LM338 based regulator

My primary goal when I started to build regulated supply was to have a GC amp with decent power (cca 25W/8 Ohms) but supplied by lower voltage than it is necessary when only 1000uF caps are used. As I noted previously I tried bigger caps in parallel with those 1000uF and did not like it. Another option was to parallel many 1000uF caps, but I did not like how it would look physically. Regulated supply appeayellow as an option since necessary bigger caps can be used before the regulators. Actually the fact that a whole capacitor setup becomes more convenient this way is not new.

Negative regulators with necessary current ratings are neither easy to find nor are cheap these days, but if each secondary has its own rectifier, the positive regulators can be used both for positive and negative rails.

Diodes D1/D2 are optional (can be used any 1N400x but if you want to have some safety margin use those 100V or higher rated) and protect regulators from discharging of stoyellow energy in the output capacitors (in the case of short at the input sides of the regulators). Transformer’s 24V secondary voltages with Schottky diodes and 2x4700uF per rail will provide enough headroom to the regulators and still won’t heat them unnecessarily. However, if one supply drives both channels, a Volt or two more on the transformer's output might be good rather than bad. Since the regulators put out stable and smooth voltage, caps following them (which are mounted directly at the chips) don't have to suppress 100/120Hz ripple but are used for HF filtering only. The LM338’s output impedance suggests that their capacitance can be seriously below 1000uF thus giving the chance for usage of seriously good caps for the modest price.

Considering said output impedance of the regulator it seems the best to use 1uF for C1/C2 but the power supply rejection is better if conventional 10uF is used and it is even better if its value is somewhat taller. (Added on 20th April 2004: Note that in the case of 10uF or more capacitance here, the datasheet suggest the usage of protection diodes on the adj pins too.) I suggest checking the graphs showing the influence of this cap on the LM338’s PSRR and output impedance. Output impedance graphs comprise the curves if a few different values of the output caps (C3/C4) are used and hence are helpful to determine their proper value. C3 and C4 should be, of course, mounted at the chip’s pins and, following the same principles, regulators should be physically as close as possible to the chips.

Power dissipated on the regulators equals to about 20% of the power dissipated at one LM3875 chip. In the case of one supply for both channels percentage will, of course, double. Hence regulators demand an adequate heatsinks.

As about the sound of the regulated supply… Let’s say I achieved my goal, and even more than that. I strongly recommend this or such regulated supply.

One similar supply based on the LM338 is described on the Class A Amplifier Site.

After this experience, with all due respect I have for Kimura, I doubt that success in usage of 1000uF capacitors has anything with the solution of the “if energy supply depends on the capacity of filter/condensers, you can easily lose the freshness of sound” problem. Simply, the LM3875 and similar chips (being significantly current sourced units) can tolerate certain ripple but are still quite supply sensitive in the HF/RF domain. View to the Power Supply Rejection Ratio vs. Frequency graph in the LM3875's datasheet gives quite clear picture of this. Attempts to keep supply's impedance low inside the large frequency bandwidth by caps bypassing introduce some problems (as bypassing I mean putting into the parallel capacitors of different values; again, while one bypass cap can be acceptable, it could be in most power-amp cases still inadequate). Smaller caps have better HF behavior. The answer was one 1000uF. I doubt that there is something more to be said about this.

© Pedja Rogic, July 2003.

     Discrete regulator without global feedback

Not much of fancy stuff. Rather like back to the roots. Or KISS. Zener referenced emitter follower. Since the current gain of one transistor is marginal for this purpose, I used the Darlington from the start. Actual part count or number of solder joints is somewhat taller than in the case of the integrated regulator (if you consider that the protection diodes might be used here too), but this circuit was enough to make me believing in the non-feedback regulator.

Comparing against the, nowadays conventional, regulators with feedback, the general shortcoming of such regulators is their relatively high output impedance. The output impedance depends on the drawn current (the higher the current, the lower the output impedance) and while it is certainly not the state of the art, it is still decent though. While the amp is in idle (35mA in the case of the LM3875 assuming separate supplies) the output impedance is about 1 Ohm; any more serious drawn current (300-400mA which is about 1W of power on 8 Ohm load) will drop it below 0.2 Ohm. Its inductive part has smooth raise i.e. it is good partner to the output capacitors and 100uF will be pretty enough. The Power Supply Rejection Ratio will depend on the dynamic impedance of the used Zener, but something about 40dB should be expected.

As a very rough rule, using Schottky diodes, a transformer’s secondary voltage should be equal to the used Zener’s voltage. A Volt or two more is recommended for safe headroom if supply is used for both channels. A higher voltage won't hurt in itself but will heat the transistors wastefully. The output voltage is equal to the Zener's voltage minus total Vbe of the used transistors. Shown transistors are the ones I had at hand so they are not critical, but something with decent hFE should be used undoubtedly.

A few things were added then. Some of them improve the objective performance of the circuit (PSRR and noise), some of them improve the sound without notable relation with any objective technical parameter. Say, once you add the low pass filter on the voltage reference, the current source won’t do much of anything (if low pass filter has low cut-off frequency, the overall PSRR of the circuit can’t be further improved by taming the voltage reference), but the sound is notably better that way.

R1 and R2 are set for 7-8mA current through the J309. Needed resistance depends on the Idss of the used samples but R1 and R2 will be probably inside 100-200 Ohms range.

Simulated power supply rejection of the active circuit (without input 2x4700uF and output 100uF capacitors) looks this way:

Z is Zener, B1 is the base of the first transistor, E1 is its emitter, E2 is the emitter of the second transistor and OUT is the output (you'd surely never realize that). And below is the PSRR including 100uF output (ideal) capacitor. No problems with the integration.

The TL431 was also tried as a voltage reference in this circuit and it indeed sounds different than Zener, but this difference is not so simple. Technically, at the point indicated as "Z", the PSRR will be about 30dB better, but, as said above, this won't reflect on the total PSRR of the circuit. Sonically, it gives better overall dynamics, a bit bigger soundstage and more fluid midrange but simple Zener has to my ears better tuned bass and (enjoyable!) cleaner highs, better defined beginning and end of the notes and has more seductive overall presentation. And yes, I used the TL431 bypassed for AC gain and in this case I’d recommend usage of the output caps (from TL431’s cathode to anode) of at least 100uF. Most simple: make the current source for 7-8mA and try both – both options have a reason to exist.

Of course, these circuits won’t work unless they are plugged to the circuits they feed. To test the circuit before that, close the outputs adding resistors to the ground.

Though it actually does not have anything with what I'd recommend here as a "layout", nothing beats point-to-point approach for the experimenting purpose. The sculptures made this way can be sometimes funny; possible ka-booms, however, won't be funny like that.

© Pedja Rogic, January 2004.