductivity of steel is about an order of magnitude lower than that of copper. At power-line frequencies, a copper shield's required thickness becomes prohibitive because of the great skin depth.
At RF, copper is a more reasonable, but never superior, shield material. At RF, both the magnetic permeability of ferrous metals and the skin depths of all metals drop dramatically. Above 10 kHz, the magnetic permeability of ferrous metals drops because of the finite time required to realign the material's magnetic dipoles. The shorter periods of high-frequency signals preclude the realignment and the resulting conversion of field energy to eddy currents. A shield must remove magnetic-field energy through ohmic absorption as reflected by skin depth. Skin depth indicates how thick a shield material must be to attenuate a magnetic field by a factor of e=2.73. Fortunately, skin depth also decreases at higher frequencies because of the signal's shorter wavelength. Thus, at high frequencies, ferrous materials' decreased permeability reduces their relative advantage over copper. The reduced skin depths make the required copper thickness more practical. However, for a given shield thickness, steel still retains about a factor of 3 absorption advantage.
Separating and shielding an amplifier from a magnetic-noise source offers the best protection against magnetic-noise coupling. However, the amplifier's physical and electrical configurations also affect this coupling. Layouts that minimize loop areas and designs that maximize common-mode rejection both reduce the resulting noise. Minimizing loop areas by placing components close to the op amp lessens the mutual inductances that couple in magnetic-noise signals. Simply reducing the lengths of component leads generally produces much of the desired area reductions. Leadless chip components and smaller op-amp packages, such as the SOIC version of the OPA132, also help to minimize loop areas. Finally, an op amp's CMR reduces magnetic-coupling effects in differential-input configurations. For these configurations, matching loop areas and distances from a noise source makes the amplifier's CMR reject the effects of some of the coupled-noise signals.
For op-amp circuits, the most confusing task in magnetic-coupling reduction can be identifying the pickup loops. The physical arrangement of the circuit's components forms these loops in several ways. Fig 4 shows the loops in a noninverting op-amp circuit. In that circuit, the op amp's connections with the source, the feedback, and the load form three loops. The op amp seems to break these loops, but the amplifier's feedback action continues them. Differing amplifier noise responses and grounding connections distinguish the three loops.
First, the ground return of resistor R1 forms loop L1 with coupled noise source, eM1; the signal source; and the op-amp input circuit. The amplifier's very high input impedance appears to break the loop by interrupting an otherwise continuous conductive path. However, feedback completes the L1 loop by forcing the voltage between the amplifier inputs to zero, just as if the inputs were shorted together. Although the ground connection seems to break this loop, it doesn't (any more than grounding a transformer secondary's center tap interrupts the transformer coupling). This loop's noise signal, eM1, drives an inverting-amplifier that provides a gain of R2/R1. This gain potentially makes the L1 loop a serious noise source and a good choice for area minimization.
Next, consider loop L3 and its coupled signal, eM3. This less obvious loop exists because eI and RL connect at the circuit's ground return, and feedback extends the signal path through the amplifier. Signal eI drives the input of a noninverting amplifier. Feedback makes the amplifier output respond to eI, continuing the loop between the top of eI and RL. The resulting noise signal, eM3, appears at the load with unity gain.
The final circuit loop, L2, depends upon the loop-continuity conditions described for L1 and L3. Feedback action maintains continuity from the amplifier's noninverting input to both the inverting input and the output. Fig 4 shows that making these two effective connections also completes the L2 loop. The resulting noise signal, eM2, also appears at the circuit output with unity gain.
Op amps form magnetic pickup loops with external components and couple magnetic field noise to the circuit output.
Some noise can remain
Minimizing the preceding loops reduces the magnetically coupled noise, but does not eliminate it. Some finite loop areas and magnetic coupling always remain. However, the difference amplifier's CMR reduces the noise further. Fig 5 shows the relevant loops and coupled-noise signals for this amplifier. The ground of the differential input acts as a center tap for the L1 pickup loop, splitting the eM1 signal into two equal parts. Unfortunately, these parts present opposite-polarity signals to the differential input's R1 resistors. Thus, instead of a common-mode input, the net eM1 presents a differential signal, which the circuit's CMR does not reject.
However, the balanced structure of the difference amplifier still permits noise reduction by matching the L2 and L3 loops. These loops produce the signals eM2 and eM3, which tend to cancel at the circuit output. Signal eM3 develops a voltage of eM3R1/(R1+R2) at the op amp's noninverting input. This input represents a common-mode signal to the op amp. The op amp's CMR reduces the effect of this signal with respect to that of a differential input. However, the circuit also amplifies this common-mode signal by 1+R2/R1, producing a noise component of eM3 at the circuit output. Signal eM2 produces an output-noise component of eM2. Together, these noise signals produce an output of eM2 eM3. Therefore, making eM2=eM3 cancels the two effects at the circuit output.
This cancellation requires matching the areas of loops L2 and L3, their distances from any interfering magnetic source, and their orientations relative to that source. Matching these three features equalizes eM2 and eM3, making their net effect a common-mode signal at the amplifier's inputs. Matching loop areas and distances equalizes the magnitudes of eM2 and eM3; matching distances produces a first-order phase equalization. Accurate phase matching, which high common-mode cancellation requires, also necessitates matched loop orientations relative to the magnetic source. Most often, this noise reduction technique aids in the rejection of low-frequency, local-noise sources, such as power-transformer interference.
At higher frequencies, RFI filtering can somewhat replace CMR to counteract increased magnetic-coupling efficiency. This high coupling efficiency can have a major effect on output noise, especially when digital circuits that radiate significant RFI are close to op-amp circuits. The amplifier's limited bandwidth can provide filtering by limiting the circuit's response to frequencies well below RF. The limited bandwidth permits adding an RFI filter at the amplifier output without restricting the circuit's useful bandwidth.
PICTURE 5
The difference amplifier fails to reject the noise signal, eM1, of the L1 loop. The amplifier potentially makes the eM3 and eM2 effects cancel.
However, output filtering does not remove one of the effects of higher level RFI. The input circuit of an op amp can act as an RF detector, separating a lower frequency envelope from a carrier (Ref 5). Larger RF signals drive the emitter-base junctions of the input stage's bipolar transistors and produce a rectifying action. The transistor's junction capacitances then store the envelope level of the RF signal at the amplifier input. The interference can completely disable the amplifier. If not, the amplifier at least transmits an amplified replica of the envelope to the circuit output. FET-input op amps significantly reduce the likelihood of this effect because they require larger voltage swings to produce rectification.
Trial and error with a difference
Because coupled noise seldom provides clear clues about its origin, successfully reducing such noise requires an unusual approach to troubleshooting. In any situation, a single remedy may produce the best results, or you may have to combine two or more techniques. Without definitive clues, trial and error prevails. The possibility of combining several techniques suggests a nontraditional approach. Normally, when a technique fails, you abandon it and try others. However, when experimenting with noise reduction, you should not be too quick to remove remedies that appear to fail. Try the "failed" techniques in combination with others. Proceed through all of the noise-reduction practices described here, leaving all remedies in place until the end. With the coupling minimized, begin removing the seemingly ineffective remedies. You will now note that some of these fixes actually are effective. Restore these remedies and omit the ones that really are ineffective.
The preceding discussions focus on a circuit's sensitivity to coupled noise, and overlook the circuit's potential for producing such noise. Current flow in a conductor produces a magnetic field capable of coupling noise back into amplifiers or other circuits. Most current that flows in op-amp circuits is too small to produce significant fields. Current flowing in output circuits is an exception. The load current supplied by the amplifier produces a potentially significant field. However, use of coaxial returns can usually make even this current insignificant.
An op amp's load current flows in a loop that begins at a power-supply output, passes through the amplifier and load, and returns to the power-supply common. To minimize the resulting magnetic field, you should minimize the area of this loop by using coaxial connections to the power supply. In such connections, the common axes of the load current's supply and return paths produce canceling magnetic fields. Usually, coaxial cable is impractical for this purpose, but a ground plane offers a good approximation. In a ground plane, a return current follows the path of least impedance to the power supply. At lower frequencies, the ground-plane resistance controls this impedance, and the return current usually follows the shortest path to the supply. This path does not constitute a coaxial return, but the lower frequency prevents the signal's magnetic field from inducing significant noise voltages.
At higher frequencies, the ground-plane inductance controls the return impedance and automatically assures a coaxial return. The inductance of the conduction path is proportional to the loop area, and the loop's magnetic field controls the direction of current flow. Seeking the path of least impedance, the return current follows a path paralleling that of the corresponding supply current. This path minimizes the inductance and the associated magnetic field. The distance between the supply trace and the ground plane separates the two conduction paths. Usually however, these paths need be no farther apart than the opposite sides of a pc board.
Bonnie Baker is manager of applications engineering at Burr-Brown Corp in Tucson, AZ, where she has worked for eight years. Her job entails developing op amps, instrumentation and isolation amplifiers, and A/D and D/A converters. She holds a BSEE from Northern Arizona University, Flagstaff, AZ, and an MSEE from the University of Arizona, Tucson, AZ.
After 29 years at Burr-Brown, Jerry Graeme has formed his own consulting firm in Tucson, AZ. At Burr-Brown, he designed and managed the design of op amps, voltage-to-frequency converters, two-wire transmitters, variable-gain amplifiers, and analog multipliers. He has written several books and has had many articles published in EDN and other technical journals. He holds a BSEE from the University of Arizona, Tucson, AZ, and an MSEE from Stanford University, Stanford, CA.
References
