What is the maximum acceptable latency in rocket control computers?

Question

Rockets require real-time computers, in that there are operational deadlines between events and system response. How quickly does the rocket's controller need to respond to sensors and events, without fail? What kind of mechanisms are in place, or what kind of practices are avoided, to make sure this happens?

Specifically:

• What was the required GNC reaction time during the STS roll maneuver, and what ADC parameters were read and polled upon at the fastest rate, that such roll program depended on?

Answer 1

The STS roll control program controlled attitude in response to sensed velocity, rather than attempting to control the trajectory directly. This portion of the flight wouldn't be particularly latency-sensitive.

Spacecraft Guidance, Navigation, and Control Requirements for an Intelligent Plug-n-Play Avionics (PAPA) Architecture :

In the first stage after lift-off, the objective of the guidance algorithm is to take the vehicle through the atmosphere while minimizing the fuel spent and keeping the structural loads in their design range. To achieve that, the algorithm provides attitude commands as a function of the vehicle velocity. These values are pre-computed and stored in the computer memory.

After the roll program completes, the shuttle does go into closed-loop control to minimize aerodynamic loads, but the overall trajectory guidance remains open-loop until relatively late in the ascent.

Space Shuttle Digital Flight Control System

Five seconds after lift-off, commands are issued to accomplish the pitch-over and roll-to-flight-azimuth maneuvers. During regions of high dynamic pressure, a load relief system in both pitch and yaw minimizes air loads on the vehicle... The load relief function is accomplished by lateral and normal accelerometer feedbacks blended into the attitude command system starting at 25 seconds into the flight.

Regarding latency:

During all flight phases, the fundamental flight control computational cycle is 40 milliseconds. A number of flight control computations take place only every 80 milliseconds, and some take place approximately once per second. In addition to the basic 40-millisecond minor cycle requirement, the total delay between sampling of the flight control sensors and transmittal of the resulting command to the effectors is constrained to be no greater than 20 milliseconds.

If I'm parsing this correctly, it means that every 40ms, there's a sensor-computer-effector cycle that takes no more than 20ms. The shuttle was notoriously aerodynamically unstable, far more so than most launchers, so I imagine that this sort of cycle time would suffice for almost all.