PGA for Low Voltage Signals

It sounds like a problem of scaling so that the middle of your available gain setting matches up with the middle of your AD range of values. From what you have provided, you need to use a gain of 32 when you have an analog signal that is in the middle of your range.

Did that help?

Dear NotUrOrdinaryJoe,

I agree that the AD values for all wavelenghts should be in the middle of my AD range, but for reach it I must to use differents gains of the available gains of the PGA(1, 8, 16, 32, 64, 128). For example, if I have a AD target from 7.000.000 to 9.000.000 (for a AD converter 24 bits) and my signal is 30mV for 340nm, I could not get a AD value inside this range. It follows below the AD values using these gains:

gain 1: 100.663

gain 8: 805.305

gain 16: 1.610.611

gain 32: 3.222.253

gain 64: 6.442.444

gain 128: overflow

Regards

Eduardo

Hello,

I think that the scope of your research is beyond my own understanding, but I guess that there is not a light sensing material which exhibits linearity between the adjacent wavelenghts, I think that you have to program an extensive series of dedicated gains for each of them .

Is this a job for a lock-in amplifier? Mecahanical chopper and a phase sensitive detector?

340nm is normally the cut off point between using a tungsten lamp >340nm and a deuterium lamp <340nm. Different manufacturers of spectrophotometers may work around this point. Similarly, glass cuvettes can be used above 340nm and below quartz cells are used. The different gains to which you refer are due to the spectral response of the lamp and measuring cell. What you do not indicate is whether you are designing a scanning spectrophotometer measuring from say 340nm to say 900nm (here you start to go into Infrared region) or whether you want to measure at several single wavelengths.

In either case measurements are preceded by running a blank. How often you run the blank depends on the overall stability of the specphoto (my abbreviation of spectrophotometer) which is lamp, electronics, detector and temperature.

My suggestion then is that you run a blank and take the signals and then store them. Run the standard and this will enable you to calculate the gains required for the different wavelengths and then run the samples.

This process is not quite as straightforward as may seem on the surface.

Dear Tony,

Thank you for your answer.

I want to measure seven singles wavelengths as follows: 340, 405, 505, 546, 578, 620 and 670nm.

The references for my system (reference blank value) are measured as soon the instrument is started for all wavelengths and then I will store them properly. I could use some standards and calculate the gains, but if my calculated gain is not near of the option gains available in PGA (1, 8, 16, 32, 64 and 128)?

Eduardo

Click on this link and watch the video. If you need more help, you might be better off to contact the manufacturer. Many times, they have tutorials or other information that will answer your questions much more completely.

I looked at the AD7190, and you are at 75% of the goal. Good work.

Depending on the precision envisioned, step attenuators can be used at the low end, but staying with the basic approach I would use a PIC lookup table / DAC / Voltage Controlled Amp approach for fine calibration for the various wavelengths. Recalculation in a PIC is slooow, even if possible.

It requires an absolutely reliable reference standard reflectivity panel. Otherwise, how would it know the right values to set?

It requires one calibration entry per wavelength (or per a few) to set the second VCA via the DAC. This include the temp. dependent correction for the whole decoding chain. Since you covered the binary amp. steps already well, This second corrector amp deals with the rest. It has to cover the 1:2 ratio and some, and the temp. compensation (1:2, 1:4, or more). For a prototype I would allow 1:8. Meaning, 1:8 for amplification, AND 8:1 for dampening. Any amplifier set up for max.8 is far more stable than the rest. Later I would reduce it to 4:1 or 3:1, as actual measurements allow.

I would select a VCA from the same general family. Set its amplification around 8, and control it per (group) of wavelength via lookup table and DAC. I would guesstimate 3 digits precision or better until drift sets in.

The critical path is calibration.

Add-on to #8.

The correcting VCA has a very limited range. Hence a limited resolution 18 bit (16, 14, 12?) DAC is needed to control it. In an actual measurement cycle the DAC has to be set up before the delta-sigma, as it is quite noisy while its status changes. The actual number of bits depends on the noise calculations.

The 7190 does not permit the insertion of the correcting VCA between the integral amp and ADC. The only place is before the 7190. Hence, its noise factor is dominant. In order to satisfy that, you may find advantageous to shift everything up, to 1:1 to 1:16 (instead 1:4, 4:1). On the other hand, the corr.VCA buffers the signals for the 7190, relaxing its input matching.

The corr.VCA output impedance is 1kOhm or much less. At such low impedance it can be colocated with the sensing diode, and if needed, connected to the main board via a shielded, balanced connection. That may allow a better thermal control too.

Thank you for the comments.

Could you clarify how to use VCA before the ADC 7190?

I would appreciate if you show me an example using two specific signals for different wavelenghts. For example, consider that for 340nm the blank (reference) signal amplitude is 30mV and for 405nm it is 70mV. How could I adjust the VCA, DAC and PIC table and calculate the gains in order to have values for both wavelenghts inside the range of 7.000.000 and 9.000.000 (considering DAC 24 bits, tha maximum value is 16.777.216)?

Best regards

Eduardo

You have already took care of binary controlled amplification and digital conversion. What remains, is filling in the gap between the digital steps, and doing some overall (mostly temperature drift, and copy by copy of the electronics, differences) corrections.

That corr.VCA needs fine control. Its max. range is set by design 1:4 / 4:1 (or for noise considerations 1:1 and 16:1) via the feedback resistors. Then the normal digital control of it is effecting only that range, instead the wide range normally affected.

Since the 7190 does not have the pins to insert the corr.VCA between the amp and the sigma/delta, It needs to go before the 7190 entirely.

I read your note carefully, and I think I understand. The rough digital values for the main amp are set fixed in PIC code. So far so good. The values for the corr.VCA has to be measured in a calibration as frequently as you establish it. THEN, it is entered into the PIC as a variable. Valid for a run, a batch, a day? There is no way to know that without many measurement runs.

IF the signal strays past the binary step amp., the calibration run allows the corr.VCA 1:4 / 4:1 range ought to bring it back into center range. After all the AD function works on a wide range of amplitudes. If not optimum, S/N deterioration will cause only loss of precision. In an AD process, it is improper to keep the signal high for best resolution. Any overflow messes up it. Centering the signal first at 10% of the overflow value is a good starting point. Too many digits in the results are nothing but a mirage anyhow.

Then, the measured value:

AD result x digital AMP setting x corrVCA setting (the DAC)

Multiplication in binary is simply shifting the appropriate bit locations.

At this point I focused on basic functioning. Refinements, like reading very low values are for later.

Do you recommend a specific VCA for using in my project?

It should include a R-2R (DAC) inside this VCA?

Regards

Eduardo

No, I do not recommend anything specific, starting with Analog. For me at least, a DAC is a separate item. Its switching is noisy, its digital control might be noisy. If it is integrated into the chip, I am willing to accept the resulting parameters.

There is always more than one way to skin a cat. Over the weekend I contemplated your original problem, and my proposed solution. While the reuse of the same amplifier architecture is an elegant and predictable solution to provide fine resolution, It may be quite taxing for you.

There is a way to get good results with the original design you picked.

That other approach I was thinking of, was that the selected ADC with 24 bits resolution is an overkill for the intended application. You need to do a comprehensive error / noise / drift evaluation for the whole system first in any case:

Light source + optics

Standard(s)

Receiver optics + grating + choppers?

Receiving diode + its driver

Amp +ADC

Postprocessing in PC

As you will not know all values, put in some guesstimated numbers first.

And the question is, what you can do about any of it. Some you can, some may be too expensive, some you may have to live with.

One of my top bet would fall on the receiving diode. No matter its biasing, its sensitivity is big time wavelength and temperature dependent. So is the noise generated. The wavelength sensitivity is exactly predictable. The noise variation is calculable and measurable. The temperature sensitivity is the sticking point. As you may decide to stabilize its housing's temperature, the chip's temperature still may vary depending on the infalling amount of radiation. That is an error factor you may guesstimate, but cannot eliminate.

When everything is said and done, you may end up with a system yielding somewhere 12 - 18 bits precision, reliably.

That has fundamental implications for your ADC. If you reduce its amplifier's output voltage by 4 - 6 (or more??) bits, The ADC will still the most reliable part of the whole scheme. The danger of the overflow is eliminated, and the binary control of the ADC amp is quite sufficient to get amplification into the right range. According, what the calibration standard's measurement pass yielded.

The digital value may be passed on to a PC to be processed in a, a say, EXCEL spreadsheet, or displayed straight in the instrument, or both.

The architecture allows handling very low measured values, but that is for later.

At first, it may be deflating to one's ego realizing, that in real life 3+ decimal digits absolute with the 4th digit relative measurement is quite an accomplishment. And sometimes fine instruments can do much less. Realizing how many bits or digits are real is a main part of your learning curve.

A very rude case popped up about what I was writing, concerning calibration, etc..

cr4.globalspec.com/thread/83109/My-Test-Stand-is-a-Calibration-Mess

It appears, calibration and all associated items are real issues for other people too.