How to improve LC-meter accuracy

[jorgen_re]

Some time ago I made a simple LC-meter which works by measuring the frequency-shift in an LC-oscillator when an unknown capacitor or inductor is placed in parallel or in series with the already known LC components. Because of frequency instabilities I made the meter do multiple measurements and then show the minimum and maximum values measured, as well as calculate the average value, standard deviation and 95% confidence interval. This seems to work quite well for my purposes, however, I want to increase the frequency stability and minimize the drift.

The "reference" frequency (when no external component is hooked up) of the LC-circuit seems to change slightly every time I turn it on. During measurements I can also clearly see that the "reference" frequency drifts. Therefore,  I switch between measuring the frequency with and without the external component wired in (using relays) and then use the difference between the two measurements (using the "frequency-shift-method") to determine the unknown value of the component. This makes the total sampling time very long when doing a lot of samples for each component.

I am open to feedback on improving the design or even ideas for a completely new one.

I am most interested in measuring inductors. Attached is a table of some measurements and results of inductors that indicates the meters accuracy. Schematics and pictures of the meter is also attached (as you can see, I did not check the footprint for the voltage regulators and placed them too close, BUMMER!)

P.S. As the menus in the photos are written in Norwegian:
"Trinn 1 av 4" -> "Step 1 of 4
"Velg L/C:" -> "Choose L/C"
"Bekreft" -> "Confirm"

"Trinn 2 av 4" -> "Step 2 of 4
"Antall samples" -> "Number of samples"

"Trinn 3 av 4" -> "Step 3 of 4
"Sampler" -> "Sampling"
"Fremgang" -> "Progress"

Resultscreen:
Min -> Lowest value measured
Max -> Highest value measured
Avg -> Average value
Std -> Standard deviation
KonfMin -> 95 % minimum confidence interval
KonfMax -> 95 % maximum confidence interval
Prefiks -> Prefix for all results

[dannyf]

even ideas for a completely new one.

Typically, you can use synchronous detectors - it is fairly involved for the analog end.

Alternatively, you can generate a sine wave form, and you then sample the V/I signals to detect the phase shift + attenuation. All of that can be done via a mcu (with pwm + adc).

Alternatively, you can use network analyzers (5933/5934) + a mcu. Those chips have some limitations but are generally good for most applications.

[Conrad Hoffman]

It sounds like you should start with a more stable oscillator. Look into temperature coefficients of parts and what the hams do to build stable LC circuits. Just to get an idea of  how some older commercial products worked, download a manual for the General Radio 1632 Inductance Bridge. It resolves 6 places with no trouble and is almost that accurate. It doesn't lend itself to computer control though, being a traditional mechanically balanced bridge. Next, download the manual for the General Radio 1657 RLC "bridge". It's the first digital bridge, not really a bridge at all, designed by Henry Hall, and established the patents that HP and others paid big $$ to use in their products. The methods used for that were the foundation for most of the very accurate digital meters since.

IMO, unless a bridge can give you both value and loss, specified with either the parallel or serial model, there isn't much point in talking about accuracy, especially with inductors, which aren't very pure examples of reactance.