The initial thought for the SPC limits was to use measurement uncertainty (MU) values of ± 0.1 dB. However, because the process effectively repeated the measurement of the RF path loss carried out during the calibration of the test system, an analysis of the original measurement equation indicated that a number of the uncertainty contributors cancel each other. A revised analysis indicated that limits of ± 0.05 dB were appropriate.

Figure 1 - Example Control Chart

As we discussed earlier, a variety of implementation problems beyond those anticipated seriously impacted the performance of the fully automated Power Test System. As soon as the methodology was introduced, an SPC failure occurred almost daily. As suspected, that confirmed that the Power Test System within the Yellowstone environment was out of control.

Our investigation revealed several causes. Let's list them, with some explanations.

1. The primary-system RF cable is attached to the DUT (device under test), or Gold Standard, using a metal plate that houses a metrology-grade push-fit N-type connector (mobile device). The robot uses an electromagnet to pick up and position the plate and then maintain the connection during measurement.

   We found that heat transfer was taking place from the plate through a Delron connector mount, then through the N-type connector itself, and finally to the Model 8482A's internal thermocouple sensor.

   When calibrating the system, DUT, or Gold Standard, this caused a temperature-related measurement drift, inducing an offset in the results. To overcome this, 12 V fans were fitted to the electromagnets.

2. Two additional power-sensor connections (Sensors A and B) to the N-type bulkhead connectors mounted on the test system were replaced with mobile devices held in position during test by passive magnets. The 8 lb force of connection with a spring-loaded N-type connector was insufficient to ensure good repeatability.

   This arrangement was replaced with the two power sensors permanently located in the test system. With switched paths for measurement and sensor zeroing (with a 50 W termination) or sensor calibration (reference source), there's no disconnection of the power sensor. As such, the nonrepeatability error is only that of the matrix RF switch.

3. Test operators initially performed a manual calibration of the Gold Standard 8482A power sensor to the 1 mW reference. This was subsequently replaced with robot-assisted calibration. It was considerably more repeatable and consistent.

4. The high-power RF amplifiers used in the Power Test System were found to be generating DC offsets when not in-circuit. This became more apparent when an earlier intermittent problem of sensor damage wasn't eliminated within the robot-controlled Yellowstone environment. Additional terminated switch paths were added to prevent destructive discharges.

5. With improving consistency and performance of the automated system, the effects of RF switch nonrepeatability, connector quality, and maintenance and cleanliness issues became more apparent. Component changes—previously deemed insignificant—were carried out, and the maintenance schedule was modified.

As problems were identified, fixes were rolled out across a large number of Power Test Systems, both within and outside Yellowstone. The performance of the systems, monitored through daily SPC runs, increased to a previously unrealizable level. Specific improvements included:

1. A yield increase from around 50% to 90% for the product tested at the Power Test System.

2. A reduction in the number of test systems from 11 to 7. That was an effective saving of 30% in capital expenditure.

3. MU was reduced by around 20%, to ± 0.072 dB (± 0.082 dB below -27 dBm).

4. The manufacturing tolerance interval was reduced by around 20% (i.e., the test limits were tightened), in line with demands for similar reductions in the customer specification.

5. The calibration intervals were now driven by SPC, with the drift in accuracy monitored through daily SPC runs. The interval is currently one month with no SPC failures. Limited experimentation indicates that an interval of eight weeks, and possibly longer, may be achievable.

6. By being able to effectively separate the performance of the test system and the product (the DUT), the number of "no fault found" conditions decreased dramatically. Passing the exacting demands of the daily SPC run meant that any DUT test failure could be confidently attributed to the product rather than to the test system.

7. Manufacturing and production resources were used more effectively.

Figure 2 - It's important to correlate problems and their resolution with system performance.

The SPC philosophy has helped identify problems associated with automating the Power Test System. Subsequently, it has provided a means of keeping the test process in control with a high degree of confidence.

This process can be applied to any test system to monitor the performance over a period of time relative to a known-good operating condition. It's the relative "drift" in the measurements and the growth in uncertainty from that point that's under examination. So if the measurement exceeds the SPC limits, then there's either an immediate and (hopefully) identifiable problem, or possible recalibration of a system is required.

Looking Forward
Back to top

Looking forward, if the reported results of the SPC test itself (rather than the calibration of the Gold Standard power meter and sensor) could be traced directly to national standards, it's unlikely that confidence in the production measurements can be further improved upon. In effect, the SPC run is calibrating the test system against a higher standard at frequent intervals through the Gold Standard (transfer device). This method could also be considered to be an inter-laboratory comparison (ILC).

The ILC Comparison
Back to top

The implications of performing a frequent ILC where one body is, for example, a national standards laboratory such as NPL or NIST, and the other is Agilent production, are twofold.

First, extended calibration intervals can be realized for test system equipment because of the increased confidence in measurement through SPC and ILC.

Second, direct traceability of measurements through the Gold Standard can be done, rather than using multiple items of test equipment in a system. This implies that your test systems may not need to leave the production line for scheduled equipment calibration.

ILC Specifics
Back to top

An ILC process is a comparison of calibration undertaken by two or more participating calibration entities. In this example, the two calibration bodies might be Agilent Queensferry production (for the Power Test System), a national standards laboratory, and (using the artifact itself), the Gold Standard unit.

We previously mentioned that one of the criteria for selecting a Gold Standard was repeatability. This is critical to achieving the most accurate comparison.

Also, specified parameters, test conditions, instrument settings, and associated uncertainties first need to be agreed on—this is essential, as the standards lab will almost certainly use different methods and equipment to perform its measurements.

The ability of the laboratory to make measurements at all of the required test points may not be possible in some instances. An ILC is needed whenever the Gold Standard is calibrated—this provides an absolute reference with the least uncertainty at that point in time.

Comparison of Results
Back to top

With ILC measurements completed by both parties, a comparison of the data generated by the Power Test System and the standards laboratory was made. But calculating the difference between these reported values would not in itself produce useful information, unless accompanied by some form of acceptance limits.

ILC Result = Results (Power Test System) – Results (Standards Lab)

These limits are calculated by combining the expanded uncertainties from both parties in some way. In doing this, the performance of both measurement systems can be accounted for in a single term.

A possible method for defining the acceptance limit involves the quadratic sum of the two uncertainties for the measured parameter, together with the Gold Standard's repeatability (refer to the following equation, where MU₁ and MU₂ are the expanded uncertainties for each party and Factor depends on the confidence required).

Note that the value of Factor must be small enough (<1.0) to reflect the fact that the acceptance limits apply to the difference of the two measurements. Additionally, the limits must assure the required confidence in the measurement traceability of the Power Test System.

If the ILC results generated by the multiple Power Test Systems fell within the acceptance limits, we could be confident that each was making absolute measurements correctly. However, if any system produced results outside of the acceptance limits, then there would be a problem with the measurement. The test system would then be put out of commission until corrective action was taken.

The results in Table 3 show the ILC data for the first Power Test System.

Table 3 - Initial ILC Result for Power Test System
(the ILC results indicated by the Difference term may be seen more clearly in the chart of Figure 3)

Freq. (GHz)	Indicated Power (mW)	Standards Lab		SQF (Manual) Power System S868		Difference S868 Power – Stds. Lab (%)	Acceptance Limits (%)
		Power Diff. (Inc – Ind) (%)	MU (%)	Power Diff. (Inc – Ind) (%)	MU (%)
0.05	0.1	0.3	0.5	-0.15	1.7	-0.45	1.2
	1.0	0.2	0.4	0.18	1.7	-0.02	1.2
	10.0	0.0	0.4	0.03	1.7	0.03	1.2
0.95	0.1	1.4	0.6	0.67	1.7	-0.72	1.2
	1.0	1.2	0.5	1.03	1.7	-0.17	1.2
	10.0	1.0	0.5	1.00	1.7	0.00	1.2
1.85	0.1	2.4	0.7	1.64	1.7	-0.74	1.2
	1.0	2.3	0.6	2.00	1.7	-0.29	1.2
	10.0	1.9	0.6	1.88	1.7	-0.01	1.2
2.70	0.1	5.3	0.7	4.54	1.7	-0.72	1.2
	1.0	5.2	0.6	4.71	1.7	-0.47	1.2
	10.0	5.0	0.6	4.64	1.7	-0.34	1.2

The following chart shows that the measurements made by the S868 Power Test System are within 0.74% (0.032 dB) of those carried out by the standards lab. The chart also shows that it may be possible to make further improvements to system performance by investigating what may be a systematic effect present at levels below 10 mW.

Figure 3 - ILC Results

Knowledge Center Main | Application Notes | Briefs | Editorial | Focus On | Guest Column | NetSeminars | Opinion | Resources | Reviews

E-mail This Article

Printer-Friendly Page

		Click here to get your listing up.