HD75DSP

Digital Signal Processor for MX Servo Controls

(It became a full blown digital computer)

Index

New Office – New Room Mate, Mal Johnson:

Digital P-92 Signal Processor Requirements:

Who would be Responsible for Servo Actuator Electronics?

Abandon Data Bus – Use Direct Wire:

Digital Arithmetic and Shaping Networks:

Digital Computer Design Book:

Multiplication by repeated addition:

Sample Data Systems:

CMOS Transistors – Transmission Gates – Analog Switches:

“H” Valve Driver Switch:

Digital Servo Valve

Speed Requirement – Close each servo loop every 2 milliseconds

Princeton Algorithm

Compute fast -- the Missile wants to Fly Backwards

Dr Ken O’Kief’s Appraisal:

Thanks To Elliott Buxton, Hydraulic Servos Win Out for the MX:

New Office – New Room Mate, Mal Johnson: We were moved into bldg 231 and Mal Johnson and I into a ground floor office. We were now on staff to Group Leader George Anderson as a part of Navigation System. Georges Group made the components used in the inertial platform like the analog servo controls and test equipment – he had many high quality people working for him.

Hydraulic Servo nil filtering – Turbine Gas Servo much filtering

Digital P-92 Signal Processor Requirements: Up to the time of the data bus demonstration and concepts review meeting I had been assuming we would be using hydraulic servos, for which there were no significant signal conditioning requirements. Now we were hearing rumblings that the Motor Contractors – with TRW Propulsion support had released two contracts for the development of solid propellant gas driven servo actuators for gimbal nozzle control. Mal Johnson and others had been contacted with regard to doing attitude control using such devices. It became obvious that these would be far different than controlling hydraulic servos. I visited with Mal about this saying I needed something to use as design criteria for making a signal processor. From this it was agreed that if I placed digital filtering capability in the forward loop as well as the feedback loop that we should be able to accommodate whatever they come up with. Any prior filtering needed could be done by the upstage Flight Control Computer.

Who would be Responsible for Servo Actuator Electronics? We had meetings with TRW G&C about what would be needed for servo actuator control electronics. They were asking what would happen if the electronics was given to the Motor Contractors along with the servo actuators. This was discussed in some length. I kept saying the responsibility for control and stability should remain with Autonetics. That from past experience with roll control stage II etc we knew motor contractors did not have the expertise to do electronics. At this point we were locked on the idea of having a data bus with servo electronics down stage. They wanted Autonetics to be the ones responsible for testing the adequacy of any system – but who would design the electronics for the vehicle was left up in the air. The designers of the gas driven servo actuators for motor contractor tests were providing their own electronics.

Abandon Data Bus – Use Direct Wire: I made layout studies on the idea of doing away with a data bus and moving all electronics up stage. I found that this would fit well as a flat cable – that it was a doable concept. I went for another visit with Lou, saying I believe we should work on the proposition that we do all the downstage electronics functions up stage – that way we can be assured of keeping that part of the business, and perhaps we could find a way to time share the electronics from one stage to the next. That it only had to handle two stages at a time for a brief period during staging. Lou agreed – adding we can probably make it a part of the Flight Control Computer.

Even after I’d worked the details on how to do this, selling the idea encountered the momentum built up for a data bus. What had been sold had to be unsold.

Downstage vs Upstage Electronics

Digital Arithmetic and Shaping Networks: It had become obvious that the system concept I had would be required to do data processing – I began to call it a signal processor, or a digital P-92, replacement for our current upstage analog P-92 controls electronics box. This became a totally new challenge for me, I had to determine how to digital computations and needed a model of what such a signal processor should do.

Digital Computer Design Book: On a week end visit to south coast plaza I came upon a book on Digital Computer Design; I bought it and poured myself into learning how to do arithmetic computations. Thankfully the industry had come out with an Arithmetic Logic Unit chip, and later with a Look Ahead Carry chip. By use of this book I learned that you need to precondition binary numbers to be either in 1’s Compliment or 2’s Compliment before or after doing an arithmetic operation. I agonized over which way was best. I also studied architecture previously used and applied it to the new chips available to me that were not covered by the book. There were no “cookbook” designs to follow but I was able to extrapolate. One of the things I found useful was a Booths Algorithm, which was the logic used associated with multiplications on what is done next.

Multiplication by repeated addition: I soon found that the Arithmetic Logic Units could add or subtract but could not do multiply or divide. I also found that I could do division by shifting decimal point and multiplying. However our applications never had a need to divide, but our multiplication demands would become tremendous and a critical factor in the design.

LaPlace, Z Transforms, and Sample Data shaping methods

Sample Data Systems: I was having a terrible time reading many books, becoming bogged down trying to determine how to perform signal conditioning function with a digital processor. I was becoming as irritated with the books and I was with my own ineptness. The Blair Bona came to my rescue. Blair would often come in to visit with Mal and observed me struggling with my face in a text book. We’d become acquainted on such things as how to rig up switched to turn on/off garage door lights from house and garage. Blair said forget that stuff in those books this is what you need. He proceeded to my black board and wrote a lines worth of differential equations, which I could follow. He then expressed the same information in Laplace transforms used for analog systems, which I didn’t understand but could track to a degree. He also showed Z Transforms then converted those to Sample Data representation. I didn’t understand right away how it worked, but I could certainly understand the data processing method. It was as if someone had written translations on a Rosetta Stone – this immediately illuminated the way to do things. I said don’t erase anything until I can write that down. Thank goodness he happened to take an interest in that because it lifted me out of a pit, I had been lost. Using combinations of information I began the design of a Digital Signal Processor – a Digital P-92.

Digital Processing required for One Gas Turbine Servo Actuator

CMOS Transistors – Transmission Gates – Analog Switches: RCA had recently come out with a family of CMOS chips, they had been #1 with vacuum tubes and were on track with their venture into CMOS but they gave up too quick. RCA, GE and Motorola literature was excellent. Motorola often used a pair of CMOS devices as a “transmission gate”. From the literature I could see that a CMOS device was quite different than a TTL logic device. A TTL transistor had a Collector input, Emitter output and a Base control – but it was one way flow. Current through the base could cause current to flow from in at collector and out at emitter. The CMOS (Complimentary Metal Oxide Semiconductor) had an Source input, Drain output and Gate control. The term complimentary indicated that when a plus signal was applied to the Base it would open the Source and close the Drain – a minus signal would close the Source and open the drain – the Source and Drain were always the opposite, the compliment, of the other. An advantage of CMOS was that it was either ON or OFF, it did not leak power to stay on as a TTL required. However the Motorola term of calling it a Transmission Gate, indicated that signal current could flow either way. It was like lifting a gate, opening either way, not like a one way swinging gate that opened only one way.

Emil Koehler helped me understand how to operate a bipolar transistor with collector-base-emitter, you “sourced” it (supplied current) to turn it on or would “sink” it (drain current away) to turn it off. He also taught me that if you turn a transistor full on, it has little voltage drop and does not heat, if you turn it off there is no current flow, thus no heat. Amplifiers operating in an analog mode are causing the transistor to throttle current flow, causing a voltage drop across the transistor causing it to heat.

“H” Valve Driver Switch: Emil also showed me what he called an H switch to reverse flow through a coil. Think of it as having a transistor in each of the four legs, that the cross bar is a coil, that current enters at the top and goes out the bottom of the H. By switching on upper left and lower right flow goes from left to right in the coil. Do the reverse thus and current flow is reversed.

Digital Servo Valve: RCA 4016 had four independent CMOS transistors. I connected them as an H switch to an 8ma servo valve coil and cause them to alternately flip flop valve current flow. It just operated cool and neat as can be for any kind of duty cycle I used. Fantastic! This caused an Analog Servo valve to behave as a Digital Servo Valve!

“H” switch lower right. 1 & 4 or 2 & 3 CMOS transistors enabled by sign bit to extend or retract command.

Status Detection added to verify valid valve driver for Silo Test

% On Time Modulation (used) vs Bang Bang Modulation

Feedback select, demodulate, Analog to Digital convert, 16 bit value ship – pre SAR method

Block Diagram of Processing for One Servo Actuator

Control Cycle Timing Protocol

Booster Control Functions

Servo Processing Timing Diagram

Data Bus Timing – Later Abandon

Princeton Algorithm Mechanization

Princeton Algorithm Multi Coefficient Processing

Speed Requirement – Close each servo loop every 2 milliseconds: I frequently checked with Mal Johnson on how fast the data had to be processed. As the design concept had operate in a nuclear event environment, I assumed a down grade of 50% in speed as compared to commercial bi-polar parts. Once I had a mechanization that worked I did speed calculations and found I couldn’t handle all seven multiplies on four servos, counting worse case staging overlap, in the allocated 2 milliseconds. I was simply not able to do all the defined functions fast enough. Again Dr Blair Bona came to my rescue.

Princeton Algorithm: Blair said what you need to do is use the Princeton Algorithm. This permits setting up coefficients in such a way that a multiplication can be done much faster. That was in 1975 and now 2002 I don’t recall exactly how that worked but it took advantage of how shifting a binary number does an instant multiply by shifting the binary “decimal point”. Anyway I modified the design to handle arithmetic in this way – now the computations could be done fast enough. I presented these findings to Dr O’Kief of TRW, he wrote a report on it and this put the subject to bed about digital being fast enough for the high speed servo loops.

Compute fast -- the Missile wants to Fly Backwards: The center of pressure is forward of the cg on a missile during stage I boost. The applied torque if miss aligned wants to flip the missile. Thus the attitude error must be kept within a limit cycle of 1 degree – this requires very fast corrective action by the servos.

Programmable Executive Program

Dr Ken O’Kief’s Appraisal: We were at TRW Redondo and had been going over the concepts for what I had always called a digital signal processor, or a digital P-92 the name of our analog processor. One day he leaned back and said, “this is fantastic – but it’s not a processor, it’s a programmable computer, it has all the attributes of a full up computer.” Since I didn’t know about computers, I’d never thought of it that way, I was only making use of the new TTL and CMOS devices to do signal processing; going another step each time to solve a problem. We knew we’d need shaping networks to handle new servos so I’d incorporated the ability to reprogram coefficients, for each stage, then change when stages changed. I’d bought a book on digital design at a book store to find out how to do arithmetic, learning of 1’s and 2’s complements and such things as booths algorithm for look ahead carry. Arithmetic logic units had just come out, I sent for sample parts and made use of them.

Shortly after that I learn that O’Kief had a Phd in computer science – he said you were so far ahead, using these new TTL parts, from what we were doing in school -- I didn’t want to tell you!! His comments came as a surprise to me as I constantly had a feeling of being behind trying to catch up, always feeling dumb, each day faced with something I didn’t know how to do. Thanks to Blair Bona, who told me about the Princeton Algorithm, I was able to get the “signal processor” to work fast enough.

When going out to lunch O’Kief would tell me about Mexican history and I’d tell him what I was learning about biology. He once called me up, asking if I’d join him in writing a book on the DNA molecule, saying we could do better than an existing book by Asimoff. I declined as I was reading a book by Watson, one of the co-discoverers of DNA, on Molecular Biology and was overwhelmed his good book on the subject. Thus in the recent year 2002 meeting with Bob Cummings I knew that ecoli cells were the most studied and ideal for making math models of their inner workings.

Thanks To Elliott Buxton, Hydraulic Servos Win Out for the MX: