Date: Mon, 22 Aug 2011 07:52:44
Thread-Topic: Filametary cathode tubes in computers
From: "Ed Lyon"
To: k2w at philbrickarchive dot org
Joe, you realize, of course, that the concept of tube-type computers is not restricted to general-purpose digital computers, but also embraces the huge analog flight simulators of the 1950s. I ran into a distant relative last week who is helping restore a flight simulator that I helped build (and was one of the check-pilots for) in 1954-55, a twin-engine A3D. As I recall, this unit had about 275 tube-type op-amps and servo amps in it. Only two of these A3D simulators were built, and this one has survived in a flight museum on the west coast. This simulator fit into two 50-foot vans with the flight trainer part (cockpit and all flight associated computer systems) in one van and the avionics plus instructors consoles in the other van.
The flight computer used 60-Hz a-c voltages to represent all variables, with a +/- 50-volt (rms) reference, i. e. 70.7/0 and 70.7/180 . Some variables used 400-Hz references, necessitated by the requirement to interface with aircraft flight instruments and avionics, and some d-c computation, using op-amps resembling Philbricks, schematic-wise, but with 1N351 silicon diodes used as zeners to couple stages and achieve the d-c voltage drop from plate to subsequent grid. These op-amps did not plug in, but were mounted to doors in the computer racks.
Pilots have reported that they could not ever have successfully flown the A3D aircraft without practice in the simulator, since that aircraft had a peculiar landing restriction: In order to land successfully (without collapsing the nose gear or damaging the tail skid) the aircraft had only 3 degrees of pitch attitude tolerance (the nose had to have between 3 degrees and 9 degrees nose-up attitude), and the flight equations we mechanized in the simulator reproduced this characteristic. We were initially very worried about this outcome, and I crashed the simulator a dozen times before I discovered the pitch limits, absolutely certain that it was an accumulation of errors in our simulation, as no aircraft could have been designed to work that way. Pilots from the Navy, however, smiled knowingly when we test-flew the simulator for their inspection, since that is exactly how the aircraft landed. Math triumphed after all.
From: Joe Sousa
Sent: Monday, August 22, 2011 9:07 PM
To: Ed Lyon
Subject: RE: Filamentary cathode tubes in computers
You had mentioned the AC based analog computers. How were Integration and differentiation done? Is that what the servo amps were for?
One engineer at work about my age remembers seeing a railroad car sized simulator for the military version of the Boeing 707 in the 1980's. This, apparently, was tube based.
Do you have any thoughts on the absence of filamentary tubes in early digital computers?
Date: Tue, 23 Aug 2011 07:11:15 -0400
From: "Ed Lyon"
To: "Joe Sousa"
Integration was done in a velocity servo. To integrate, the inputs to the servo amplifier are the derivative and an answer (feedback) voltage from a drag-cup generator mounted on the servo motor shaft. A large number of precision potentiometers, most of them having custom-located taps and shunt/series resistors (to create nonlinear functions of the integral) are usually mounted on the servo motor shaft, via a gearbox. The function called Mach Number, for example, in the A3D went from M=0 to M=1.2 in 320 degrees of pot rotation (geared down about 600:1 from the motor-generator shaft) required about 60-70 potentiometers mounted on it to produce the many functions of Mach that enter the flight equations, but so many potentiometers would create sufficient drag that the integrator would suffer static friction (stiction) at very low inputs. To reduce stiction, we would have only one potentiometer on the Mach servo shaft, a very linear one, and use its output to drive a position servo having a very high gain amplifier. This position servo would carry all the Mach function pots. Since Mach is a velocity term, the input to the Mach integrator must be an acceleration term, generated by the net X-axis force (mainly thrust minus drag, plus gravity times cosine of pitch angle) divided by the aircraft mass plus cross products of pitch rate X vertical velocity and turn rate X lateral velocity, all these modified by nonlinear functions of pressure-altitude and temperature. Mach was then resolved into aircraft velocities if several flavors: true airspeed, indicated airspeed, true velocity oner the ground, altitude change rate, etc., each of which has influence on to where the aircraft moves, and what the instruments say, as a result. Nearly all the functions are in closed loops, some, like engine RPM being fairly short loops involving throttle position, altitude, airspeed, fuel density, and afew other variables, while others, like airspeed itself, close only when the long loop developing thrust and the equally long loop developing drag answer each other as they input the Mach servo, perhaps settling on M=0.866. In a few minutes, though, all controls remaining unchanged, the Mach number creeps upward a hair, because the fuel load (and aircraft mass) reduces, and with the slight positive pitch-up angle fixed, the speed increases slightly, and the altitude creeps upward slightly. These changes may affect the engine thrust, depending on the initial speed and altitude, as well, either tending to increase or decrease the Mach creep. These are second-order differential equations of extraordinary complexity
Another servo shaft (also an integrator) that had many nonlinear functions was pressure altitude, which also needed a follower servo to carry all the mechanical potentiometer loading. In the A3D, pressure altitude went from some mildly negative altitude to 50,000 feet above sea level. Some servos (also mostly integrators) had no stops, rotating freely, and therefore carried no potentiometers (except, of course the reset pot for instructor use in resetting everything to zero). Such functions (like aircraft heading angle, pitch angle, roll angle) could be position servoes under some circumstances (autopilot ON, for example) or free integrators, while others that were integrators ran always as integrators (like position along the earths latitude and longitude axes) except for instructor resets.
Differentiation was seldom required, since most variables originated as derivatives of other functions we would eventually need. But multiplication!! there was plenty of that. Every potentiometer was a multiplier, and the overall simulator must have had 200 potentiometers, most of them nonlinear, by way of taps and shunt resistors. Critical multiplier variables used ten-turn Helipots, but most were 320-degree TIC (Technology Instrument Co.) pots. Servo motor-generators were all Kollsmann. Gear-boxes were made in house by our instrument shop in about 20 or 30 different ratios, all in identical cast housings. In checkout, all variables were measured using a slide-back voltmeter that measured voltage ratio with respect to the same +/- 50-volt a-c references as were used in the simulation. A helipot ten-turn pot on the slide-back voltmeter was used to null out the function being measured, as each variable making up that function was run through its gamut of values by hand-turning each servo or cockpit control. The checkout workbook for the A3D comprised about 500-700 pages of such measurements.
Dont get me started.
Date: Tue, 23 Aug 2011 07:28:26
From: "Ed Lyon"
To: "Joe Sousa"
Oh, regarding filamentary tubes in computers, the use of them usually forces all cathodes to be at the same potential, and does provide some inter-tube feedback due to filament-to-filament coupling, which must be taken into consideration in design. I believe they felt unipotential cathodes were easier to work with despite their inefficiency and heat problems. To get the needed Gm (for the required computational speed and consequent gain-bandwidth product) at the relatively low impedances found necessary by the extensive wiring throughout the computer, big cathodes were found to be needed, also, in some stages, making it yet harder to use filamentary tubes (and it made cathode followers more difficult to work with). They did not like to put power supplies on each computer board because of power-line transformer magnetic pickup, and so they tended to like the notion of 6.3-volt heater wiring in twisted pairs to each socket, with hash filters at the input ports of each board, rather than d-c power supplies or at least regulators for filament power on each board.