Issue 152 March 2003
2-D Optical Position Sensor

---

CIRCUITRY

If you decide to build this system, the layout is important for the detector pre-amps and ADC. A ground plane-type board is recommended.

We decided to design a PCB because it makes development much easier (see Figure 6). To limit noise pickup, the distance between the PSD and pre-amps should be no more than 1" (see Figure 7). The four-channel 12-bit ADC should be well bypassed and located near the gain-stage amps. Our circuit was laid out on two PCBs. We used a small pre-amp module for the pincushion 10 mm × 10 mm Hamamatsu S5991-01 lateral-effect photodiode and the larger one for the rest of the circuitry.

(Click here to enlarge)

Figure 6—The 2-D optical position sensor uses two op-amps in each leg of the PSD. The first stage is located near the PSD, and the voltage generated is negative with respect to ground. The second stage is located on the main board. It inverts the signal from the pre-amp stage back to a positive voltage for the 12-bit ADC.

 

(Click here to enlarge)

Figure 7—The first and second stages must use micropower rail-to-rail op-amps (LT1490s are used here). A programming port is located at J3 for downloading and flashing the PIC16C873 processor. The gain resistors in the second stages R12 through R19 and those in the pre-amp stages should be 0.1%-tolerance types to achieve the better than 1% accuracy that’s possible with the PSD.

Power comes from the two AA alkaline batteries that we mounted underneath the main board. The batteries drive an efficient charge-pump regulator and voltage inverter for operating the rail-to-rail op-amps. We recommend Linear Technology LT1490s.

All of the op-amps use 0.1% precision resistors (these tolerances must be used for measurement accuracy); the rest of the component tolerances are standard. For biasing this PSD, R9 is omitted, and R8 is 100 W. The design of this instrument assumes a laser pointer of less than 5 mW of power (most operate in the 2- to 3-mW range) and a PSD wavelength responsivity of 0.40 A/W.

The first-stage pre-amp gain must be set so that a 5-mW pointer can’t saturate it. Also, this type of PSD divides the total photocurrent seen by each lead by four. Finally, the rail-to-rail pre-amps will saturate when the output reaches 5 V. The feedback resistor value is determined by the equation depicted in Figure 4g. Therefore, the pre-amp gain resistors should be no larger than 10 kW. You must allow for ambient light, which will force Rf to be smaller. Four 0.1%-tolerance, 2.4-kW resistors were finally selected.

The second gain stages U7 and U8 invert the negative output from the pre-amps and offer a means to change the overall gain if a lower-power pointer is to be used. Note that the PSD used here will become nonlinear if the power density on its surface exceeds 3 W/cm2. The typical 2- to 3-mW laser level has a beam diameter of roughly 3 mm and a power density 70 times less than this limit. Saturation will occur when milliwatt power levels are focused to small spots on the PSD surface.

The outputs of the second stages go through 10-Hz, anti-aliasing, low-pass filters and are input to a Microchip MCP3204 four-channel, 12-bit analog-to-digital converter. Tactile push buttons serve to initiate the nulling functions and to change display modes. A MAX232 sends position data to an external device. Finally, a 2 × 16 LCD that’s addressed in nibble format and based on the Hitachi HD44780 controller is used as the display.

SIGNAL PROCESSING

The four photocurrents are turned into voltages by the pre-amps and then converted into 12-bit values. Remembering that the normalized position value (n) always ranges from –1 to 1, the problem is how to convert it to dimensional units. So, the first question to ask is: what resolution is desired?

Our goal for this instrument was 0.0001" resolution (2.5 µm). Because the measurement range from the center of the PSD is L/2, 5 mm or 0.1968", multiply the normalized position value by 1968—the largest position that can be sensed in units of tens of thousandths of an inch.

Now, we’ll recap. To display data with 0.0001" resolution, the PSD signals are processed by: obtaining X1, X2, Y1, and Y2 from the 12-bit ADC and bounds check for high and low; calculating the numerators in Figures 4e and f; multiplying the numerators by 1968; dividing both results by the sum signal (i.e., S = X1 + X2 + Y1 + Y2); and discarding the remainder.

POWER DATA

The power of the incident beam is proportional to the aforementioned sum signal, S = X1 + X2 + Y1 + Y2. The optical power (in milliwatts) is obtained via the equations in Figures 4h and i, where S is in A/D counts, G is the gain of the second stage, and R is the responsivity of the PSD in amps per watt. Rf is the value of pre-amp feedback resistors.

FIRMWARE

The firmware for this project was written in C using a CCS compiler. The main task is interrupt driven from Timer0 and continually acquires data from the four-channel ADC, solves the position equations, and then displays the data. The Mode and Null push buttons are polled to determine if the operator has pressed them.

There are additional points that must be emphasized. The low-pass filter on the outputs of all four gain stages won’t completely remove the strong 120-Hz optical noise signal created by room lights. We also wanted the PSD to be able to produce good results without the use of an optical filter. The firmware implements a first-order infinite impulse response (IIR) low-pass filter. The algorithm for this type of low-pass filter is given in Figure 4j. Note that C_n is the current filter output, and C_n–1 is the previous filter output. R_n is the current input to the filter, and R_n–1 is the previous input to the filter. Finally, f is the 3-dB breakpoint frequency of the filter in hertz, and T is the sample period in seconds.

For this filter, the sample period was 10 ms, and the 3-dB breakpoint was 5 Hz. Thus, the algorithm becomes a simple one with two coefficients (see Figure 4k). The actual code fragment is:

filter = ((signed int32)

olddata * coefs.a) + (((signedint32) newdata + (signedint32) lastsample) * coefs.b); olddata = (signed long int) (filter >> 15);

The two coefficients were multiplied by 32,768 before being multiplied by the delayed terms. Then, the result was shifted to the right by 15. This software filter effectively reduces the induced optical noise to low levels. Even under extremely intense lighting, the displayed position is rock steady.