The design is pretty much a standard one, and Im sure there are commercial units out there that are similar. To my knowlwdge, it is not an exact copy of any commercial unit, nor am I aware of any patents on the topology. To experienced builders: I realize that many improvements and refinements can be made, but the idea was to keep it simple, and should be do-able by anyone who can make a circuit board and has the patience not to do a sloppy job.
The input stage is an LF351 op amp which provides most of the open loop gain as well as stabilizes the quiescent dc voltage. This feeds a level shift stage which references the voltage swing to the (-) rail. The transconductance stage is a darlington, to improve high-frerqency linearity. The 2SC2344 by itself has a rather large collector-base capacitance which is voltage dependent. The MPSA42 presents this with a low-z and has a C(ob) of only a few pf that is effectively swamped by the 33pF pole-splitting cap. The stage is supplied by the 2SA1011 active load (current source) which is about 20 ma. The current to the stage is limited by the 2N3094 to about 70 ma under worst case.
The output is a full complementary darlington with paralleled outputs. Although you could "get away with" only one if only 8 ohm easy-to-drive loads are used, this is not recommended. The use of parallel devices increases the ability to drive reactive loads (which can pull a significant current while the voltage waveform crosses zero and puts a high voltage and a high curent across the transistor simultaneously), gives the amp a higher damping factor, and reduces the maximum current each transistor has to supply to peaks (remember, the gain of a power transistor drops as the current increases).
Compensation is two-pole and one zero. The op-amps pole and the pole generated by the 33pf cap and the 470 ohm bias resistor of the MPSA42 dominate. (the 33pF gets multiplied by the stage gain.) The 22 pf feedback capacitor provides lead compensation, and is taken from the output of the tranconductance stage rather than the output itself. In this way, the phase lag introduced by the output transistors is not seen by the high-frequency feedback. This intorduces a closed-loop pole which limits the high-frequency response. The two compensation capacitors must be type 1 creamic (NPO) or silver mica - with ZERO voltage coefficient.
The amp was designed to run 2 channels off a +/- 55 volt unregulated supply, reducing to +/- 48 volts under full load. It used a 40-0-40 volt, 5 amp toroid transformer, a bridge rectifier, and 10,000 uf of filter cap per side. If a standard EI transformer is used, a 6-amp rated unit should be used. With this power supply, it produces 100 watts continuous, both channels driven into 8 ohms resistive with no clipping. Dynamic headroom is about a db and a half. For more headroom, unloaded voltages to +/- 62 volts can be used with no circuit modification.
By the way, the schematic is in Postscript.
Limitations:
With no modifications the amp will drive 4-ohm speaker systems with no current limiting. The short-circuit current limit is set to about 4.5 amps peak, which will handle conventional speaker loads.(It will, of course, produce higher peak currents as the output voltage swing approaches the rail.) If you are going to be running some of those high-end speakers with impedance minima of half an ohm, or that stay reactive throughout most of the audio band ( ie, 0.5 +j3.2 ohms) you will probably already own a better amp than this. If the higher-power Motorola power transistors are used, it will drive a 2-ohm resistive load without problems (except heat).
I have never heard any slew-induced distortion on this amp with a CD players band-limited (22KHz) signal. I suppose that real high-end freaks could pick it to pieces by hitting it with a TTL square wave mixed with a 19KHz stereo pilot tone and crank it up. I guarantee that there will be spurs all over the spectrum, but who listens to that?
Possible Modifications: (What if I want mo power???)The Toshiba output transistors (2SD424/2SB554 pair) shoud not be used with supply voltages above +/-60 volts. If you plan on cranking it up, use more in parallel or use the 250 watt Motorola pairs (MJ15024/MJ15025). If very low impedances are expected, raise the bias in the transconductance stage to give more base drive to the output darlingtons or add another current gain stage. Higher-Beta (and faster) power transistors cant handle reactive loads worth a crap. Dont substitute high-fT parts unless you are sure they have adequate second-breakdown capability.
The NE5532 op-amp can be used in the input stage. If more than one are used off the +/-15 volt shunt regulators (balanced ins, anti-slew Bessel filters, etc.) the 2.7K dropping resistors may need to be reduced to say, 1.8K ohm to maintain regulation. The 2.7K resistors will allow up to 4 LF351 type op amps off the regulator (I used a quad 347 for balanced inputs to avoid hum in a DJ setup).
Construction tips:The output transistors and thermal compensator (2SC1567) will need to be mounted on a common heat sink - a finned unit measuring 5 in. high by 8 in. wide with 1.25 in fins should do nicely for one channel. (They look nice if you make the sides of the case out of them). Most normal applications wont require more cooling than this. The reason the 2SC1567 was chosen for the output bias regulator is because it is fully insulated - the ECG version will require additional mounting hardware. TO-3 hardware for the outputs is cheap and easy to get.
The driver transistors and voltage amps (2SC3344/2SA1011 pairs) will all require heatsinking as well. Individual TO-220 heat sinks on the circuit board will suffice - the voltage amps dissipate about 1.4 watts each. A common piece of 1/8 in. thick 1 in. wide X 4in. long angle aluminum will suffice for all 4 on each channel, but bear in mind that it must be oriented to take advantage of natural convection, and the transistors must be insualted.
Keep the imput grounds separate from everything else, and return them at ONE point. Failure to do so WILL result in high distortion (5% or so), or even oscillation.
The output stage bias should be set to about 25 milliamps in the output transistors. This value takes a while to stabilize, and you may have to monitor it over an hour or so during initial setup. To measure it, measure the voltage across the emitter resistor and use Ohms law. This way, you can check the current sharing in the parallel output transistors at the same time and change them if there is a serious discrepancy. With parts of the same date code, they should not be off by more than 10% after it has warmed up. Higher output stage biases can be used, but it takes more care in setting it. If you want an idle current of more than 50 milliamps per side, increase the value of the emitter resistors.
Initial Checkout:DO NOT just plug something like this in! A seemingly insignificant error can set your house on fire! (As well as blow out $30 worth of transistors in a microsecond.) A variac will work in theory, but the amp may latch to the rail if the supply drops too low. I suggest the use of a ballast resistor - a 60 to 100 watt light bulb in series with the AC mains. You get a bright flash when the caps charge, and then it goes (almost) out as the idling supply current reaches its nominal low value. The amplifier will then work normally at low volumes. If the amp draws too much current for whatever reason, the lightbulb will glow brightly, increase resistance, and limit the power to the circuit. Usually, there will either be a mis-wire (use your DMM) or oscillation (will show up on a scope or RF power measuring device). If the bulb goes dim-bright-dim-bright... then the amp is marginally stable and the grounding layout should be checked. Compensation capacitor values may need to be adjusted if any significant changes were made. Mine is stable the way it is.
Additional Notes:The schematic is in postcript, so it should just be able to be printed out. The emitters of the transistors are labelled by an "e". I was too lazy to put arrows on the transistor symbols - and Ive been using it that way for over a year now.
Trouble finding parts? MCM (1-800-543-4330) has all the transistors. Total cost for a stereo version should be between $150 and $250, depending on what kind of bargains you can find on the case, transformer, and heatsinks. If you have to pay "list" for everything, it will likely cost about $1000 to build.
The information included herin is provided as-is, with no warranties express or implied. No resposibility on the part of the author is assumed for the technical accuracy of the information given herein or the use or mis-use of said information.
The equipment described in this article was designed, fabricated, and tested on my own personal time using my own personal resources.
Author:, warren@eggo.csee.usf.edu
