Manufacturing and testing of devices at microwave and mmWave frequencies have many inherent challenges. These devices have traditionally been built in small quantities and sold for high average selling prices. The next area of communications growth, however, is expanding into frequencies beyond 50 GHz and will require greater optimization and efficiency of manufacturing operations for high accuracy and measurement speed to meet cost and volume objectives.
Uncertainties in the signal path that were insignificant at lower frequencies, now become dominant factors with up to 20× the magnitudes (see Table 1). Some examples include: socket and/or probe card losses and repeatability, connection cable losses, mismatch errors, connector repeatability, modulation accuracy and repeatability, signal dynamic range and low signal measurement sensitivity. To put this into perspective, the estimated total path loss at 2 GHz is 0.55 to 0.75 dB including 0.2 dB peak-to-peak mismatch ripple:
When making a return loss (RL) measurement the round-trip loss must be overcome and removed to calculate the actual RL of the device under test. At 2 GHz, that could be as much as 1.5 dB, with 0.75 dB to the device and 0.75 dB reflected. A well-matched device and test setup will typically have a 25 dB RL or better.
Contrast this with the estimated total path loss at 50 GHz. It is significantly larger at 6.7 dB to 7.5 dB, including 0.8 dB mismatch ripple:
When making RL measurements, the round-trip loss at 50 GHz could be as much as 15 dB, 20× higher than at 2 GHz. That must be overcome and removed to de-embed the actual RL of the device.
A typical well-matched device and test setup at 50 GHz has 20 dB RL or better. When the median 14 dB round-trip loss is added, however, a dead short at the device will look like 14 dB RL, which is equivalent to a VSWR of 1.5. The variance in measurements over frequency will be almost as high as 1 dB.
A signal into a well-matched device of 18 dB RL will first see 7 dB loss, then the device’s 18 dB RL, then 7 dB loss on the return trip, returning a signal that is 32 dB down from the original test signal level. This reduces the effective measurement dynamic range by about 15 dB. In linear terms, this means that the measurement system must measure a 39.8× smaller voltage signal and still maintain the usable accuracy and repeatability required to test the device.
Losses have a logarithmic effect on the ability to make accurate and repeatable measurements. If the total losses to the device were only 2 to 3 dB greater, the round-trip loss in the system would approach 20 dB. In linear voltage terms that is 100× smaller, which approaches the limit of what can be error corrected.
Since loss in the device connection path has a logarithmic effect on the ability to make accurate and repeatable measurements within reasonable testing times, the biggest impact and the first place to direct improvement efforts is in reducing loss in the socket and connection.
This example illustrates why scalar measurements are not sufficient when testing at 50 GHz and why vector measurement techniques are required. Vector measurements are more precise when used with de-embedding algorithms to characterize and compensate for physical parasitic elements present at 50 GHz; however, there are two significant challenges when using vector de-embedding:
1. With losses between the measurement instrument and the device approaching 7 to 10 dB, the error vectors are often greater than the device’s measurement vectors. While it is possible to remove large error vectors to see the smaller device vectors, it requires a measurement system that is repeatable.
2. Since most of the communications devices in this frequency range are also frequency translation devices, the fact that there are different input, local oscillator (LO) and output frequencies further complicates the measurement process when attempting to use traditional S-parameter testing methods to achieve these accurate results.
Either one of these challenges would make bringing these devices into production difficult but combining them could make it seem insurmountable.
Much work has been done to address these issues by embracing new methods to determine the linearity and distortion introduced by the devices that meets optimization and efficiency goals for building cost-competitive mmWave devices at high volumes. Some of these alternate methodologies are:
1. Power measurements with vector error correction to overcome mismatch losses inherent in typical scalar power measurement methods using traditional spectrum analyzers and power meters.
2. Generating multiple signals and combining these sources with mmWave power amplifiers for enough power to overcome high losses and still cause the devices to produce measurable nonlinearity effects for intermodulation distortion (IMD), second order intercept point (IP2) and third order intercept point (IP3) measurements.
3. S-parameter measurements with indirect phase reference methodology that uses multiple one-port models at different input and output frequencies with accurate calibration processes to characterize and de-embed the measurements.
4. Using vector voltage averaging to measure signals below the noise floor to increase measurement sensitivity.
5. Model-based distortion measurements to overcome modulation inaccuracies at millimeter frequencies for accurate adjacent channel power ratio (ACPR) and error vector magnitude (EVM) measurements with direct amplitude (AM-to-AM) and amplitude frequency (AM-to-FM) measurements for providing accurate phase and amplitude information to the design team for model and process compensation and alignment.
