RF testing for devices such as amplifiers and RFICs can be tedious work. Such devices work over a wide range of frequencies and power levels, and they must meet specifications over temperature and power-supply ranges. Testing for all of those conditions can generate loads of data. Fortunately, automation can cut test time and help you make sense of all that information.
Just because you have a spectrum analyzer, network analyzer, or power meter with features that can improve testing, you may not be able to use the instrument if you need to maintain compatibility with older models. If you’re testing cutting-edge RF products, your test equipment may not have the necessary dedicated features and you’ll have to develop your own.
Test engineer Bill Drago of L-3 Communications Narda Microwave East supports production of RF amplifiers, downconverters, upconverters, and transceivers that operate in the C band through the Ka Band. These products often remain in production for years, which is why Drago can’t take advantage of features in newer test equipment that can automate many of the measurements he needs to make—he developed his tests before such automation was available, and he needs all of his test instruments to continue to follow identical procedures. Thus, he has written software that performs automated measurements such as amplifier gain, 1-dB compression, IMD (intermodulation distortion), return loss, spurious noise, and noise figure. Drago explained why spurious-noise testing is important. “Downconverters and upconverters mix an input signal with an LO [local oscillator],” he said. “The converter’s LO must be tuned to customer specifications within a certain range and step size. The converter needs a frequency synthesizer that’s programmable with specified steps over its frequency range. The frequency synthesizer can’t introduce any spurs into the converter, so we must test for that.”
To measure spurious noise, Drago uses an Agilent Technologies’ spectrum analyzer to perform a frequency sweep through an approximately 1-GHz band around the carrier. He usually breaks that sweep into a number of smaller sweeps, each consisting of 601 frequency points. Each step might be a few kilohertz wide.
If Drago used one large sweep, its step size and the instrument’s RBW (resolution bandwidth) would be too wide and might miss a spur. He adjusts sweep span and RBW so that spurs don’t fall between the points in a sweep.
Using a number of smaller sweeps is also faster than a single sweep for the RBW that Drago needs. He noted that a single sweep could take as long as an hour, depending on RBW and frequency. Furthermore, smaller sweeps can reveal failures sooner than waiting for a large sweep to complete.
Although some of Drago’s spectrum analyzers have built-in test applications for measuring spurious noise, he doesn’t use them, because not all of his spectrum analyzers have that function. If he were to use that feature, he might not have a replacement instrument for the production line should that instrument fail. Instead, he wrote his own applications, and by keeping the test applications outside the instrument, he can use any spectrum analyzer that’s available.
Drago has written several other test programs, including one that measures an amplifier’s 1-dB compression point. This test is built into some, but not all, of his VNAs (vector network analyzers). The algorithm uses a binary-search process, similar to that used by a successive-approximation analog-to-digital converter. He starts with an input signal from an Agilent RF signal generator that’s the highest possible value for the amplifier under test. He then measures output power with an Agilent RF power meter. If the output signal is compressed by more than 1 dB, he cuts the input signal in half, then increases it or decreases it by half of that value until he finds the 1-dB compression point (Figure 1).
Finding harmonics Michael Ford is a test engineer at Comtech PST, a manufacturer of RF amplifiers that operate from 500 MHz to 6 GHz at power levels from 100 W to 10 kW. Ford’s typical test station contains an RF signal generator, a spectrum analyzer, a network analyzer, a power meter, and RF switches (Figure 2). A USB digital I/O module from Measurement Computing controls the RF switches. The amplifiers mount on an environmental plate that changes their temperature. The test station measures gain, output power, harmonic distortion, IMD, efficiency, spurious noise, and harmonics.
On occasion, Ford will use an instrument’s built-in functions. For example, he might measure spurious noise with an Agilent spectrum analyzer’s built-in application. But he also writes his own applications to make those measurements if his instrument doesn’t have that feature.
Ford supports both engineering and production. When making measurements for engineering evaluations, he uses built-in functions such as spurious noise and harmonic analysis. For production, Ford always uses his own software routines.
“We write our own routines using a modular format because it lets us use equipment from multiple vendors such as Agilent or Rohde & Schwarz. We use the same test routines and just change instrument command libraries.” For example, his routine for spurious noise measurements works with spectrum analyzers from either manufacturer.
Ford’s routines for measuring harmonics of a carrier frequency use variables for parameters such as center frequency, span, RBW, and VBW (video bandwidth). After receiving those parameters, the routine runs sweeps at multiples of the carrier frequency to find the power of its harmonics. The results go to a spreadsheet for analysis.
Multiple cellular networks While Drago and Ford must support products that remain in production for years, engineers developing tests for RFICs face different problems and have different reasons for not always using an instrument’s automation features.
Joe Flynn is a staff engineer at fabless semiconductor company Sequoia Communications. He evaluates RFICs that support several cellular wireless standards (Ref. 1). He has developed several test stations for evaluating the devices. “We characterize transceivers for gain, noise figure, IMD, cross modulation, and EVM [error-vector magnitude].” Flynn also develops his own automation tools, but measurement speed was his critical feature in choosing equipment.
Because RFICs transmit and receive modulated signals, Flynn’s test stations include spectrum analyzers for characterizing frequency content, and they include modulation analyzers for characterizing modulated content. Figure 3 shows a simplified diagram of a system that tests the receivers in Sequoia’s ICs. The receiver test bench lets Flynn evaluate how a receiver performs in the presence of undesired blocking signals such as those from simulated cellular base stations, other cellphones, and broadcast radio stations.
As part of the blocking-signal test, Flynn must measure SNR (signal-to-noise ratio) over a frequency range of 100 kHz to 12.7 GHz in 200-kHz steps. That’s approximately 60,000 measurements per channel, and the RFIC has 1300 channels over seven frequency bands. An SNR test generates loads of measurement data, and it’s just one of many tests that Flynn must run on a preproduction lot of parts.
To help analyze the data, Flynn developed a data converter that produces data plots. The tool lets him use production test analysis tools to view engineering data. For example, he might want to see the distribution of parameters such as gain, return loss, and current consumption across the 100 parts in a preproduction run. “When we look at the data, we get a feel for how production parts will behave,” he said.
Sequoia also developed an in-house Visual Basic Web-based tool that manipulates the bench-characterization data. The tool uses .netCharting software (dotnetcharting.com) to create some 1200 data plots on the parts. “When we have a test review, the tool lets us find any out-of-spec measurements.” The tool lets him select data parameters to plot and refine the data by selecting certain test conditions. For example, it lets him look at receiver gain and noise figure versus temperature or power-supply voltage.
Because he must make so many measurements, speed was key for Flynn when he selected a spectrum analyzer and developed a test method. To cut measurement time, he uses two Agilent MXA spectrum analyzers, one each for a receiver’s I and Q channels. The instruments are frequency and phase locked, which synchronizes the measurements. He then had to optimize RBW and sweep time to minimize test time. “When I started, a single SNR measurement on one channel at one blocker frequency took about a second. Now, I make each measurement in 18ms.”
Flynn also learned how to minimize test time by using both LAN and GPIB communications. Most instruments in the test system use GPIB because, according to Flynn, “You can’t beat GPIB when sending a series of short commands. The overhead needed to use Ethernet is only apparent when transferring large blocks of data.” Thus, he uses Ethernet for the logic analyzer, which collects data on digital baseband signals. He also uses a dedicated Gigabit Ethernet LAN card to avoid packet collisions between the test equipment and the corporate network. He uses three GPIB cards to cover all of the test equipment.
Mix and match CSR is a fabless semiconductor company that develops wireless communication RFICs for personal area networks such as Bluetooth and WiFi. James Blackwell heads the company’s applications engineering group. He helps customers evaluate CSR’s RFICs and develop products based on the company’s devices.
CSR engineers use a mixture of in-house and purchased test equipment, usually starting with in-house testers. “Because we’re often on the leading edge,” said Blackwell, “we have to develop our own test suites until the test-equipment companies catch up.”
One example is a Bluetooth tester for performing loopback tests. Blackwell noted that the Bluetooth specification defines a loopback test where the tester wirelessly controls the DUT (device under test). With loopback testing, a manufacturer can test a Bluetooth product wirelessly as it moves along a production line.
When CSR’s engineers developed a Bluetooth RFIC several years ago, engineers had to build their own RF test system using RF signal generators, spectrum analyzers, and vector signal analyzers. They wrote Matlab scripts to control the instruments, process the data, and produce test reports.
In a loopback test, the tester will command the DUT to produce a 2.405-GHz tone with a specified modulation signal, for example. The tester will measure as many as 20 parameters such as transmitted RF power (peak and average). It will also perform receiver tests such as sensitivity and bit-error rate.
Some Bluetooth devices transmit at controlled power levels, so the DUT must operate in steps in its power table while the tester measures differences in power. It will also measure frequency tolerance and drift. A test of the frequency response of the DUT’s modulation filter uses 10101010 and 11110000 bit patterns.
As time went on, test-equipment makers developed Bluetooth testers, and CSR was able to use them. Today, the company’s engineers use Bluetooth testers from Rohde & Schwarz, Agilent, and Anritsu. CSR has all three instruments, so the engineers always have one that their customers use. That’s crucial when an engineer needs to reproduce a customer’s test.
CSR engineers didn’t switch to a dedicated Bluetooth tester immediately. “We work with the test-equipment manufacturers to develop test applications for their equipment,” Blackwell explained, “but sometimes we must wait for a second or third generation of a tester before we can use it. Even after we adopt a commercially available tester, we may still use our own test suites for certain tests.”
Blackwell noted that dedicated Bluetooth testers may perform some tests faster or more accurately than CSR’s in-house testers, but the company’s engineers still use the in-house tester when they feel that it performs the tests better than a dedicated tester can.
Test equipment manufacturers respondToday’s RF test equipment has many features that automate tests, but the engineers interviewed for this article don’t often use them, at least not in production. Bill Drago of L-3 Communications Narda Microwave East and Michael Ford of Comtech PST both noted that they need flexibility and thus prefer to write their own test algorithms. CSR’s James Blackwell had to write his own test software because the automated tests he needs often aren’t available in test equipment. And Sequoia Communications’ Joe Flynn chose his spectrum analyzer based on raw speed, not on its features.
Yet, test equipment has automation features for a reason. To find out what measurements the features in test equipment can help you automate, I spoke with representatives of Agilent Technologies, Giga-tronics, Keithley Instruments, Rohde & Schwarz, and Tektronix.
Industry standards for wireless products often define tests, and instruments such as spectrum analyzers have options for the spectral masks required by these tests. But these features are often used in design evaluations rather than in production. ”Production tests give manufacturers confidence that each unit works,” said Darren McCarthy, product manager for spectrum analyzers at Tektronix. “The features built into test equipment are more for conformance testing, performed during engineering evaluations.”
RF test instruments often must generate or capture signals across one or more frequency bands. Some instruments let you download a frequency table into the instrument, then call the table with one command. “Using list mode, an engineer can preload up to 4000 parameters into our frequency synthesizers,” said Malcolm Levy, director of sales and marketing at Giga-tronics. “You can program the dwell time at each frequency.”
Rohde & Schwarz network analyzers have eight ports that operate without switching, which cuts test time. The instruments also have a user-control port. Product manager for network analyzers Yassen Mikailov noted that this rear-panel port lets you run custom test sequences stored in the instrument, triggered by external events. The company’s spectrum analyzers also have predefined measurement sequences such as ACP (adjacent channel power). “One GPIB command lets you measure power in the main channel and one or two adjacent channels, said Mikailov.”
Bob Nelson, applications engineer at Agilent Technologies, observed, “Some tests, such as EVM [error vector magnitude] for WiMax devices, are difficult or impossible to do without application software. Customers often choose the equipment that has the applications they need.”
Nelson also noted that measurement time is important. In ”RF engineers automate tests,” Flynn explained that speed was the most important parameter for his choosing a spectrum analyzer. Nelson explained that built-in functions let an instrument take measurements at multiple frequencies, then send all the data to a host PC at once. “You don’t need to send multiple commands to get test data,” he said. “In addition, instruments know what power levels they need for specific standards. You don’t have to program them.”
Mike Millaem, principal applications engineer at Keithley Instruments, supports customer measurement applications that range from basic data acquisition to fully automated measurement and analysis. For example, some customers develop waveforms with Matlab, then download them into test equipment for playback, and then analyze their measurements in Matlab. Engineers such as CSR’s Blackwell do just that, creating their own tests with Matlab.
Programming compatibility among instruments is also important, especially to engineers who must support a product line for many years, as Drago and Ford must do. They have, on occasion, run into instances where they have had to change command libaries when they change instruments. Agilent’s Nelson acknowledged that issue by saying “The functionality of test equipment changes over time based on customer requirements. Code compatibility should be better in the future.”
About the AuthorMartin Rowe is a Senior Technical Editor fo Test & Measurement World.
ReferencesRowe, Martin, “The RFIC evaluator,” Test & Measurement World, March 2009, p. 9.
Click here for the illustrations:
, ,
Caption
Figure 1: A binary-search algorithm finds an amplifier’s 1-dB compression point. Courtesy of L-3 Communications.
Figure 2: A typical automated tester for RF amplifiers uses a power meter, spectrum analyzer, network analyzer, and signal generator. Courtesy of Comtech PST.
Figure 3: An RFIC test station for testing receivers uses two GPIB cards and a LAN to communicate with instruments. Courtesy of Sequoia Communications.
