- Buyers Guide
The Impact of Power Amplifier Turn-On Characteristics in Cognitive Radio Networks
Cognitive radio networks (CRN) often assume that the transmit chain can be instantly switched on and off. In practice, however, there is a delay between DC power being applied to a radio frequency (RF) power amplifier (PA) and the achievement of full output power and linearity requirements. There is also excess power consumption as the PA reaches its static operating conditions. The findings presented here as a result of PA measurements suggest that to guarantee correct operation without any spurious distortion products interfering with other users, a non-negligible delay is required after applying power and before beginning transmission.
CRNs are often regarded as viable options for future mobile communications standards operating over very wide frequency bands.1,2 In wideband CRNs with multiple channels available for sharing, user throughput can be significantly increased by performing a “spectrum handoff,”3 in which the user switches to a vacant channel to continue data transmission if the current channel becomes busy (see Figure 1).
Spectrum sensing is used to check that a particular channel is vacant. If it is, spectrum handoff occurs. As shown in Figure 1, periodic spectrum sensing and access requires intermittent transmission and occasional switching between channels.3 It is assumed that in the transmit chain, the PA consumes the majority of battery power, particularly if high peak-to-average power ratio (PAPR) standards like 3GPP Long Term Evolution (LTE)4 are used. To conserve power, the PA is switched off between transmissions instead of simply being placed into a standby mode. Generally, however, PAs are designed to transmit continuously. By being periodically switched on and off, the PA is operating outside of its normal operating mode, and hence static specifications supplied by the manufacturer no longer apply.
Figure 2 illustrates our measurement approach using a digital sampling oscilloscope. This allows the PA’s RF output voltage and current consumption to be simultaneously sampled. A function generator (a) produces a trigger signal controlling the transistor switch (d) that applies power to the RF PA (k). The trigger signal can be fed either directly (e) to the digital oscilloscope (i), or via a 10 µs to 100 ms 555 timer delay generator (b) that introduces a variable delay (f). For delays greater than one second, a stopwatch is used and the oscilloscope is manually triggered. The delay enables the PA’s output voltage to be sampled at a time offset.
For this work, the PA is an Analog Devices ADL55365 with a 1 dB compression point (P1dB) of 20 dBm. An operating frequency of 600 MHz is chosen, which is in the TV white space band frequently touted as a suitable location for CRNs due to its propagation characteristics and usage profile.6 The PA’s current consumption is measured by sampling the voltage across a current sensing resistor (g) with a differential amplifier (h). An RF signal generator (j) produces a two-tone test signal. A photograph of the test board is shown in Figure 3. The RF input is applied to the SMA connector on the bottom left, with the output on the bottom right. The SMA connector at the top left is the trigger input and the one on the top right the current sense output. The ADL5536 is at the bottom of the board and above it is the transistor that applies power when a trigger signal occurs.
Third order intermodulation distortion (IMD3) generated by the two-tone test7 is assumed to be equivalent to adjacent channel power (ACP) generated by a wideband signal, i.e., LTE. To make an accurate ACP measurement, it must be averaged over multiple frames. In the case of LTE, the frame length is 10 ms. As will later be seen, the PA’s linearity characteristics vary considerably over a 10 ms period. For this work, data is collected over a 2 µs period and the IMD3 is derived by applying a Fast Fourier Transform (FFT) to the samples. It is assumed that the IMD3 does not change appreciatively over 2 µs. A two-tone test with tones at 600 and 610 MHz resulting in IMD3 products at 590 MHz (2 × 600 – 610 MHz) and 620 MHz (2 × 610 – 600 MHz) are accurately resolved by the FFT.
Turn-On Delay and Power Consumption
The amplifier’s turn-on characteristics with a two tone test are shown in Figure 4 over the first 9 µs of operation after the trigger. The trigger signal (green trace) occurs at 1 µs. The RF input level is adjusted so that under static conditions, the amplifier’s IMD3 meets the WCDMA emission mask of –33 dBc.8 Under these conditions, the output power (Pout) is 17.7 dBm. Although the PA starts to produce an output voltage (blue trace) 2 µs after the trigger (at time = 3 µs) it takes a total of 4 µs (at time = 5 µs) before the output voltage reaches its full amplitude. The PA turn-on delay (Td) for full output power is therefore determined to be 4 µs. The large current spike (red trace) is due to the PA biasing networks requiring time to settle and charging of the supply decoupling capacitors necessary for stable operation.
Under static conditions, the ADL5536 consumes 106 mA, resulting in a static power consumption (Pstatic) of 530 mW with a 5 V supply. With 17.7 dBm Pout the efficiency is therefore 11 percent, which is common for an amplifier of this type under these conditions. By integrating the current consumption between 1 and 5 µs and subtracting Pstatic, the excess power required to turn-on the PA (Pturn-on) is calculated to be 590 mW. The total power consumption (Ptotal) for the PA during turn-on and data transmission is given by:
where Ton is the time the PA is switched on. Compared to scenarios published in the literature where a Ton of 40 ms9 or 100 ms10 is specified, the impact of Td and Pturn-on is insignificant. Only at very small values of Ton (say less than 100 µs) does Td and Pturn-on have an impact. It should be noted that the numbers published here are only for the ADL5536. Other PAs will have different values of Td and Pturn-on, which must be determined by measurement.
Intermodulation Distortion During Turn-On
Lower and upper IMD3 fluctuations over time are compared with the –33 dBc WCDMA emission mask in Figure 5. Note that the result at 3 µs is made up from data recorded between 2 and 4 µs, while the PA output power is still ramping up.
The large fluctuation in IMD3 shown in Figure 5 is quite likely due to PA memory effects,7 which would also account for the asymmetry. Memory effects are distortions influenced not only by the present signal amplitude, but also by its past values. These can include thermal effects due to the amplifier’s bias conditions changing as its temperature increases due to power dissipation.
The lower IMD3 fluctuates by over 12 dB and the upper IMD3 fluctuates by over 9 dB before static conditions are achieved. The emission mask is breached by 0.25 dB at 10 µs and at 100 ms it is nearly breached again. Because the data was recorded with a large granularity, it is possible that the emission mask is also breached between measurement points. Based on these results, a CRN should wait at least 10 µs before commencing transmission to avoid breaching the emission mask. However, for complete assurance it should wait 100 ms, which suggests that this particular PA is unsuitable for some published scenarios.9,10
Harmonic Distortion During Turn-On
During turn-on (see Figure 4), the output voltage exhibits asymmetric positive and negative excursions. This asymmetry introduces even order harmonic distortion at multiples of the input frequency. Conventional narrowband systems remove harmonic products by filtering, but this is not possible in a broadband CRN operating over multiple octaves2 where harmonically related spurious products can easily interfere with another user.
The three second-harmonics of the two-tone test were measured: 1200 MHz (2 × 600 MHz), 1210 MHz (600 + 610 MHz) and 1220 MHz (2 × 610 MHz). Their combined power is calculated along with the combined power of the two fundamental tones to determine second harmonic suppression (see Figure 6). As in Figure 5, the first result (at 3 µs) is data recorded between 2 and 4 µs after the trigger signal. Figure 6 shows that the second harmonic is extremely large between 2 and 4 µs after the trigger. It settles down afterward, but continues to fluctuate by almost 6 dB. Even if the PA meets the spurious harmonic emission mask under static conditions, the spike at 3 µs probably does not. A delay of 10 µs would be required before transmission could commence, in agreement with the IMD3 results of Figure 5.
This article examines the turn-on characteristics of an RF PA. Measurements show that the PA took 4 µs to achieve full output power, and during that time consumed an extra 590 mW of power in addition to the PA’s 530 mW static power consumption. Such an additional delay and power consumption are insignificant when considering examples published in the literature; however, significant spurious distortion products are generated by the PA which cannot be ignored. Both IMD3 and second order harmonic distortion are examined. The data indicates that, at a minimum, a user should wait 10 µs after applying power to the PA before commencing transmission. A conservative practice would increase this to 100 ms. These resultsare PA dependant; any PA used in a CRN where it is periodically switched on and off (e.g., periodic spectrum sensing and spectrum handoff) should be measured as described.
This work shows that the PA can no longer be treated as a black box. When designinga CRN, an appropriate delay should be included in the protocol for the PA to sufficiently “warm up” and achieve static conditions. In order to prevent spurious products from interfering with other users, no signal should be transmitted during this “warm up” period. The findings in this work are applicable to other applications as well, where PAs are periodically switched on and off.
The authors would like thank the University of Bristol for the use of test equipment in these measurements.
- Y. Zou, Y.D. Yao and B. Zheng, “Cooperative Relay Techniques for Cognitive Radio Systems: Spectrum Sensing and Secondary User Transmission,” IEEE Communications Magazine, Vol. 50, No. 4, April 2012, pp. 98-103.
- Rohde & Schwarz, Wireless Communications Standards, poster retrieved February 18, 2013, from www.rohde-schwarz.com.
- S. Wang, Y. Wang, J.P. Coon and A. Doufexi, “Energy-Efficient Spectrum Sensing and Access for Cognitive Radio Networks,” IEEE Transactions on Vehicular Technology, Vol. 61, No. 2, February 2012, pp. 906-912.
- 3GPP LTE Homepage, www.3gpp.org/article/lte, viewed 22 January 2011.
- ADL5536 Datasheet, www.analog.com, viewed April 10, 2012.
- N.S. Shankar, C. Cordeiro and K. Challapali, “Spectrum Agile Radios: Utilization and Sensing Architectures,” Proceedings of the First IEEE International Symposium on New Frontiers in Dynamic Spectrum Access Networks, November 2005, pp. 160-169.
- P.B. Kenington, High-Linearity RF Amplifier Design, Artech House, Norwood, MA, 2000.
- C.W. Liu, “Modeling 3G/WCDMA/HSDPA Handset Transmit System,” Microwave Product Digest, February 2007.
- Y. Pei, A.T. Hoang and Y.C. Liang, “Sensing-Throughput Tradeoff in Cognitive Radio Networks: How Frequently Should Spectrum Sensing be Carried Out?” Proceedings of the IEEE 18th International Symposium on Personal, Indoor and Mobile Radio Communications, September 2007, pp. 1-5.
- Y. Liang, Y. Zeng, E. Peh and A. Hoang, “Sensing-Throughput Tradeoff for Cognitive Radio Networks,” IEEE Tranactions on Wireless Communications, Vol. 7, No. 4, April 2008, pp. 1326-1337.