• Welkom bij Audiofreaksforum, Audio, Hifi, Luidsprekers, Buizen. Graag inloggen of registreren.
 

ChatGPT snapt het ècht niet!

Gestart door Regenpak, 18 maart 2023, 18:12:54

Vorige topic - Volgende topic

Regenpak

Ik lach me een kriek. Ik ben die Accuphase aan het torture testen. De dissipatie van de eindtorren bij een sinussignaal heeft een maximum bij een bepaalde amplitude. Die is niet de maximale onvervormde amplitude want dan staat er maar weinig spanning over de eindtorren bij de maximale stroom in de dummyload. Ik kon online zo gauw niet vinden hoe je op dat maximum komt, ik dacht iets van 2/3 van de voedingsspanning, dus ik dacht laat ik het aan ChatGPT vragen. Dat ging niet best:

Tjerk:
Given an analog class B amplifier driven with a pure sine wave. Its power rails are +10 V and -10 V. What is the output peak-to-peak voltage that causes the highest power dissipation in the amplifier?

ChatGPT:
For an analog class B amplifier, the output stage consists of two complementary transistors, one NPN and one PNP. Each transistor conducts for half of the input signal cycle, which means that each transistor only handles half of the signal.

The output voltage swing of the amplifier is limited by the power rails, which are at +10 V and -10 V. The maximum output voltage swing is equal to the voltage between the power rails, which is 20 V.

When the amplifier is driven with a pure sine wave, the maximum output peak-to-peak voltage occurs when the input signal is at its maximum peak-to-peak voltage, which is twice the amplitude of the sine wave.

The maximum output peak-to-peak voltage of the amplifier is, therefore, 2 times the voltage swing or 2 times 10 V, which is equal to 20 V.

However, this maximum output voltage swing will not cause the highest power dissipation in the amplifier. The highest power dissipation occurs when the amplifier is driven to produce the maximum output power, which occurs when the output voltage swing is half the maximum possible value.

In this case, the output peak-to-peak voltage that causes the highest power dissipation is equal to half the maximum output voltage swing or 10 V.


This seems incorrect. If I calculate the power dissipation in a spreadsheet, the maximum dissipation of the amplifier is between 12 and 13 Vpp output. What is the exact value? Accuracy 4 digits.

You are correct, I apologize for my mistake. The output peak-to-peak voltage that causes the highest power dissipation in a class B amplifier driven with a pure sine wave is not necessarily half the maximum possible value. Instead, it depends on the specific design of the amplifier and the load it is driving.

To calculate the output peak-to-peak voltage that causes the highest power dissipation in this case, we need to determine the voltage at which the output stage switches from one transistor to the other. This switching voltage is known as the "crossover distortion voltage" and is typically around 0.5-1 V for a class B amplifier.

Assuming a crossover distortion voltage of 0.5 V, the output voltage swing will be limited to 9.5 V (10 V - 0.5 V) peak-to-peak before the amplifier starts to clip. Therefore, the output peak-to-peak voltage that causes the highest power dissipation in this amplifier is 9.5 V.

To calculate the power dissipation in the amplifier, we can use the formula:

P = V^2 / (2*R)

Where P is the power dissipation, V is the output peak-to-peak voltage, and R is the load resistance. Assuming a load resistance of 8 ohms, the power dissipation is:

P = (9.5 V)^2 / (2*8 ohms) = 56.40625 W

Therefore, the output peak-to-peak voltage that causes the highest power dissipation in this amplifier is 9.5 V, and the corresponding power dissipation is 56.40625 W.


My class B amplifier is considered perfect so the crossover distortion is not an issue. Still, the voltage now is even less than before whereas my spreadsheet calculation shows a value between 12 and 13 Vpp. Because this is complex math using an integral I could not figure out the correct formula. The spreadsheet just adds discrete values approximating the end results. Which functionally of course is the same as integrating. What is the correct formula to calculate this elusive output voltage that creates maximum dissipation in my hapless perfect amplifier?

If we assume a perfect class B amplifier with no crossover distortion, then the output voltage swing is limited by the power supply rails, which are at +10 V and -10 V. The maximum output voltage swing is equal to the voltage between the power rails, which is 20 V.

To determine the output peak-to-peak voltage that causes the maximum power dissipation in the amplifier, we can use the formula for power dissipation in a resistor:

P = V^2 / R

where P is the power dissipated in the load resistance, V is the output voltage, and R is the load resistance.

In a class B amplifier, each transistor conducts for half of the input signal cycle. Therefore, the maximum output power occurs when each transistor conducts for the same amount of time. This happens when the output voltage swing is centered around zero, or when the output voltage is zero on average.

To find the output voltage that results in zero average voltage, we can integrate the output voltage waveform over one cycle and set the result equal to zero:

∫[0 to T] V(t) dt = 0

where T is the period of the input sine wave.

For a sine wave with peak amplitude A, the output voltage waveform is a square wave with peak amplitude A. The integration of a square wave over one cycle is zero if the waveform is centered around zero.

Therefore, the output voltage that results in zero average voltage is half the maximum output voltage swing or 10 V.

This means that the output peak-to-peak voltage that causes the maximum power dissipation in the perfect class B amplifier is 20 V, which is the maximum voltage swing.

Using the formula for power dissipation in a resistor, we can calculate the power dissipated in an 8 ohm load resistance:

P = (20 V)^2 / 8 ohms = 50 W

Therefore, in the case of a perfect class B amplifier, the output peak-to-peak voltage that causes the maximum power dissipation is 20 V, and the corresponding power dissipation in an 8 ohm load is 50 W.


Hm. I think I'll stick to my spreadsheet.

I understand, using a spreadsheet can be a very useful tool for simulating and analyzing circuit behavior, especially when dealing with complex mathematical formulas. Just be sure to validate the results of your spreadsheet simulations with real-world measurements and testing, as there can be many factors that may affect the behavior of a circuit that are difficult to model accurately in a simulation.
ChatGPT Mar 14 Version


Gaat echt he-le-maal nergens over. Crossover? Boeit niet. Belastingsweerstand? Irrelevant. Vermogen daarin? Ditto. Nauwkeurigheid vier cijfers? Wordt genegeerd. "Just be sure to validate the results of your spreadsheet simulations with real-world measurements and testing" is toch wel de kicker. Ik moet eerst een theoretisch model hebben en pas dan kan er worden getest en gemeten! Overigens, m'n spreadsheet (met cellen als "=(1-0.65*B3)*(0.65*B3)" voor de momentane dissipatie) gaf een maximum ergens tussen 60 en 65% van de voedingsspanning. Dus 12-13 Vpp in mijn theoretische perfecte klasse B versterker. Die helemaal geen NPN en PNP transistoren hoeft te hebben! In dit geval IRFP(9)240 MOSFETs. Maar dat terzijde...

Tjerk
It's been Monday all week long!

Sytse Fliege

Toch komt 't al wel een heel eind, voor een AI programma.

Dat je er nu niets aan hebt is evident, maar als dit doorzet zou het toch wel eens significant interessant kunnen worden.

Ik heb een dag of wat terug eveneens een teleurstellende ervaring gehad met ChatGPT, op de vraag of Henk kon terugkeren vanachter de waarnemingshorizon in een zwart gat kreeg ik alleen maar ontwijkende antwoorden.

Als ik daar toch een keer een concreet antwoord op krijg, dan ga ik mijn petje er wel voor afnemen, maar tot die tijd zie ik het toch maar even als spielerei.





Ach, kikker toch op!

Regenpak

Het concrete antwoord zal niet komen. Er komt hooguit een lulantwoord. Net als op mijn vraag. Alleen als het concrete antwoord al bestaat (42) zal dat worden opgelepeld.

Tjerk
It's been Monday all week long!

Vincent Kars

Het is de nieuwe hobby, vragen stellen over audio aan ChatGPT
Uiteraard hangt de kwaliteit van het antwoord af van de mate waarin dit model is getraind gegeven het onderwerp.
Vaak getuigd het antwoord van
- kennis van zaken
- nuance
- ontbreken van opinies
Kom daar maar eens om bij het gemiddelde lid van een audio forum 😁

Sytse Fliege

Wat er bij mij ondertussen niet meer in gaat is dat leerlingen hier leraren mee kunnen bedonderen.

Alhoewel, mogelijk is de kwaliteit van het huidig lerarendom ook below par.

Maar hé, laat ik het ChatGPT eens vragen!

Oh, aha, veelzeggend;

Ach, kikker toch op!

Regenpak

Citaat van: Vincent Kars op 18 maart 2023, 22:37:00Vaak getuigd het antwoord van
- kennis van zaken
- nuance
- ontbreken van opinies
Ik ben wel benieuwd naar jouw onderbouwing hiervan. Doorgaans getuigt het antwoord van ChatGPT juist niet van kennis van zaken. Zie hierboven. Of Herr Fliege's voorbeeld. Kijk, dat je de computer in de juiste richting moet douwen, allà, maar allerlei irrelevante zaken erbij halen (nuance?) slaat nergens op.

Tjerk
It's been Monday all week long!

Tom!

Garbage in, garbage out en een hoop omhaal van woorden om te verbergen dat het inhoudelijk niets voorstelt.
Geluid is een van de meest overschatte eigenschappen van muziek.
--- E. Schönberger

morph

Het programma kan grammaticaal kloppende teksten samenstellen uit een gigantische verzameling woorden en onderwerpen.
Maar de beoordeling of de inhoud correct is zal afhankelijk blijven van de bronnen waaruit geput kan worden.
Waarbij het maar de vraag is hoe die bronnen beoordeeld worden en welke criteria er daarvoor door mensen aan toebedeeld zijn.

Algoritmes worden ingezet als rigoureuze graafmachines terwijl niet zelden juist een pincet gewenst is voor een goed resultaat.

Vincent Kars

CitaatIk ben wel benieuwd naar jouw onderbouwing hiervan.

Veel hangt af van de training en er is uiteraard ontzettend veel waar het niet op getraind is. Dat laat onverlaat dat het probleemloos slaagt voor " Uniform Bar Exam, the Law School Admission Test (LSAT), the Graduate Record Examination (GRE) Quantitative," zie https://arstechnica.com/information-technology/2023/03/openai-announces-gpt-4-its-next-generation-ai-language-model/

Kortom vertoont intelligent gedrag.


morph

Nou, "intelligent" gedrag...

AI models in the GPT series have been trained to predict the next token (a fragment of a word) in a sequence of tokens using a large body of text pulled largely from the Internet. During training, the neural network builds a statistical model that represents relationships between words and concepts. Over time, OpenAI has increased the size and complexity of each GPT model, which has resulted in generally better performance, model-over-model, compared to how a human would complete text in the same scenario, although it varies by task.

As far as tasks go, GPT-4's performance is notable. As with its predecessors, it can follow complex instructions in natural language and generate technical or creative works, but it can do so with more depth: It supports generating and processing up to 32,768 tokens (around 25,000 words of text), which allows for much longer content creation or document analysis than previous models.

morph

19 maart 2023, 13:03:28 #10 Laatste wijziging: 19 maart 2023, 13:05:31 door morph
Citaat van: Vincent Kars op 19 maart 2023, 12:14:40dat het probleemloos slaagt voor " Uniform Bar Exam, the Law School Admission Test (LSAT), the Graduate Record Examination (GRE) Quantitative,"

Kortom vertoont intelligent gedrag.

Je zou dus ook kunnen concluderen dat de advocatuur en andere menselijke onderdelen binnen het plaatselijke juridische systeem géén intelligent gedrag vertonen maar telkens simpel oplepelen wat ze ergens gelezen hebben.

Tom!

Dat is een rechtenstudie toch ook voornamelijk? Goed kunnen stampen en feiten oplepelen?
Geluid is een van de meest overschatte eigenschappen van muziek.
--- E. Schönberger

morph

19 maart 2023, 13:36:26 #12 Laatste wijziging: 19 maart 2023, 13:38:08 door morph
Ja, uit bestaande regels en jurisprudentie een "conclusie"  leren construeren, zonder emoties mee te laten spelen.

Sytse Fliege

Het juridisch systeem is toch wel iets gecompliceerder dan dat hoor. Ik heb wel wat rechtszaken meegemaakt en de uitkomsten wilden nog wel eens heel anders zijn dan wat ik er van verwacht had, maar in mijn optiek wel altijd plausibel. Zodra ChatGPT onverwacht uit de hoek kan komen wordt het pas echt AI, nu is het wat mij betreft nog niet veel meer dan een "language programm" wat in staat is data uit een enorme hoeveelheid bronnen te kunnen oplepelen in een vorm die er uitziet alsof het door een taalkundig capabel mens is gedaan.

Ach, kikker toch op!

Regenpak

Citaat van: Vincent Kars op 19 maart 2023, 12:14:40Kortom vertoont intelligent gedrag
Daar ben ik het totaal mee oneens. Het is slim geprogrammeerd en is in staat om uit een enorme berg informatie iets min of meer zinvols te destilleren. In mijn geval viel het nogal tegen. Intelligent is dat absoluut niet.

Tjerk
It's been Monday all week long!