What makes RF test so fast with PXI?
16 January 2009
If you’ve ever considered using a PXI RF measurement system, you may have wondered: what makes the measurements faster? Is it the data bus? Is it the CPU? Is it the instantaneous bandwidth of the RF analyser? While each of these factors can affect overall measurement time, the CPU has the biggest impact, as David Hall investigates.
There is just something about speed that excites us and captures our attention. For engineers involved in RF validation or production test, measurement speed is actually just as practical as it is exciting. Facing tougher time-to-market requirements and handling higher production volumes, the task of making automated test systems faster is often one of the most critical challenges for a test engineer to solve.
If you’ve ever considered using a PXI RF measurement system, you may have wondered: what makes the measurements faster? Is it the data bus? Is it the CPU? Is it the instantaneous bandwidth of the RF analyser? While each of these factors can affect overall measurement time, the CPU has the biggest impact on overall measurement time. Now lets investigate the contributing factors to overall measurement time and evaluate the effect of the CPU on our results.
Timeline of a Spectrum Measurement
When testing RF devices, the spectrum measurement is used as a basis for a variety of tests including: spectrum mask (WLAN and ZigBee), adjacent channel leakage ratio (WCDMA), adjacent channel power, and intermodulation distortion (Power Amplifiers). In a PXI Vector Signal Analyser (VSA), measurement time can be divided into 1) acquisition time, 2) data transfer to the processing unit (CPU), and 3) CPU processing time. In order to determine which portion takes the longest, we can theoretically calculate the times for the first two steps and compare it to the overall measurement time.
Signal Acquisition Time
For any spectrum measurement made with a vector signal analyser, acquisition time is inversely proportional to the Resolution Bandwidth (RBW) of the measurement. Given the RBW, you can roughly determine the acquisition time (assuming no windowing) from the equation shown:
Generally, each wireless standard will define a specific RBW that should be used for spectrum measurements. For example, W-CDMA Adjacent Channel Leakage Ratio (ACLR) measurements require a RBW of 30 kHz. By contrast, the Wireless LAN (WLAN) spectrum mask requires a RBW of 100 kHz.
Data Transfer Time
Next, we can estimate the time required to transfer data from the analogue-to-digital converter (ADC) to the CPU. As a general rule, we must sample an RF signal at approximately 1.25 times the desired span. For a 50 MHz span, the sample rate should be 62.5 MS/s.
Given our acquisition time and sample rate, we can easily calculate the acquisition size in bytes. Knowing the size of the acquisition, we can finally determine the time required to transfer data from the digitiser (ADC) to the CPU. Research data has shown that the typical latency for instruments on the PCI and PCI Express data bus is less than 0.7 µs. Moreover, typical sustained transfer rates using highly optimised instrument drivers can be as high as 100 MB/s for PCI instruments and 800 MB/s for PCI Express instruments. Using a PXI Express RF vector signal analyser, we can therefore estimate the total data transfer time.
We find that the acquisition and waveform transfer time for a 50 MHz spectrum measurement (100 kHz RBW) is just under 14 µs. However, latency and data transfer time are only part of the story for RF spectrum measurements. On a traditional box instrument, the same measurement can be performed in about 100ms. Therefore, even when using higher latency data buses such as LXI/LAN (1 ms latency), measurement time is dominated by processor performance - and not bus latency. Through experimentation, we can better understand the effect different processors have on overall measurement performance.
Evaluating Overall Measurement Time
Comparing the measurement time of a traditional spectrum analyser, to a PXI-based system can be achieved by taking spectrum measurements with different PXI embedded controllers. By understanding the impact of a CPU on overall measurement time, we can better understand potential bottlenecks to achieving faster measurement speeds. In this test, 50 MHz spectrum measurements (100 kHz RBW) were made using four PXI Express controllers – three with Intel-based CPUs and one with an AMD Processor.
• PXIe-8103 Embedded Controller (2.0 GHz Pentium M 760 processor)
• PXIe-8105 Embedded Controller(2.0 GHz dual-core Intel Core Duo processor T2500)
• PXIe-8106 Embedded Controller (2.16 GHz Intel Core 2 Duo T7400 dual-core processor)
• PXIe-8130 Embedded Controller (2.3 GHz dual-core AMD Turion 64 X2 processor)
To make our experiment more valuable, we can also compare the effect of resolution bandwidth on measurement time. Theoretically, measurements with a finer RBW should take longer given the larger acquisition, transfer, and processing times. As RBW widens, acquisition and transfer times reduce significantly, revealing the time due to software overhead.
From the results, we observe that the PXIe-8106 controller (the fastest Intel model) produces the fastest measurement times in all cases. At a 100 kHz RBW, we observe that a 50 MHz span measurement takes only 2.1 ms, compared to 3.3 ms for the PXIe-8130 (AMD) controller. Not only do these results show that PXI instruments can perform many measurements significantly faster than traditional boxes, but they also show that CPU time is clearly the biggest contributor to overall measurement time.
With engineers facing increasing pressure to improve measurement throughput, continued attention must be focused on reducing measurement time. Fortunately, engineers using PXI measurement systems are already ahead of the curve. Because software-defined instrumentation allows the user to select a CPU, measurement times can be reduced simply by using a faster CPU. Today, PXI RF measurement systems are already performing many RF measurements significantly faster than traditional boxes. Whats more important however, is that PXI systems deployed today will get increasingly faster over time with the simple step of upgrading the CPU.
David Hall is Product Manager at National Instruments.
Contact Details and Archive...
Related Articles...
Most Viewed Articles...