AN-83: Performance Verification of Low Noise, Low Dropout Regulators

Regulator Noise Measurement

Regulator noise measurement begins with attention to test setup details. The extremely low signal levels require attention to shielding, cable management, layout and component choice.⁶ Figure 6a is the bench arrangement. The photo shows the completely shielded environment required to obtain faithful noise measurements. The metal can⁷ encloses the regulator under test and its internal battery power supply. A BNC fitting (photo lower center) connects the regulator output to the noise filter test circuit (black box). Note that the monitoring oscilloscope and voltmeter are not simultaneously connected to the output, precluding ground loops which would corrupt the measurement.

Figure 6b details the regulator enclosure with its cover removed. The battery supply is visible; the regulator occupies the can center. The BNC fitting connecting the noise filter box (lower left) eliminates triboelectric disturbances a cable might contribute.

Figure 6c is the noise test circuit box. Functions are as labeled in the photo. The two capped BNC connectors (box lower) are unused box entries.

Figure 7’s oscilloscope photo shows an LT1761 regulator’s noise measured at the filter output. Monitoring this point with the RMS voltmeter shows a 20µV_RMS reading.

Figure 8’s spectral plot of this noise indicates diminished power above 1kHz, in accordance with expected regulator noise density. Figure 9 shows more complete spectral noise density data for three regulator types. Noise power decays uniformly with increasing frequency, although the three regulators show some dispersion below 200Hz.

Bypass Capacitor (C_BYP) Influence

The regulator’s internal voltage reference contributes most of the device’s noise. The reference bypass capacitor filters reference noise, precluding it from appearing, in amplified form, at the output.⁸ Figure 10 is a study of regulator noise vs various values of C_BYP. 10a shows substantial noise for C_BYP = 0µF, while 10d displays nearly 9× improvement with C_BYP = 0.01µF; intermediate values of C_BYP (10b and 10c) produce commensurate results.

Interpreting Comparative Results

Figure 11’s photos compare an LT1761-5’s output noise (11d) with three other regulators (11a, 11b and 11c). These three devices are manufacturer specified for low noise performance, but the photos do not indicate this. The seeming contradication is probably due to ambiguity in testing methods or specifications. For example, inappropriate choice of test equipment (see Appendix C) or measurement bandwidth can easily cause huge (5×) errors. This uncertainty mandates the noise testing described to insure realistic conclusions.

Note 6: Capacitor choice is discussed in Appendix B, “Capacitor Selection Considerations.”

Note 7: The cookies were excellent, particularly the thin ones with sugar on top.

Note 8: See Appendix A, “Architecture of a Low Noise LDO,” for details.

APPENDIX A

Architecture of a Low Noise LDO

Noise Minimization

The low noise LDOs use Figure A1’s scheme, with special attention to minimizing noise transmission within the loop and from unregulated input. The internal voltage reference’s noise is filtered by C_BYP. Additionally, the error amplifier’s frequency response is shaped to minimize noise contribution while preserving transient response and PSRR. Regulators which do not do this have poor noise rejection and transient performance.

Pass Element Considerations

Extremely low dropout voltage requires considering the pass element. Dropout limitations are set by the pass elements on-impedance limits. The ideal pass element has zero impedance capability between input and output and consumes no drive energy.

A number of design and technology options offer various trade-offs and advantages. Figure A2 lists some pass element candidates. Followers offer current gain, ease of loop compensation (voltage gain is below unity) and the drive current ends up going to the load. Unfortunately, saturating a follower requires voltage overdriving the input (e.g. base, gate). Since drive is usually derived directly from V_IN, this is difficult. Practical circuits must either generate the overdrive or obtain it elsewhere. Without voltage overdrive, the saturation loss is set by V_BE in the bipolar case and channel on-resistance for MOS. MOS channel on-resistance varies considerably under these conditions, although bipolar losses are more predictable. Note that voltage losses in driver stages (Darlington, etc.) add directly to the dropout voltage. The follower output used in conventional 3-terminal IC regulators combines with drive stage losses to set dropout at 3V.

Common emitter/source is another pass element option. This configuration removes the V_BE loss in the bipolar case. The PNP version is easily fully saturated, even in IC form. The trade-off is that the base current never arrives at the load, wasting power. At higher currents, base drive losses can negate a common emitter’s saturation advantage. As in the follower example, Darlington connections exacerbate the problem. Achieving low dropout in a monolithic PNP regulator requires a PNP structure that attains low dropout while minimizing base drive loss. This is particularly the case at higher pass currents. Considerable effort was expended in this direction in the LT176X through LT196X designs.

Common source connected P-channel MOSFETs are also candidates. The do not suffer the drive losses of bipolars, but typically require volts of gate-channel bias to fully saturate. In low voltage applications this may require generation of negative potentials. Additionally, P-channel devices have poorer saturation than equivalent size N-channel devices.

The voltage gain of common emitter and source configurations is a loop stability concern but is manageable.

Compound connections using a PNP driven NPN are a reasonable compromise, particularly for high power (beyond 250mA) IC construction. The trade-off between the PNP V_CE saturation term and reduced drive losses over a conventional PNP structure is favorable. Also, the major current flow is through a power NPN, easily realized in monolithic form. The connection has voltage gain, necessitating attention to loop frequency compensation. Regulators utilizing this pass scheme can supply up to 7.5A with dropouts below 1.5V (LT1083 through LT1086 series).

Readers are invited to submit results obtained with our emeritus thermionic friends, shown out of respectful courtesy.

Dynamic Characteristics

The LT176X through LT196X’s low quiescent currents do not preclude good dynamics. Usually, low quiescent power devices are associated with slow dynamics and instability. The devices are stable (no output oscillation) even with low ESR ceramic output capacitors. This contrasts with conventional LDO regulators that often oscillate with ceramic capacitors.

The internal architecture provides an added bonus in transient performance when the 0.01µF noise capacitor is added. Transient response for a 10mA to 100mA load step with a 10µF output capacitor appears in Figure A3 with the capacitor deleted. Figure A4 shows the same situation with the 0.01µF bypass capacitor in place. Settling time and amplitude are markedly reduced.

APPENDIX B

Capacitor Selection Considerations

Bypass Capacitance and Low Noise Performance

Adding a capacitor from the regulators V_OUT to BYP pin lowers output noise. A good quality low leakage capacitor is recommended. This capacitor bypasses the regulator’s reference, providing a low frequency noise pole. A 0.01µF capacitor lowers the output voltage noise to 20µV_RMS. Using a bypass capacitor also improves transient response. With no bypassing and a 10µF output capacitor, a 10mA to 500mA load step settles within 1% of final value in under 100µs. With a 0.01µF bypass capacitor, the output settles within 1% for the same load step in under 10µs; total output deviation is inside 2.5%. Regulator start-up time is inversely proportional to bypass capacitor size, slowing to 15ms with a 0.01µF bypass capacitor and 10µF at the output.

Output Capacitance and Transient Response

The regulators are designed to be stable with a wide range of output capacitors. Output capacitor ESR affects stability, most notably with small capacitors. A 3.3µF minimum output value with ESR of 3Ω or less is recommended to prevent oscillation. Transient response is a function of output capacitance. Larger values of output capacitance decrease peak deviations, providing improved transient response for large load current changes. Bypass capacitors, used to decouple individual components powered by the regulator, increase effective output capacitor value. Larger values of reference bypass capacitance dictate larger output capacitors. For 100pF of bypass capacitance, 4.7µF of output capacitor is recommended. With 1000pF of bypass capacitor or larger, a 6.8µF output capacitor is required.

Figure B1’s shaded region defines the regulator’s stability range. Minimum ESR needed is set by the amount of bypass capacitance used, while maximum ESR is 3Ω.

Ceramic Capacitors

Ceramic capacitors require extra consideration. They are manufactured with a variety of dielectrics, each with different behavior across temperature and applied voltage. The most common dielectrics are Z5U, Y5V, X5R and X7R. The Z5U and Y5V dielectrics provide high capacitance in a small package, but exhibit strong voltage and temperature coefficients as shown in Figures B2 and B3. Used with a 5V regulator, a 10µF Y5V capacitor shows value as low as 1µF to 2µF over the operating temperature range. The X5R and X7R dielectrics have more stable characteristics and are more suitable for output capacitor use. The X7R type has better stability over temperature, while the X5R is less expensive and available in higher values.

Voltage and temperature coefficients are not the only problem sources. Some ceramic capacitors have a piezoelectric response. A piezoelectric device generates voltage across its terminals due to mechanical stress, similar to the way a piezoelectric accelerometer or microphone works. For a ceramic capacitor the stress can be induced by vibrations in the system or thermal transients. The resulting voltages produced can cause appreciable amounts of noise, especially when a ceramic capacitor is used for noise bypassing. A ceramic capacitor produced Figure B4’s trace in response to light tapping from a pencil. Similar vibration induced behavior can masquerade as increased output voltage noise.

APPENDIX C

Understanding and Selecting RMS Voltmeters

The choice of AC voltmeter is absolutely crucial for meaningful measurements. The AC voltmeter must respond faithfully to the RMS value of the measured noise. The majority of AC voltmeters (including DVMs with AC ranges) are not capable of doing this. This includes instruments with “true RMS” AC scales. As such, selecting an appropriate instrument requires care. The selection process begins with a basic understanding of AC voltmeter types.¹

AC Voltmeter Types

There are three basic AC voltmeter types. They include rectify and average, analog computing and thermal. The thermal approach is the only one that is inherently accurate, regardless of input waveshape. This feature is particularly relevant to noise RMS amplitude determination.

Rectify and Average

The rectify and average scheme (Figure C1) applies the AC input to a precision rectifier. The rectifier output feeds a simple RC averager which is gain scaled, providing the output. In practice, gain is set so the DC output equals the RMS value of a sine wave input. If the input remains a pure sine wave, accuracy can be quite good. However, nonsinosoidal inputs cause large errors. This type voltmeter is only accurate for sine wave inputs, with increasing error as the input departs from sinosoidal.

Note 1: Another way to approximately measure RMS AC noise using an oscilloscope is the tangential method. See Reference 14.

Analog Computation

Figure C2 shows a more sophisticated AC voltmeter method. Here, the instantaneous value is (ideally) continuously computed by an analog computational loop. The DC output follows the equation noted, resulting in much better accuracy as input waveshape varies. Almost all commercial implementations of this approach utilize logarithmically based analog computing techniques. Unfortunately, dynamic limitations in the ZY/X block dictate bandwidth restrictions. These circuits typically develop significant errors beyond 20kHz to 200kHz.

Thermal

The thermally based AC voltmeter is inherently insensitive to input waveshape, particularly suiting it to noise RMS amplitude measurement. Additionally, high accuracy at bandwidths exceeding 100MHz is achievable. Figure C3 diagrams the classic thermal scheme.2 It is composed of matched heater-temperature sensor pairs and an amplifier. The AC input drives a heater, warming it. The temperature sensor associated with this heater responds, biasing the amplifier. The amplifier closes its feedback loop by driving the output heater to warm its associated temperature sensor. When the loop closes, the heaters are at the same temperature. This “force-balance” action results in the DC output equalling the input heater’s RMS heating value—the fundamental definition of RMS. Changes in waveshape have no effect, as they are effectively downconverted to heat. This “first principles” nature of operation makes thermally based AC voltmeters ideal for quantitative RMS noise measurement.

Performance Comparison of Noise Driven AC Voltmeters

The wide performance variation of the above methods, and even within a method, mandates caution in selecting an AC voltmeter. Comparing AC voltmeters intended for use in RMS noise measurements is illuminating. Figure C4 shows a simple evaluation arrangement. The noise generator drives the external input of text Figure 2, producing a suitably band passed input at the voltmeter under test.³

Note 2: See the References section for publications covering thermal AC–DC conversion.

Note 3: Noise generators are worthy of study and considered in References 4 and 5.

Figure C5 shows results for 20 voltmeters. Four of the voltmeters are thermal types; the remainder utilize logarithmic analog computing or rectify and average AC-DC conversion. The four thermal types agreed well within 1%. In fact, three of the thermal types were within 0.2%, while the fourth (HP3400A), a metered instrument, is only readable to 1%. The other 16 voltmeters showed errors up to 48% relative to the thermal group! Note that the errors cause lower readings than are actually warranted; a poorly chosen voltmeter will give unfairly optimistic readings.

The lesson here is clear. It is essential to verify AC voltmeter accuracy before proceeding with RMS noise measurement. Failure to do so may cause highly misleading “results.”

Thermal Voltmeter Circuit

It is sometimes desirable to construct, rather than purchase, a thermal voltmeter. Figure C6’s circuit is applicable to noise measurement. Text Figure 2’s filter feeds A1. A1’s output biases A2, which provides additional AC gain. The LT1088 based RMS/DC converter is made up of matched pairs of heaters and diodes and a control amplifier. The LT1206 drives R1, producing heat which lowers D1’s voltage. Differentially connected A3 responds by driving R2, via Q3, to heat D2, closing a loop around the amplifier. Because the diodes and heater resistors are matched, A3’s DC output is related to the RMS value of the input, regardless of input frequency or waveshape. In practice, residual LT1088 mismatches necessitate a gain trim, which is implemented at A4. A4’s output is the circuit output.

Start-up or input overdrive can cause A2 to deliver excessive current to the LT1088 with resultant damage. C1 and C2 prevent this. Overdrive forces D1’s voltage to an abnormally low potential. C1 triggers low under these conditions, pulling C2’s negative input low. This causes C2’s output to go high, putting A2 into shutdown and terminating the overload. After a time determined by the RC at C2’s input, A2 will be enabled. If the overload condition still exists, the loop will almost immediately shut A2 down again. This oscillatory action will continue, protecting the LT1088 until the overload condition is removed.

To trim this circuit, connect its input to a 10mV_RMS, 100kHz signal. Set the 500Ω adjustment for exactly 100mV DC out. Next, apply a 100kHz, 100mV_RMS input and trim the 10k potentiometer for 1V DC out. Repeat this sequence until the adjustments do not interact. Two passes should be sufficient.

APPENDIX D

Practical Considerations for Selecting a Low Noise LDO

Any design has particular requirements for a low noise LDO; every individual situation should be carefully examined for specific needs. However, some general guidelines apply in selecting a low noise LDO. Significant issues are summarized below.

Current Capacity

Insure the regulator has adequate output current capacity for the application, including worst-case transient loads.

Power Dissipation

The device must be able to dissipate whatever power is required. This effects package choice. Usually, the V_IN-V_OUT differential is low in LDO applications, obviating this issue. Prudence dictates checking to be sure.

Package Size

Package size is important in limited space applications. Package size is dictated by current capacity and power dissipation constraints. See the preceding paragraphs.

Noise Bandwidth

Insure that the LDO meets the system’s noise requirement over the entire bandwidth of interest. 10Hz to 100kHz is realistic, as information usually occupies this range.

Input Noise Rejection

Insure that the regulator can reject input related disturbances originating from clocks, switching regulators and other power bus users. If the regulator’s power supply rejection is poor, its low noise characteristics are useless.

Load Profile

Know the load characteristics. Steady state drain is obviously important, but transient loads must also be evaluated. The regulator must maintain stability and low noise characteristics under all such transient loads.

Discrete Components

The choice of discrete components, particularly capacitors, is important. The wrong capacitor dielectric can adversely effect stability, noise performance, or both. See Appendix B, “Capacitor Selection Considerations.”

Williams, J., “A Monolithic Switching Regulator with 100µV Output Noise,” Linear Technology Corporation, Application Note 70, October 1997.
Sheingold, D. H. (editor), “Nonlinear Circuits Handbook,” 2nd Edition, Analog Devices, Inc., 1976.
Kitchen, C., Counts, L., “RMS-to-DC Conversion Guide,” Analog Devices, Inc. 1986.
Williams, J., “Practical Circuitry for Measurement and Control Problems,” “Broadband Random Noise Generator,” “Symmetrical White Gaussian Noise” Appendix B, Linear Technology Corporation, Application Note 61, August 1994, pp. 24–26, pp. 38–39.
General Radio Company, Type 1390B Random Noise Generator Operating Instructions, October 1961.
Hewlett-Packard Company, “1968 Instrumentation. Electronic—Analytical—Medical,” AC Voltage Measurement, Hewlett-Packard Company, 1968, pp. 197– 198.
Justice, G., “An RMS-Responding Voltmeter with High Crest Factor Rating,” Hewlett-Packard Journal, Hewlett-Packard Company, January 1964.
Hewlett-Packard Company, “Model HP3400A RMS Voltmeter Operating and Service Manual,” HewlettPackard Company, 1965.
Williams, J., “A Monolithic IC for 100MHz RMS/DC Conversion,” Linear Technology Corporation, Application Note 22, September 1987.
Ott, W.E., “A New Technique of Thermal RMS Measurement,” IEEE Journal of Solid State Circuits, December 1974.
Williams, J.M. and Longman, T.L., “ A 25MHz Thermally Based RMS-DC Converter,” 1986 IEEE ISSCC Digest of Technical Papers.
O’Neill, P.M., “ A Monolithic Thermal Converter,” H.P. Journal, May 1980.
Williams, J., “A Fourth Generation of LCD Backlight Technology,” “RMS Voltmeters,” Linear Technology Corporation, Application Note 65, November 1995, pp. 82–83.
Tektronix, Inc., “Type 1A7A Differential Amplifier Instruction Manual,” Check Overall Noise Level Tangentially, 1968, pp. 5-36 and 5-37.