Report - June, 1995
By Bob Carver
The Sunfire Amplifier: A Power Supply Story
The Sunfire amplifier had its genesis over fifteen years ago. After I sold Phase Linear, which I founded in the early 70's, and decided to start Carver Corporation, I wanted to come out with a new amplifier that would be light years ahead of anything currently available. I began work on a signal tracking power supply. Successfully implemented, an amplifier that incorporated such a power supply would he able to deliver lots of power, would run stone cold, would be incredibly efficient; essentially all of the input power would become output power, would be able to deliver massive amounts of current, and would drive almost any impedance down to one ohm and below. It would have the potential of ultra-high reliability because it would be running cold, would not require heat sinks, and because it would be so efficient, the power supply could be much smaller for the equivalent output power (in a conventional amplifier, only 20% to 30% of the input power actually appears at the output of the amplifier as usable audio power). I toiled for over a year trying to make this into a reality but couldn't get it to work. And so, after a year of working until two in the morning, I finally gave up and instead developed a different power supply called the Magnetic Field power supply. That power supply and its power amplifier became the original Carver "Cube". I used that to start Carver Corporation.
Fast Forward 13 Years
A little over two years ago, while still at Carver Corporation, I decided to have another go at it. I pulled out my notes from years past, including the old patent. This time I succeeded and did so in spades. The resulting amplifier was able to deliver both massive power as well as current. It could operate down to one ohm, and it didn't get hot! In short, it fulfilled the original dreams I had years ago. I called that amplifier the Lightstar, and on December 17, 1992, I turned over the design to my engineering department for packaging (having completed about 95% of the work) and went on a sabbatical with the intention of final tweaking and voicing when I got back. Upon my return, I had a falling out with Carver Corporation and early last year left Carver to form the Sunfire Corporation. At first it was Zeus Audio, named after my puppy, but I received a letter from an attorney who said, "No, you can't name it Zeus because we represent an amplifier company, and we have names for our products like Hercules, Aphrodite, Apollo, and Zeus." Enter Sunfire.
How it Works
In order to understand how the Sunfire amplifier works, it would he helpful to review a conventional amplifier and illustrate some of the very difficult engineering problems associated with powerful and very high current amplifiers. As you know, a conventional amplifier has a power supply, and for a 300 watt amplifier, the power supply voltage is approximately 90 volts. Those 90 volts are parked way up in the sky at 90 volts above ground zero. The audio signal varies under that voltage, and as long as the amplitude of the audio signal remains below 90 volts, the amp will not clip or run out of power. Of course, if the audio signal is required to be larger than the 90 volt power supply, the amplifier will clip. Now, In a conventional amplifier, when the amplifier is delivering power to the loudspeaker load, the current flows out of the power supply, through the output transistor or transistors, and then into the load.
As an example, assume the output voltage at the loudspeaker is 30 volts and that 10 amperes of current are flowing. The current starts at the power supply and flows through the transistors and in doing so, it makes them get hot. How hot? The measure of hotness is power (in watts) which is equal to voltage times amperage. Remember, there are 10 amperes flowing, and if there are 30 volts on the loudspeaker with a 90 volt power supply, that leaves 60 volts across the transistors. Recalling our formula for power, that is 60 volts times 10 amps, which is 600 watts! That is not the power going to the load; that's the power going into the transistors as heat and must be dissipated. Hence, the transistors are mounted on a large heat sink; the heat is transferred to the heat sink and ultimately to the atmosphere.
Now, since the amplifier is only about 20% to 30% efficient, a lot more power has to go into the amplifier than comes out because 600 watts are going up in heat. Since it's inefficient, there must he lots of output transistors, lots of heat sink, and the power supply has to be much larger than would ordinarily be required in order to make up for all the power that's being wasted. Instead of a 30 pound power supply, it has to be 80 pounds. Well, so what? It's not difficult to add power supply and heat sink necessary to allow the amplifier to deliver the power. However, a problem that is very insidious exists! The problem is this. The output transistors that amplifier designers use are big 20 ampere output transistors. I use them, they are used in small amplifiers and large amplifiers. They are used in high end amplifiers and even used in most of the big receivers these days. It's a standard part in our industry. It's the big Motorola, Toshiba, the Sanyo, or Sony equivalents. This transistor is rated at 20 amperes. However, it's only able to deliver 20 amperes if there are 10 volts or less across it. That's because it's a 200 watt part and can never dissipate more than 200 watts, or its rating is exceeded. At 50 volts, for example, it can deliver only 4 amperes because 4 amperes times 50 volts is 200 watts. At 90 volts it can deliver only 2.2 amperes. Going back to the earlier example with 60 volts across it, it can deliver only 3.3 amperes. Not very much current. If a designer wants to have an amplifier that's able to deliver plenty of current into very low impedance loads, to deliver current in an unvarying way, no matter how difficult the loudspeaker impedance, no matter what the phase angle, many paralleled output transistors must be used - lots and lots of them.
Remember, the transistors are not good for 20 amperes, they are really only good for a small portion of that, especially when driving low impedance loads. The transistors have to be mounted on huge heat sinks, and, because the amplifier is not very efficient, it must have a huge power supply. Since each transistor draws its own idling current, the amplifier tends to run hot when it is just sitting there at idle. Biasing issues become very severe problems. To this day, solutions are still being sought.
For example, one manufacturer uses the sliding biasing circuit, and anaother uses a four-tiered switchable dynamic biasing circuit.
Engineers and designers forever fret over whether they're going to bias it Class A or Class AB or use a sliding bias scheme. Big problem. Amplifiers that can deliver this current do exist, but to get there you have to reach up to the big bucks amplifiers. Such amplifiers can deliver the performance, but they start at about $8,000. There is another way.
The Tracking Downconverter
In the Sunfire amplifier, that 90 volt power supply voltage that I mentioned earlier is removed from being parked 90 volts above ground and is brought down and parked at only 6 volts above ground. At any moment in time, regardless of what the output of the amplifier is, that power supply voltage will always be 6 volts above the output signal. If the output signal is zero, the output of the Tracking Downconverter will be 6 volts. If the output of the power amplifier is 30 volts, as in the previous example, the output of the Tracking Downconverter will he 36 volts. The voltage across the transistors remains a constant, unvarying 6 volts.
Now, consider the previous example. The amplifier was delivering 30 volts to the load and 10 amperes of current were flowing. That example resulted in 600 watts of power in the output transistors. In the Sunfire amplifier, that same 10 amperes is not dropping across 60 volts. It's dropping across 6 volts, and so the power is only 6 volts times 10 amps --- 60 watts wasted compared to 600 watts.
Going from 600 watts of energy dissipated as heat to 60 watts is an order of magnitude less. It's so little power that the amplifier does not have a heat sink. It doesn't need one. Yet it can deliver well over 2,000 watts into 1 ohm. And because of its increased efficiency, the power supply doesn't have to weigh 80 pounds. The power supply can be a more reasonable 30 pounds. But here's the best part. Remember that a 20 ampere transistor can only deliver the full 20 amperes if there are 10 volts or less across it (because of its 200 watt limit). In the Sunfire, since there are only 6 volts across the transistors at all times, the full output current of 20 amperes can be delivered from each output transistor instead of 2, 3, or 4 amperes as would be the case in a conventional amplifier.
Because the output transistors can deliver their full 20 amperes and there is never high voltage across them, they can deliver very high current into low impedance loads. In the Sunfire I use 12 output transistors per channel, each capable of 20 amperes, representing a peak-to-peak output current of over 240 amperes! And it can do so into very low impedance loads. That's a staggering amount of current. And that's what is required in order to have an amplifier that meets the performance needs of today's home theater and full orchestral audio consumer.
A remarkable feature of the Tracking Downconverter is its intrinsic and unique ability to transform high voltage and low current into low voltage and high current. For example, if the input power to the downconverter is being delivered at a very high voltage, the output power can be delivered at a very high current. The transformation ratio, i.e., how much the current is increased is in the same proportion that the voltage is decreased. In the case of the Sunfire, the power supply voltage is 2 times 125 volts, approximately 250 volts. Therefore, if the input current is 10 amperes and the output voltage is 25 volts (corresponding to a difficult or low load impedance), the output current will be 100 amperes (250 divided by 25 is 10, with the 10 ampere input current multiplied at the output by 10 for 100 amperes). It's this remarkable property of a Tracking Downconverter that allows the amplifier to deliver tons of current into vanishing low load impedance's. It is also the property that allows the amp to run cold, to have a smaller power supply than would be required conventionally, and to possess a very flat output voltage characteristic. Whenever the load impedance is halved, the power doubles (this is called "load invariant").
At this point in the design, the Sunfire is an amplifier that can deliver high current as well as voltage, delivering both simultaneously for tremendous output power, and runs cold. However, the design is not yet complete. The amplifier needs to he voiced. Voicing of an amplifier is the last 5% in its design (potentially the most time consuming) and is where the art of amplifier design is truly based. It is accomplished by altering the values of resistors and capacitors in various parts of the circuit, and then listening to the effect that a particular value has. When I voice an amplifier, I first use a female vocalist so that she can be accurately located in an acoustic area between the speakers, and in such a way that a believable halo of space surrounds her in three dimensions. Also, I fine tune the amp so that her voice is soft, musical, lyrical, and has a great deal of believability. (Not too much to ask, right!?)
After I finish the female voicing, I work on the male voice using baritones to get the chestiness that is a prominent feature here. When that part of the amplifier voicing is completed, I go to the symphony. I have in my head a template of what a symphony orchestra should sound like. I close my eyes and fit the sound of that symphony orchestra in my head, to the sound that my amplifier is making through the loudspeakers. (Baked fresh Salmon and a nice glass of Chenin Blanc, followed by the Seattle Symphony under the stars. Ahhh ... that's got it.) In the case of the Sunfire, after finishing the human voices, I found that the symphony orchestra locked in, and I had only minor adjustments to make, sort of like getting the flesh tones correct on a color television receiver, once that is done, all the other colors often lock in with very little further effort.
Current Source - Voltage Source
At that point I had an amplifier that was tremendous - lots of current, lots of voltage, incredible performance, and then I added a unique feature: a choice of outputs - voltage source output and current source output. Let me explain. A transistor is inherently a voltage source device; whenever an amplifier designer builds an amplifier with transistors, the result is a solid state amp that will typically have a very low output impedance approaching zero. A vacuum tube, on the other hand, is intrinsically a current source device. If an amplifier designer builds an amplifier out of vacuum tubes, the amplifier has a current source output characteristic, i.e., a higher output impedance. In my opinion, it's this high output impedance that is primarily responsible for at least 80% to 90% of what makes a vacuum tube amplifier sound the way it does: a glow to the midrange, a soft high end, typically a layered stage depth, and often a sound stage that is wider than it would be with a solid state amplifier. This musical presentation is very sumptuous and lovely to listen to, is quite captivating, and is the main reason many people love vacuum tube amplifiers.
Now, back to the Sunfire. There are two sets of output terminals on the back. One is a voltage source output, with a very low impedance (about 0.01 Ohm). The other is a current source output with a higher impedance (1 Ohm) current source output characteristic. The choice of which to use is up to the listener. If you want a solid state kind of sound, use the voltage source output terminals. If you want the vacuum tube sound, use the current source output terminals. Or - and this is the best part - you can bi-wire your speakers. Use the voltage source to the woofer, and wire the current source to the upper range of the system. That way you have the tight slam impact bass that a solid state amplifier can deliver, and you have the glow to the mid-range, the sumptuous sound stage, and soft, delicately detailed highs that current source amplifiers, i.e., vacuum tube amplifiers, typically deliver. This is the best of both worlds.
Sunfire Circuit Description, Amplifier Section
The input stage is a low noise FET operational amplifier operated in forced Class A single ended mode. The output of this stage drives balanced Class A level shifters and a balanced Class A voltage stage that swings the full rail of 250 volts peak to peak. The remainder of the current gain stages run full balanced with a constant VCE (collector-to-emitter voltage) of 6 volts to the loudspeaker. It is heavily biased into the Class A region for small signals and Class AB for large signals. Since the power dissipation in the output stages under simple quiescent bias conditions is 15 times less than a regular amplifier for the same output power, much more idle current can be used. The issue of how to bias this amplifier thus becomes moot - all but irrelevant - because of the 6 volts. Even though it has a vacuum tube output characteristic on the current source output terminals, there is not a vacuum tube inside at all except for the meter pilot lamp. It's fully solid state.
Coming in from the outside world, we find a conventional main power supply; a large power transformer, and filter capacitors. The output of this power supply feeds the Tracking Downconverter. The output of the downconverter is fully regulated and tracks the audio, receiving its input signal from the same signal that drives the main amplifier. Essentially, the downconverter is another power amplifier because its output voltage is in synchrony with, and tracks the audio signal, always above it a constant 6 volts. The input to the downconverter is a small signal Class A Motorola transistor. The output of this transistor drives a Texas Instrument PWM digital comparator. The output of the comparator drives a Hewlett Packard precision optocoupler which level shifts the digital control pulses to the gates of 12 International Rectifier Hexfets. The final output is smoothed into a continuously varying tracking voltage by the main energy storage downconverter inductor wound with #12 wire (this is heavy duty stuff) on a low loss non saturating ferrite inductor. The final energy storage capacitor is a 6.8 microfarad low FSR unit, and 12 dB of negative feedback are taken from this capacitor to the input stage. Finally, a Shotky free wheeling diode provides the energy return path for the Hexfet side of the downconverter inductor.
Many amplifier testers will operate an amplifier into an essentially dead short circuit and give it a pulse of 500 microseconds, 20 microseconds, or even 1 microsecond and measure the output current. This test is not a true estimate of the amplifier's capability since the output current can be very large when the output impedance is zero. As power is the current (I) squared, times the resistance (R), no matter how large the current measurement, the output power is actually zero under these circumstances. It is a parlor trick. The amplifier could never sustain those huge currents for more than a few hundred microseconds because if it did, the transistors would burn up. Take any conventional amplifier and do such a test with it, and you can have incredibly high current for a few hundred microseconds but not for any longer. Otherwise, the amplifier would blow up, because for high voltages that exist across the transistors during that small instant in time, the transistors are rated for only a few amperes (not tens or hundreds of amperes). However, this procedure does tell the amplifier tester a lot about the protection circuits. A skilled tester can determine whether it has current limiters, or power-fold back protection circuits, or whether it doesn't have any protection circuits at all and relies on fuses alone. The test does not tell anything about how much useful current the amplifier can deliver. A conventional amplifier may deliver 60 amperes for 100 microseconds but could not, under these conditions, ever deliver more than 8 amperes of current for longer than that. Not exactly a high current amplifier! By comparison, the Tracking Downconverter allows an amplifier to deliver those huge currents all day long without getting hot.
General Thoughts on Amplifier Design
Integrated Circuit Operational Amplifiers:
In the past, integrated circuit op amps have received a bad rap for use in audio circuits and for a simple reason. My experience has been that in a sampling of op amps from any manufacturer, I find that about one in fifteen will exhibit crossover notch distortion. The reason for this is that most op amps operate with a Class AB output stage, but they do not have a control for adjusting the idling current. However, most of the op amps will exhibit none of this distortion, and a few of them will actually run slightly warmer than intended. In high speed mass production, the op amp idling current is set by the design of the circuit, but it does not come with an adjustment to allow for variations in idling current. This problem may be completely eliminated by operating op amps in what's known as forced Class A operation. This is very easy so do. All that is required is a current source or a simple pull up resistor installed at the output of the amplifier. This forces one transistor to be always off and the other transistor to be continuously operating as a single ended Class A output device. As long as the op amp is operated within the new current source limit, the output will be totally free of crossover notch distortion. The practical result is that any family of op amps can be used with absolute assurance that all of them, time after time again, will not have crossover non-linearities. Thus, the crossover notch distortion problem gave op amps a very bad name in the audio world, but from my perspective, unnecessarily so, because it can be corrected.
In my designs, whenever I use an op amp, I always use a current source at its output. The choice of whether to use an op amp or to use discrete components is a matter of application. For example, for low distortion small signal requirements, an op amp is definitely the way to go. Normally, an op amp will have better power supply rejection and will be far more linear. In the case of FET input amplifiers, vanishingly low offset voltages and great immunity to input rectification accrue. The slew rate can be as high as we please and distortion as low as we please depending on the choice of op amps. However, in other applications, for example, one with large signal swings, a discrete circuit is the proper choice when higher current is required than is normally available from integrated circuit op amps. In conclusion, for a small signal amplifier operating on plus and minus 15 volts, I would always choose a good op amp. I would never build a discrete one unless I had a very special application, such as explained below.
I design with discrete circuits whenever I have complex feedback or signal processing issues in which control voltages must be developed for muting circuits, protection circuits, or dynamic control circuits as in a pro logic decoder, and of course, in the output stages and driver stages of high power, high current audio amplifiers.
I prefer to use film capacitors for coupling capacitors and to use electrolytic and/or film capacitors in bypass applications. I prefer to use ceramic capacitors in high frequency feedback systems and for certain high frequency bypass applications. I use electrolytics for energy storage and will use an electrolytic capacitor as a coupling capacitor provided that under no conditions is the voltage across the capacitor allowed to vary at all. This means that a very large coupling capacitor has to be used at the lowest frequency of interest, and it must be approximately 100 times larger than normally required. Hence, an electrolytic can't be used in a filter circuit or timing circuit. In that case, I would use either a film capacitor or a precision ceramic capacitor. Further, I believe that ceramic capacitors are best for high frequency stabilization in feedback loops and that the use of film capacitors in such applications is something that relatively inexperienced designers do. When you examine one of our circuit designs, you will find a mixture of electrolytic capacitors, ceramic capacitors, tantalum capacitors, film capacitors, low ESR film capacitors, and high current capacitors depending on the particular application. Each type of capacitor has its advantages and disadvantages when used in any particular circuit. The choices you will see in our circuits are the ones that I believe yield the best results and the best sound.
Transistors - Bipolar or Mosfet
I believe that the output stage of a power amplifier is best served by designing and building it with bipolar transistors simply because bipolar transistors are more linear, can deliver more current, and will typically have better SOA (safe operating area) specifications for simultaneous voltage and current when compared to an equivalent mosfet. If a very high performance amplifier is desired, bipolar transistors basically are the exclusive way to go (in my opinion), and you can see this by simply surveying the amplifiers on the market. The very expensive, very high current, high performing amplifiers in the $8,000, $10,000, $15,000 price range use bipolar transistors. Not one is designed using mosfets. Bipolars are best in audio output stages, but still, excellent results can be obtained in lower powered amplifiers using mosfets.
Mosfets or Hexfets (Brand name of International Rectifier Mosfets)
I design high power clocking circuits using mosfets because that's where their advantage concentrates. If a device is going to be on or off, then a mosfet is definitely the way to go because SOA considerations are not an issue, and their high speed and lack of storage time can be incredibly efficient. In those applications, they are extremely rugged, in fact, far more rugged than bipolar transistors - just the opposite than when used as a linear output device, in which case bipolars are more rugged than mosfets. To summarize, I use bipolars when the circuit is to be operated in a linear fashion and mosfets when operated in digital applications.
My choice of using precision parts is based on my scientific view of the world. It is not based on myths or fads. For example, in the Sunfire, I use the latest, lowest transition time, highest precision digital comparator available, which is the Hewlett Packard HCPL-2611. I ordinarily use 1% precision resistors, because assembly and manufacturing efficiencies are vastly increased. Provisions for adjusting the circuit to bring it into specifications are not required. Each circuit performs the same as the previous circuit time after time after time in a manufacturing environment.
There are many, many unusual amplifier designs in the audio community, some of them costing incredible amounts. Although I would not personally design a $50,000 amplifier, I love to talk about them, because that is what makes audio so much fun.
Other related articles
© Copyright 1995, 1996, 1997 Secrets of Home Theater & High Fidelity
Return to Table of Contents for this Issue.