Newsroom

Dec 1, 1993

PALO ALTO, CA--- Hewlett-Packard Journal reports about the high-resolution direct-drive solution from COPI for an optical spectrum analyzer

Dec 1, 1993

PALO ALTO, CA--Dec 1, 1993--A high-resolution direct-drive diffraction grating rotation system - design of controller mechanism used in HP's 71450A and 17451A optical spectrum analyzers - Technical

Joseph N. West

Creating a high-resolution, high-speed positioning system that can provide over two million data points per revolution of the diffraction grating required a design that is much different from the gear-reduction positioning systems typically used in optical spectrum analyzers.

The wavelength tuning of the double-pass monochromator used in the HP 71450A and 71451A optical spectrum analyzers is controlled by the angular position of the diffraction grating. After the input beam is collimated by the lens, it strikes the diffraction grating, where each wavelength is dispersed at a different angle. For each angle of the diffraction grating a corresponding wavelength is passed back through the optics and focused on the center of the first-pass aperture (slit). The width of the slit determines the resolution bandwidth of the wavelengths that pass through the remainder of the system to the detector. Rotating the grating causes the dispersed wavelengths to sweep across the slit, making the monochromator act as a tunable filter. Fig. 1 shows the housing that contains the optical and electromechanical components that make up the double-pass monochromator assembly.

The angular resolution requirements for the grating positioning system in the monochromator can be determined by calculating the relationship between the angular position of the diffraction grating relative to the collimated light and the spatial dispersion of the light at the resolution bandwidth slit.

Using the grating equation* for a scanning monochromator like the one used in the HP 71450A and 71451A optical spectrum analyzers, roughly 600 microradians of diffraction grating rotation per nanometer of optical dispersion (at 1300 nm) can be calculated. To represent narrow signals, it is desirable to have at least sixteen data points across the narrowest resolution bandwidth of the instrument (0.08 nm). This translates to 200 data points per nanometer of dispersion. Dividing 600 microradians by 200 points gives an angular resolution requirement of about three microradians (0.00017 degree) per point, or about 2,100,000 data points per revolution of the diffraction grating.

Conventional Methods

The traditional approach to building such a high-resolution positioning system is to use large amounts of gear reduction (see Fig. 2). This is the approach used in most older optical spectrum analyzers. In these systems there are commonly two stages of gear reduction. The first stage might consist of a planetary gearhead with a reduction of about 20:1 which would be followed by a worm drive with an additional reduction of about 30:1 for a total gear reduction of about 600: 1. The advantage of this approach is that it reduces the resolution requirements of the primary feedback device, often an optical encoder. It is possible to get away with using a fairly low technology, 1,0004ine TTL-output encoder. Looking at every zero crossing from the two quadrature channels of this encoder gives a resolution of 4,000 counts per revolution of the encoder which combined with the 600:1 gear reduction provides sufficient resolution to position a diffraction grating.

However, the gear reduction approach has several drawbacks. One of these is speed. With a 600:1 gear reduction, the diffraction grating is rotating at only 1/600 the speed of the motor. Similarly, the acceleration of the diffraction grating is only 1/600 the acceleration of the motor. Moving the diffraction grating at any significant speed requires that the motor and gear train be accelerated to very high speeds. Reversing the direction of the grating requires decelerating the motor and gear train and accelerating them to high speeds in the opposite direction. In the past this speed penalty came to be accepted as inevitable because no alternative was generally available for such a high-resolution system.

A second drawback to the gear reduction approach is backlash. Backlash is the slop exhibited by a gear train when the direction of rotation changes and the gears change from contact on one gear face to contact on the opposite gear face. A number of techniques can be applied to minimize backlash in a system. These techniques generally consist of a method for compliantly loading the gear train to ensure that the same gear faces remain in contact regardless of the direction of rotation. Fig. 3 shows a simple spring used to reduce backlash in a system that rotates over a small angle. Obviously, in a full-rotation system a more complicated scheme is required. These antibacklash techniques help, but do not eliminate backlash. Even though the same gear faces remain in contact, as the gear train reverses direction lubricants are smeared in an opposite direction and the bearings that support the gears are deflected in an opposite direction. For a high-resolution system, some degree of backlash will still be evident.

A third drawback of geared systems is susceptibility to errors caused by wear or by changes in environmental conditions. In a typical gear-reduction system, the angular position of the drive motor is monitored with an optical encoder. The angular position of the diffraction grating is not measured directly, but is inferred from the motor position and the gear ratio. As gears wear or expand and contract with temperature, or as lubricant viscosities increase over time, the actual position of the diffraction grating relative to the motor position will change. Periodic recalibration is needed to correct these errors.

Direct-Drive Grating Rotation System

Based on the drawbacks mentioned above, the decision was made to use a direct-drive, direct-readout grating rotation system in HP's double-pass scanning monochromator. Fig. 4 shows this system. A drive motor and a rotary optical encoder are directly attached to the shaft that holds the diffraction grating. This system has the following advantages:

Direct measurement of the angle of the diffraction grating
No errors from backlash or "wind-up" deflections in the gear train
No susceptibility to wear and much less sensitivity to environmental changes
Speed, compactness, and ruggedness

Implementing a high-resolution direct-drive, direct-readout system placed some stringent requirements on the components used and the design process. Two things were required. The first was a high-torque motor for fast starting and stopping and high-speed scanning. The motor we chose is a frameless, brushless dc motor. The permanent magnet rotor uses strong rare-earth magnets and mounts directly to the diffraction grating shaft. The stator mounts to the fixed outer housing where heat can be dissipated in a controlled manner. The motor is brushless so there will be no debris generated by brushes wearing and no maintenance (brush replacement) over the life of the instrument. In addition, brushless dc motors have an advantage over conventional brush-type motors in that there is no friction because of rubbing between the commutator and the brushes. Friction is a problem in the control of high-resolution systems.

A second, and more difficult requirement is a high-resolution, optical-encoder-based measurement system that is able to resolve directly more than two million points per revolution of the diffraction grating. Building such a system involved searching for the latest in optical encoder technology and then applying considerable design effort to accomplish the necessary resolution goals.

The encoder used in our system is a sine wave output incremental rotary optical encoder with 9,000 lines on the rotating disk. In an incremental encoder a single light source, typically a light-emitting diode, shines a beam of light through a rotating disk that contains radial slits which alternately transmit or block the light. The light passing through the slits is detected by two sets of photodetectors that convert the light into electrical signals. Before hitting the photodetectors, the light also passes through a phase plate containing two additional patterns of slits. These two patterns are offset slightly relative to one another so that the signals received by the two sets of photodetectors are 90 degrees out of phase. The quadrature relationship of the signals makes it possible for the user to know the direction of rotation of the encoder by looking at which channel is leading by 90 degrees and which channel is lagging. There is also a third channel that provides an index pulse once per revolution for determining absolute position. The outputs of the main A and B channels are very close to sinusoidal (see Fig. 5).

Each zero crossing of the A and B channels increments or decrements a position counter to provide coarse position information, depending on the relative phase of the A and B channels. To increase the resolution beyond the usual 4 x linecount value, a process called interpolation is used. Because the signals are sinusoidal, there is additional analog information between zero crossings which can be extracted. Commercial circuits are available that perform the interpolation function, but they typically have interpolation ratios that are convenient numbers in the base 10 number system, such as 5x or 50x For a digital control scheme, it is more convenient to have an interpolation ratio that is an integer power of two. The HP 71450A and 71451A optical spectrum analyzers use an interpolation ratio of 64:1 (actually 256:1 for data acquisition, but the two lowest-order bits are not used for control). This gives a resolution of 9000 x 4 x 64 = 2,304,000 counts per revolution determined directly from the encoder.

Interpolation in this design is achieved by amplifying the two sinusoidal outputs of the encoder until the minimum and maximum values are just within the range of an analog-to-digital converter (ADC). The outputs of the ADC are then a digital representation of the sine and cosine signals. The ratio of the two digitized outputs is the tangent of the angle. By looking up the ratio in an arctangent table, the angle that is the interpolated fractional position between the sine and cosine zero crossings can be found. The accuracy of the interpolation is dependent upon the degree of distortion in the sine and cosine signals, the phase angle between them, and the number of resolvable bits in the analog-to-digital conversion process.

One of the problems encountered in controlling high-resolution systems such as this is the change in behavior of the friction in the system as the system changes from moving to a fixed position. This happens when the grating is either tuned to a fixed wavelength in zero span or it is momentarily stopped while changing directions at the beginning or end of a sweep. The degree of resolution is so fine that the difference between the static case and the dynamic case becomes readily apparent.

While the system is in motion, that is, servoing to a moving target position, the friction is a nicely behaved linear damping term. The rotational inertia of the system interacting with the motor winding resistance forms a simple pole. There is also a pole at zero frequency since a constant input voltage to the motor gives a steadily increasing angular position. This is a fairly simple system to close a servo loop around (see Fig. 6). Fig. 7a shows the openloop frequency response of the system while in motion. The noisy measurement at frequencies below 1 Hz is because of a signal-to-noise ratio problem in this particular measurement and not an indication of the actual low-frequency response.

The spring-mass resonance in the fixed position is readily apparent from the peak in the magnitude response at 30 Hz.

At very low rotation rates, the system rapidly jumps between servoing to a moving target position or to a fixed target position. If the loop is not compensated to take this change of behavior into account, the result can be a system that is stable when moving but that oscillates when it servos at a fixed position.

In our design we found a single set of loop compensation values that provide a stable response for either operating mode, ensuring that the system works well under all conditions.

Conclusion

Direct-drive technology has been applied with great success in a number of industrial applications ranging from phonograph turntables to industrial robots and military gun turrets. Applying these techniques to an optical spectrum analyzer produces a system that provides fast, accurate, and reliable rotation of the diffraction grating, and with regard to motion control, brings the latest technology to optical spectrum analysis.

* See "Diffraction Grating" on page 170