DEVELOPMENT OF WIDEBAND RADIO CHANNEL

for directional measurements were studied in [P1]–[P4]. The system was realized at 2.154 and 5.3GHz, [P1], [P4], and [P7] and has been tested at 60 GHz. Measurements were performed at 5.3 GHzin both indoor and outdoor environments [P4]–[P7]. Indoor propagation mechanisms were studiedby rotating a directive antenna in a channel sounder receiver [P5]. The indoor measurements werethoroughly analyzed, and empirical models were developed for wireless local area network systemsimulations [P6].22.1Wideband radio channel measurement techniquesReview of the measurement techniques for wideband radio channelsRadio channel characterization can be basically done with any transmitter-receiver configuration.Wideband measurement techniques have traditionally been divided by the used waveform into pulseand continuous-wave techniques[10], [11].The main drawback of pulse sounding techniques (e.g.[12]) is that because the transmitted energy ispulsed they require a high power amplifier (PA) in the transmitter (TX), which increases the complexity of the system. Moreover, performing mobile measurements presumes licenses, which havepower spectrum limitations of the transmitter.Network analyzer is an obvious choice for wideband measurements [13]–[16]. The frequencysweeping waveform, which is rectangular in the frequency domain, has –13 dB sidelobes in thesinc-shaped delay-domain autocorrelation function. By using windowing functions to shape the signal, (e.g. [15]), trade-off between the sidelobe level and the mainlobe width can be applied. However, cable connection makes the network analyzer unpractical for large distances. Incharacterization of the radio channel, comprehensive measurement campaigns require a practicalmeasurement device. The slowness of the frequency sweeping limits the Doppler range ofmeasurement. Thus the measurement requires in practice a frozen environment. However,measurements campaigns with limited number of datasets have been performed even with virtualarrays (e.g.[17]) using optical antenna feed.Direct sequence (DS) based methods, where phase-modulated pseudo-noise (PN) sequences areused as the sounding waveform, are nowadays widely used in radio channel characterization. Usingthe sliding correlator (SC) [18] principle, time/bandwidth scaling can be done in the receiver, whichallows to perform wideband measurements with a relatively slow A/D conversion speed. Instead ofa sliding correlator, a stepping correlator can be used, as in [19] and [20]. Today, A/D conversioncan be done with several hundred megasamples per second, and direct sampling of the receivedsignal [21] has advantages over the SC, especially in measurements with antenna arrays, whichrequire rapid sampling of several antenna elements. The bottleneck in a system using rapidsampling is the speed of the bus between the sampling unit and the mass memory. In the DS-basedchannel sounder, energy is transmitted continuously. So, compared to the pulse method, less peakpower equal to the effective operation time of the PA is required in the TX. The application of thismethod is presented in detail in this section.The further development of DS based measurement techniques involves finding the optimal sounding waveforms with maximum spectral efficiency [22]. In this technique, combined with nonlinearpredistortion, the rectangular spectrum shape is retained even after the RF power amplifier; hencethe link budged is improved, and requirements for the TX-filters are relaxed.9

2.2HUT sounderIn this section, the sounder developed at Helsinki University of Technology partly as a part of thiswork is described. The sounder, which is based on the DS method, operates at 2.154 and 5.3 GHzcenter frequencies, and the extension to 60 GHz frequency range has already been tested.2.2.1 TransmitterIn the transmitter (Fig. 2-1) the microwave carrier is modulated by the m-sequence generated in thePN generator. The PN generator consisting of feedback shift registers can generate L 31-2047chip m-sequences. The chip rates fchip between 2.5 MHz 30 MHz can be generated by a digitalphase-locked loop (PLL). A double-balanced microwave mixer is used as a 2-PSK modulator. Mechanical realization is such, that the 2 and 5.3 GHz transmitters are separate units. Transmittedpower is 40 dBm at 2 GHz with a separate power amplifier and 30 dBm at 5 GHz.PN-gen.PSK-drivePoweramp.10 MHzRubidiumstandardPLL2.154 /5.3GHz PLLFig. 2-1. Transmitter of the measurement system.The chip rates higher than 30 MHz have a separate PLL and a code generator, which are realizedusing ECL-circuits. The maximum operating frequency of a code generator utilizing the linear feedback shift register is(f )chip max τ SR1, n τ gate Στ tr τ mux(2-1)where τSR, τgate, τtr, τmux are the shift register (SR) flip-flop, gate, transmission line, and multiplexerpropagation delays (Fig. 2-2), respectively, and n is the number of gates in series in the feedbackloop having the longest delay.XORτgateτtr2τtr1τSRshift register12345678output9Fig. 2-2. Delays in the feedback shift register (the multiplexer to change the feedback paths is notshown).10

Other topologies of high-speed code generators can be realized using e.g. FPGAs. Chip rates higherthan 1 GHz have been realized [23]. The performance of the DS channel sounder is evaluated on thebasis of waveform autocorrelation properties. The autocorrelation Rs(τ) of the periodic signal s(t)with period Ts is defined as1 TsRs (τ ) s(t )s (t τ )dt .(2-2)Ts 0The autocorrelation of the waveform is related to the power spectrum Sp(f) of the signal by the Wiener-Kinchine-relation and hence to the Fourier transform of the signal S(f) by (e.g. [24] )Rs (τ ) ) (S-1p( f )) ) (S ( f ) S ( f )).*-1(2-3)Autocorrelation of the high-speed code generator output waveform with fchip 166.7 MHz is shownin Fig. 2-3. The waveform was sampled with a digital oscilloscope, and the autocorrelation wascalculated off-line by (2-3). The envelope of the spectrum of a signal with 53.75 MHz chip rateupconverted to 5.3 GHz is shown in Fig. 2-4. Because the signal is periodic, the spectrum is a linespectrum, where the lines are separated by fchip/L. This is not seen in Fig. 2-4 due to the limitedresolution of the spectrum analyzer. In the radio channel measurements, filters must be applied inTX, because the sidelobes of the spectrum are not allowed by the transmission regulations.Furthermore, the sidelobes reduce the power efficiency of the measurement. Moreover, signalcomponents that remain outside of the information bandwidth of the A/D converter are aliased, andincrease the noise level in the measurement process.0 Rs(τ) [dB]-1 0-2 0-3 0-4 0-5 0-6 005 001 00 01 50 02 00 02 50 03 00 0Sample no.Fig. 2-3. Autocorrelation of a PN code waveform, fchip 166.7 MHz, sampling rate is 1 GHz, L 511 (realized with two ECL-ICs). Spurious level is -45 dB.11

Fig. 2-4. Output power spectrum of the transmitter without filters at 5.3 GHz, fchip 53.75 MHz, L 511.2.2.2 ReceiverIn the receiver (Fig. 2-5) RF front-end, the antenna signal is filtered (B 100 MHz), amplified in alow-noise preamplifier and downconverted to 300 MHz intermediate frequency (IF) with the totalsingle-sideband noise figure of 2.5 dB. The IF-stage includes automatic gain control (AGC) withcomputer-controlled digital step attenuators having a dynamic range of 72 dB, and a major part ofthe signal amplification. The 2 and 5 GHz versions are realized with two interchangeable plug-inunits. The major improvement to the system that is not included in [P1] is the separate samplingunit for inphase (I) and quadrature (Q) channels. Also, microwave switch arrays in front of the RFfront end to measure antenna arrays with two polarizations have been realized [25].RF-front endIF-stage1.854/5 GHz10 MHzRubidiumstandardSynt. unit300 MHzAGCIIQQdemodulatorISCPCQfc'PN-gen.Fig. 2-5. The receiver block diagram.12

2.2.2.1 Matched filteringIn [P1], basic limitations of the sounder concept are discussed. In this work, the post processingwith super resolution algorithms is not covered. It is assumed that a matched filter (MF), realizedeither by digital signal processing (DSP) or by SC, is used as a detector. The impulse response ofthe matched filter hMF is the mirror image of the signal waveform in the time domainhMF (τ ) s ( τ ) .(2-4)In the frequency domain, (2-4) can be written using the corresponding Fourier transformsH MF ( f ) S * ( f ).(2-5)The impulse response of the radio channel is estimated from the received complex signal r(t) byhˆ(τ , t ) hMF r (t ).(2-6)Using the frequency domain notation, (2-6) can be written ashˆ(τ , t ) ) (R( f ) S ( f )),*-1(2-7)where R(f) is the Fourier transform of the received signal over the time of the signal period. Thisimplies, that the IR estimation is averaged over the period of the measurement waveform. In [P1],the use of deconvolution as enhancement of delay resolution is discussed. Then IR is estimated byhˆ(τ , t ) )-1 R( f ) S * ( f ) , S ( f ) γ p (2-8)where γ adjusts the trade-off between resolution and dynamic range.2.2.2.2 Bandwidth limitationThe RF parts of the measurement system, described by their impulse response hsys in [P1] affect thesignal autocorrelation and hence the dynamic range and the resolution of the system. The impulseresponses of the band limiting filters of the system mainly contribute to hsys. As demonstrated in[P1], filters with sharp cutoff have high transient response (e.g. [26]). Filters must be applied toreject the sidelobes of the sinc-spectrum in the TX (Section 2.2.1), and the interference, imagefrequency and spurious signals in the RX, and to prevent the aliasing of noise and interference tothe information band in the sampling process. To reduce the transient responses, the phasedistortion of the filters can be compensated in off-line processing by multiplying the received signalR(f) with the complex conjugate of the baseband transfer function of the filter Hf(f)hˆ(t , τ ) ) (R( f ) S ( f ) H-1*f( f )* ).(2-9)Filter effects can be further reduced by deconvolution [27], which requires high SNR. Obviously,linear phase filters are preferable in channel sounding applications, since they do not cause transient13

responses [26]. In the filter design procedure, Bessel polynoms approximate such filter. However,conversion from baseband prototype to band-pass filter does not completely preserve the linearphase characteristics [26]. In Figures 2-6 and 2-7, the system response of the 5 GHz version of thesounder is presented, measured with L 511, fchip 60 MHz, and sampling rate is twice the chiprate. The narrowest filter is the anti-aliasing filter at baseband, which is a 5-stage Bessel filter withthe 3 dB bandwidth of 35 MHz. Filter requirements especially in the TX can be relaxed by applyingwaveforms described in [28] and [22] -instead of the conventional m-sequence generator an offline optimized waveform with rectangular spectrum is generated from digital memory.0 5 10 τ 17 ns 15 h(τ) [dB] 20 25 τ 33 ns 30 35 40 45 50 5500.10.20.30.40.5τ [s]0.60.70.80.91 7x 10Fig. 2-6. The mainlobe of the measurement system response.0 5 10 15 h(τ) [dB] 20 25 30 35 40 45 50 5501234τ [s]5678 6x 10Fig. 2-7. The measurement system response.14

2.2.2.3 Sliding correlatorIn the SC, a replica of the transmitted m-sequence is generated in the receiver with a chip rate fchip’.The scaling factor K is defined by the difference of the TX and RX chip rates:K f chip'f chip f chip.(2-10)The RX sequence is upconverted to the IF and correlated with the received signal in an analog correlator, which consists of double-balanced mixers and integrators for I and Q channels [1]. Thecomplex output of the SC r'(t) is the convolution of the IR h(τ,t) and the crosscorrelation R’s(τ) ofthe TX and RX baseband signalsr ' (t ) h(τ , t ) Rs' (τ ) ,(2-11)where τ is a function of time: τ t K and ' ' denotes convolution. Bandwidth compression by theamount of K is achieved at the cost of the measurement time. Thus fDmax is limited to a few tens ofHz at maximum. However, in an analog correlator, the delay range can be zoomed by resetting orshifting the reference m-sequence generator. Thus, Doppler range can be enhanced to hundreds ofHz.Response of the system with parameters fchip 53.75 MHz, L 511 and K 2150 is shown in theFig. 2-8. The I and Q time/bandwidth scaled IR components are A/D-converted to 12 bits withmaximum 2 500 kHz sampling

The services provided by future mobile radio networks, e.g. Internet and video transfer, require data rates substantially higher than those in use today. For example, capacity up to 54 Mbit/s has been . Wideband radio channel models are empirical, semideterministic or deterministic. Empiri