Computing Spherical Trigonometry with Spinning Metal: The B-52's Electromechanical Angle Computer

Ken Shirriff has published a detailed teardown of the electromechanical angle computer from the B-52 bomber’s star tracker system, and it is the kind of engineering artifact that makes you stop and reconsider your intuitions about what computation requires. The device solves spherical trigonometry continuously, in analog, using rotating shafts and electromagnetic induction rather than arithmetic units and clock cycles. Understanding how it works requires a short detour through celestial navigation and a class of electromechanical components that barely appear in modern engineering curricula.

Why a B-52 Needs to Know Where Stars Are

The B-52 entered service in 1955, and its primary navigational challenge was flying at high altitude over featureless terrain (or ocean) for missions lasting ten or more hours. Inertial navigation platforms, the kind that integrate accelerometer readings to track position, accumulate drift at a rate of roughly one to two nautical miles per hour. Over a ten-hour mission that becomes a circle of uncertainty 10 to 20 miles in radius, which is professionally awkward when your mission involves precision delivery.

The solution was periodic celestial fixes: observe a known star, compare its measured altitude and azimuth against the star’s computed position for a given assumed location and time, and use the difference to correct the inertial platform. This is the same technique navigators had used on ships for two centuries, called the Marcq St. Hilaire intercept method in its systematic form. The mathematics involved is spherical trigonometry, because the Earth is a sphere and the celestial sphere wraps around it.

The fundamental equation is:

sin(Hc) = sin(L) · sin(Dec) + cos(L) · cos(Dec) · cos(LHA)

Here Hc is the computed altitude of the star, L is the observer’s latitude, Dec is the star’s declination (its celestial latitude), and LHA is the local hour angle, which captures the star’s east-west position relative to the observer as the Earth rotates. A navigator with sight reduction tables solves this equation manually in about two minutes. A bomber crew flying at 500 knots does not have two minutes; they need a continuous, automated solution.

Resolvers: Electromagnetic Function Generators

The angle computer’s core component is the resolver, an electromagnetic device that computes sine and cosine as a side effect of its physical geometry. A resolver looks like a small motor: it has a rotor and a stator, wound with coils. The primary winding sits on the rotor. The secondary windings sit on the stator, at exactly 90 degrees to each other.

When you apply an AC reference voltage to the rotor and rotate the shaft to angle θ, the two stator outputs are:

V_sin = V_ref · sin(θ)
V_cos = V_ref · cos(θ)

This is not an approximation. It is an exact consequence of how mutual inductance depends on the cosine of the angle between two coils. The resolver does not estimate a trigonometric function; it instantiates one. Its shaft angle is the argument, and its output voltages are the values.

Resolvers can also be operated in reverse, as a coordinate rotation device. If you feed two voltages representing a 2D vector into the stator windings and servo the rotor until one output nulls, the rotor shaft angle settles at the vector’s angle. That is a mechanical arctangent computation. Chained together, resolvers implementing coordinate rotations can evaluate the full spherical triangle equation in one continuous mechanical pass.

The Spherical Triangle in Metal

The celestial fix computation requires three sequential coordinate transformations. Starting from a star’s position in the equatorial coordinate system (right ascension and declination), you must rotate through the Earth’s current orientation (applying local hour angle), then rotate into the local horizon coordinate system (applying latitude), arriving at altitude and azimuth. Each rotation maps cleanly to a resolver stage.

The first resolver receives the star’s declination on its shaft and the reference signal on its input. Its outputs represent the star projected into an intermediate frame. The second resolver applies the hour angle, effectively rotating the celestial equator to account for the Earth’s spin and the aircraft’s longitude. The third resolver applies the observer’s latitude, completing the transformation into the local horizon frame. The final shaft angle and output amplitude encode azimuth and altitude respectively.

This is essentially a mechanical implementation of the direction cosine matrix used in modern attitude computation. The resolver chain multiplies through a rotation matrix whose elements are sin/cos of the navigation angles, except the multiplication happens electromagnetically rather than arithmetically. The values are represented as voltages, not bits; the function evaluations happen in continuous time, not in clock cycles.

Synchros, the close relatives of resolvers used for angle transmission, fed shaft position data from the astro-tracker telescope gimbals into the computer. As the telescope tracked a star across the sky, the synchros reported the gimbal angles continuously, and the resolver chain updated its outputs continuously. The system was computing a celestial fix at every instant, not on demand.

Why Not Digital?

The question is worth taking seriously because early digital computers existed in the 1950s. The Univac 1 had shipped in 1951. The short answer is that airborne digital computers of sufficient capability were not yet small enough, reliable enough, or power-efficient enough for this application. The early IBM 650 weighed nearly two tons. The electromechanical angle computer fit inside a rack unit and ran on aircraft power.

The analog approach also had a latency advantage that mattered for closed-loop control. A resolver chain responds to input changes within microseconds, constrained only by the AC reference frequency (typically 400 Hz in aircraft systems, chosen because aircraft electrical systems run at 400 Hz rather than the 60 Hz used in ground installations). A digital computer of the era would have taken tens of milliseconds to complete a trigonometric computation, which is fine for once-per-second position updates but awkward for the servo loops that kept the telescope pointed at the star.

The AN/ASQ-38 bombing navigation system, which flew in B-52Ds and later variants, combined an inertial platform, an astro-tracker, a Doppler radar ground-speed sensor, and a digital computer. By the time the digital computer appeared in the system, it handled mission planning and weapon delivery math, while the analog and electromechanical sections continued to handle the fast inner loops. The two paradigms coexisted for years.

Precision and Its Limits

The precision achievable with a resolver chain depends on the angular accuracy of the mechanical components and the signal-to-noise ratio of the analog voltages. Military-grade resolvers of this period were specified to roughly 10 arc-minutes of angular error, meaning the computed altitude and azimuth were accurate to that level absent other errors. The astro-tracker itself, using a photomultiplier tube to detect stars even in daylight by operating in a narrow spectral band, contributed its own pointing error.

The end-to-end celestial fix accuracy in operational use was on the order of 0.5 to 2 nautical miles, depending on atmospheric conditions and the number of stars observed. That was sufficient to update the inertial platform and keep the mission within acceptable delivery tolerances for the weapons of the era.

Wear and temperature were the persistent enemies. Resolvers are precision mechanical components; their bearings degrade, their coil resistance shifts with temperature, and their transformer coupling varies as materials expand and contract. Airborne systems running at the temperature extremes of high-altitude flight needed careful thermal compensation and regular calibration. The maintenance manuals for these systems are substantial documents.

What Survives

Resolvers are still manufactured and still used, primarily in applications where the environment is too harsh for optical encoders: motors in heavy industry, landing gear sensors, and control surface position feedback in aircraft. The resolver-to-digital converter IC, which samples the sine and cosine outputs and computes the shaft angle digitally, is a standard catalog part. The analog computation is gone but the angle transduction remains.

The deeper lesson from the angle computer is architectural. The engineers who designed it were not working around a missing digital capability; they were solving a real-time coordinate transformation problem with the best available substrate for continuous-time function evaluation. Resolvers are genuinely good at evaluating sine and cosine of physical angles because that is literally what they are. The elegance is not nostalgic; it is appropriate use of material properties.

Shirriff’s teardown is worth reading for the photographs alone. The construction quality of these components, the hand-wound coils, the precision machining, the shielding, reflects what it cost to build reliable analog computation in the 1950s. It also makes vivid what the transition to digital meant: not just cheaper arithmetic, but the decoupling of computation from physical substrate.