Modulated waveform (COFDM)
Most bench setups use a modulation technique that resembles the real-world application, such as one containing several packets or frames of data. Due to test-time constraints and memory constraints of the hardware, sourcing and measuring this data at ATE rates is unrealistic. A full frame of data for this application would represent 96 ms of transmission time and require 12 Mbytes of memory. Therefore, we decided to use two symbols of data for the modulating signal. That equates to approximately 320 kbytes of memory, which is still quite a large capture array for ATE. We selected two symbols with the greatest PAR (peak-to-average ratio) in order to detect the greatest effect on the device.
The RF generator needs to be adjusted based on the PAR of the signal in order to achieve the required output power. Figure 1 shows the constellation of the two DQPSK (differential quadrature phase-shift keying) symbols that were extracted from the bench waveform.
FIGURE 1. This DQPSK constellation diagram shows all of the constellation points that were extracted from the bench waveform.
Setting the generator to the same specified power level for the bench ACR test setup makes it possible to measure the power density in the specified bandwidth in terms of dBm/Hz at the center frequency of the DUT (device under test). The RX design in this example demonstrated a power density of approximately –156 dBm/Hz in a 1.3-MHz bandwidth for a device with an ACR of 35 dB. Figure 2 shows a typical OFDM (orthogonal frequency domain multiplexing) modulated spectrum from which the power was measured.
FIGURE 2. The OFDM receiver under test demonstrated typical in-band power levels for the wanted signal.
Keeping the device set to the same channel, the tester must shift the RF generator frequency to the adjacent channel frequency and set the power to the expected ACR plus the original input power. The tester then needs to measure the power density again in the same specified bandwidth—that is, dBm/Hz at the center frequency of the device.
In this example, the tester measured a power density of approximately –169 dBm/Hz in a 1.3-MHz bandwidth for a device with ACR of 35 dB. Figure 3 shows the power from the adjacent channel breaking into the “wanted” band.
FIGURE 3. Unwanted signals from adjacent channels break into the desired band.
The ACR measurement is the difference between the wanted channel and the adjacent channel measurements (Figure 4). Subtracting the two results reveals the SNR: –156 dBm/Hz – (–169 dBm/Hz) = 13 dBm/Hz.
FIGURE 4. The ACR measurement is the difference between the wanted power and the adjacent break-through power
After running the test 100 times, we calculated the standard deviation of this measurement and found it was 0.3 dB, which is extremely stable for a noise measurement. This low standard deviation was achieved by using the Unique Test Period averaging function of the LTX-Credence DIG-HSB digital-signal processing instrument in the company’s X-series testers.
A correlation graph that compares the actual measured difference between the in-band and out-of-band SNR shows that there is a good correlation to the ACR measured on the bench (Figure 5).
FIGURE 5. Correlation data compares the ACR measured on the bench to that measured on the ATE system.
This is due to accurately replicating the bench test setup, using the same excitation signal, and measuring the output in the same bandwidth that is being used in the application. This approach captured all of the influences, from phase noise to image rejection, that we saw on the bench. By using the same modulation technique and measuring the in-band power while the received signal is out-of-band, it is possible to achieve a good correlation to the bench test for ACR.
About the Author
Peter Sarson is the test development manager for austriamicrosystems’ Full Service Foundry business unit. He received his BEng (Hons) from Sheffield University, UK, in 1998 and his chartered engineer status from the Institution of Engineering and Technology (formerly the Institution of Electrical Engineers) in 2003. He has worked in automated test engineering for 11 years.