22 Marzo 2012 IFEMA, Madrid spain.ni.com/nidays.

Transcription

1 22 Marzo 2012 IFEMA, Madrid spain.ni.com/nidays

2 The Art of Benchmarking Speed PXI Versus Rack-and-Stack Test Equipment Filippo Persia Systems Engineer Automated Test Mediterranean Region National Instruments

3 Agenda Timeline of a Factors Affecting Speed Initialization, setup, and measurement personalities Accurately measuring measurement speed Tips and Tradeoffs Processor, traces, and composite measurements Repeatability versus averaging Benchmark: NI PXIe-5665 and NI PXIe-5663 versus Agilent PXA and MXA Final Comparison Results

4 The Traditional Spectrum Analyzer Uses swept-tune approach Automated through GPIB The Traditional Spectrum Analyzer Attenuator Mixer IF Amp Detector Video BW Digitizer Display Local Oscillator (LO)

5 The Modern RF Signal Analyzer Uses digital IF approach Fundamentally software defined PXI RF Signal Analyzer Attenuator Mixer IF Amp IF Filter Digitizer Display CPU Local Oscillator Result

6 Does Speed Matter? Automated s: YES! In characterization speed gates time to market In production speed can gate throughput Interactive s: Somewhat When times exceed 100 ms, speed affects productivity Most modulation measurements performed in <100 ms speed affected by many variables Configure Acquire Data Execute Algorithm Return Result

7 Optimizing and Measuring Test Time

8 Timeline of an Automated Configure Acquire Data Execute Algorithm Return Result Instrument Settings Frequency and attenuation changes incur settling times Tools such as list modes enable faster configuration Settings PXI: All measurement settings can usually be set in <1 ms Rack-and-stack: LAN latency is 1 ms

9 Timeline of an Automated Configure Acquire Data Execute Algorithm Return Result Acquisition time depends on measurement type EVM: Acquisition time must be longer than burst Spectral measurements: Acquisition mostly depends on measurement settings (IF bandwidth, FFT mode, and so on) Averaging requires multiple acquisitions

10 Timeline of an Automated Configure Acquire Data Execute Algorithm Return Result Algorithm execution speed depends on: Horsepower of processor (faster CPU = faster measurements) Existence of other processes (measurements and traces) Complexity of signal (OFDM, MIMO, and so on) Dominates total measurement time

11 Timeline of an Automated Configure Acquire Data Execute Algorithm Return Result Rack-and-stack: results transferred via GPIB, LAN, and so on Require both a data bus (internal) and a result bus (external) PXI: results are computed onboard Instrument only uses a data bus Result transfers driven by bus latency Typical LAN latency 1 ms

12 Accurately Measuring Time For Loop: 100 to 1,000 iterations Configure Acquire Data Execute Algorithm Return Result Get Timestamp Get Timestamp Configure times can be measured independently Acquire to Return Result generally measured together Software timestamps have inherent jitter Compute mean time of 100 1,000 measurements Use Read versus Fetch command to ensure new data results

13 4 typical Configurations Comparison: PXI and Traditional PXIe PXIe-5663 Agilent PXA N9030A-526 Controller PXIe 8133 controller Integrated CPU Analysis Bandwidth Options Installed Quad Core Intel i7 Q GHz 4 cores Intel Core 2 Duo T2500 2GHz 2 cores Agilent MXA 50 MHz Option B40: 40 Mhz 25 MHz GSM/EDGE WCDMA/HSPA+ LTE a/b/g/n GSM/EDGE WCDMA/HSPA+ LTE GSM/EDGE* WCDMA* LTE a/b/g*

14 Best Practices for Benchmarking Evaluate and adjust appropriate measurement settings Default settings do not always yield fastest measurements Minimize acquisition time Turn auto detection off Select fast spectrum modes Set appropriate number of averages Not all instruments use same default settings Turn off traces and front panel displays Use composite measurements for fastest measurements All results computed on same set of I/Q data Introduces dynamic range tradeoff for spectrum measurements

15 Benchmark: WCDMA Composite Signal characteristics WCDMA = Wideband Code Division Multiple Access Modulation Type: QPSK (5 MHz Channel) Signal Source: NI PXIe-5673E RF Signal Generator Composite measurement characteristics EVM-QPSK EVM = Error Vector Magnitude Simpler modulation quality metric (versus modulation accuracy) ACLR = Adjacent Channel Leakage Ratio ACPR = Adjacent Channel Power Ratio Best metric of dynamic range

16 DEMO- Agilent PXA vs PXIe 5665 ACPR FlexRIO Co-Processor

17 DEMO- Agilent PXA vs PXIe 5665 ACPR Averaging ON Averaging OFF Agilent PXA NI 5665 Time (ms) ACPR (dbc) Time (ms) ACPR (dbc) Fast ACPR IBW NA -81 Fast ACPR IBW NA -81

18 NI PXIe-5663 Versus Agilent MXA (WCDMA) Accuracy Agilent MXA NI PXIe-5663E Difference EVM-QPSK 0.40% 0.32% 0.08% EVM Repeatability (1 Slot) 0.02% 0.01% 0.01% ACLR dbc dbc 3.8 db Speed Agilent MXA NI PXIe-5663E Difference WCDMA Composite: 1 Time Slot 67.3 ms 26.5 ms 2.5X WCDMA Composite: 3 Time Slots ms 65.8 ms 1.9X

19 Additional Major Test Time Factors Tuning and Settling Times Example: change center frequency from 1 to 2 GHz Dependent on synthesizer technology used personality load time Example: switching from GSM/EDGE to WCDMA Large factor in multistandard testing settings configuration Usually is a function of bus latency Best to minimize number of bus transfers Parallel testing implementation

20 Understanding PLL Loop Bandwidth (BW) RF signal analyzers use 2 types of LOs Voltage Controlled Oscillator (VCO) Yittrium Iron Garnet (YIG) Loop bandwidth affects settling time on VCOs Wide BW: faster settling and degraded phase noise Narrow BW: longer settling and improved phase noise Phase Detector Filter VCO 10 MHz Reference φ LO N Divider

21 PLL Loop BW Versus Phase Noise (NI PXIe-5663E) Integrated Phase Noise = dbc Integrated Phase Noise = dbc

22 Tuning Time PXIe 5665 Mode Tuning Speed (1 GHz Step) Frequency Accuracy Phase Noise (10 KHz offset) Normal 9.3 ms 0.1 ppm -134 dbc/hz Fast 2.3 ms 1 ppm -130 dbc/hz

23 RF List Mode

24 Tuning Time: NI PXIe-5663E Versus Agilent MXA graph Band change at 3.6 GHz incurs longer tune and transfer time NI PXIe-5663E tuning time <500 µs when using wide-loop BW

25 DEMO PXIe 5665 List Mode Test

26 DEMO PXIe 5665 List Mode Test Time for List Mode + Peak Detect Test Agilent PXA Time (ms) NI 5665 Time (ms)

27 Other Benchmark Data Wireless LAN (802.11g) Signal Type NI PXIe-5663E Time Agilent MXA Time* Speedup a/g: EVM and Power 7.0 ms 93.9 ms 13.4X *WLAN s performed using N9077A-XFP, single acquisition composite measurement application (fast mode). LTE (3GPP Long Term Evolution) Signal Type NI PXIe-5663E Time** Agilent MXA Time** Speedup 5 MHz BW: EVM and Power 71 ms 541 ms 7.6X 10 MHz BW: EVM and Power 92 ms 703 ms 7.7X ** All measurements made on a single subframe with auto detection turned OFF

28 Best VSA Comparison NI PXIe-5665 Agilent PXA R&S FSU Base Price (14 GHz) $50,000 $70,000 $80,000 Typical Phase Noise dbc/hz dbc/hz dbc/hz Noise Floor (no pre-amp) -154 dbm/hz dbm/hz dbm/hz Third Order Intercept +24 dbm +20 dbm +25 dbm ACLR WCDMA -87 dbc - 89 dbc - 87 dbc EVM WCDMA 0.4% 0.5% 0.5%

29 What all this means Test Times

30 Summary speed is tricky to benchmark Requires careful attention to measurement settings Many instrument settings affect speed (traces and more) Many tradeoffs: speed versus accuracy versus repeatability PXI designed for extremely fast measurements Speed is a function of CPU performance (multicore) Usually 3X to 20X faster than rack-and-stack NI PXIe-5663E has accuracy similar to MXA NI PXIe-5665E has accuracy similar to PXA