That you only use 48 codes, my thought was that a slight shift would use a different 48 codes and the THD could have different discrete levels depending on small changes in amplitude, for instance. Like testing A/D's with 997Hz which is prime and on long term average exercises all the codes. Say you had a multi-bit A/D with one bad DNL transition or missing code, it might never show up.
Okay. Good point. So how does it not matter with sigma delta?
Digital Audio has fixed resolution according to bit depth and audio standards.
For what we use them, (the ADC), for we are not bound to this. Removing the fundamental or simply reducing it's amplitude and then applying gain buys us a lot more resolution. So missing codes is not such a big deal. For direct measurement it's a different story.
On a delta sigma there aren't discrete steps in the same way. As long as the clock is steady all the bits will intrinsically be the same.
Sent from my LG-H811 using Tapatalk
Sent from my LG-H811 using Tapatalk
If were were setting up such a test system today, I would use a D/A on the same clock as the measurement system driving two of Victors oscillators at around .1% drive in series.
As I don't think anyone is really 24 bits perfect, the requirement would only require 150 dB ish suppressed harmonics.
As to measuring resistor distortion fancy test gear is no longer required! Using the bridge method to suppress the fundamental makes it fairly easy. But you do need 10 resistors to test. For not as good resistors you can of course use reference resistors but you loose the ability to use reciprocity to advantage.
BTY the input signal must be balanced and floating. A transformer for this is fine if you don't have a balanced signal generator.
Attachments
Have you considered using the quadrature technique that Scott and Samuel used for the capacitor bridge?
No that doesn't use reciprocity as I understand it. If you analyze the bridge method there is virtually no lower limit to the measurement.
No, you can move the resistors around in the circuit and determine which ones are better. So from a large population you can select the best. With some math you can narrow in on each resistors exact values.
If you take two say Vishay S102 resistors in series and drive the top with your oscillator and force the center tap to ground with a second identical amplifier connected to the bottom you should get a pretty good balanced output. 🙂
EDIT - It's a joke
Last edited:
I meant the variance in the distortion, anyway seems like a lot of work. And what about mechanisms that don't follow the parallel series rules?
The work is easily done by a spreadsheet. If the mechanism is related to the voltage drop then one device sees V and the others V/2.
What kind of mechanism do you see where this doesn't hold? If the behavior is the same at V/2 as at V then how could there be distortion?
The work allows you to continue to improve the state of the art.
For less strenuous measurements reference parts are fine.
Yes, I assumed that resistance would be chosen to optimize performance.
The AD797 works well down to around 600 ohms, where the noise performance is clearly superior.
Different if you just want to swap it in as a replacement for an existent circuit.
Have to reassemble a band-saw, to cut steel to widen my hydraulic press, to bend the aluminium plate to finish some speakers, to free up the workshop space to...you know how it is.
In the meantime I have been at work on theory, by coincidence yesterday I realized why the AD797 works so well😉 So that has inspired me a bit to finish it.
Like Samuel before me, I have started to realize that the Sin^2 + Cos^2 leveler may look attractively simple but has a few hidden issues.
Need to decide if I will switch to some sort of sample-and-hold, as you recommended from the start.
Best wishes
David
In that case check out the OP615. Limited to +/-5V but this is enough for any good controller. Just scale it to your needs.
Power goes as V^^2, but in any case what about a distribution in TC of +-X ppm/C. A lot of maths to de-embed all the resistors.
Simple enough if you use say 3 fours vs a single. Of course with enough parts you can use groups of 16.
Much more work than using a reference. But how do you get a reference? Change temperature and measure resistance? Change voltage? Then how do you measure temperature, voltage and resistance?
Reciprocity is neither fast nor easy, just able to be very accurate. The greater the sample size the better you can get.