Wideband 4 4 MIMO over the air transmission at 2.45 GHz

Transcription

1 Diplomarbeit Wideband 4 4 MIMO over the air transmission at 2.45 GHz Ausgeführt zum Zwecke der Erlangung des akademischen Grades eines Diplom-Ingenieurs unter Leitung von Dipl.-Ing. Robert Langwieser Dipl.-Ing. Dr. techn. Werner Keim und Ao. Univ.Prof. Arpad L. Scholtz E389 Institut für Nachrichtentechnik und Hochfrequenztechnik eingereicht an der Technischen Universität Wien Fakultät für Elektrotechnik und Informationstechnik von Lukas W. Mayer Matrikelnummer: Adresse: 1090 Wien, Hahngasse 12/17 Wien, November 2005

2 1 I hereby certify that the work reported in this diploma thesis is my own, and the work done by other authors is appropriately cited. Lukas W. Mayer Vienna, November, 2005

3 I would like to thank my parents for giving me the necessary support to finish my studies and this diploma thesis. Furthermore, I would like to thank my sister and my brother for many hours of relaxing and funny activities that balanced some exhaustive weeks during my studies. 2

4 Contents 1 Introduction 5 2 Description of the testbed hardware IF receiver and transmitter hardware Upconverters Power amplifier and antennas Low-noise amplifiers Downconverters IF amplifiers Local oscillators Development of power amplifiers Specification First drafts and measurements Final design amplifier and filter Measurements on the power amplifier Pre power amplifier Setup of the radio frequency frontend RF transmitter Channel setup RF receiver The MIMO transmission experiment MIMO UMTS HSDPA baseband processing SISO STTD

5 CONTENTS DSTTD Measurement results Summary 40 A Measurement programs 41 A.1 Interfacing the network analyzer A.2 Interfacing the spectrum analyzer A.3 Automated intermodulation measurement B Power splitters 45 B.1 Basic structure of the microstrip implementation B.2 Design and simulation B.3 Construction B.4 Measurements C Power-supply unit 50 C.1 Requirements C.2 Design

6 Chapter 1 Introduction A base station transmitter of a conventional mobile telecommunication system is put into operation. The radio frequency signal that is emitted by this transmitter is sent into a metropolitan area. Objects like house walls, vehicles, or the ground scatter the incident wave and apply a certain attenuation and phase shift to the outgoing waves. As a result, the space is filled with many propagating waves approaching from numerous directions and with manifold amplitude and phase. A receiver picks up the superposition of these field components with an antenna. The received signal shows a strong variation in signal strength while the receiver is moving. This variation is because the field components may sum up and thus provide a strong receive signal at one point or might cancel each other bringing signal strength to a minimum at another point. This phenomenon is called fading. Since a small receive signal is very sensitive to noise and thus hard do decode correctly, fading has an impact on the quality of a wireless data link. One approach to regain a stable transmission is to use more than one antennas at the receiver. It is very unlikely that two or more appropriately positioned antennas all receive weak signals. In fact, simulations have shown that the performance of a wireless data link can be significantly improved with additional antennas at either the transmitter or the receiver or both. Special transmit and receive algorithms are necessary to get the most out of such a wireless channel. It is obvious that computational and hardware complexity of such a MIMO (multiple input multiple output) system rise. On the other hand, higher datarates, better coverage, or a more efficient use of the spectral resources might be reasons to put effort into a MIMO system. However, the performance of these systems has not yet been extensively measured and approved. To advance research on the field of MIMO com- 5

7 CHAPTER 1. INTRODUCTION 6 munications, a testbed is set up at the Vienna University of Technology, Institute of Communications and Radio-Frequency Engineering. This diploma thesis describes the development of hardware components that are needed to put a 4 4 MIMO radio transmission at 2.45 GHz into practice. Based on the concept described in the diploma thesis of R. Langwieser [12], the upconverters and downconverters were built. The two stage power amplifiers, a Wilkinson power splitter, and a power supply unit were designed from scratch. Furthermore, automated measurement procedures were created in Matlab to allow fast, accurate, and reproducible measurements on the devices. After verification of the individual radio hardware components, the four transmission paths were set up and the measurements on MIMO channels were started. An example for a 4 4 MIMO transmission is given at the end of this thesis.

8 Chapter 2 Description of the testbed hardware Please note that this chapter briefly describes the components used in the testbed hardware. For a complete and detailed description of the hardware please refer to the diploma theses of Sebastian Caban [8] and Robert Langwieser [12]. The following sections only provide the necessary information to understand the MIMO radio hardware and have been added for the sake of completeness. The setup of the MIMO transmission hardware and the MIMO experiment are described in Chapter 4 and Chapter 5. The block diagram in Figure shows the basic hardware configuration of the testbed. The transmitter host generates four IF 1 signals at 70 MHz with D/A 2 converter modules. These four signals are then passed to the four radio transmission hardware paths that consist of upconverters, power amplifiers, transmit and receive antennas, downconverters, and IF amplifiers. An oscillator and a power splitter provide the LO signal (2.38 GHz) to the upconverters and downconverters. The receive antennas are mounted on an xy-positioning table. By moving the antennas, different channel realizations are created. To compensate for the radio transmission loss, the gain of the IF amplifiers can be adjusted by the measurement computer. The received IF signals are fed into the A/D 3 converter modules of the receiver host. The generation and acquisition of the IF signals are controlled by a measurement computer situated anywhere in the LAN 4. Matlab is used to generate and process the signal samples. A detailed block diagram of one transmission path is shown in Figure Intermediate Frequency 2 Digital to Analog 3 Analog to Digital 4 Local Area Network 7

9 Signal Samples 70 MHz 2.45 GHz 70 MHz Transmitter Host IF RF LO LAN GPIB IF RF IF RF IF RF IF RF Power Splitter Intermediate Frequency Radio Frequency Local Oscillator Local Area Network General Purpose Interface Bus Power Amplifier LAN Measurement PC RF IF RF IF RF IF RF IF LO IF Amplifier Voltage Source 2.38 GHz GPIB xy-positioning table Figure 2.0.1: Block diagram of the MIMO transmission hardware. LAN GPIB Control Voltage Receiver Host Signal Samples CHAPTER 2. DESCRIPTION OF THE TESTBED HARDWARE 8

10 CHAPTER 2. DESCRIPTION OF THE TESTBED HARDWARE IF receiver and transmitter hardware The basic elements for the MIMO transmission are the transmitter and the receiver host. These computers contain fast D/A and A/D converter modules to generate and acquire the IF signals at a frequency of 70 MHz. For maximum flexibility, both converter-cards are interfaced by Matlab and transmit/receive pure signal samples. Thus every modulation scheme and receiver structure can be rapidly implemented in software. The two computers are connected to a gigabit LAN and can be remoted by a measurement PC that controls the transmission sequence. A detailed software and hardware description on generating IF signals with this equipment can be found in [8]. 2.2 Upconverters To put a radio transmission into practice, the IF signals have to be transposed to a frequency where the wave propagation properties are adequate for mobile communications. The frequencies that suit best are between a few 100 MHz and several GHz. The reason for this is that multipath propagation and diffraction enable good coverage, especially in metropolitan areas. Higher frequencies usually enable smaller devices, but the electromagnetic radiation behaves more and more like optical beams, which means that reflection losses rise and shadowing occurs. These higher frequencies (>5 GHz) will be used for short range mobile communications (WLAN,...). For our testbed the transmission frequency of 2.45 GHz was chosen to minimize equipment size while preserving practical propagation properties. MIMO antennas at this frequency are small enough to be integrated into notebooks etc. Furthermore, there is a wide range of electronic components readily available for this frequency band. The first task of the upconverter is to suppress all out-of-band signals which result from digital IF signal generation (Section 2.1). This is done by an SAW filter 5 produced by Amplitronix. The measured electrical data of the filter is shown in Table As one can see in Figure 2.2.1, the filter separates the IF frequency band very well (>50 db) from conversion artifacts, but leaks as frequency rises above 120 MHz. These products are then well suppressed by the transmit filter after upconversion. Considering the output power spectrum of the D/A converter and all filters implemented, any unwanted signal is suppressed by more than 50 db. 5 Surface Acoustical Wave filter

11 CHAPTER 2. DESCRIPTION OF THE TESTBED HARDWARE 10 Parameter Insertion loss Input return loss Output return loss Center frequency 3dB-bandwidth 6dB-bandwidth: 20dB-bandwidth: Value 13.6 db 4.2 db 7.2 db MHz MHz MHz MHz Table 2.2.1: Electrical data of the IF filter C produced by Amplitronix S 21 / db f / MHz Figure 2.2.1: Frequency response of the IF filter ABFF015C produced by Amplitronix. The next task of the upconverter is the IF to RF 6 conversion. This is done by the Maxim MAX2671 double balanced mixer. This IC 7 features buffer amplifiers at the inputs to reduce leakage between the LO 8 input and the IF input. At an intermediate frequency of 70 MHz, a radio frequency of 2.45 GHz, and an LO power of -18 dbm, the mixer offers a conversion gain of 6 db and an LO to RF isolation of 25 db. A Hittite HMC308 buffer amplifier is applied after the mixer. To suppress out of band signals like the LO signal and the remaining 6 Radio Frequency 7 Integrated Circuit 8 Local Oscillator

12 CHAPTER 2. DESCRIPTION OF THE TESTBED HARDWARE 11 D/A conversion artifacts, two 6 th order microstripline bandpass filters with a center frequency of 2.45 GHz and a bandwidth of 60 MHz are used. 2.3 Power amplifier and antennas The power amplifier consists of a pre power amplifier and a power amplifier stage. A detailed specification and design description of these components can be found in Chapter 3. The antennas used for the transmission are quarter wavelength monopole ground plane antennas. Matching to 50 Ω was done by choosing the length and thickness of the antenna rod. The four antennas were placed on a common ground plane. To experiment with antenna spacing, some ground planes making possible different antenna positions were built. Figure shows a picture of the λ / 2 -spaced antenna configuration. Figure 2.3.1: 4-element MIMO-antenna. The elements are 28.8 mm long and 4 mm thick. The spacing between the monopoles is 62 mm ( λ / 2 ). 2.4 Low-noise amplifiers Since the testbed is intended to be used for modulation scheme testing and channel sounding, the LNA 9 is omitted. This may sound weird because the LNA takes care of better SNR 10 and thus provides good quality of the received signal. This approach is misleading in this special case, because the 9 Low Noise Amplifier 10 Signal to Noise Ratio

13 CHAPTER 2. DESCRIPTION OF THE TESTBED HARDWARE 12 signal quality is impaired most by interference from radio frequency equipment situated in the close environment. Thus, the MIMO testbed is not a noise limited system and that is why the LNA can be omitted in these first experiments. A certain signal to noise ratio can be achieved by attenuation of the received signal. Decreasing the transmit power would enhance interference from other radio frequency equipment and can only be done if man-made noise is low. Some measurements of this kind have been carried out in the basement of the Institute of Communications and Radio-Frequency Engineering where disturbing signals are very weak. A detailed description of these measurements can be found in [9] 2.5 Downconverters For downconversion of the received RF signal the Maxim MAX2682 integrated circuit is used. This mixer offers a conversion gain of 7.9 db and has a noise figure of 13.4 db. A Minicircuits MAR-4SM amplifier IC is used to enhance the signal by 8.3 db. Finally, the IF signal is fed through an SAW filter which selects the IF band. In Figure the four downconverters are shown. Figure 2.5.1: Downconverter modules. In the opened downconverter the mixer (chip on the right), the first IF amplifier stage (transistor in the middle) and an IF filter (left) is seen.

14 CHAPTER 2. DESCRIPTION OF THE TESTBED HARDWARE IF amplifiers The IF amplifier which was custom built by Kreuzgruber GmbH is used to compensate the path loss of the radio transmission. The amplifier offers a gain between 4 db and 94 db. The gain is controlled via a voltage. Latter is provided by a GPIB 11 interfaced voltage source. Before the actual measurements some test blocks are sent and the input power at the A/D converters is adjusted iteratively. All IF amplifiers are set to the same gain. This is necessary for channel sounding where the differences between the channel path losses are of interest. 2.7 Local oscillators The local oscillator units custom built by Kuhne Electronic GmbH are used to supply the up- and downconverters with a crystal stabilized, low phase noise CW 12 signal. To convert the 70 MHz IF band to 2.45 GHz, an LO frequency of 2.38 GHz is needed. To simulate a frequency offset between transmitter and receiver, the LOs can be fine tuned in the range of some khz. Each local oscillator has four outputs plus one auxiliary output which provide -5 dbm each. 11 General Purpose Interface Bus also known as the IEEE-488 interface bus. 12 Continuous Wave

15 Chapter 3 Development of power amplifiers 3.1 Specification Since the MIMO testbed is intended to be used for code and modulation scheme testing, the most important property of all RF components is linearity. According to the ISM-Band definition 1 where transmit power is limited to 20 dbm EIRP 2 (which yields a little less output power than 20 dbm due to the gain of the λ / 4 -antennas), the desired output power for each channel of our MIMO system was 17 dbm. The Fairchild RMPA2453 power amplifier chip was the best choice because of its high 1 db compression point and high third order intercept point. Furthermore, input and output are matched to 50 Ω. The basic electrical characteristics of this InGaP HBT 3 MMIC 4 are compiled in Table Parameter Frequency range Power gain Output 1 db compression point Supply voltage Supply current at P OUT = 24 dbm Value GHz 26 db 26.5 dbm 3.3 V 200 ma Table 3.1.1: Electrical specifications of the RMPA2453 power amplifier. To suppress the LO signal, which is still present after upconversion, a 1 The Industrial, Scientific, and Medical radio bands are defined by the ITU-R in and of the Radio Regulations 2 Equivalent Isotropically Radiated Power 3 Heterojunction Bipolar Transistor 4 Monolithic Microwave Integrated Circuit 14

16 CHAPTER 3. DEVELOPMENT OF POWER AMPLIFIERS 15 bandpass filter is used at the amplifier output. Here a helix-filter from Neosid (Table 3.1.2) is used. Parameter Value Center frequency 2.45 GHz Insertion loss 2 db 3 db bandwidth 81.2 MHz 6 db bandwidth MHz 20 db bandwidth MHz Input / output return loss > 10 db within 20 MHz bandwidth Table 3.1.2: Measurement results of the NEOSID helix-filter. To accelerate prototyping I decided to manufacture the printed circuit boards by myself at the TU-Wien. I chose Rogers RO4003 (Table 3.1.3) double sided substrate material. Parameter Value Substrate thickness 0.81 mm Relative dielectric constant ε r 3.38 ±0.05 Dielectric loss factor tan(δ) Copper plating thickness 34µm Copper conductivity S/ m Table 3.1.3: Electrical specification of Rogers RO4003 substrate material. 3.2 First drafts and measurements A simple test board was first designed to check the basic charakteristics of the IC. The application schematic presented in the datasheet was used. To match input and output, the supply pins of the amplifier have to be connected to 3.3 V via 10 nh and 2 nh inductors respectively. The AVX L08056R8DEW thin film surface mount inductor (L=9.4 nh, Q=50 at 2.4 GHz) was chosen for 10 nh. The 2 nh inductor was implemented as a short-circuited microstripline (length=3.37 mm, width=0.3 mm). The parameters of the stub were determined with RF simulation software tools. To ensure a good ACground, bypass capacitors with SRF 5 near 2.45 GHz were chosen. Based on the simulations, the PCB of the test circuit was then designed with CadSoft Eagle. 5 Self Resonant Frequency

17 CHAPTER 3. DEVELOPMENT OF POWER AMPLIFIERS 16 Since the PGA 6 -like package of the RMPA2453 has very small pin spacing of 0.25 mm, it was positioned using a microscope with 1:4-drive tweezers. All parts were then soldered in an infrared oven. Figure shows the test board. The measurements carried out on the test board confirmed the matching as well as the specification of the RMPA2453. Figure 3.2.1: RMPA2453 power amplifier test board (scale 2:1). 3.3 Final design amplifier and filter As soon as the performance of the amplifier was verified, I designed the final amplifier layout which included the transmit filter. The schematic is shown in Figure The PCB layout top-side as well as the layout diagram are shown in Figure Pin Grid Array: Package where pins are located at the bottom and so impossible to solder by hand.

18 CHAPTER 3. DEVELOPMENT OF POWER AMPLIFIERS 17 Figure 3.3.1: RMPA2453 power amplifier schematic. (a) Top-side layout (b) Layout diagram Figure 3.3.2: Topside layout and layout diagram of the final RMPA2453 board 3.4 Measurements on the power amplifier The first step was to verify the DC supply current of the power amplifier with terminated input and output. The current of 96 ma met the description in the datasheet. Then I checked whether DC voltage at the input and output was zero. To detect oscillations, I connected the output of the amplifier to the spectrum analyzer via a 30 db attenuator. The amplifier was stable with 50 Ω terminations at input and output. After this basic sequence the circuit was connected to the network ana-

19 CHAPTER 3. DEVELOPMENT OF POWER AMPLIFIERS 18 lyzer 7. The first impression of the frequency response (S 21 ) was surprising. It showed some periodic resonances in the cut-off region which obviously resulted from casing. The resonances could be damped by filling the empty space in the case with absorbing foam. After this measure the helix-filter could be tuned to its center frequency (2.45 GHz). The final frequency response of one of the power amplifiers is depicted in Figure The drop at 3.15 GHz is still a relict of the casing resonances. Next, the linearity of the device was determined. Therefore the amplifier had to undergo the intermodulation measurement described in Section A.3. As an extension the PAE 8 was also determined by the automatic measurement program. The results were satisfying and are shown in Figure S 21 / db f / GHz Figure 3.4.1: Frequency response of the power amplifier. Maximum gain: 24.2 db, 3 db-bandwidth: 71.2 MHz. 7 HP-8753B 8 Power Added Efficiency

20 CHAPTER 3. DEVELOPMENT OF POWER AMPLIFIERS 19 P OUT / dbm fundamental signal power 3 rd order IM power 5 th order IM power P IN / dbm (a) Intermodulation plot of RMPA2453 power amplifier (output 1 db compression point: 21.7 dbm). PAE / % P OUT / dbm (b) Power added efficiency of RMPA2453 power amplifier. Figure 3.4.2: Intermodulation plot and power added efficiency of RMPA2453 power amplifier. The dual tone experiment is a good method to compare and measure linearity properties and approximate the intermodulation distance for a specified output power level. However, the behavior of a real device also depends on the signal it has to cope with. In mobile communications signals are mostly noise-like and thus are disturbed in a different way compared to harmonic signals. To limit distortion, the IEEE 9 fixed a spectral mask which must not be exceeded by the power spectral density of the output signal. To reliably measure the PSD 10, the IEEE specified a table of valid RBW 11, VBW 12 and averaging settings for the spectrum analyzer. Figure shows the output spectrum of the amplifier which is operated at the maximum power level where the spectral mask requirements are fulfilled. 3.5 Pre power amplifier Since the upconverter output power level is around -15 dbm a pre power amplifier stage is needed to drive the power amplifier. This stage contains the Hittite HMC313 GaAs HBT MMIC amplifier. The basic electrical characteristics are summarized in Table Institute of Electrical and Electronics Engineers 10 Power Spectral Density 11 Resolution Bandwidth 12 Video Bandwidth

21 CHAPTER 3. DEVELOPMENT OF POWER AMPLIFIERS 20 Parameter Frequency range Power gain Output 1 db compression point Supply current Noise figure Value DC - 6 GHz >17 db 14 dbm 50 ma 6.5 db Table 3.5.1: Electrical characteristics of the HMC313 amplifier. To put the pre power amplifier in operation, the basic design task was to match its output to 50 Ω at 2.45 GHz. As recommended in the application notes, the input was equipped with a capacitor to block DC voltage (see C2 in Figure 3.5.1). An AVX 15 pf capacitor with self resonant frequency near 2.45 GHz was the best choice. The output matching network was designed to provide the optimal pull load (S 22) to the amplifier output. This was done by a series and a stub microstripline. The variables of this network were determined using RF simulation software tools. The schematic of the amplifier is presented in Figure Figure 3.5.1: Schematic of the HMC313 pre power amplifier. The PCB layout and the layout diagram of the amplifier was created in Cadsoft Eagle and is shown in Figure The transmission line that leads from the amplifier output towards the SMA-connector X2 is wider than a 50 Ω line. This has the advantage that length can be reduced. The serial transmission line adjusts the real part of the output impedance to 50 Ω. The compensation of the imaginary part is then done by a shunt line which

22 CHAPTER 3. DEVELOPMENT OF POWER AMPLIFIERS 21 terminates to AC ground via a 15 pf capacitor C3 13. Since AC voltage is at a minimum at this point, it is very convenient to apply the DC supply voltage there. The bias point of the HMC313 is adjusted by a resistor (R4) which is placed between the positive supply voltage (10V) and the transistor. To avoid oscillations, the DC supply current is additionally fed through a 0.22 µh inductor (L1) and blocked with a 1 µf tantalum electrolytic capacitor (C4) and a 10 nf ceramic capacitor (C5). Since the MIMO testbed involves four separate hardware transmit paths, it is necessary to adjust the gain of all paths to the same value. The power amplifiers in particular show strong dependence of gain due to component tolerances and the tolerances of PCB fabrication. Furthermore the microstrip filters contained in the upconverters contribute to the variation in gain by fabrication tolerances. To equalize the gain in the four paths, I matched the components accordingly. The remaining deviation was cancelled with four attenuators included in the pre power amplifiers. The T-shaped attenuator can be found in Figure (R1, R2, R3). (a) Top-side Layout (b) Layout diagram Figure 3.5.2: Topside layout and layout diagram of the final HMC313 board. The results of the measurement procedure described in Section 3.4 revealed that matching and gain was as expected (Table 3.5.2). 13 To bypass AC signals, this capacitor has its self resonant frequency near 2.45GHz.

23 CHAPTER 3. DEVELOPMENT OF POWER AMPLIFIERS 22 Parameter Value 3 db Bandwidth 870 MHz Power gain 20 db Output 1 db compression point 12 dbm Input return loss 7.5 db Output return loss 10 db Table 3.5.2: Measurement results of the HMC313 amplifier board at 2.45 GHz. The input attenuator (R1, R2 and R3) was bypassed for the measurement.

24 Chapter 4 Setup of the radio frequency frontend In Chapter 2 and Chapter 3 the components of the MIMO transmission system were described. This chapter focuses on properly joining the individual components to the testbed to obtain best performance. The testbed (Figure 2.0.1) is designed to enable flexible testing of various MIMO modulation schemes. The radio transmission hardware exchanges IF signals with a transmitter and a receiver host computer. These computers use fast D/A and A/D converter modules interfaced by Matlab to generate and receive the IF signals [8]. Complex baseband signal samples are interchanged with a measurement PC via the LAN. This PC processes the received signal samples by applying the receiver algorithms implemented in Matlab. The testbed also offers the possibility of real-time processing by the use of fast FPGA 1 chips. Since this work focuses on the design of the MIMO testbed, the receiver structures were implemented in Matlab only. While signals processed in software do not undergo unwanted nonlinear distortion, this is always the case when signals pass hardware components as mixers and amplifiers. This important issue was considered right from the beginning of the testbed design. On the one hand, there are maximum input and output power levels that may not be exceeded. On the other hand, low input power makes the signal more sensitive to noise. To overcome those difficulties I evaluated the maximum input power level and the gain of all components. These measurements are described in Section A.3. The most critical element in the radio transmission chain is the power amplifier. Its linearity can e.g. be specified by the spectral mask requirement of the UMTS 2 1 Field Programmable Gate Arrays 2 Universal Mobile Telecommunications System 23

25 CHAPTER 4. SETUP OF THE RADIO FREQUENCY FRONTEND 24 Input Power Output Power Component Maximum Operated Maximum Operated D/A converter dbm -13 dbm Upconverter -11 dbm -23 dbm -10 dbm -22 dbm Pre power amplifier -9 dbm -22 dbm +5 dbm -6 dbm Power amplifier -6 dbm -6 dbm +17 dbm +17 dbm Downconverter -12 dbm -12 dbm -10 dbm -10 dbm IF amplifier -16 dbm -16 dbm 4 dbm -5 dbm A/D converter -5 dbm -5 dbm - - Table 4.0.1: Maximum and operating power levels of the UMTS-signal at the testbed components. Attenuators are put between the D/A converter outputs and the Upconverter inputs to set the output power of the transmitter. standard. Therefore, all other components have to be operated at an input power level that leads to a lower distortion than the distortion introduced by the power amplifier. In Table the maximum input and output power levels (of a UMTS signal) that provide sufficient linearity of the devices are shown. Furthermore, Table shows the power levels at all devices while operated in the testbed. 4.1 RF transmitter The radio frequency hardware basically consists of four equal transmission paths (Figure 4.0.1). In the transmitter, the IF signals (center frequency: 70 MHz, maximum radio bandwidth: 20 MHz) provided by the transmitter host are converted to 2.45 GHz. The output power of each of the succeeding amplifiers is 23 dbm at the 1 db compression point. Due to the high peak-toaverage ratio of modern modulation schemes (CDMA 3 signals used for HS- DPA 4 ), the achievable average output power is below 23 dbm. Figure shows the measured output power spectral density of one power amplifier according to the UMTS standard [2]. Furthermore the limiting base station spectral mask (P MAX 31dBm) is plotted. The maximum achievable output power is determined by increasing the signal amplitude, which results in nonlinear distortion, until the spectral mask is still fulfilled. Therewith, a 3 Code Division Multiple Access 4 High-Speed Downlink Packet Access

26 70 MHz 2.45 GHz 14 bit f S = 200 MHz DAC l = 2 m a = 2.2 db a = 10 db 2.45 GHz f C = 70 MHz B = 20 MHz a = 14 db f LO = 2.38 GHz P OUT = -5 dbm LO g = 8 db g = 8.3 db a LO RF = 25 db g = 6 db l = 25 m a = 27 db Power Splitter f C = 70 MHz B = 20 MHz a = 14 db g = 17 db f C = 2.38 GHz a = 10 db B = 100 MHz g = 26 db 70 MHz f C = 2.45 GHz B = 60 MHz a = 8 db f C = 2.45 GHz B = 81 MHz a = 2 db l = 25 m a = 3.6 db f C = 70 MHz B = 20 MHz g = db g = 16 db 14 bit f S = 100 MHz ADC Figure 4.0.1: Block diagram of one transmission path. B = 100 MHz g = 26 db f C = 2.45 GHz B = 81 MHz a = 2 db P OUT = 17 dbm CHAPTER 4. SETUP OF THE RADIO FREQUENCY FRONTEND 25

27 CHAPTER 4. SETUP OF THE RADIO FREQUENCY FRONTEND 26 mean output power of 17 dbm was obtained. 10 power spectral density / dbm/30khz frequency / GHz Figure 4.1.1: Power spectral density of a 17 dbm UMTS signal measured at an amplifier output (resolution bandwidth = 30 khz). The spectral mask for a UMTS base station (P max 31dBm) is also plotted. An important issue when transmitting four signals at the same time is to avoid crosstalk between the transmit channels. The high signal power at the antennas (4 17 dbm) yields strong electromagnetic fields in the area where the transmission equipment is located. In particular the high gain of the two stage power amplifier brings small signals that intrude into the casing of the upconverter to a formidable level. To avoid this, the upconverter cases were sealed with copper foil shielding tape. Furthermore, the LO line was equipped with a band-pass filter to reduce interference with RF signals picked up along cables or IF signals that are leaking from the downconverters and are passed through the LO. With these measures a channel separation of more than 40 db could be achieved. Figure shows a photograph of the IF and RF hardware at the transmitter. The four upconverters are located on the left, the two stage power amplifiers are on the right. In the middle you can see the IF amplifiers which are part of the receiver. On top of the IF amplifiers the LO amplifier is mounted. The power supply unit is shown in the back.

28 CHAPTER 4. SETUP OF THE RADIO FREQUENCY FRONTEND 27 Figure 4.1.2: Photo of the IF and RF hardware at the transmitter. 4.2 Channel setup The transmit and receive antennas are λ / 4 monopole ground plane antennas arranged in a linear array configuration with spacing of λ / 2. The spacing can be changed easily allowing for investigation of its influence on the MIMO channel on the performance of various transmission schemes. Figure shows a floor plan of the 5 th storey where the measurements are carried out. It can be seen that the transmitter and receiver are located in different rooms and are separated by walls and doors. Since there are a lot of scattering objects (e.g. measurement equipment, computers, cables, etc.) situated in these rooms, multipath propagation is enhanced. The bee-line between the antenna arrays is approximately 13 m. This distance is fairly sufficient to provide multipath propagation in this indoor setup. The receive antenna array is mounted on an xy-positioning table that moves the antennas within 1 m in each direction. This enables measurements on a time varying multipath channel as commonly observed in mobile communication. The movement of the receive antennas can be controlled from the measurement PC using an RS-232 interface. By moving the whole receiver equipment to

29 CHAPTER 4. SETUP OF THE RADIO FREQUENCY FRONTEND 28 appropriate spots, LOS 5 scenarios can also be created. Coaxial cables lead the received and downconverted signals to the receiver host. The following experiment evaluates the performance of MIMO in such an environment. 1.9m RX y x 5.8m 7m 3.5m Corridor TX 1.9m 7.7m Figure 4.2.1: Floor plan of the laboratory. The position of the xy-positioning table carrying the TX antenna array and the position of the RX antenna array are shown. To categorize the propagation properties, a channel sounding experiment was carried out before the MIMO measurements. Figure shows the SISO channel coefficient versus the receive antenna position. The spacing between the positions was λ / 10 and the covered area was 2λ 2λ. The channel statistics was evaluated by measuring the flat fading 4 4 MIMO channel coefficients at 400 positions of the receive antenna array ( = 6400 coefficients). Figure shows the magnitude (in db below maximum) of 5 Line of Sight

30 CHAPTER 4. SETUP OF THE RADIO FREQUENCY FRONTEND 29 these coefficients versus the receive antenna position. It can be seen that the signal strength strongly varies when the antenna is moved along. Such a behavior is typical in a multipath NLOS 6 SISO scenario and leads to lower channel capacity compared to an AWGN 7 scenario without fading (e.g. LOS). 0 estimated channel coefficient / db Y position / λ X position / λ Figure 4.2.2: Magnitude of measured channel coefficient between two antennas versus receive antenna position. The receive antenna array was moved within a square of 2λ 2λ with a resolution of ( λ / 10 ). probability estimated channel coefficients Figure 4.2.3: Weibull plot of measured channel coefficients of the 4 4 MIMO channel (4 antennas 4 antennas 400 positions = 6400 channel coefficients). 6 Non Line of Sight 7 Additive White Gaussian Noise

31 CHAPTER 4. SETUP OF THE RADIO FREQUENCY FRONTEND 30 Figure shows the Weibull plot of the 6400 channel coefficients. A Weibull distribution is characterized by a straight line in the Weibull plot. Rayleigh- and Ricean-fading are special cases of the Weibull distribution. For convenience the line that represents Rayleigh-fading is also plotted in Figure Figure 4.2.4: xy-positioning table with receive antenna array, downconverter units, and local oscillator. 4.3 RF receiver The four signals picked up by the receive antennas are fairly weak and thus sensitive to noise. Cables of approximately 2 m length lead these signals from

32 CHAPTER 4. SETUP OF THE RADIO FREQUENCY FRONTEND 31 RF frontend parameter Value Number of transmitter channels 4 Number of receiver channels 4 Intermediate Frequency 70 MHz Radio Frequency 2.45 GHz Bandwidth of RF Frontend 20 MHz Maximum supported signal bandwidth 6.25 MHz Transmit power per antenna (UMTS) 17 dbm IF amplifier gain 4-94 db Table 4.3.1: Testbed parameters. the antennas that are mounted on the xy-positioning table to the downconverters (Figure 4.2.4). The cables that lead the signals back to the receiver host would further impair the noise figure of the receiver by their attenuation (27.8 db at 2.45 GHz for 25 m cable). To avoid an intolerable loss of SNR, the downconverters are mounted on the xy-positioning table. Once the signal is converted to 70 MHz, the cable attenuation reduces to a tolerable level (3.6 db at 70 MHz for 25 m cable). To provide an identical LO signal to the upconverters and downconverters, a fifth cable is used between the movable receiver and the fixed transmitter. The local oscillator is placed on the movable receiver unit. An amplifier accounts for the LO cable insertion loss of 27 db. A bandpass filter suppressing the 70 MHz signal is used to avoid crosstalk between the transmitter and the receiver IF signals via the LO connection. To benefit from the full dynamic range of the A/D converters, four adjustable gain IF amplifiers are used. Each of these units offers a gain between 4 db and 94 db. The gain is adjusted via a GPIB controlled voltage source. It is possible to adjust the input power of the A/D converter cards in an iterative process. For each measurement sequence, the gains are kept constant. Fast fading due to multipath propagation is therefore not compensated by the IF amplifiers. The dynamic range of the A/D converter cards is about 84 db (14 Bit) and thus sufficient to convert weaker signals as well.

33 Chapter 5 The MIMO transmission experiment As soon as the testbed was put into operation several measurement runs were started. The following example shows that MIMO can be advantageous in mobile communication. At this point, I would like to acknowledge the work of Christian Mehlführer, who expertly implemented the Matlab code to put the MIMO UMTS HSDPA transmission into practice. A detailed description of the whole experiment was published in [7]. A description of the digital signal generation and reception in Matlab can be found in [11]. 5.1 MIMO UMTS Several standardization workgroups are considering MIMO techniques as mandatory for their next generation systems. One of these workgroups is the 3GPP 1 which is responsible for UMTS standardization. In Release 99 of the UMTS standard a so-called STTD 2 mode is specified mandatory for the UE 3 receiver. This STTD mode incorporates Alamouti coding [13] at the transmitter, thus achieving a diversity order of two. However, this STTD mode is not a true MIMO transmission since it only requires two transmit antennas and one receive antenna. The use of multiple antennas on both the transmitter and the receiver side is expected to be standardized as an extension of the HSDPA mode. One corporate proposal is DSTTD-SGRC 4 which will be evaluated and 1 3 rd Generation Partnership Project 2 Space-Time Transmit Diversity 3 User Equipment 4 Double Space-Time Transmit Diversity with Sub-Group Rate Control 32

34 CHAPTER 5. THE MIMO TRANSMISSION EXPERIMENT 33 compared to STTD and SISO HSDPA in this section. The evaluation is performed by free space transmission experiments that reveal the block error ratio and the system throughput. 5.2 HSDPA baseband processing In this section the baseband processing of the various transmitters and receivers implemented is described. The STTD and DSTTD schemes are based on the SISO HSDPA scheme. Therefore, the SISO scheme will be explained in more detail before the MIMO systems are discussed SISO HS-PDSCH Data Stream CRC and Bitscrambling Turbo Coding R=1/3 Rate Matching and Interleaving Symbol Mapping Figure 5.2.1: SISO HSDPA signal generation. The SISO transmitter, in our case completely implemented in Matlab, generates HSPDA subframes as specified by the 3GPP [6]. It performs the following operations on randomly generated data bits: Add 24 CRC (Cyclic Redundancy Check) bits allowing for frame error detection in the receiver. Turbo coding with rate 1/3. Adaptive rate matching to the code rate as determined by the CQI 5 value. Adaptive symbol mapping (QPSK, 16-QAM), also determined by the CQI. Spreading with spreading factor 16. Scrambling with a specific scrambling sequence of a Node B. 5 Channel Quality Indicator

35 CHAPTER 5. THE MIMO TRANSMISSION EXPERIMENT 34 After generation of the HS-PDSCH 6 chipstream, the pilot channel and the synchronization channel are added. These channels are used for channel estimation and for synchronization (i.e. frame synchronization and symbol timing recovery) at the receiver. The resulting chip stream is then transmitted using a previously developed interface between Matlab and D/A converters [5]. During the SISO Transmission the total transmit power of 17 dbm was sent to a single antenna. The Matlab interface also provides the sampled receive signal after completion of the radio transmission. This signal is then correlated with the synchronization channel to obtain the optimum sampling time instant. After decimation the resulting synchronized chip stream is further processed. In the digital baseband receiver, a matched filter is applied on the signal samples. Since the delay spread is low in our indoor scenario, flat fading occurs during the transmission and an equalizer is not needed. After the matched filter all operations (i.e. descrambling, despreading, demapping, decoding) done in the transmitter are performed inversely. By evaluating the CRC bits, the block error rate is determined STTD HS-PDSCH Data Stream CRC and Bitscrambling Turbo Coding R=1/3 Rate Matching and Interleaving Symbol Mapping STTD Coding Figure 5.2.2: STTD HSDPA signal generation. The SISO HSDPA mode can be extended to two transmit antennas by Alamouti coding [13] of the transmit symbols. This so-called STTD mode achieves a transmit diversity order of two allowing for a more reliable data transmission. The receiver for this transmission is again the matched filter receiver that multiplies the received symbols with the conjugated, transposed channel matrix. To do a fair comparison to the SISO transmission, the total transmit power of the two antennas was set to 17 dbm (14 dbm per antenna). 6 High Speed Physical Downlink Shared Channel

36 CHAPTER 5. THE MIMO TRANSMISSION EXPERIMENT DSTTD HS-PDSCH Data Stream SG1 CRC and Bitscrambling Turbo Coding R=1/3 Rate Matching and Interleaving Symbol Mapping STTD Coding S/P SG2 CRC and Bitscrambling Turbo Coding R=1/3 Rate Matching and Interleaving Symbol Mapping STTD Coding Figure 5.2.3: DSTTD HSDPA signal generation. The DSTTD scheme transmits two independent data streams on four transmit antennas. The input data is first separated into two subgroups, each subgroup is further processed as in the SISO scheme. After symbol mapping, Alamouti coding is performed on every subgroup individually. The resulting four symbol streams of the four transmit antennas are spread and scrambled with the same spreading and scrambling sequence. In our experiment each antenna was supplied with 11 dbm to maintain the total transmit power of 17 dbm. Since the UMTS standard does not specify pilot channels for four transmit antennas, a novel pilot pattern presented in [4] was used in our experiment. The space-time receiver used for the measurements is an MMSE-SIC 7 receiver. This receiver decodes the subgroup with higher SINR 8 first and performs the interference cancelation with error corrected symbols obtained from the Turbo decoder. A more detailed description of the receiver implemented and the determination of the subgroup decoding order is given in [3]. 5.3 Measurement results This section provides block error rate and throughput measurements of SISO, STTD, and DSTTD HSDPA transmissions. An overview of the measurement parameters is given in Tables 5.3.1, 5.3.2, and The measurements were 7 Minimum MeanSquared Error with Successive Interference Cancelation 8 Signal to Interference and Noise Ratio

37 CHAPTER 5. THE MIMO TRANSMISSION EXPERIMENT 36 Parameter Pilot channel power (E c /I or ) Synchronisation channel power (E c /I or ) Control channel power (E c /I or ) Channel coefficient estimation Turbo decoding Retransmissions Value 10 db 12 db 12 db least squares max-log-map - 8 iterations none Table 5.3.1: Common measurement parameters. Parameter Value Modulation 16-QAM Transport block size 2880 Coding rate 3 / 4 No. of channelization codes 2 Peak data rate 1.44 Mbps Table 5.3.2: SISO and STTD measurement parameters. performed with varying HS-PDSCH 9 E c /I or 10. The E c /I or value basically is the amount of energy that is assigned to the user. Figure shows the measured BLER 11 for the different HSDPA schemes. In [1] 3GPP specifies a BLER of 10 % as target for the CQI reporting. When using this BLER value for comparisons between the different schemes we observed that STTD (N R = 1) outperformed SISO by about 6 db. STTD also needed about 2 db less of E c /I or than DSTTD to achieve 10 % BLER. Note, that the peak data rate of STTD is only half that of DSTTD. Due to non equal data rates of the MIMO schemes, a fair comparison can only be drawn in terms of throughput to account for different peak data rates. Such a throughput comparison is shown in Figure We observed that STTD with N R = 1 performed almost as well as the SISO system for E c /I or < 15 db, but achieved its peak data rate of 1.4 Mbit/s at 6 db 9 High Speed-Physical Downlink Shared Channel 10 E c /I or denotes the ratio between the energy of a specific chip stream (data, pilot, synchronisation, control) and the total energy available at the transmitter. 11 Block Error Ratio

38 CHAPTER 5. THE MIMO TRANSMISSION EXPERIMENT 37 Parameter Value CQI 0 Modulation 16-QAM Transport block size 2880 Coding rate 3 / 4 No. of channelization codes 2 Peak data rate 2.88 Mbps Table 5.3.3: DSTTD measurement parameters. Each of the two data streams encountered in the DSTTD mode has transport block size and coding rate as shown in this table. less E c /I or. throughput / bps x SISO STTD N R =1 STTD N R =3 DSTTD N R =3 STTD 100% DSTTD 100% HS PDSCH E /I / db c or Figure 5.3.1: SISO, STTD, and DSTTD throughput comparison.

39 CHAPTER 5. THE MIMO TRANSMISSION EXPERIMENT BLER 10 1 SISO STTD N R =1 STTD N R =3 DSTTD N =3 R HS PDSCH E /I / db c or 8 6 Figure 5.3.2: SISO, STTD, and DSTTD Block Error Ratio comparison. Figure shows a throughput comparison of DSTTD for different numbers of receive antennas. A significant gain of 4 db between the three and the two receive antenna scheme was measured. The third receive antenna favors the channel matrix to be of full rank, thus increasing the data throughput. A fourth receive antenna on the other hand does not significantly (approx. 1 db E c /I or gain) change the throughput. This can be explained by the small antenna spacing of λ / 2. throughput / bps 3 x 106 DSTTD N =2 R 2.5 DSTTD N =3 R DSTTD N =4 2 R % 90% HS PDSCH E /I / db c or Figure 5.3.3: DSTTD throughput for different numbers of receive antennas. In this section indoor measurements for SISO and MIMO HSDPA were presented. The capabilities and important design criteria of the MIMO testbed were discussed. Measurements revealed for DSTTD that three antennas should be used at the receiver striking an optimum tradeoff between

40 CHAPTER 5. THE MIMO TRANSMISSION EXPERIMENT 39 data throughput and receiver complexity in this scenario. It was furthermore shown that the DSTTD scheme works very well with an antenna spacing of λ/ 2 in this indoor scenario. More measurements have to be performed to investigate the dependence of the transmit and receive antenna array dimensions on the system throughput.

41 Chapter 6 Summary In my work at the 4 4 MIMO testbed I designed and built the last components necessary to perform an over the air transmission at 2.45 GHz. The upconverters and downconverters were built based on an existing design. The two stage power amplifiers, the LO power splitter and the power supply unit were designed from scratch. All testbed components were characterized with self made automated measurement routines implemented in Matlab. These measurements gave detailed knowledge about the linearity and the performance of the individual devices. The four transmitter and receiver chains were then set up and tested. Finally 4 4 MIMO over the air transmissions were possible. Several measurements were and are still done with this equipment. In a first measurement campaign basic space time block codes were tested [11]. Then the very first 4 4 MIMO UMTS HSDPA transmission that is described in this thesis and published in [7] was done. Measurements on Alamouti codes are described in [9]. In a new measurement campaign the testbed is used to determine the influence of antenna spacing on a MIMO transmisison [10]. 40

42 Appendix A Measurement programs All components in the MIMO-Testbed have to be inspected and classified very meticulously. It is obvious that in a MIMO system there are similar components which have to be tested exactly the same way. To efficiently compare the acquired data, it is a major advantage to make the measurement results available in digital form. The basic approaches and methods for this are described in the following chapter. A.1 Interfacing the network analyzer To acquire S-Parameters of various testbed components and for qualifying filter and gain characteristics, a network analyzer is used. I decided to keep the data acquisition as simple as possible by just querying the numerical S-Parameters using the GPIB 1 interface. A Matlab function (NetAnal.m) was implemented that returns the displayed S-Parameter data from the network analyzer 2 as a data array. This array contains n rows (where n is the number of measured points) and 2 columns. The second column represents the complex S-Parameter value which was measured at the frequency specified in the first column. The function also generates a bode plot output on the screen to double-check the data transmission. 1 General Purpose Interface Bus also known as the IEEE-488 interface bus. 2 HP-8753B 41

43 APPENDIX A. MEASUREMENT PROGRAMS 42 A.2 Interfacing the spectrum analyzer I started out interfacing the spectrum analyzer 3 by writing the Matlab function CapSpec.m. This function generates a picture output (EPS 4 -file) of the data displayed at the spectrum analyzer. The settings that were made by the user at the spectrum analyzer are also retrieved and plotted on top of the picture. These settings are resolution bandwidth (RBW), video bandwidth (VBW) and the setting of the internal attenuator (ATT). A sample of a picture generated by CapSpec.m is shown in Figure A.2.1. The function also returns a data array containing the measurement values in the second column and the corresponding frequency in the first column. 0 RBW=1kHz VBW=1kHz ATT=10dB 20 P/dBm f/mhz Figure A.2.1: Picture generated by CapSpec.m. The measurement parameters are shown on top of the picture. A.3 Automated intermodulation measurement Since linearity measurements are a very time consuming task if done manually for each testbed component, I decided to create an automated measurement program in Matlab (IM.m). This program exchanges data with the measurement devices using the GPIB interface. 3 ADVANTEST R Encapsulated Postscript

44 APPENDIX A. MEASUREMENT PROGRAMS 43 P IN P OUT P MEAS Signal Generator 1 Signal Generator 2 Wilkinson Combiner Stepable Attenuator DUT Cable and Attenuator Losses GPIB Bus Attenuator (optional) Spectrum Analyzer GPIB/LAN Bridge Figure A.3.1: Block diagram of the automated intermodulation measurement. According to the block diagram (Figure A.3.1) the intermodulation measurement consists of the following procedure: The dual-tone signal (provided by the two signal generators and the Wilkinson combiner) is set at a defined power level by the step attenuator 5. This signal is fed through the DUT 6. If the DUT is not strictly linear, the odd-numbered terms of the characteristic curve will cause parasitic frequencies next to the two main signals. The power spectrum P MEAS is measured with the spectrum analyzer 7 and evaluated in the measurement program. The attenuation inserted by the optional attenuator and cables between the DUT output and the spectrum analyzer input is corrected by software. Therefore it is necessary to determine the losses between P OUT and P MEAS (by using a network analyzer), and passing the loss (in db) to the IM.m function as a parameter. Figure A.2.1 shows a typical power spectrum of two intermodulating signals measured at a DUT output. The program finds and saves the power of every peak in the spectrum. In the subsequent steps the signal level at the DUT-input is automatically increased by setting the step attenuator. After this procedure an array containing the attenuation value, the power of the fundamental signals as well as the power of the 3 rd and 5 th order intermodulation products 8 is available. Now, the DUT output power of the fundamental signals is compared to the output power of the 3 rd and 5 th order intermodulation products. The 5 Rhode und Schwartz, 0-2.7GHz, DPSP step attenuator 6 Device Under Test 7 ADVANTEST R Usually it is sufficient to consider the 3 rd and 5 th order intermodulation signals because higher order signals are very weak. This could also be confirmed during various measurements.

45 APPENDIX A. MEASUREMENT PROGRAMS 44 differences are presented in a P OUT versus P IN plot as shown in Figure This plot can be used to determine 1 db compression, 3 rd and 5 th order intercept point as well as saturation power and gain. I also used these results to determine the maximum input power that still provides a certain suppression of intermodulation products. Especially when characterizing power amplifiers, it is important to pay attention to the power level at the spectrum analyzer input. This is crucial not only because of the maximum input power level, but also because of the intermodulation produced by the spectrum analyzer itself. A test measurement with no DUT revealed the maximum power level at the spectrum analyzer input that provides sufficient linearity. The second task of IM.m is to calculate the power added efficiency. The DC power consumed by the DUT can be measured by the DC voltage source 9. Hence the power added efficiency PAE = 100% P OUT P IN + P DC can be determined and plotted (see Figure 3.4.2). 9 Agilent E3633A (GPIB interfaced voltage source)

46 Appendix B Power splitters The LO power splitter contained in the testbed is used to distribute the LO signal to the upconverters. Since four components have to be supplied, the splitter consists of three equal 1:2 Wilkinson dividers. For setting the desired power level at the upconverter inputs, a resistive Π attenuator is set in front of the divider structure. B.1 Basic structure of the microstrip implementation A Wilkinson divider splits a signal into two equal parts with identical phase shift. One advantage compared to impedance dividers is, that the outputs of a Wilkinson divider are decoupled very well. Bad matching at one output leads to bad matching of the input, but does not alter matching or signal amplitude at the other output. This is important because the LO power level has to remain constant for all upconverters. Different LO powers would lead to different conversion gains and thus to different signal levels. The microstrip implementation of a Wilkinson divider consists of two λ / 4 transmission lines that transform the 50 Ω output impedance to 100 Ω 1. The input matching to 50 Ω is now achieved by connecting the two 100 Ω ends in parallel. Any signal that approaches the circuit from an output (this can also be reflection due to output mismatch) undergoes a 180 ( λ / 2 ) phase shift while passing to the other output. There the signal is cancelled by the signal passed through a 100 Ω resistor that is connected between the two outputs. 1 This is the basic property of a λ / 4 -transformer (quater wave transformer). The impedance of the quater wave transmission line is given by Z L = Z 1 Z 2 and evaluates to 70.1Ω in this special case. 45

47 APPENDIX B. POWER SPLITTERS 46 However, this behavior only applies to periodic signals with one specified frequency. The Wilkinson dividers are therefore optimized for 2.38 GHz. B.2 Design and simulation To split the LO power into four equal parts three Wilkinson dividers were used as shown in Figure B.2.1. The divider parameters were determined by use of standard microwave design software. To determine the insertion loss of each divider a simulation that considers resistive and dielectric losses was set up. According to this simulation every divider contributes 3.5 db to the insertion loss. To achieve the necessary overall insertion loss of 10 db 2 a resistive 3 db Π attenuator was inserted in front of the divider. A positive side effect of this measure was an improvement of the input matching. The schematic shown in Figure B.2.2 represents the whole 1:4 divider. It was used for simulation and PCB 3 generation. Figure B.2.1: Layout of 1:4 Wilkinson power splitter (scale 1:1) for 2.38 GHz. On the left there are the pads for the SMD resistors that form the Π- Attenuator. The three Wilkinson dividers successively split the power into four equal parts. The 100 Ω resistors are fitted between the parallel transmission lines at each divider. See Figure B.3.1 for a photograph. 2 The upconverters require -18 dbm of LO power for optimum operation (high linearity). Since the available LO power at the transmitter was -8dBm, the insertion loss of the power splitter was fixed to 10dB. 3 Printed Circuit Board

48 Figure B.2.2: Schematic of 1:4 Wilkinson power splitter for 2.38 GHz APPENDIX B. POWER SPLITTERS 47

49 APPENDIX B. POWER SPLITTERS 48 B.3 Construction To reduce costs and speed up prototyping, low cost FR4 substrate was used for the 1:4 divider. Input and outputs were terminated with female SMAconnectors for best compatibility with the MIMO testbed. The PCB was enclosed in a metal box to minimize electromagnetic interference with other devices or the environment. Figure B.3.1 shows a photo of the Wilkinson divider which was then integrated into the testbed. Figure B.3.1: Photo of boxed 1:4 Wilkinson power splitters. B.4 Measurements To determine the performance of the power splitter a network analyzer 4 was used. Figure B.4.1 represents the return and insertion-losses measured for various ports. Note that the output pair 2, 3 shows better decoupling from output pair 4, 5. This is because the crossing signals are decoupled by the inner divider and have to pass the other two dividers as well (remember the insertion loss of 3.5 db per divider). 4 HP-8753B