processing circuit for two-dimensional position sensitive detector (2d-PSD)

Hi everybody!

I want to build a processing circuit for two-dimensional position sensitive detectors (in my case type Hamamatsu S5990-01). With the spot intensity I use, the PSD delivers signals I1-I4 in the order of 20-40mV.

The evaluation board should deliver the following outputs:

x-coordinate of the spot along the active area of the PSD:

amp_factor * [ I1+I2 - (I3+I4) / (I1+I2+I3+I4) ]

and similarly for the y-coordinate:

amp_factor * [ I2+I3 - (I1+I4) / (I1+I2+I3+I4) ]

Furthermore, the circuit should operate a at least 120kHz, which is probably not easy because of the analog division by the sum of the four signals. (.../(I1+I2+I3+I4)).
Has anyone of you done something like this before and could help me with the design and the choice of the right components?
Thanks a lot!

AD make some nice log amps that will handle orders of magnitude more bandwidth then that, or do it with the exponential relationship between voltage and current in a diode junction (Often actually the base/emitter junction in a transistor).

Alternatively, there is a devious trick using an ADC and a DAC, in which you feed the ADC reference input with the denominator and get your division that way. then convert back to the analogue domain with the DAC.

Personally I would probably use a quad opamp to scale things up a bit and then just stick a small micro in there to do the math and have done with it, lazy maybe, but quick and easy and two small chips.

Regards, Dan.

hey dan!
many many thanks for your help! could you be more precise on the idea with the adc/dac trick or tell me where i can find more information?
i also like your last idea with a micro. unfortunately i don't know a lot about micros and programming.

i have tried to do the division with a ad633jr earlier (see attachment), but my output signals were not stable (oscillating and noisy) even at dc-inputs.

That is NOT the logic your math asks for (Division should bind more tightly then subtraction so your eqn should be something more like (adding brackets to show the binding) a + b - ((c+d)/(a+b+c+d)), what your circuit gives is more in the line of (a+b-(c+d))/(a+b+c+d), not the same thing at all.

Which one is it you are trying for?

The AD633 is quite fast, and needs good decoupling to work well, consider also what happens when the sum is small, you can end up demanding a huge amount of gain and that is never good for stability.

For that matter, with ~40mV would it not be a good idea to provide some frontend gain (maybe 20dB or so)? It will make your noise problems go away.

The AD trick is simple, you feed the denominator into the reference intput (scaled so it is always larger then the numerator) and apply the numerator to the adcs input, the resulting value will be 1/(2^Bits) * numerator/denominator, easy. Feed this to a DAC having a fixed reference and you get a voltage proportional to numerator/denominator, job done (Old BB app note as I recall).

Regards, Dan.

Your circuit is not a correct design for that sensor. Instead of summing the signals directly from the sensor, which thus splits the signal and increases noise, you need to first boost each signal with an op amp preamp with gain for each output and then sum the signals from the preamp outputs.

Since the sensor is designed as a current output device not voltage, the preamp op amp should have a transimpedance connection (sensor directly to the (-) op amp input with a feedback resistor from the output to the (-) input).

Select the value of the feedback resistor to give about a 1V peak output. Note that the OPA4277 has a slew rate of only 0.8V/µs thus it's maximum sine-wave output voltage at 120kHz is about 1V peak (sine-wave slew rate = 2π*f*Vp), thus you will need a faster op amp to get a higher voltage for the multiplier input. One way is to use the low-noise OPA4277 for the preamp to generate about 1V peak voltage and then use a faster op amp for the subsequent summing and amplification. Those op amps don't need to have such low noise.

Of course the connection from the sensor to the preamp input must be short (the preamp right at the sensor if possible), using shielded wires if there's a problem with hum/noise pickup.

You likely will need to add a small compensation capacitor across the preamp feedback resistor to minimize peaking in the frequency response due to the input capacitance from the sensor. The optimum value will need to be determined during testing.

That is NOT the logic your math asks for (Division should bind more tightly then subtraction so your eqn should be something more like (adding brackets to show the binding) a + b - ((c+d)/(a+b+c+d)), what your circuit gives is more in the line of (a+b-(c+d))/(a+b+c+d), not the same thing at all.

Which one is it you are trying for?

Click to expand...

sorry, this was a mistake in my first post. i mean (a+b-(c+d)) / (a+b+c+d) of course.

The AD633 is quite fast, and needs good decoupling to work well, consider also what happens when the sum is small, you can end up demanding a huge amount of gain and that is never good for stability.

Click to expand...

ideally, the sum should stay more or less constant at about 120mV (if one signal is about 30mV) since it represents the total light intensity. it is used to normalize the positions in such a way, that fluctuations in light intensity don't have any influence upon the real position. do you think 120mV is already critical? what would be a good range for the ad633 to work with? is there any alternative device i should prefer and how exactly should the ad633 be decoupled?

For that matter, with ~40mV would it not be a good idea to provide some frontend gain (maybe 20dB or so)? It will make your noise problems go away.

Click to expand...

i will definitely try do that. it was also suggested by crutschow

The AD trick is simple, you feed the denominator into the reference intput (scaled so it is always larger then the numerator) and apply the numerator to the adcs input, the resulting value will be 1/(2^Bits) * numerator/denominator, easy. Feed this to a DAC having a fixed reference and you get a voltage proportional to numerator/denominator, job done (Old BB app note as I recall).

Click to expand...

i will also try this. thank you Dan!

Regards, Dan.

- - - Updated - - -

Your circuit is not a correct design for that sensor. Instead of summing the signals directly from the sensor, which thus splits the signal and increases noise, you need to first boost each signal with an op amp preamp with gain for each output and then sum the signals from the preamp outputs.

Since the sensor is designed as a current output device not voltage, the preamp op amp should have a transimpedance connection (sensor directly to the (-) op amp input with a feedback resistor from the output to the (-) input).

Select the value of the feedback resistor to give about a 1V peak output. Note that the OPA4277 has a slew rate of only 0.8V/µs thus it's maximum sine-wave output voltage at 120kHz is about 1V peak (sine-wave slew rate = 2π*f*Vp),

Click to expand...

this makes sense. i will post a new diagram soon.

thus you will need a faster op amp to get a higher voltage for the multiplier input. One way is to use the low-noise OPA4277 for the preamp to generate about 1V peak voltage and then use a faster op amp for the subsequent summing and amplification. Those op amps don't need to have such low noise.

Click to expand...

any idea which fast op amp i could use for that? and is it a good idea to stick with the ad633?

Of course the connection from the sensor to the preamp input must be short (the preamp right at the sensor if possible), using shielded wires if there's a problem with hum/noise pickup.

Click to expand...

i am doing this already. thank you for your great help!

lastfreename said:

..................................

any idea which fast op amp i could use for that? and is it a good idea to stick with the ad633?

........................

Click to expand...

Just about any op amp with at least a 4MHz GBW and a slew rate of 8V/µs or greater should work, for example the MC34074 quad. What parts do you have access to?

The AD633 looks adequate for your purposes.

i have access to whatever is sold on the web i have checked rs-components and farnell. both sell the mc34074. i will follow your advice and order some.. thanks again!

- - - Updated - - -

dan, i would also want to try the devious ADC/DAC trick that you mentioned 3 days ago. could you recommend any ADC and DAC devices for my purpose? i would like to order them together with the op-amps that were suggested by crutschow and try both ways.

I'll jump in for Dan in case he's not around. Here's a 12-bit multiplying DAC that you could use (you can also use a multiplying DAC with fewer or more bits if you want). Figure 7 on page 8 shows it being used for gain control of a signal which really gives a division of -Vin / (input bit value). Note that the minimum circuit gain is one. I'm not sure if that's exactly what Dan was referring to but it should work for your purposes.

TI, Analog Devices, and Microchip all sell ADCs so just pick one appropriate for you application. For a 128kHz signal you should have a sample rate (not clock rate) of at least 5 times 128kHz or 640kHz.

You will also may need some simple logic and clock circuitry to control the ADC. Some, I believe, have a built-in clock, simplifying the amount of external circuits required.

But it's not clear to me this more complex scheme would have any advantage over just using an analog multiplier.