Part 4

REGULATORS FORHIGH-PERFORMANCE AUDIO

Real-World Implementations and Sonic Evaluations

By Gary A. GaloContributing Editor

In the first three issues of 1995,TAA readers were given a wealthof information on state-of-the-art

regulators for low-level audio applica-tions. Now that you've seen WaltJung's circuits (1/95, p. 8), analyzedhis measurement data (2/95, p. 20),and built Jan Didden's clean, easy-to-assemble printed circuit board (3/95,p. 20), you're ready to drop a few ofthese regulators into real-world proj-ects. Perhaps you've been reluctant tobuild them until a report on how theyaffect the sound was published. This,and more, is what Part 4 is all about.

Over the past several months I'veamassed a great deal of experiencebuilding and implementing these reg-ulators using Jan's PC boards. I haveinstalled and evaluated ±14V versionsin an extensively modified AdcomGFP-565 preamplifier, along with +5V,-5V, and -15V versions in a PhilipsDAC960 digital-to-analog convertermodified to Pooge 5.5 standards.1'2

From reading Parts 1-3, you knowthat these fast, ultra-wideband regula-tors require careful attention to partsselection and layout in order to per-form to their potential.

As Jan's article so aptly illustrates,there are more "components" in a cir-cuit than those shown in the schemat-ic diagram, and much more to theseregulators than input, ground, andoutput. Jan's Fig. 1 shows the othercomponents that exist within a real-world regulator circuit, and evenmore phantom inductances, capaci-tances, and resistances can enter thepicture once the regulator is installedin a real-world device. With a featureas sophisticated as remote sensing, theassembled boards can't be considereddrop-in replacements for the old 7815and 7915 three-terminal types.Connecting these regulators to realelectronic circuits requires consider-

34 The Audio Amateur 4/95

ably more effort—and expertise—onthe part of the builder.

Test Gear NecessitiesAt several points throughout the pre-vious articles, Walt and Jan have men-tioned the potential for oscillation ofthese regulators. With such sophisti-cated circuitry, we can no longer relyonly on a digital voltmeter for regula-tor testing (though we certainly doneed one). A wideband oscilloscope,20MHz minimum, is an absolute mustfor verifying their proper operation. Iknow many of you will be tempted tobuild and install these devices with-out a scope, but you must avoid thistemptation at all costs.

Jan's excellent PC board is extreme-ly simple to assemble. Thanks to aclean layout and clear instructions, itresembles a "One Evening" noviceHeathkit project, but its simplicity isincredibly deceiving. Implementingthese regulators in an overall systemdesign is no less than an advancedproject. Whether you use the regula-tors as part of a new design or modifyexisting equipment, there's no substi-tute for proper test equipment.

A few of my experiences may helpemphasize this point. In issue 2/86 Ireviewed Phoenix Systems' P-94-SRParametric Equalizer Kit.3 This was aHeathkit-style project, in whichbuilders assembled the factory-madeprinted circuit boards and mountedeverything in factory-supplied cases.Since this was a finished kit design,you would expect it to operate per-fectly when it was completed. Minedidn't.

Ever since I began working withfast, wideband op amps in the late1970s, I have always checked powersupply rails for oscillations. I found aproblem when I checked these. Farfrom sophisticated, wideband regula-

tors, this equalizer used the 7815/7915pair. I fixed the problem by addingmore local supply bypassing near theTL-074 quad op amps. Even foolproof,easy-to-use, three-terminal regulatorscan oscillate under the right (wrong?)conditions. You'll never know unlessyou check them with a scope.

A problem I had with a Nelson PassA-40 power amplifier is also worthrelating. Pass did everything he couldto make its construction possible forthe novice, even if no test equipmentwas available. If you didn't have avoltmeter to check DC offset, youcould put a 470Q, 1W resistor acrossthe speaker terminals. No heat = safeoffset.4 When I measured harmonicdistortion, one channel was excessive-ly high above 12W output; below 10Wthe amplifier measured just as Passspecified.

The problem was amplifier oscilla-tion caused by insufficient local sup-ply bypassing. The Old Colony kitcontained much larger heatsinks thanthose used by Pass in his prototype, sothe leads to the output transistorswere rather long. Adding bypassingfrom the output transistor cases tolocal ground solved the problem, andthe amplifier sounded excellent.5

Without the proper test equipment, Iwould never have known the amplifi-er was malfunctioning, and wouldprobably have blamed the bad soundon the design. The design was not theproblem, however; my particular lay-out created a situation the designerdid not foresee.

Since Walt, Jan, and I are well awareof the potential for oscillation, itwould be irresponsible of us to avoidstressing the test equipment issue.Every hobby requires an investmentin the right tools and equipment, andgood test gear is more 'affordable thanever. MCM Electronics' Tenma line

includes a dual-trace "Trainer" scopewith a 20MHz bandwidth (#72-905)for $335. They also have a completeline of digital voltmeters.

HeatsinksIn Part 3 Jan makes some suggestionsfor heatsinking the pass transistors,Ql and Q2 (Ql in Walt's diagrams).Jan uses a TO-220 heatsink made inSwitzerland by Fischer Electronic.Since there isn't an exact replacementin the US, he recommends Aavid'sHS-112. When I installed the +14Vregulators in my modified Adcom 565preamp, I found the HS-112 inade-quate for this load, which is approxi-mately 100mA.

There's a fairly easy solution to thisproblem. The HS-112 has three finsper side. Aavid also makes the HS-114, with six fins per side. For evenmore heat dissipation, Aavid's HS-113"booster" is used in conjunction witheither the HS-112 or HS-114. It mountson the top of the transistor, so heat canbe dissipated from both sides of themetal tab. I recommend this combina-tion for higher current applications.

The 1.5-inch-long HS-114 over-hangs Jan's board, which may notwork in some physical layouts. I findthat it's very easy to cut off one or twopair of fins with a band saw or hack-saw to custom fit the heatsink to myown requirements. Be careful not tobend or twist the heatsink when youmake the cut.

If you use a hacksaw, I recommendclamping the heatsink to a woodblock with a #6 sheet metal screw andflat washer. Use one of the two holesin the heatsink, and tighten the screwjust enough to hold the heatsink inplace: you don't want to bend it. Thenmake the cut with a sharp hacksawblade on the opposite end of theheatsink. When finished, remove anyburrs or rough edges with a small file,and be sure the heatsink is completelyfree of metal filings. This is really asimple process and takes only a fewminutes.

For the Adcom 565 preamp regula-tors, I trimmed HS-114s to five pair offins. Four pair works fine for the 5Vregulators I built. In both cases I usedtrimmed HS-114s along with the HS-113 booster. Your exact currentrequirements will determine just howmuch heatsinking is needed. Withsome ingenuity, you can easily fabri-cate parts from readily availablesources. For example, Digi-Key stocks

all three Aavid products and ships thesame day.

If you ensure that there won't beany electrical contact between theheatsinks and any other point, partic-ularly ground, you shouldn't need toinsulate the pass transistors. Thisshouldn't be a problem in most cases,but make sure the ground plane onthe PC board doesn't touch theheatsink. They were perilously closeon the board samples I got from Jan,so I had to trim the positive groundplane. A single-edge razor bladeworks fine for this.

Use the insulators if you have anydoubts. (You'll get slightly better heattransfer without them, however.) Jansuggests placing a plastic spacerbetween heatsink and board to solvethis problem. This also isolates theboard and other components from theheat, which is worth considering inhigh-current situations. Always usewhite silicone thermal compound,such as GC Electronics 8109 (availablefrom Mouser or Newark). Metallicoxides contained in the white com-pounds facilitate heat transfer.

Preliminary TestsIt is worthwhile to bench test yourregulators prior to installing them, asa malfunctioning device is much easi-er to troubleshoot before it is buried ina chassis. The raw supply I built forthis purpose is shown in Fig. 1, with aparts list in Table 1.

This +13V supply can be used fortesting 5V and 14V regulators. To testthe latter, use the full 26V availablebetween the positive and negativerails; 5V regulators can be fed fromthe 13V rails. When conducting yourtests, be careful to observe correctinput polarity: you can damage the opamp and transistors with reversedpolarity. The Ik bleeder resistors areoverrated at 1W, but they stay cool,and the stiff leads make solid connec-tion points for clip leads.

For bench testing, omit the remotesensing by jumpering load output tosense, and load ground to senseground. You should also put a resistoracross the output to pull 75-100mAfrom the regulators. Use 150Q, 1W forthe 14V versions, and 50ii, 1W for the5V ones. The load resistor values canbe tailored to match the current drainin your specific application.

I find a Variac (VARIable AC trans-former), shown in Fig. 1, extremelyuseful for testing. If you put a digitalvoltmeter across the regulator out-puts, you can slowly increase theVariac's AC output while monitoringthe regulator's DC output. When test-ing a positive regulator, be sure theoutput is actually going positive asyou begin turning the Variac. If itisn't, power down and find the prob-lem. The output of a negative regula-tor should begin to go negative assoon as AC is applied.

It's easy to reverse polarity on abench setup connected with clip leads,but if you make a mistake the Variacavoids potential disasters. It also allowsyou to adjust the DC input to the regu-lator, duplicating the voltages whichwill be present in your equipment.

Once you have verified correct DCperformance, check the regulator foroscillation. Set your oscilloscope for

TABLE 1

RAW DC TEST SUPPLY PARTS LIST

TRANSFORMER18VCT.2.0A

BRIDGE6A, 100PIVCAPACITORS0.47nF/100V

2,200uR25V

RESISTORS1k/1W

VARIACTenmalOA

(Mouser 41FJ020)

(Digi-Key PB61-ND)

Panasonic V-series(Digi-Key P4733-ND)Panasonic HFQ(Digi-Key P5716-ND)

Yageo metal oxide(Digi-Key P1.0KW-1BK-ND)

Variable Autotransformer(MCM72-110)

6A BRIDGE

\tA

II 1 IIII I II±/\.

^XXi ±S

*Y*HhfHH* X O.47UF

22OOuF/23V ,-

22OOUF/25V ~

tk/ IW

1 k/1 w

FIGURE 1: Raw supply for bench testing regulators prior to installation. The Variac is recommend-ed to avoid damage if the circuit is not operating properly.

The Audio Amateur 4/95 35

maximum sensitivity and minimumtime base. On my scope these are 5mVand 0.2us per division. Always set thescope for AC coupling in these tests.Also, if your scope's input couplingselector has a ground position, makesure it is not in this position. Check thatno oscillations are present after theregulator has reached its rated DCoutput voltage.

You can check dropout voltage byconnecting your digital voltmeterfrom the input to the output of theregulator. Set your scope to20-50mV/division sensitivity and2.0ms/division time base, and connectit to the regulator output. As youdecrease the Variac output from 117VAC, you'll see the voltage differencebetween the regulator input and out-put decrease proportionally.

The scope trace will remain astraight line until the regulator "dropsout" of regulation. At this point,120Hz sawtooth ripple will appear onthe scope, and will quickly rise inlevel as the input voltage is droppedfurther. The dropout voltage is theinput/output voltage differential atthe point where the ripple just barelyappears. With a 100mA load, dropoutcan be anywhere from IV to 1.8V in aproperly funct ioning regulator.(There's more on the dropout issuelater in this article.)

Transformers and Raw SuppliesWhether you use these regulators in anew design or modify existing equip-ment, you must make some decisionsregarding the raw, unregulated por-tion of your power supply. In the past1 have used toroidal transformers forpractically all of my audio projects.These devices are extremely efficient,and, since they concentrate the mag-netic field in the core, radiate a lowhum field. Rick Miller, author of thesidebar on rectifier diode noise whichaccompanies my Pooge 5.5 article,6

has been measuring power trans-former bandwidths. He concludesthat we are barking up the wrong treewith power toroids.

As it turns out, toroidal transform-ers are wideband devices which areextremely effective at transferringpower line noise to equipment. Figure2 (prepared by Rick on an AudioPrecision System 1) is a comparativefrequency response plot of the twotransformers. It illustrates the prob-lem quite dramatically. The top,dashed trace is an Avel-Lindberg D-

Toroid 1 split bobb10 000

*J) !! AUC 95 1? n 1

- 1 3 03

li 03

- J O 3J

-J^ "0

-30 DC

- 3 5 CO

HO 00 L

FIGURE 2: Bandwidth measurements on con-ventional and toroidal power transformers:dashed trace (top) is an Avel-Lindberg D-3022toroid, flat to nearly 200kHz; solid trace (bottom)is a Magnetek FD7-36, nearly 35dB down atthis frequency. (Courtesy of Rick Miller).

3022 toroidal transformer, which isnearly flat to 200kHz. The bottom,solid trace is a Magnetek FD7-36, asplit-bobbin design which is part oftheir Quick Pack series. The Magnetekis nearly 35dB down at 200kHz, witha -3dB point around 4kHz, while theAvel-Lindberg's -3dB point is wellabove 100kHz. I don't mean to singleout Avel-Lindberg in this example;toroidal transformers from othersources have similar characteristics.

Rick's measurements show thatSignal Transformers' A-41 series offerseven more effective high-frequencynoise attenuation. These dual-bobbindesigns have two independent prima-ry and secondary bobbins. This great-ly reduces the capacitive couplingbetween them, which is extremelyimportant for attenuation of commonmode noise. Split- and dual-bobbinconstruction eliminates the need forexpensive—and not nearly as effec-tive—electrostatic shielding.

For more information on the effectsof transformer construction on noisetransfer, Rick suggests TopazElectronics' Noise Suppression ReferenceManual.7 It makes two important

points: "physical separation of coilsplaced side by side on separate legs ofthe magnetic core of a transformerwill provide far less capacitive cou-pling than coils wound directly overone another"; and related to electro-static shielding, "capacitance aroundthe Faraday shield will still coupleenough noise from the primary to sec-ondary to cause problems in sensitiveequipment."

The capacitive coupling betweentransformer windings is inverselyproportional to the transformer's hi-pot rating. ("Hi-pot" is short for highpotential, the point at which the dielec-tric material—in this case the enamelinsulation on the transformer wireand the bobbin itself—breaks down.)The higher the rating, the lower thecoupling. Magnetek Quick-Pack trans-formers have a hi-pot rating of 2.5kVRMS; Signal's A-41 series is rated at4kV RMS. (Magnetek transformers areavailable from Mouser; Signal sellsfactory-direct.)

Beyond the selection of the powertransformer, we now recommend rawsupplies even more sophisticated thanthose of Pooge 5.5. A raw supply for±14V supplies is shown in Fig. 3, witha parts list in Table 2. A unique featureis the common mode chokes on theDC side of the rectifier bridges, anoth-er Rick Miller innovation. Thesechokes are 56mH Panasonic types,carried by Digi-Key. The 0.47uF capac-itors on the input line filter are special250V AC Panasonic InterferenceSuppression caps.

Further sonic improvements arenoticeable when common mode chokesare used between the rectifier diodesand the input filter capacitors, as in Fig.3. Note the absence of large film capaci-tors directly across the transformer sec-ondaries. They are unnecessary withthe common mode chokes, and can

SIGNAL A4I-BO-2B

0+19VDC UNREC

OUNREG GROUND

OUNREC GROUND

O- 1 9VCO UNREG

FIGURE 3: Raw supply suitable for use with high-performance regulators. Common mode chokesare used for AC line filtering and DC filtering after the rectifier bridges.

36 The Audio Amateur 4/95

,

RAM

8PP

s?

5

,

*

POS REG

T6-T3-T2O

1

T1 1 -Tl 3-T19

NEC REG

_

\/

9

OM

OtoD

m-

OUJM

ZO aoiu

ouo

.I/ .

.1C

A. B.FIGURE 4: Two methods for connecting the regulators in equipment with a common raw DC ground. A: Incorrect method; B: correct method, recom-mended to avoid a ground loop.

cause high-Q resonance problems withthese transformers. As outstanding asthese regulators are in terms of linerejection, effective power line filteringand low-noise rectifier diodes are stillsonically beneficial.

In Part 3 Jan shows the best methodfor connecting these regulators to theraw supplies and the powered circuit-ry. Figure 3 is consistent with his rec-ommendations, since it has separatebridge rectifiers for the positive andnegative raw supplies; the unregulat-ed supplies do not share a common

ground. If you are working with exist-ing equipment as part of a modifica-tion project, you may be forced to usea raw supply with a single rectifierbridge and a common ground.

Figure 4 shows two options for con-necting these regulators in such cases."A" is not recommended, since thereare two ground paths between theraw supply and the powered circuitry.The correct method is shown in "B,"where one ground lead is run fromeach regulator board to load ground,or the existing equipment 's main

TABLE 2

HIGH-PERFORMANCE, DUAL-POLARITYRAW SUPPLY PARTS LIST

L1, L2, L3 Panasonic 56mH common mode line filter{Digi-Key PLK1017-ND)

C1,2 Panasonic 0.47|aF/250V AC interferencesuppression(Digi-KeyP4614-ND)

C3-12 Panasonic 0.47u.F/100VV-series(Digi-Key P4733-ND)

C13,14 Panasonic 2,200|iF/35V HFQ(Digi-Key P5751-ND)

D1-8 General Instrument GI-851(Digi-Key GI851CT-ND)

T1 Signal Transformer A41-80-28

The Audio Amateur 4/95 37

power supply ground bus. I used thismethod for the Adcom GFP-565 pre-amp, since it eliminates the possibilityof a ground loop.

Op Amps and DecouplingWalt offers builders a choice of two opamps: Analog Devices' AD848 andAD797. The graphs in Part 2 show thelatter to be superior in virtually allaspects of performance. You maywonder just how audible these effectsare, but in my listening tests I alsofound the 797 to be the superior sonicperformer. (I elaborate on this later inthe article.)

Remember that the op amp is pow-ered from an unregulated supply. Soregardless of which one you choose,its own power supply can affect theoverall regulator performance. In Part1 Walt provides an optional low-passdecoupling, consisting of a 22.1Qresistor in series with the op amp sup-ply, and a 120uF electrolytic capacitoradded for local bypassing. This R/Ccombination produces a first-order,low-pass filter with a corner frequen-cy of 60Hz.

In Part 3 Jan discusses the plusesand minuses of powering the op ampthrough a series resistor. He concludesthat, in this application, op amp sup-ply current is neither frequency- nor

DIDDEN PC BOARD

(OUTPUT) LOAD ._LT

TLSENSE

SENSE SHIELD

SENSE GROUND *

GROUND *

O .01 uF "̂ v.10 1

—.— LOCAL DECOUPLING

^NOTE: CONNECTIONS ARE THE SAME FOR + AND - REGULATORS

FIGURE 5: Remote sense decoupling scheme. Connections and parts values are the same forpositive and negative regulators.

load-dependent; therefore, no ripplecurrents will develop across the 22.1Qresistor. I have heard ±14V regulatorsboth with and without decoupling,and the former is sonically superior.Walt's Fig. 12b shows a dramaticimprovement in line rejection with it,with the AD797 performing substan-tially better than the AD848 at low fre-quencies. Jan's board accommodatesthe decoupling, and I highly recom-mend its use.

Oscillations and DecouplingThe remote sensing capability is ahighlight of these regulators, but thepotential for oscillation in the mega-hertz region exists in any wideband

op amp. The AD797 is particularlysusceptible. Supply oscillation is gov-erned by any number of layout-relat-ed issues, including length and induc-tance of the remote sensing wiring,shielding characteristics, and theamount of low-ESR local supplybypassing. There's no way to predictwhether the supply will oscillate—eachimplementation must be tested.

Once the regulator has beeninstalled, connect your scope probebetween load and load ground on theregulator PC board. Use the sameoscilloscope settings as previouslynoted in the preliminary tests. Rotatethe triggering until a stable traceappears. A virtual straight line indi-

±5V Supplies Using the Bidden PCBBy Walt Jung

The original 5V regulator published as Fig.9 in Part 1 is a good positive device, capableof very low noise when used with theAD780. Unfortunately, no board design hasspecifically addressed this circuit for 5V use.In addition, this type of design, which isbased on a three-terminal 1C reference, can-not be "flipped" to provide negative or -5Voutputs. Many modern designs do needhigh-quality ±5V sources, for example, theYoutsey et al mods to the DAC-in-the-Box(TAA 4/94, p. 8).

Fortunately, some fairly simple adap-tations to the original article's Fig. 8a or8b can provide the desired functionalityto achieve ±5V operation. What is evenmore fortuitous is that these changes canbe readily implemented by simple partsubstitutions on the Jan Didden PCBdesign in Part 3.

This adaptation implements the loweroutput voltage (+) or (-) 5V versions byusing a 2.5V reference diode in place ofthe original 6.9V LM329. Specifically, theindustry standard LM336, a two-terminal

2.5V reference 1C, allows this. It can sim-ply be substituted in the same footprintas the LM329 on Jan's board. As long asthe surrounding support circuitry is fullycompatible with the lower voltage opera-tion, this can implement a very high qual-ity +5V or -5V regulator, with the sameease of construction as noted by Gary.

The specific changes to thepositive/negative regulator circuitswhich accomplish these goals are listedhere. Note: In adapting the original cir-cuits to 5V output, change only the fol-lowing items; leave all other details asoriginally published. (Complete partnumbers and order information appear inTable 3.) In each step, the first referencedesignation pertains to Part 3, Fig. 2 (pos-it ive regulator); the designations inparentheses refer to Part 3, Fig. 3 (nega-tive regulator); the original reference des-ignations to Figs. 8a and 8b from Part 1are in brackets.

1. Change Dl (D6) [Dl] to an LM3362.5V TO-92 diode type, polarizing it as

shown in the original schematic. Specialnote: Do not connect the adjust pin.

2. Change R4 (R14) [R6] to a 2.49k 1%metal film type.

3. Change R20 (R21) [R5] to a 4.99k 1%metal film type.

4. Change XI (X2) [Ul] to an AD797.This is optional in terms of basic func-tionality, and the circuit also works withthe AD848. Do not substitute other opamps, as the input CM range must becompatible with 2.5V operation!

5. Delete the D2 and D3 (D4 and D5)[D2 and D3] 1N4148 diodes when/ifusing the AD797. Retain them if using theAD848.

6. Low-dropout operation is highlyrecommended for the ±5V regulators, andshould be implemented using all three ofthe steps outlined above.

These changes affect DC operation forthe most part, so AC performance cangenerally be expected to be consistentwith what has already been published forthe original Fig. 9 circuit. -

38 The Audio Amateur 4/95

cates the supply is working correctly;an oscillating one won't be subtle. Ifthe regulator is powering digital cir-cuitry, random digital hash on the sup-ply lines is normal. An oscillation willbe repetitive, at a specific frequency. Ascope with good triggering shouldenable you to get a firm sync on anyoscillation which may be present.

Don't be surprised to see some verylow level ripples toward the left edgeof the screen, even if the supply isworking properly. The probable causeis the test setup's ground lead induc-tance. To verify, connect the scopeprobe to the chassis ground near thesame point as the ground lead. Youshould see the same low-level ripplesas before, particularly on a widebandscope, but you now know that this is afunction of the test setup rather thanthe regulator.

While I found that 797-based regu-lators oscillate in some cases and notin others, I was determined to makethis op amp work properly. Walt and Iagreed on the need to ensure that the797 could be reliably implemented—with remote sensing—in a variety of lay-outs. To solve the oscillation problem,Walt devised an implementation fordecoupling the remote sensing lines atvery high frequencies (Fig. 5). Thepolarity of the regulator isn't defined,since both positive and negative usethe same decoupling topology. Thelocal AC bypass removes the remotesense feedback path, and its associatedphase shift, at very high frequencies.

First, solder a O.OljiF Panasonic V-series stacked film capacitor betweenthe load (output) and sense pads on thePC board. You can solder this small capto the board's foil side. Use insulatingsleeving, particularly on the negativeboard where the cap must jump over aPC trace. Next, insert a lOii, %W resis-tor in series with the remote sense lineat the load. This R/C network results ina -3dB point of 1.6MHz; the regulatorstill qualifies as a wideband audiodevice, but the chance of oscillation isgreatly minimized.

With remote sensing, I recommendthis decoupling regardless of whichop amp you choose. Since no one canpredict the effect of every possible lay-out and implementation, you muststill check the supplies with a scope.The 10Q resistor actually changes theDC gain of the op amp and raises theoutput voltage. The change is veryslight, though -10Q is the tolerance ofthe Ik feedback resistor. Even with

worst-case tolerances, a 5V regulatoris well within safe operating limits forlogic circuits.

A final note on oscillation: even reg-ulators built with the AD848 can oscil-late if a low-Z film cap is placeddirectly across the regulator output(as noted in Part 2, p. 34). It is veryimportant to build the PC board as speci-fied! Don't be tempted to add any filmbypassing to the input or output. Alow-Z film cap across the output elec-trolytic will virtually guarantee regula-tor oscillation. If the regulator is with-in 2" of the powered circuitry, don't

use any local film bypassing, either.Local film bypass capacitors shouldbe at least 3-4" from the PC board. Indifficult circumstances a ferrite beadbetween the regulator output and theload can be helpful. Jan's sidebaroffers some helpful suggestions foroscillation problems.

Dropout WarningsWhen I first tested regulators builtwith the AD797, I found that thedropout voltages were not as low asthose noted by Walt (his measure-ments were based on the AD848).

The Audio Amateur 4/95 39

With a 100mA load, my positive regu-lators measure as high as 2V, whereas1.5V or less is typical of the AD848,even with loads of several hundredmilliamps. The 797 requires moreinput headroom than the 848, primari-ly because of differences in its outputstage design.

The op amp in Walt's Fig. 8a mustbias up to nearly the same DC potentialas the output DC voltage, since the VBEof Ql and the VF of D4 essentially can-cel. As the input voltage Vg is lowered,the output swing limitation of Ul canlimit the regulator dropout if the opamp voltage limit with respect to itssupply rail is significantly higher thanIV (the dropout of the current source,Q2, D5, and R7). The 797's output can'tswing as close to its rail voltage as the848, which results in higher dropoutvoltage for the regulator.

The following enhancements aresuggested to improve the dropoutvoltage with both the AD797 and theAD848 op amps. (Part numbers arelisted in Table 3.) These changes arelisted in order of decreasing sensitivi-ty. In each step, the first reference des-ignation pertains to Part 3, Fig. 2 (pos-itive regulator); the second designa-tion in parentheses refers to Part 3,Fig. 3 (negative regulator); the originalreference designation to Figs. 8a and8b, from Part 1, appears in brackets.

1. Change D7 (D9) [D4] to aPanasonic 30mA green LED. Thisenhances the output swing of the op

amp. Don't be tempted to substituteanother LED: the 2V drop across thespecified part serves as a level shift,and is critical.

2. Lower R20 (R21) [R5] to 10k or12k. This yields a small improvement,typically 0.05-0.1V. Note that Jan hasalready made this change in Part 3.Don't worry about this unless youbuilt a regulator from Walt's Figs. 8aor 8b, and low dropout is an issue inyour application.

3. Change current source Q3 (Q4)[Q2] to a device with lower VSAT Thisimproves dropout by about the sameamount as the resistor change in step2. Recommended transistors arePN2907A (PNP-Q3) for positive regu-lators and PN2222A (NPN-Q4) fornegative.

With all three changes, the dropoutvoltage will be close to IV, but the firsttwo steps will get you most of theway. You don't need to change Q3 andQ4 unless you absolutely must squeakout another 0.1V or so. Low dropoutwon't be important if your raw inputrail voltages are high enough. The flipside of this coin is heat dissipation.One of the advantages of low-dropoutregulators is they enable you to uselower raw DC voltages. Most of thepass transistor's heat is produced sup-plying current to the load, rather thandropping voltage. Remember that fora given current drain from the regulat-ed output, the heat will increase as theraw DC voltage is raised.

TABLE 3

PARTS LIST FOR MISCELLANEOUS REGULATOR CHANGES

REMOTE SENSE DECOUPLING0.01U.F/50V Panasonic V-series capacitor (Digi-Key P4513-ND)lOflWW, 1% Yageo metal film resistor (Digi-Key 10.0X-BK-ND or Roederstein MK2)LOW-DROPOUT MODIFICATION30mA Panasonic green LED (Digi-Key P309-ND)

PN2907A transistor, TO-92 case (Digi-Key PN2907A-ND)PN2222A transistor, TO-92 case (Digi-Key PN2222A-ND)

10k, %\N, 1% Yageo metal film resistor (Digi-Key 10.0KX-BK-ND or Roederstein MK2)±5V VERSIONS

National Semiconductor LM336BZ-2.5 (Digi-Key LM336BZ-2.5-ND)2.49k, ttW, 1% Yageo metal film resistor (Digi-Key 2.49KX-BK-ND or Roederstein MK2)4.99k, Vt W, 1 % Yageo metal film resistor (Digi-Key 4.99KX-BK-ND or Roederstein MK2)

SOURCES FOR OTHER REGULATOR PARTSAnalog Devices AD797AN and AD848JN (Newark Electronics, both items listed in current catalog #114)Panasonic 120nF/25V HFQ capacitors (Digi-Key P5698-ND)0.1|iF/50V Panasonic V-series capacitors (Digi-Key P4525-ND)LM329 Reference (Digi-Key LM329DZ-ND)1N4148 Diodes (Digi-Key 1N4148-ND)2N5087 Transistor (Digi-Key 2N5087-ND)2N5089 Transistor (Digi-Key 2N5089-ND)D44H11 Transistor (Mouser 570-D44H11)D45H11 Transistor (Mouser 570-D45H11)Roederstein MK3 series, 1k, 0.5W (Michael Percy)

Other VoltagesThese regulators can be adjusted forother output voltages, though ±14Vand ±5V should cover most audioapplications. Table 1 in Part 2 sup-plies alternate resistor values forvoltages from 10V to 18V.

Some of you may be tempted toraise the rail voltages beyond ±14V forpreamp power supplies. With the gain

So It Oscillates—Now What?By Jan Bidden

As Gary explains, you should definite-ly check your supplies for oscillations.They are not inherently unstable, butwith so many variables there is alwaysa chance. Use the following checklist tosystematically review and remove thepossible causes.

1. Assuming you use the proposedPCB, did you build it as described inParts 3 and 4? No extra film capsshould be placed at the regulated out-put!

2. Have you limited the lead lengthsas much as possible? Preferably, theyshould be no longer than 4 or 5". Besure to twist the raw supply lines andthe regulated load connections, but notto each other. Do not connect the senseshield to the ground point at the load,only at the PCB.

3. Mount the board(s) over a metalenclosure wall or partition, as close aspossible. This will decrease any oscilla-tory tendencies, and also improve thenoise figure.

4. Check the circuit to be poweredfor excessive decoupling capacity. Withthese very low impedance regulators,more than 100|iF or so is overkill, andpromotes instability. If you use filmcaps at the load, don't make them larg-er than luF or so. I know it goesagainst the grain to actually remove afilm bypass cap. Although they are asolution to many problems, in thisapplication they can actually causeproblems.

5. The remote sense decoupling fil-ter should take care of any remainingoscillations, but in persistent cases youcan increase the resistor value to 22Qand the capacitor to 0.015uF, with neg-ligible impact on performance.

6. The AD848 in this application is abit more stable than the AD797, so if allelse fails this could be a solution. AsGary notes, this is not as good sonically.

7. Finally, don't get discouraged. Wehave built many of these regulators,and every one could be persuaded towork as advertised.

40 The Audio Amateur 4/95

determining resistors set at Ik each,the op amp has a voltage gain of 2.When the reference is the 6.9V LM329,this actually produces 13.8V. The 10Qresistor in the remote sense decou-pling brings it up to around 13.9V.

In most cases, there's no need tooperate preamp power supplies athigher rail voltages. Even with thesevalues, my modified Adcom GFP-565outputs close to 8V RMS before clip-ping. Since most power amps clip with2.5V input, it is pointless to raise thepreamp rail voltages. Contrary to whatsome believe, higher rail voltageswon't give you more "headroom" ifthe next device (i.e., your power amp)can't handle the input signal.

Listening EvaluationsA set of ±14V regulators has beeninstalled in my extensively modifiedAdcom GFP-565 preamp for over sixmonths. Throughout most of the mod-ification process (subject of a futurearticle), I have used a second, unmod-ified GFP-565 for comparison.Installing the high-performance regu-lators in the 565 affected nearly everyaspect of performance, with the moststriking improvement in the area ofdynamics. I was amazed to find that,even though the line stage gains inboth units were identical, the modi-fied 565 actually played louder thanthe stock preamp. This may seemstrange at first, but there is a logicalexplanation.

The new power supplies offer asense of unrestricted dynamics; fullorchestral crescendos are renderedwith a sometimes overwhelmingimpact. Subjectively, the original sup-ply regulators compress orchestraltutti passages, whereas the modifiedpreamp releases them with full force. Iwas repeatedly lowering the volumerelative to the stock preamp to achievethe same subjective playback levels.This was frustrating, since I was nowplaying most of my reference CDswith the volume around 9:00, leavinglittle room for adjustment. Walt and I

ACKNOWLEDGMENTS

Many thanks to Walt Jung and Jan Didden for their excel-lent and hard work throughout this project. Their involve-ment did not end with the publications of Parts 1,2, and 3,but continued until the completion of this article, including(but hardly limited to) proofreading of this manuscript.Special thanks to Rick Miller for providing the transformermeasurements, and considerable information on relatedissues, and to Walt for preparing the laser printout fromRick's ,HGL file. Rick's important work on common modechokes in raw DC supplies is also greatly appreciated.

have actually lowered the voltagegains in our line stages from 11 (a10k/Ik feedback combination) to 5(4k and Ik) to allow a greater range ofvolume control adjustment.

There's more to the dynamicimprovements than sheer volume,however. With the new regulators, thepreamp sounds effortless no matterhow taxing the source material. Evenwith the most demanding recordings,it remains clean and detailed, free ofharshness or edge.

These super-quiet regulators alsolower the subjective noise floor: low-level dynamics aren't artificially ele-vated, they simply descend effortless-ly. In "Siegfried's Funeral March"from Wagner's Gotterdammerung(Solti, London CD 414-115-2), it is easyto miscalculate the dynamic contrasts.If you adjust the volume so the softtimpani notes (CD 3, Track 7) are at acomfortable level, the fortissimo at theclimax can be unbearably loud. Withthe climaxes set at a realistic level, the

The Audio Amateur 4/95 41

entire dynamic range sounds muchcloser to a real concert hall experience.

The new analog regulators alsoincrease soundstage size, both left-to-right and front-to-back. Prior toinstalling them, I had difficulty sepa-rating the bass drum from the timpaniin "The Hut on Fowl's Legs" fromMussorgsky/Ravel's Pictures at anExhibition (Reiner, RCA Victor LivingStereo CD 61958-2, Track 14), not interms of timbre but of localization.The placement of these instruments isnow reproduced with pinpoint accu-racy, and considerably deeper in thesoundstage than before. Inner detailand articulation are also improved.

To compare the AD848 and AD797op amps, I soldered machined-pin,gold-plated sockets to the ±14V regu-lator board. I then soldered the opamps to gold-plated headers (withCaig ProGold contact conditioner onthe header pins) for a gold-on-goldcontact.

The 797 reveals greater inner detailfrom recordings than the 848. Its sonicpresentation is also a bit more "laidback," more natural and musicallyconvincing. The soundstage is notonly deeper, it seems to have beenmoved back slightly. The 848's presen-tation is closer and more "forward."The line rejection measurements bearout these differences, and it's not sur-prising that the results of improvedline rejection are similar to those ofbetter power line filtering.

DAC RegulatorsI installed and evaluated the newdigital regulators in the DAC960 inthree phases: a +5V regulator for thedemodulator board; +5 and -5V reg-ulators for the TDA1541A DAC chip;and a dedicated -15V supply for theTDA1541A. All digital regulators usethe AD797. Instruct ions for thisprocess are beyond the scope of thisarticle. If there is sufficient interest,I'll prepare an "Ask TAA" column onthe subject. Write to me (c/o TAA) ifyou would like to see this published.

Based on my experience upgrad-ing the original digital supplies, Iexpected similar improvements fromthese changes, and the DAC regula-tors would offer "more of the same."My assumptions were wrong. Eachof the upgrades produced differentresults. All listening evaluationswere conducted with Analog DevicesAD1890 evaluation board for jittersuppression.8

42 The Audio Amateur 4/95

The real surprise is the demodula-tor regulator, which yields animprovement in dynamics and basssimilar to the analog regulators inthe 565 preamp. The effect on weightand impact in the bass region is likegetting better subwoofers or a newpower amp. The bass drum inReiner's Pictures is deeper and morepowerful. In Ernest Ansermet'srecording of Ravel's Alborada del gra-cioso (London CD 433-717-2), thebass drum, while always quiteimpressive, is now even heftier thanI had realized.

I was completely surprised at theability of a digital regulator to makesuch a striking improvement in thebass. This is undoubtedly jitter-relat-ed, since the demodulator board'sinput switching circuitry and inputreceiver should have a significanteffect on jitter performance.

The Ansermet recording is alsoincredibly clean and well-defined, notjust in the bass but across the entirespectrum. The ±5V supplies for the1541A DAC result in improved articu-lation, detail, and the sense of air andspace around the instruments. Ravel'scolorful orchestrations are reproducedwith an openness and transparencywhich are uncanny, and the delicacyof his scoring is far more evident herethan in my British-pressed LondonTreasury Series LP, which sounds dulland lifeless by comparison (most ofLondon's "Made in England"Richmond and Treasury Series LPswere extremely good; when theybegan pressing these LPs in the US inthe mid-1970s, the sound quali tybecame abysmal).

Track 2 of Reiner's Pictures, which Iregularly use as reference material,has a series of four string glissandibowed close to the fingerboard. Priorto installing the ±5V DAC regulators,the effect sounded merely like fingerssliding up and down the strings. NowI can clearly hear the subtle articula-tions of the bow. Such a soundstagesubtlety often goes unnoticed. Thefour glissandi begin with the violas,move to the celli, then the second andfirst violins. On a less refined D/Aconverter, there appears to be a gener-al movement from right to left. Withthese regulators, the exact placementis clear: right, far right, left, far left.Musical subtleties, carefully notatedby Ravel and superbly executed bythe Chicago Symphony, are revealedin all their glory by the DAC960.

I have noticed an amazing numberof tape edits on CDs made from ana-log sources, which, prior to installingthe DAC regulators, had escaped myattention. One example is the EMIreissue of Boris Christoff's 1962 stereoremake of Mussorgsky's BorisGodunov (Andre Cluytens, CDS747933-8), during Boris' Monologue inAct II (CD 2, Track 6). Another occursin the Solti Gotterdammerung, in thefinal scene in Act II (CD 2, Track 12).Quite a number of tape edits are audi-ble on the Decca/London operas pro-duced by John Culshaw. Some ofthese edits were not obvious until Ireplaced the DAC regulators. In fact,the DAC960's ability to resolveminute details is so great that I some-times hear off-axis microphone col-orations as the singers move aroundthe soundstage.

Icing on the CakeThe sonic effect of the new digital reg-ulators is nothing short of dramatic,every bit as important as the analogregulators in my preamp. After spend-ing nearly a week with the DAC960 inthis state, I decided to give theTDA1541A a dedicated -15V supply.While this effect was more subtle, itwas nonetheless worthwhile. Low-level resolution and detail wereenhanced a bit further, with the lastounce of performance squeaked fromthe 1541 A.

The -15V supply is critical, since itis the voltage source for the DAC'scurrent outputs. If you check this sup-

SOURCESDigi-Key Corp.701 Brooks Ave.S,PO Box 677Thief River Falls, MN 56701-0677(800) 344-4539, FAX (218) 681-3380

MCM Electronics650 Congress Park Dr.Centerville, OH 45459-4072(513) 434-0031, FAX (513) 434-6959

Michael Percy Audio ProductsBox 526,170 HighlandInverness, CA 94937(415) 669-7181, FAX (415) 669-7558

Mouser Electronics958 N. Main St.Mansfield, TX 76063(800) 346-6873

Newark Electronics(312)784-5100(Call for branch nearest you)

Signal Transformer500 Bayview Ave.Inwood, NY 11696(516)239-5777, FAX (516) 239-7208

ply line with a scope, you'll see somelow-level digital hash. This will alsoappear on the -15V rails to the analogcircuitry if they share a common sup-ply. Part of the perceived improve-ment may be due to removal of noisefrom the analog supply rail.

Some of you may wonder why Icontinue to modify a digital-to-analogconverter that uses an "obsolete" digi-tal chip set. Some truly wonderfuldigital filters and D/A converter chipsare now available, with 20-bit resolu-tion, 8x oversampling, and such.Several manufacturers produce D/Aconverters equipped with HDCDdecoding capability.

Having upgraded the regulators inthe DAC960 and heard their soniceffects, I honestly believe no one real-ized the full potential of the Philips16-bit chip set while it was in produc-tion—least of all Philips. I recentlyauditioned an $800 HDCD D/A con-verter made by a leading Americanmanufacturer which was inferior tomy DAC960 in every respect, evenbefore the high-performance regula-tors were installed. It used three-ter-minal adjustable regulators for theanalog circuitry and 7805 types for thedigital circuits.

There is a lesson to be learned fromall of this: digital circuitry, no matterhow sophisticated, will never per-form to its potential with cheap, low-end power supply regulators.Manufacturers of both digital record-ing and playback equipment need toseriously reconsider the criticalness ofpower supply regulation to the per-formance of high-end digital circuitry.If you own a DAC960 modified toPooge 5.5 standards, I believe the dig-ital supply upgrades are well worthinstalling. You may find that the

REFERENCES1. G. Galo, "Pooge-5: Rite of Passage for the DAC960,Parts I and II," TM (2/92,3/92): 10,34.

2. G. Galo, "Pooge 5.5: More DAC960 Modifications," TM(1/94): 22.

3. G. Galo, "P-94-SR Stereo Parametric Equalizer," TAA(2/86): 41.

4. N. Pass, "Pass/A40 Power Amplifier," TM (4/78): 4.

5. G. Galo, "Distributor Delight," TM (3/86): 52.

6. R. Miller, "Measured RFI Differences Between RectifierDiodes in Simple Capacitor-Input Power Supplies," TM(1/94): 26.

1. Noise Suppression Reference Manual. TopazElectronics, 1660 Scenic Ave., Costa Mesa, CA 92626,(714)557-1636.

8.G.Galo,"Ask TM" TM(4/94):41.

DAC960—once again—outperformsmany expensive products with moresophisticated digital circuitry.

ConclusionsA great deal of discussion about thevirtues of shunt regulators hasoccurred in the audio press over thepast few months. Since these devicesrequire a series resistor terminated bya shunt capacitor, they automaticallyprovide low-pass filtering of powerline and rectifier noise. As importantas this may be, several other criticalperformance areas must be addressed.

Before choosing shunt regulators foryour next project, I suggest verifyingtheir measured performance in everyarea, as discussed in Part 2 of thisseries.

Walt Jung's high-performancepower supply regulators set new stan-dards of performance in both analogand digital applications. They requirea great deal from the builder, yet Ihave found the sonic improvementsworth every hour spent on this proj-ect. I hope that you agree. The cumu-lative improvement in my audio sys-tem has truly been a revelation. •

The Audio Amateur 4/95 43