A Quadrature Downconversion Autocorrelation Receiver Architecture for UWB

Transcription

1 A Quadrature Downonversion Autoorrelation Reeiver Arhiteture for UWB Simon Lee, Sumit Bagga, Wouter A. Serdijn Eletronis Researh Laboratory, Faulty of Eletrial Engineering, Mathematis and Computer Siene (EEMCS) Delft University of Tehnology, Delft, the Netherlands {w.k.lee; s.bagga; ABSTRACT In this paper, a new UWB reeiver arhiteture is proposed. Unlike a rake reeiver, it does not suffer from the timing and template mathing problems, and it irumvents proessing at high frequenies, thereby reduing the on-hip iruit omplexity and power onsumption and offering simple but effetive narrowband interferene rejetion. Simulations show that with urrent IC tehnology, the reeiver only shows a slight, aeptable performane loss with respet to the ideal ase. 1. INTRODUCTION Ultra-wideband (UWB) tehnology has gained muh interest during the last few years as a potential andidate for future wireless short-range data ommuniation. Reently, the FCC has alloated the spetrum from 3.1 GHz to 10.6 GHz for UWB appliations. Due to its large bandwidth UWB has the promise of high data rates [1]. A partiular type of UWB ommuniation is impulse radio [], where very short transient pulses are transmitted rather than a modulated arrier. Consequently, the spetrum is spread over several gigaherz, omplying with the definition of UWB. Currently, the rake reeiver is onsidered to be a very promising andidate for UWB reeption, due to its apability of olleting multipath omponents [3]. Rake reeivers perform detetion by orrelating the inoming pulse with a loally generated template pulse, whih is taitly assumed to be perfetly synhronized to the inoming pulse. However, perfet synhronization an never be aomplished. Another issue is the mathing of the template with the reeived pulse. Sine the antennas and the hannel are frequeny seletive, thereby ausing dispersion and ringing, the transmitted pulse beomes severely distorted and hene it is unlikely that the reeived pulse orresponds to the template pulse. Not muh is published on the atual implementation of rake reeivers. Moreover, often a lot of rake fingers are required to aommodate the wireless hannel, rendering it not favourable from an implementation point of view. Indeed low omplexity rake reeivers are being investigated []. The transmitted referene sheme proposed by Hotor and Tomlinson [5] does not suffer from the above problems and requires fewer RF building bloks ompared to the multiple finger rake reeiver. The ore part of the transmitted referene sheme reeiver, alternatively known as autoorrelation reeiver, is shown in Fig. 1. integrate & dump In the transmitted referene sheme, two pulses per symbol are sent with a ertain hosen delay τ d between them. The first pulse ats as the referene and the seond pulse is the modulated one. The reeiver delays the first pulse by the delay τ d, multiplies it with the seond pulse and integrates the result over one delay length, whih in fat orrelates the two pulses. When using polar NRZ modulation, for a logial zero, a pulse g(t) is transmitted, subsequently followed by a polarity reversed pulse g(t). To send a logial one, two pulses with the same polarity are sent sequentially. Instead of transmitting two pulses for eah symbol, it is also possible to use the previous pulse as the referene, resulting in differential oding. The absolute value of the output after integration is in fat the energy of the pulse while the polarity of the output ontains the data. If the output is negative, this orresponds to a logial zero, while a positive output orresponds to a logial one. Thus, the information is in the relative polarity of the two pulses and the delay between them ats as a synhronization mehanism. As long as the two onseutive pulses have orresponding waveforms exept for their polarity, the autoorrelation reeiver an detet them properly. Yet, diretly proessing at UWB frequenies, due to the inherent influene of on-hip parasiti reatanes at these frequenies, requires relatively large urrents, or the use of expensive high speed tehnologies (SiGe, GaAs) that are not ompatible with mainstream CMOS IC tehnology for mass market use. Another important issue is narrowband interferene. Sine UWB systems transmit at a low spetral density, it is very likely that existing narrowband systems with relatively high power will jam the UWB system, even though UWB systems are laimed to have high proessing gain [6]. Most solutions in literature propose MMSE ombining of the rake fingers [6, 7] to ombat narrowband interferene. Both the normal rake reeiver and the autoorrelation reeiver suffer from narrowband interferene. It is expeted that the IEEE80.11a and HiperLAN wireless LAN systems around 5 GHz, whih is in the middle of the UWB spetrum, generate most of the interferene [8, 9]. Additionally, although the UWB antenna may do some out-of-band filtering, due to their relatively high power levels, it is even likely that signals originating from wireless LAN systems operating around. GHz and even GSM an penetrate into the system, ausing jamming of the UWB system. This paper suggests a new reeiver arhiteture based on the transmitted referene sheme from Hotor and Tomlinson [5] for its mentioned advantages with respet to the rake reeiver. It deals with interferene rejetion by filtering and avoids high frequeny on-hip proessing by using frequeny onversion. Setion gives a detailed analysis of the system. Setion 3 presents the simulation results, followed by the onlusions in Setion. Figure shows the proposed reeiver arhiteture. Figure 1: An autoorrelation reeiver with a typial waveform

2 S(f) a) 90 o integrate & dump 0 5 LO 10 S(f) frequeny [GHz] b) Figure : The proposed quadrature downonversion autoorrelation reeiver. SYSTEM ANALYSIS The proposed arhiteture employs frequeny onversion. Its osillator frequeny is hosen suh that the spetrum wraps around DC. In onventional narrowband systems downonversion is applied to have the proessing at lower frequenies, reduing on-hip iruit omplexity and power onsumption. Normally, downonversion systems suffer from limited image rejetion. In the ase of UWB there is not suh a thing as an image sine the spetrum is already a ouple of gigaherz wide and the image is part of the desired UWB spetrum. Hene, if the osillator frequeny is hosen suh that the UWB spetrum is down-onverted and even wraps around the origin, the signal bandwidth is lowered and redued. Moreover, interferers below 3.1 GHz an be uponverted, making it possible to remove them with a low-pass filter. In traditional narrowband systems frequeny wrapping is not possible, as the resulting bandwidth must equal the original bandwidth. Here, sine we are dealing with shorttransient pulses, frequeny wrapping is allowed. Of ourse, after downonversion the waveform is hanged, but as long as the two reeived onseutive pulses are distorted equally and we are able to disriminate individual pulses, whih is the ase, the autoorrelation reeiver is able to detet the signal orretly. We will now make a proper hoie for the frequeny of the loal osillator (LO). If the LO frequeny is set to the enter frequeny of the pulse, whih for simpliity is hosen to be in the middle of the band alloated by the FCC for UWB (6.85 GHz), the system ats like a zero IF system and the bandwidth is halved, i.e., the band from 3.1 GHz to 10.6 GHz is transformed into a baseband signal from DC to 3.75 GHz. However, the wireless LAN originally residing at around 5.5 GHz is shifted to 1.35 GHz, so it is still in band. Yet, at these frequenies it is muh easier to filter out the interferer on hip than at its original frequeny. Another option is to set the osillator frequeny to this wireless LAN frequeny suh that this interferer is shifted to around zero, making it possible to remove it with a simple high-pass or band-pass filter. A possible disadvantage is that the down-onverted bandwidth of the UWB signal extends up to approximately 5.1 GHz and the onverted. GHz interferer falls in band. A third option is to set the LO to 5.5 GHz and filter out below ( =) 0.35 GHz and above (5.5. =) 3.1 GHz. By this we remove the interferers ompletely with a simple bandpass filter, albeit at the expense of the loss of part of the inoming frequeny band, from (5.5 + (5.5.)) = 8.6 GHz tot 10.6 GHz and from 5.15 to 5.85 GHz, whih is a bandwidth of.7 GHz, being only 36%, or equivalently 1.9 db. The UWB spetrum before and after downonversion is illustrated in Fig. 3. The loal osillator is also shown frequeny [GHz] Figure 3: The UWB spetrum with narrowband interferers at. GHz and 5 GHz a) before downonversion and b) after downonversion.1 Time Domain Analysis Beause UWB systems rely on timing information, we will use a time domain analysis throughout the remainder of this paper. Any physial band-pass waveform an be represented by: A()os( t ω t + ϕ()) t (1) where A(t) is the amplitude envelope, ω is the arrier frequeny and ϕ(t) is the phase modulation. This desription is ommonly used for arrier-based signals but an also be applied to pulse-based signals as long as they have a bandpass spetrum. Although (1) is appliable to the generally used first and seond derivative of the Gaussian pulse, for simpliity of the analysis to ome, the pulse used, g(t), is hosen to be the real part of a Morlet, defined to be: t /σ g() t = e os( ω t) () where σ determines the pulse width and ω is the enter frequeny of the spetrum. In Fig., a Morlet is shown with a pulse width of 1 ns and a enter frequeny of.5 GHz, along with its spetrum. Figure : A typial Morlet a) waveform with a pulse width of 1 ns and enter frequeny of.5 GHz and b) its frequeny spetrum Equating (1) to () shows that A(t) is just the Gaussian envelope, ω is the enter frequeny and ϕ(t) = 0. Now onsider the situation where two pulses of equal sign are transmitted. For the moment, assume that we only have the upper path of the iruit shown in Fig.. This path is denoted here as the in-phase path. After mixing with a osine of angular frequeny ω os and a normalized amplitude of one and using (1), we obtain for the first pulse: g( t)os( ω t) = A( t)os( ω t)os( ω t) (3) os os Assuming an ideal low-pass filter and using os(a)os(b) = ½[os(a + b) + os(a - b)], the signal after filtering beomes: At () os(( ω ωos ) t ) ()

3 After delaying this first pulse in the delayed path: At ( τ d ) os(( ω ωos )( t )) (5) To ease the analysis, a hange of variable is made by letting t = t - τ d : At (') os(( ω ωos ) t ') (6) This is the signal in the delayed path before the multiplier. For the seond pulse the same analysis holds but the input is now a pulse delayed by τ d in the transmitter: g( t τ )os( ω t) = A( t τ )os( ω ( t τ ))os( ω t) (7) d os d d os The signal after the low-pass filter equals: At ( τ d ) os( ω ( t ) ωost ) (8) Again, after a hange of variable t = t - τ d : At (') os(( ω ωos ) t ' ωos ) (9) The signal before integration is obtained by multipliation of (6) and (9), resulting in: A (') t os(( ω ωos ) t ')os(( ω ωos ) t ' ωos ) (10) This an be rewritten as: 1 A ( t') [ os(( ω ωos ) t ') + os( ωos ) ] (11) For a transmitted logial zero, the output is just (11) with a minus sign in front. From (11) it an be seen that only if ω os τ d = πn, where n, the output is fully positive or fully negative, depending on the transmitted symbol. The output after the integrate-and-dump filter is thereby in absolute sense maximized. This means that if ω os τ d πn, the performane will degrade. E.g., ω os τ d = ½π even auses the output to be zero, irrespetive of the transmitted data and detetion is not possible. Together with the above frequeny onversion analysis, this puts a seond onstraint on the hoie of the osillator frequeny.. The Quadrature Downonversion Autoorrelation Reeiver Arhiteture The previous analysis assumed perfet synhronization between the osillator and the pulse arrier. In reality this is not the ase. Unless the osillator is being loked onto the inoming signal there is always a relative phase between the osillator and the pulse arrier. Denoting this relative phase by ϕ 0 this means that in (3) and (7) ω os t has to be replaed by ω os t + ϕ 0. Following the same analysis as before, it an be found that the signal before integration now equals: 1 A ( t') [ os(( ω ωos ) t ' ϕ0 ) + os( ωos ) ] (1) If ω os = ω, this means that after integration the result depends on ϕ 0. The output an even be zero whereas a positive value was expeted. This is a well-known phenomenon in oherent detetion. In oherent detetion, the osillator an be loked to the arrier but sine in this situation there is a suppressed very weak arrier that is only present when the pulse is present and there is also narrowband interferene, this is not possible. A possible solution is to add a similar path but now mixed with a sine instead of a osine and add the outputs after autoorrelation, resulting in the arhiteture shown in Fig.. This lower path is alled the quadrature path from now on. For this quadrature path, it holds that the signal after multipliation and before integration beomes: 1 A ( t') [ os(( ω ωos) t ' ϕ0) + os( ωos) ] (13) Now adding the two quadrature autoorrelation signals given by (1) and (13) results in the signal before integration: A (') t os( ωosτ d ) (1) (1) shows that the output does not depend on ϕ 0 anymore. The influene of the relation between ω os and τ d is obvious. For a transmitted logial zero it an be found that the output is: A (') t os( ωosτ d ) (15).3 Analysis of the Quadrature Downonversion Autoorrelation Reeiver Subjet to Mismath The previous analysis assumed ideal phase and amplitude relations between the two paths. When implemented, these relations will never be exat due to omponent mismath. Also, the delays and the osillator period will show some variation. It is important to analyze the performane of the arhiteture when these errors are taken into aount. The following errors are onsidered: The mismath between the delay in the transmitter and in the reeiver. This an be taken into aount by introduing τ tx and τ rx, where τ tx denotes the delay in the transmitter and τ rx indiates the delay in the reeiver. The phase mismath between the osine and sine, originating from an imperfet quadrature relation. This phase error is denoted by ϕ e. The amplitude mismath in the two paths, whih an be modelled by assigning the osillators in the in-phase and quadrature path a different amplitude A I and A Q respetively. The error in the ratio of the delay and the osillator period, n. Hene, all these errors an be taken into aount by replaing os(ω os t + ϕ 0 ) by A I os(ω os t + ϕ 0 ) and sin(ω os t + ϕ 0 ) by A Q sin(ω os t + ϕ 0 ± ϕ e ) in the previous analysis, where ϕ e is defined to be positive. In the previous derivations the delay in the transmitter was assumed to be the same as in the reeiver, both being equal to τ d. With the introdution of mismath between these delays, if the delaying ation is due to the transmitter delay, τ d has to be replaed by τ tx, while τ d has to be replaed by τ rx if the delaying is due to the reeiver. With these substitutions, for the in-phase path, the signal after multipliation and before addition beomes: 1 AAt ( ') ( ' ) os(( ) ' 0) I At+ τ ω ωos t ϕ (16) os( ω τ ωos) where τ = τ rx - τ tx. For the quadrature path, the orresponding signal beomes: 1 AAt Q (') At (' + τ) os(( ω ωos) t ' ϕ0 (17) os( ω τ ωos) Adding both paths results in:

4 + + AQos(( ω ωos) t' ϕ0 AI os(( ω ωos) t ϕ0) ( AI AQ ) os( ω τ ωosτrx) (18) Consider the first part of (18). Suppose A Q = (1 ± )A I to denote the amplitude mismath, where 0 < < 1. Then the first part of (18) results in: ( AI + (1 ± ) AI ) os( ω τ ωosτrx) = ( + ± ) AI os( ω( τrx τtx) ωosτrx) = (19) ( ) At (') At (' + τ) ω ωos τrx (1 + ± ) AI os = ω(1 ± β) τrx At (') At (' + τ) (1 + ± ) AI os( ωosτrx ± βωτ rx) where β denotes the relative mismath between the delay in the transmitter and reeiver. It holds: 0 < β < 1. For the argument in the osine in (19), it an be written: τ rx τ rx ωosτrx ± βωτ rx = π( ± β ) (0) T T Where T os is the osillator period and T is the inverse of the pulse enter frequeny. Now onsidering the seond part of (18): AQos(( ω ωos) t' ϕ0 = AI os(( ω ωos) t ϕ0) AI os(( ω ωos) t ϕ0) = (1) (1 ± ) AI os(( ω ωos) t' ϕ0 os(( ω ) ωos t ϕ0) AI os(( ω ωos)' t ϕ0 ( ± ) AI os(( ω ωos) t' ϕ0 Using os(a) - os(b) = -sin(½(a + b))sin(½(a - b)), we get: AI sin(( ω ωos) t ϕ0 ± )sin( ± ) = () ( ) AI os(( ω ωos) t' ϕ0 ) ± ± sin(( ω ωos) t ϕ0 ± )sin( ± ) + At (') At (' + τ) AI ( ± )os(( ω ωos) t ' ϕ0 os Now adding part 1 and part yields the output before integration to be larger or equal to (1 + ) os( ω τ ± βω τ ) At (') At (' + τ) (5) os rx rx ± AI sin( ) ( + ) Note that this is a worst-ase situation given a ertain, β and ϕ e. In the ideal ase (, β, ϕ e = 0) the result would be (if the nominal osillator amplitude would be A I ): A (') t ± AI (6) Compared to (6) the result of (5) shows some degradation due to the phase mismath, the amplitude mismath and the mismath between the delays in the transmitter and the reeiver. In ase of a Gaussian envelope, the degradation due to the term A(t )A(t + τ ) is relatively small. 3. RESULTS To verify the funtional behaviour of the proposed reeiver, the system was simulated in Matlab/Simulink, assuming interferene and multipath to be absent. These assumptions are justified as interferene an be dealt with by means of a band-pass filter. Sine the multipath omponents are resolvable, they will add up. A Morlet was used with a enter frequeny of.5 GHz, an effetive pulse width of 1ns and a peak amplitude of one. The osillator frequeny was set to.6 GHz. A pulse repetition time of 10 ns was used, whih orresponds to a bit rate of Mb/s. The input onsisted of five pulses with random polarity, representing a sequene of five differentially oded random data bits. Also to verify the onept, a sub-optimal seondorder low-pass Butterworth harateristi with a ut-off frequeny of.5 GHz was hosen for both filters. First, the ideal situation was simulated. The initial phase of the osillator was set to zero; the delay in the reeiver mathed the delay between the pulses in the transmitter (equal to the pulse repetition time) and the osillator period was a perfet integer of this delay; no amplitude mismath was present in the two paths and the osillator signals had a perfet quadrature relation; the osillator amplitude in both paths was set to one. The absolute value of the integrator output in this ideal ase serves as the referene value, as from the analysis in Setion.3 it followed that any mismath appearing in the system will degrade the output value. All subsequent results will be normalized to this value to show the performane loss. Fig. 5 shows the pulses in both paths after downonversion and filtering and the integrator output in this ideal ase. Sine for any t: sin(( ω ωos) t ϕ0 )sin( ) + sin( ) + ( + ) (3) ( ± )os(( ω ωos) t ' ϕ0 ) it holds that the absolute value of the seond part is always smaller than At (') At (' + τ) ± AI sin( ) + ( ± ) ()

5 this, one an lok the delays to a highly aurate time referene suh as a rystal. The delay mismath then relies on the mismath between the rystals, whih an be as small as 0.001, yielding an aeptably small performane loss. A worst-ase ombined error simulation with = 0.0, ϕ e = and β = 0.00 shows that the normalized output is still as high as 80%, equivalent to a loss of only 1.9 db. Figure 5: Two onseutive pulses in both paths after downonversion and filtering (left) and integrator output for the data sequene 0 (right) in the ideal ase Seondly, the initial phase was varied from 0 to 360 in steps of 10. In line with the analysis of Setion. the output shows no degradation with respet to the ideal value. Thirdly, the influene of the amplitude mismath was simulated by adding a gain blok after the osillator in the quadrature path, its gain being equal to (1-), ranging from 0 to 0.05 in steps of The normalized output value as a funtion of is shown in Fig. 6. The effet of the phase error between the two paths on the output was evaluated by adding an additional phase shift of 0 to 5 degrees in steps of a half degree to the osillator in the quadrature path (after removing the amplitude mismath). Fig. 6 shows the normalized output as a funtion of the phase error. The mismath between the delay in the transmitter and the reeiver was simulated by setting the pulse repetition time in the generator to (1 + β) of its nominal value (10 ns) and varying β from 0 to in steps of The normalized output as a funtion of β is shown in Fig Normalized Output vs. Amplitude Mismath 0,00 0,01 0,0 0,03 0,0 0,05 alpha Normalized Ouptut vs. Phase Error phase error [degrees] Figure 6: Normalized output as a funtion of the amplitude mismath (left) and the phase error (right) Normalized Output vs. Delay Mismath 0,000 0,001 0,00 0,003 0,00 0,005 Figure 7: Normalized output as a funtion of the relative delay mismath With urrent IC tehnology,, β and ϕ e an be readily made smaller than %, 5% and, respetively. From Fig. 6 and Fig. 7 it an be seen that the mismath between the delay in the transmitter and in the reeiver plays a dominant role in the performane loss. Compared to this delay mismath, the performane degradation due to the amplitude mismath and phase error is negligible. Therefore, the design of mathing delays should get most attention. Without additional measures, in urrent IC tehnology, β is one order in magnitude larger than allowed for aeptable performane loss. To overome beta. CONCLUSION A new UWB reeiver arhiteture has been introdued. It avoids high frequeny proessing, allowing for redution of the on-hip iruit omplexity and power onsumption and rejets narrowband interferene in a simple but effetive way. Assuming omponent mathing that an be readily ahieved in today s IC tehnologies, Matlab/Simulink simulations show that the output only shows a performane loss up to 1.9 db ompared to the ideal ase. The mismath between the delay in the transmitter and the reeiver is the largest ontributor to this degradation. ACKNOWLEDGEMENTS This work is part of the AIRLINK projet of Delft University of Tehnology ( and partially sponsored by the Ministry of Eonomial Affairs under the Duth Initiative Freeband. 5. REFERENCES [1] J. Foerster, E. Green, S. Somayazulu and D. Leeper, Ultra-Wideband Tehnology for Short- or Medium-Range Wireless Communiations, Intel Tehnology Journal Q, 001 [] M.Z. Win and R.A. Sholtz, Impulse Radio: How it works, IEEE Comm. Letters, vol., pp , Feb [3] J.G. Proakis, Digital Communiations, 3rd ed. MGraw- Hill, 1995 [] D. Cassioli, M.Z. Win, F. Vatalaro and A.F. Molish, Performane of Low-Complexity Rake Reeption in a Realisti UWB Channel, Proeedings of the IEEE International Conferene on Communiations, vol., pp , May 00 [5] R.T. Hotor and H.W. Tomlinson, Delay-Hopped Transmitted-Referene RF Communiations, Proeedings of the IEEE Conferene on Ultra Wideband Systems and Tehnologies, pp , May 00 [6] I. Bergel, E. Fishler and H. Messer, Narrow-band Interferene Suppression in Time-Hopping Impulse-Radio Systems, Proeedings of the IEEE Conferene on Ultra Wideband Systems and Tehnologies, pp , May 00 [7] Q. Li, and L.A. Ruh, Hybrid RAKE/Multiuser Reeivers for UWB, Proeedings of the Radio and Wireless Conferene, pp , Aug. 003 [8] M. Hämäläinen, V. Hovinen, R. Tesi, J. Iinatti and M. Latva-aho, On the UWB System Coexistene with GSM900, UMTS/WCDMA, and GPS, IEEE Journal on Seleted Areas in Communiations, vol. 0, pp , De. 00 [9] M. Hämäläinen, R. Tesi and J. Iinatti, On the UWB System Performane Studies in AWGN Channel with Interferene in UMTS Band, Proeedings of the IEEE Conferene on Ultra Wideband Systems and Tehnologies, pp , May 00