field test unit no 001 "model T"

[moodz]

...the 24 bit goodness is coming.........

[6666]

Hi hdphilip
I understand that D6 and D7 will clip down the fly back spike coming from the coil, thats pretty standard
but I dont get why we need the extra D2,D4, D11, D12, going off to vpp, when the signal is all ready clipped
please enlighten me

[mickstv]

You don't need D6 or D7 but you do need to have the other four diodes connected and referenced to the coil VPP.

D2,D4, D11 and D12 are the ones that clip the flyback.

[6666]

Thanks Mick

[moodz]

small but important update.

...been slaving over a hot processor for a couple of days now ...the good news is that the 6 channel 24 bit ADC code is now stabilised ( had some comms issues ... solved ) and the initial DSP filtering code is working .... see below ... the old resolution is the top reading ... the new resolution is stable voltage readings to 9 decimal places ....nanovolt country.
The dithering, oversampling and narrowband measurement technique allows effective measurement to little shy of 28 bits and could probably do better with a little clock fiddling.

There are 6 independant 24 bit channels that with a filter cut off of approx 32 hertz down to DC ( think of 6 ultra precision voltmeters ).

The next steps are to build the samplers / demodulators and construct the code to do the (V1 - 0.77V2 + tan(V3)) * 0.006V4 ... .... or whatever the algorithm is going to be for target extraction.


moodz.

[hdphilip]

Paul,
very impressive

in order to have the last digits being stable, were you measuring a battery?


Philip

[moodz]

Paul,
very impressive

in order to have the last digits being stable, were you measuring a battery?


Philip

Good question .... actually I have just realised what you are looking at is a bug .... but its a good bug . If you refer to the schematic posted previously you can see that every channel has protection diodes to AGND at the differential channel inputs. Channel 1 was reading a precision 5 volt reference coupled via 1 K resistor .... but of course this will forward bias the protection diode and the channel is actually reading the voltage drop across the conducting diode.

So actually that reading is supposed is supposed to be 0.4980447769 volts .... the bug means I have a decimal point in the wrong place so I need to check the printf formatting in my LCD code.

Thats 10 decimal places not 9 or 0.1 nanovolt resolution.

There is nothing too fancy about measuring these sort of voltages ... we have a sampling voltmeter here at the shop and it measures +/- 10 nanovolts full scale.

moodz

[hdphilip]

humm,
so with that sort of percision, the front end opamp would have to be a very low gain, about 10, in order to preserve the noise floor of the system correct?

philip

[moodz]

humm,
so with that sort of percision, the front end opamp would have to be a very low gain, about 10, in order to preserve the noise floor of the system correct?

philip

Absolutely .... a low gain diff amp followed by a switched demod then into the ADC. The first amp is mainly for common mode rejection.

moodz

[hdphilip]

so the new frontend will be fully differential from the coil back to the adc, maybe a fully differential intergrator followed by a buffer to feed the adc? maybe a 4066 tossed in their to have a "Quiet" window to look through?

philip

[moodz]

so the new frontend will be fully differential from the coil back to the adc, maybe a fully differential intergrator followed by a buffer to feed the adc? maybe a 4066 tossed in their to have a "Quiet" window to look through?

philip

I think so ... may as well take advantage of the differential inputs on each channel .... dont forget the on chip 30 db digitally controlled diff amp that fronts up each channel.

With 6 channels for sampling we should be able to come up with some good configurations.


paul.

[Davor]

The first amp is mainly for common mode rejection.

A plain old transformer would do it even better. Look for 10mH per coil types.

[moodz]

A plain old transformer would do it even better. Look for 10mH per coil types.

It could but in the configuration I use a transformer would upset the bias on the blocking fets and kill them ...worth thinking about though.

[moodz]

... YAU

...been slaving over a hot processor for a couple of days now ...the good news is that the 6 channel 24 bit ADC code is now stabilised ( had some comms issues ... solved ) and the initial DSP filtering code is working .... see below ... the old resolution is the top reading ... the new resolution is stable voltage readings to 9 decimal places ....nanovolt country.
The dithering, oversampling and narrowband measurement technique allows effective measurement to little shy of 28 bits and could probably do better with a little clock fiddling.

There are 6 independant 24 bit channels that with a filter cut off of approx 32 hertz down to DC ( think of 6 ultra precision voltmeters ).

The next steps are to build the samplers / demodulators and construct the code to do the (V1 - 0.77V2 + tan(V3)) * 0.006V4 ... .... or whatever the algorithm is going to be for target extraction.


moodz.

[ATTACH]18789[/ATTACH]

The long term DC accuracy can be maintained at 1 microvolt except channel 5 which is a bit noisier ...must be the layout or something....short term drift is around 200 nanovolts per second over a 32 Hertz bandwidth .... however cleaning up the power supply and shielding should improve things .... I will start on the demods now since I have some spare time.

moodz

[hdphilip]

hello moodz,

maybe it's the chip itself that could be a little noisy on channel 5, are ypu going to use the internial ref, or would a externial vref be more stable ?

philip