Talk:Integrator

Attention: Make sure the sum of your inputs does not exceed the machine unit (±10V), otherwise an overload will occur

I don't think this is correct. An overload occurs when the output exceeds the machine unit. However it should be no problem if the sum of the inputs exceeds the machine unit. — Preceding unsigned comment added by MKoch (talk • contribs)

Well, if the sum of the inputs exceeds the machine unit, the output will also do so, isn't it? You could think of cancellation of overloads (like in a sum with three inputs, two in overload but opposite sign, like +15V-15V+5V = +5V). This is a border case similar to saying +infinity - infinity + 5 = 5, because actually an overload value is typically an illegal value on the computer and the upcoming computation is most likely wrong. --Sven (talk) 13:51, 1 October 2021 (CEST)

No, if the sum if the inputs exceeds the machine unit, that doesn't mean the integrator's output will exceed the machine unit. At least not immediately. Your example with +-infinity isn't applicable, because every signal comes from another output and is always in the -1 to +1 range (or slightly above).--MKoch (talk) 14:04, 1 October 2021 (CEST)

When you connect a -1 signal to the input then the output will ramp from 0 to 1 in 1ms. Overload after 1ms. When you connect two -1 signals to two inputs of the same integrator, then the output will ramp from 0 to 1 in half the time. Overload after 0.5ms.--MKoch (talk) 14:15, 1 October 2021 (CEST)

The integrator has three "1" inputs with 1MOhm resistance and two "10" inputs with 100kOhm resistance. If all inputs are connected to one machine unit (10V), the input current into the summing junction is 23 * 10µA = 230µA. The output of the TL074 can drive about 25mA into the summing junction (via the capacitor), that's means there is a safety factor of more than 100. We could connect many more inputs to the same summing junction and it would still work fine.--MKoch (talk) 22:15, 1 October 2021 (CEST)

Thank you for your input. From my perspective, again, these are border cases, as the computing element which outputs a saturated, or overloaded, signal, also triggers the overload signal on the whole board. That is, a LED will pop up, notifying the user that there is something wrong. I made the analogy to a floating point exception in classical digital IEE754 computing. You can instruct your algorithms to stop on such an exception, or you can deal with NaNs and infinities on a one-per-one basis. Of course one can imagine a code where such values correspond to some special, intended meaning and all this is fine. In a similar spirit, inserting an "overloaded" value into some integrator input might be fine, especially for a short time. However, at the moment, I cannot think of a computing circuit where one wants to integrate overloaded values. Do you have one?

All the best, --Sven (talk) 09:51, 4 October 2021 (CEST)