Feature Archives | Feedback

Meeting Walter

A couple of days ago I had the privilege of meeting for a few hours with my old boss Walter. In 1985, Walter hired FGL, (Frank Greenhalgh Labs) to consult for CEAG Corporation of Germany. CEAG was at the time the largest manufacturer of mainframe power supplies in the world. IBM was their largest customer. I had previously co-founded and worked as VP of Engineering for CEAG Corp. in the US. Walter provided me with the opportunity to start my own consulting business, working as a design engineer and not a manager. For ten years I worked in my lab developing power supply designs for CEAG, Germany.

As Walter still circles the globe calling on customers and visiting plants from China to San Jose, it is not easy to meet up with him. This is especially true since he lives in Germany. The day we met he had a few hour layover at JFK, just enough time to drink a beer and talk a bit about old times. During the conversation I asked him if CEAG (now actually ASCOM) is still using the droop system on their large power systems. He smiled and said "yes and so is everyone else." Walter told how all the competition was using the "Droop System." He had considered stopping them with a patent infringement suit but decided not to. Wow, I thought, the droop system has become widely accepted in industry. I was very proud. I had conceived it in 1981, CEAG eventually patented the concept in 1983, and now it is in use universally, patent or no.

The Pitch to Amdahl

It all started in 1980, as Jack and I were flying to San Francisco from New York. Jack was my partner and president of CEAG, US. I was VP of engineering. We had just introduced our first commercial standard product, a 5-volt, 300-ampere power supply. Amdahl Computer, IBM's rival, had evaluated it and liked what they saw. We were flying to Sunnyvale to discuss their requirements. Jack said our sales representative had told him they were concerned about current sharing when large numbers of supplies were connected in parallel. The specification asked for 5% current sharing for up to fifty power supplies. I took a sip on my martini (this was 1980) and started thinking.

During the rest of the flight I scribbled and calculated and finally told Jack. "Don't worry, tomorrow we will present "The Droop System."

The next day at a meeting at Amdahl we presented our product. First we answered questions about our 300-amp design, and then we went through the Amdahl requirement specification, line by line. When we came to the line that said:

"When connected in parallel the supplies shall share current equally within 5%."

I said, "We are going to use The Droop System". "What is the Droop System?" they asked.

I answered: "'The Droop System' is the simplest and most effective way to parallel power supplies." On the blackboard I drew three batteries. Each battery had a series resistor from the positive terminal. The batteries represented three power supplies. All the negative terminals were tied together for the negative output and the ends of the resistors were connected together to become the positive output. The pitch went like this:

If all the batteries have the exact same output voltage, and source resistance, then the currents will share perfectly. "This is EE101, Mesh Equations, Norton Equivalents," I said. "The secret is to create power supplies that have an equal output voltage and equal output impedance."

I proposed that the power supply we would custom design for them would have a no load output voltage of 5.03 volts (thirty millivolts high) and output impedance of .1 milliohms. As the load on the power supply is increased its nominal output voltage will 'Droop' from 5.03 volts to 5.00 volts (1-millivolt for every 10 amperes), resulting in an output voltage of 5.00 volts at 300 amperes. As all supplies will have the same output voltage and impedance, all currents will be equal and the actual output voltage will never be greater than 30 millivolts (.6 per cent) of nominal. Problem solved.

They were impressed, but skeptical. What I had proposed was to make a power supply that had an output voltage accuracy of one millivolt out of 5000 millivolts or .02% (20PPM) and an output impedance of .1 milliohms +/- 1%. I argued that although this had never been done before it certainly could be. I spoke with supreme confidence, even though I had no idea at the time if it could be done.

They bit, they believed me. As we flew home I sat with my martini wondering if I could build a power supply that would have an absolute output accuracy of 20 ppm (parts per million) and an output impedance of .1 milliohms, +/-.01 milliohms, over the industrial temperature range and five years of operation. I had faith that we could pull it off. That faith came from my experience at Julie Research.

Working in Parts per Million

In 1962, I spent a year working as an instrument engineer for Julie Research Laboratories (http://loebejulie.com/jrl/index.html ). The company's major product was wire wound resistors. Not just ordinary resistors, but the most stable and accurate resistors made at the time. Accuracy's of 10 PPM were normal with 1PPM possible. Although the major customers for resistors were analog computer manufacturers, I worked developing highly accurate and stable instruments, and power supplies, using Julie resistors. My first project was a constant current power supply for an ultra stable gyro. It had a stability requirement of 1PPM (.001%) per week. This was achieved this by using 3 standard cells as the reference, and a chopper stabilized null detector as an error amplifier.

When I left Julie, I knew just what accuracy and stability could be achieved if you used exotic methods. Of course that was twenty years before I made the pitch to Amdahl. I really wasn't sure if the present state of the art would allow me to achieve with silicon what I did in 1962 with batteries and chopper amplifiers, but I thought with a little luck I will.

I Guessed Right

Over the next six months we worked designing the first high current power supply that current shared using the "Droop System." I had guessed right. It was now possible to obtain temperature stabilized references on a chip. Op-amps with low drift and offset voltages and film resistors of high accuracy and stability were also available. Using these wonderful but still relatively expensive components we were able to build a 1500 watt power supply with the accuracy of a standard cell and a guaranteed output impedance of .1 milliohms.

The "Droop System" worked very well. Fifty power supplies, delivering 50 to 15,000 amperes would share their current within 5%. It worked without the necessity of having a sharing bus. It worked without increasing the noise and it worked as the output was programmed from two to six volts depending upon jumpers in the input connector. I thought the concept of knowing how much current was being drawn per supply by simply measuring the output bus voltage (i.e.: 5.010 volts means 200 amperes) was very cool. Over the next years over 100,000 of these supplies were built and the demonstrated MTBF (Mean Time Between Failure) exceeded two million hours. The increased cost for accuracy and stability had paid off.

CEAG Patents Droop System

In November of 1983, CEAG filed for a Patent on the droop system. The Patent (http://patent.womplex.ibm.com/details?patent_number=4635178) #4,635,178 was issued to CEAG and I was named as inventor. After I left CEAG in 1995, I always wondered what would become of the "Droop System". Now I know. The "Droop System" is used as the standard for paralleling large numbers of high current power sources. I feel very proud to have developed it. I am also glad that Walter decided not to decrease its use by his competitors.

Frank Greenhalgh
Oct. 31, 1999

Comments Frank's column hit a nerve and the responses we are receiving are proof of that. Read the comments as well as Frank's response

About the Author

Frank Greenhalgh has been working in power supplies and systems for 38 years. He has many impressive accomplishments and patents. Over the years he has made significant contributions to Trio Laboratories where he held the position of Chief Design Engineer and was then promoted to Vice President.