Accuracy considerations and error correction techniques for 20-GHz lightwave component analysis

An understanding of factory calibration techniques, system capabilities, and device-under-test sensitivities can result in more accurate and repeatable measurements using the HP 8703A lightwave component analyzer.

THE ACCURACY OF MEASUREMENTS made with a lightwave component analyzer is the result of instrument performance, measurement technique, and consideration of the particular device being measured. To ensure the most accurate characterization of wide-bandwidth components, it is important to understand the measurement capabilities of the instrumentation, decide on the optimum configuration and calibration for a particular measurement, and be aware of potential impact the test device can have on the resulting measurement. The accuracy of the result is a combination of multiple sources of errors that must be carefully considered.

Factory Calibration

The measurements made by the HP 8703A lightwave component analyzer can be divided into four categories: optical-to-optical (O/O), electrical-to-electrical (E/E), optical-to-electrical (O/E), and electrical-to-optical (E/O). Optical and electrical calibrations are done in an HP standards laboratory using state-of-the-art equipment. This laboratory has NIST (U.S. National Institute for Standards and Technology) traceability for all of these measurements. However, there are no E/O or O/E standards that are directly traceable to NIST. The accuracy of these calibrations is based on theoretically derived performance of electro-ptical devices as described below. Although the techniques are not directly traceable to NIST, the measurements used in the calibrations (optical power, RF power, etc.) are traceable.

One technique used at Hewlett-Packard for calibrating O/E devices is known as heterodyne calibration. The outputs from two temperature-tunable YAG (yttrium aluminum garnet) lasers are combined in an optical coupler and applied to a photodetector under test. The YAG lasers are tuned so that the difference between their frequencies sweeps the microwave band of interest. The output of the photodetector is measured on a calibrated receiver such as an HP 71210C spectrum analyzer. The optical power output of each YAG is measured and the sum of the two is computed. From this information the frequency response of the photodetector can be derived. Transfer standards calibrated on the HP heterodyne system are used to calibrate the HP 83410/11A receivers and the HP 83400/01/02/03A sources.1 Although this method does not directly calibrate the HP 8703A, 20-GHz photodiode receivers measured on HP's heterodyne system are used as a second verification of the frequency response of the HP 8703A.

The technique used to calibrate the HP 8703A's E/O frequency response is known as the two-tone technique (Fig.

This technique applies two microwave signals (tones) to the optical modulator, which is biased off quadrature. It can be shown that the magnitude of the tones' mixing product (difference frequency) is proportional to the square of the modulator's response times the product of the magnitudes of the tones. If the tones are swept across the frequency band of interest (difference frequency kept constant) and the output of the modulator is monitored with a low-frequency detector, the square root of the ratio of the detector output to the tones' magnitudes will be a measure of the modulator's frequency response. The modulator can then be used to measure the frequency response of an optical receiver. Some advantages of this method are that it does not require accurate frequency response calibration of the detector (the detector only sees one frequency), it can be performed at any wavelength within the range the optical modulator simply by changing the laser, and with the addition of the low-frequency photodetector it can be performed with equipment found in most microwave test stations. For the HP 8703A calibration this method provides improved accuracy over measuring a transfer receiver calibrated on the HP heterodyne YAG system. It also eliminates the need for a calibrated E/O source transfer standard.

There are also no standards for measuring the phase of E/O or O/E devices. For phase calibration, a 20-GHz photodiode receiver was modeled on an HP 85150B microwave design system. To model the device, its signal path was divided into sections: optical, optical-to-electrical, and RF (see Fig. 2). The known physical properties or theoretical properties of each section were converted to a model which was entered into the computer. In addition to phase response, the computer predicted several characteristics of the device that could be measured, such as the magnitude of the response and electrical parameters such as S22- Comparing the predicted values to measured values gave information that was used to refine the model. When good correlation between the measured and predicted values was obtained for measurable parameters, we had high confidence that the model would also predict correct phase information. As a final check, an optical modulator was also modeled and the receiver and modulator were measured together. Their combined phase response was compared to the sum of the models (this is an E/E measurement for which standards exist). Agreement was within 5 degrees above 5 GHz and within 15 degrees at lower frequencies. The latter is nearly all attributable to the optical modulator, whose rapidly changing response at low frequencies is more difficult to model. The receiver has a smoother response, and therefore its phase response is more predictable.

Factory Calibration Sequence

Fig. 3 shows simplified signal paths for O/E and E/O measurements. For accurate measurements the HP 8703A RF and optical paths including the RF and optical cables must be calibrated out. To eliminate the need for an expensive reference receiver and source at the customer site, the HP 8703A is calibrated for these measurements at the factory and calibration constants are stored in the HP 8703A memory. The customer need only connect the measurement cables to the HP 8703A front-panel ports during calibration and the HP 8703A will automatically correct for systematic errors including the cables.

From Fig. 3 we see that the responses of four paths need to be corrected inside the HP 8703A: the port 1 electrical output, the port 2 electrical input, the optical output, and the optical input. The sequence of factory calibration for these paths is as follows: Step 1. The HP 8703A's built-in synthesizer signal is combined with an external synthesizer and applied to the optical modulator via a rear-panel port to perform the twotone calibration described above. The difference frequency is measured with a low-frequency detector and a spectrum analyzer at each frequency point and stored. This measures the frequency response of the optical output. Step 2. The optical output is connected to the optical input through a previously calibrated optical cable and this path is measured and the data stored. Step 3. Port 2's electrical response is measured with a previously calibrated source and stored. This gives a measure of Port 2's absolute response. Step 4. Port 2 is connected to Port 1 with a previously calibrated RF cable and the measured response is stored. Port 2's response and the RF cable's response are subtracted from this measurement to compute the absolute response of Port l's output.

Step 5. A reference receiver is connected between the optical output and Port 2 and measured. This receiver has had its absolute responsivity calibrated at 130 MHz and has been phase modeled as outlined above. Step 6. From the above data the computer generates calibration constants which are then stored in the HP 8703A's memory.

Step 7. As a final check a photodiode receiver that was previously calibrated on the HP heterodyne YAG system is measured with the HP 8703A and the results of the two systems are compared. Correlation between the systems is typically better than (+ or -) 0.5 electrical dB (see Fig. 4).

User Calibration

As an example of how the HP 8703A corrects E/O measurements, let us look again at Fig. 3. Four parameters need to be corrected to measure the DUT accurately: the port 1 response (PORT1), the optical input response (OPTIN), the RF cable response (RFCBL), and the optical cable response (OPTCBL). To calibrate the system the operator connects the measurement cables to the HP 8703A input ports as shown in Fig. 5a. Pressing the appropriate front-panel keys will cause the HP 8703A to calibrate the system automatically by measuring the following signal paths:

CAL1 = (PORT1)(RFCBL)(PORT2)

CAL2 = (OPTOUT)(OPTCBL)(OPTIN).

The factory-stored E/O constants are:

FACTE/O = (OPTOUT)(PORT2).

The HP 8703A combines this data to form correction constants

(CAL1)(CAL2) CORRECTION =

FACTE/O

(PORT1)(RFCBL)(PORT2)(OPTOUT)(OPTCBL)(OPTIN) /

(OPTOUT)(PORT2)

= (PORT1)(RFCBL)(OPTCBL)(OPTIN).

The operator then connects the cables to the DUT as shown in Fig. 5b and the HP 8703A measures the DUT. MEASURE = (PORT1)(RFCBL)(E/ODUT)(OPTCBL)(OPTIN).

The displayed data is:

MEASURE / CORRECTION = (PORT1)(RFCBL)(E/ODUT)(OPTCBL)(OPTIN) / (PORT1)(RFCBL)(OPTCBL)(OPTIN)

= E/ODUT.

A similar correction process occurs for O/E, O/O, and E/E measurements.

Verification

With the calibration stored in the instrument for E/O or O/E applications, the measurement of a device under test is accomplished by a process of substituting the DUT for system calibration data. Each instrument makes its measurements accurately because the factory calibration process generates absolute calibration data unique to that set of hardware. The quality of measurement provided is a direct result of the calibration performed at the factory. The long-term stability and repeatability and therefore the integrity of the product are direct results of the stability of the system hardware and the original calibration. For user verification of the system performance, the customer is provided with a measurement of the absolute characteristics of the critical instrument measurement paths made at the time of calibration. The integrity of the measurement system is then monitored over time by testing its ability to reproduce the original absolute characteristics measured at the factory. This provides the user with a simple measurement that monitors the system's status and is linked directly to its original calibration.

Specifically, at the time of the factory calibration of an HP 8703A, a measurement is made on the hardware that represents an absolute measurement of the optical and electrical portions of the measurement system that are used in E/O or O/E measurements. In each case, the only external connections to the instrument are RF and optical through connections to close the measurement paths. The absolute measurement data along with instrument setup information and prescribed limit lines (which account for the allowable drift and repeatability of the measurements) are stored on a verification disc and shipped with the instrument. To verify the instrument, the customer makes the simple through connections, the data and limit lines are recalled, and a new measurement is compared to the original. If the new measurement is outside the allowable limits, verification has failed and the customer will be advised to cheek the quality of the cable and connections being used. If that does not alleviate the out-of-limits condition, the instrument will be considered out of specification and must be returned for repair and/or recalibration.

It is important that this process does not require special electrooptical verification reference devices. The only key elements besides the instrument are appropriate interconnect cables. This avoids the need for an expensive optical verification kit. In addition, the simplicity of the procedure makes it easy for the user to test the integrity of the system upon receipt and at frequent intervals during its use.

In addition to the frequency response verification, a full performance verification of the HP 8703A must also include absolute optical measurements of power and wavelength. All E/E microwave performance verification is done with a standard verification kit similar to other microwave analyzers.

System Capabilities

A thorough understanding of the HP 8703A's hardware characteristics and limitations is essential to achieving the best measurement results.3

The electrical stimulus is a 130-MHz-to-20-GHz synthesized microwave source with 1-Hz frequency resolution. The synthesizer's frequency accuracy and repeatability play a key role in achieving accurate and stable calibrations. A + 5-dBm output power level provides a high signal level (compared to other microwave network analyzers) to achieve typical drive levels used for laser sources. Since the analyzer receiver is tuned to the source drive frequency, relatively high source harmonics (< 10 dBc) are not a cause of measurement error. The typical source match varies over the frequency range from 20 dB at 130 MHz to only 10 dB at 20 GHz. This can result in measurement mismatch errors that may require user calibration.

The electrical receiver is a microwave sampler tuned to the source frequency. Its typical operating range is between 0 dbm-the largest signal input before receiver compression begins-and typically - 105 dbm-the noise floor in a 10 Hz bandwidth. The receiver bandwidth can be increased in steps to 3 kHz. This allows faster response measurements but increases the noise floor of the receiver accordingly. Similar to the source, the receiver has an input return loss that varies from 26 dB to 1 2 dB (typically) over the operating frequency range. Fig. 6 shows the error of the receiver detection capability as a function of input power level. The high-power errors occur as a result of compression in the receiver samplers. The low-level errors occur as the detected signals approach the noise of the system. The detection response over the middle region from 0 to - 80 dBm is very linear.

The optical output signal is intensity modulated at the measurement frequency rate. It has an average optical power of 250 microwatts and operates at a fixed modulation level of approximately 130 microwatts peak-to-peak intensity change. Laser options result in different optical source linewidths ranging from 3 nanometers for the Fabry-Perot structure to less than 50 MHz for the DFB sources. The optical output match is determined by the output match of the lithium niobate modulator and typically corresponds to greater than 15 dB of return loss. All the laser sources are isolated to avoid problems caused by sensitivities to reflections at the source output port.

The optical receiver is an amplified photodiode receiver in front of a tuned electrical receiver. Fig. 7 is a curve of the typical dynamic accuracy of the optical detection path. The high-level distortion comes from compression in the receiver samplers. The low-level noise is about 15 dB higher than in the electrical receiver because of noise added by the amplifiers in the photodiode structure. Also, since the photodiode is a power sensing device, the electrical detected output signal varies by 2 dB for every 1-dB optical change. Thus the optical levels shown represent an electrical detection dynamic range that is twice as large.

In conjunction with optimization of the basic measurement hardware, several levels of user calibration are provided to allow correction for some aspects of the system. Table I shows the various user calibrations available for each type of measurement.

The response calibration is the process of transferring the stored calibration of the reference receiver or source to the current measurement (in E/E or 0/0 measurements it is a simple through calibration). The port calibrations and match calibrations improve equivalent electrical source and load matches to greater than 30 dB. This is particularly important when test port and device port matches are of such a level that their multiple mismatch errors could be a significant source of error (typically for low-return-loss devices). Calibration of electrical ports for reflection measurements results in an equivalent directivity of >36 dB. Isolation calibration removes measurement system crosstalk to ensure full access to the system dynamic range.

Device-Under-Test Sensitivities

Besides understanding the primary characteristics of the measurement hardware, it is also important to pay careful attention to any device-under-test characteristic that could impact measurement accuracy or cause unstable or nonrepeatable measurements.

First, the primary DUT characteristics (gain, match, distortion, etc.) must be considered in setting the measurement signal levels for maximum dynamic range and accuracy. These characteristics may establish a need for extra padding in the measurement path or determine the initial choice of user calibration. Not a small issue is the choice of connectors for the DUT. Optical connectors in particular can have match characteristics (for a mated pair) that vary from less than 10 dB return loss for noncontacting connectors to 40 dB for precision connector types. Similarly, 3.5-mm RF connectors can vary from 20-dB-match, industrial-quality connectors to >45-dB-match instrument-grade connectors. In addition to the right choice of connector type, the proper use and care of the test port and DUT connectors are essential for making good and repeatable measurements. The cleanliness of optical connections is very frequently the source of significant repeatability difficulties.

More interesting than these issues are some of the more subtle sensitivities optical portions of the DUTs tend to display. One such characteristic that can often lead to confusing and unrepeatable measurements is the sensitivity of laser sources' responsivity-versus-frequency characteristics to optical load characteristics. A particular optical load can couple back into the laser and alter the modulation response of the structure. The diagram in Fig. 8 shows a test configuration used to measure reflection sEnsitivity. A low-coupling-factor coupler is used to monitor the modulated output of a laser while the main coupled path at the test laser output is exposed to various load conditions. Fig. 9 is a polarization sensitivity measurement of a FabryPerot laser with a high-reflection load that has been set with a polarization controller for maximum and minimum sensitivities. The responses change from 1.5-dB peak-to-peak ripples to greater than 4-dB peak-to-peak ripples. While this configuration can be used to measure such sensitivities, in other measurement configurations the same phenomenon can be the source of unexplained fine-grain variations in a device measurement. The sources used in the 20-GHz lightwave component analyzer products all have optical isolators to avoid any measurement system sensitivity of this type.

Another interesting device characteristic that can influence measurements is that of interferometric "noise" effects. These result when optical reflections in a system interact in a time varying manner. Since the wavelengths of light are so small, a fiber can easily change length by many wavelengths with small changes in temperature. Such changes will cause the vector sum of the various light components in the system to vary over time. A method of reducing this condition is to add a length of fiber between the sources of optical reflections that is longer than the coherence length of the source spectrum. This will cause the multiple reflections in the system to be uncorrelated so that they no longer add or subtract coherently. Fig. 10 shows a simple fiber measurement with two different lengths of fiber. The longer length is greater than the coherence length of the 50-MHz-linewidth DFB laser source in the measurement system. Fig. 11 is a plot of the fiber measurements at a fixed modulation frequency of 1 GHz. The two traces are for the different lengths of fiber: the top trace is for 1 meter and the bottom trace is for 50 meters. The drift is somewhat random with time and the long-line trace shows about half that of the shorter line. The remainder of the effect is caused by multiple reflections in other parts of the system that were not separated by the longer line. Measurement Example

Consider a typical device measurement as shown in Fig. 12. This is a photodiode receiver with a slope responsivity of approximately - 10 dBe (0.316 A/W). Its electrical input match and optical output match both correspond to approximately 14 dB of return loss.

Table II shows estimates of the expected measurement accuracy for the sample receiver. The sources of error identified relate to the accuracy factors discussed. One term or factor not discussed is the measurement calibration error term. This represents the errors introduced during the user calibration; the user calibration step is equivalent to measuring a device with no error-inducing characteristics.

The non-RSS estimate is a classical estimate of overall accuracy obtained by adding the systematic terms linearly to the RSS (root sum square) combination of all of the random terms. Since the accuracy of this system is influenced by numerous sources of error that have similar magnitudes and independent distributions, a better estimate of the typical expected performance is given by the RRS combination of all of the error terms, both systematic and random. In this case, the classical (non-RSS) value of expected accuracy is 1.7 dB and the typical (RSS) value is 0.75 dB.

Fig. 13 shows the measurement repeatability of a single lightwave receiver measured on ten different HP 8703A systems. There are two aspects that should be discussed: (1) the offset repeatability observed, which relates to the difference in the device's slope responsivity (i.e., conversion efficiency), and (2) the modulation frequency response flatness differences observed. In some cases measurements on any two systems agree to within tenths of a dB. The ten measurements all fall within about a 2-dB band over the frequency range.

A comparison of the two extreme measurements is shown in Fig. 14. The minimum trace was stored in memory and the maximum trace in the data register. A Data/Memory function was performed on the two traces to show the fine-grain agreement. The offset difference between the two measurements is about 2.0 dB, which means that the low-frequency slope responsivity for the two extreme cases varied between - 21.5 dB and - 23.5 dB. The frequency flatness difference between these two extreme measurements is about (+ or -) 0.5 dB if the fine-grain ripple or noise is neglected. If the finegrain ripple is included in the measurement, the frequency flatness is no longer (+ or -) 0.5 dB but almost (+ or -)1.5 dB. The multiple measurements represent a sampling of the distribution of potential random measurement error.

In the accuracy calculations using the non-RSS and RSS methods, the RSS method yielded a typical flatness uncertainty of (+ or -) 0.6 dB, which agrees well with the measurement results when the fine-grain ripple is ignored, and the nonRSS method yielded a flatness uncertainty of (+ or -) 1.7 dB, which agrees well when the fine-grain ripple is included.

Acknowledgments

Contributions to the calibration and verification process came from multiple HP entities, including HP Laboratories, the Signal Analysis Division, the Microwave Technology Division, and the Network Measurements Division. In particular, Paul Hernday and Roger Wong from the Network Measurements Division were the architects of the measurement system and calibration concepts. Dave McQuate and Roger Jungerman from the Microwave Technology Division provided the device modeling to substantiate the calibration technique. Evelyn james from the Network Measurements Division developed the test system software to implement the factory calibration.

References

1. R. Wong, P. Hernday, M. Hart, and G. Conrad, "High-Speed Lightwave Component Analysis," Hewlett-Packard journal, June 1989, Vol. 40, no. 3, pp. 35-51.

2. R. Wong, P. Hernday, and M. Hart, "Technical Innovations for the Design of a New 20-GHz Lightwave Component Analyzer," Hewlett-Packard RF and Microwave and Lightwave Measurement Symposium, October 1989.

3. D. Harkins, P. Hernday, and R. Wong, "Accuracy Considerations and Error Correction Techniques for 20-GHz Lightwave Component Analysis," Hewlett-Packard RF and Microwave and Lightwave Symposium, July 1990.

4. C. Hentschel and I. Muller, "Single-Mode Fiber Optic Connector Technology and Performance," Hewlett-Packard RF and Microwave and Lightwave Symposium, August 1988.

COPYRIGHT 1991 Hewlett Packard Company
COPYRIGHT 2004 Gale Group