CHANNEL ESTIMATION PERFORMANCE FOR ZERO-OVERHEAD CHANNEL ACCESS IN MOBILE SENSOR NETWORKS

Transcription

1 CANNEL ESTIMATION PERFORMANCE FOR ZERO-OVEREAD CANNEL ACCESS IN MOBILE SENSOR NETWORKS John E. Keider, Ghassan Maaoui, Steve Gifford, and Scott Chuprun Genera Dynamics C4 Systems Scottsdae, AZ, 8557 Brian Sader U.S. Army Research Laboratory Adephi, MD, 0783 ABSTRACT In this work we study the effects of channe estimation/equaization on the BER performance of RF carrier frequency hopped OFDM (F-OFDM) when using synchronization information that is embedded directy into the OFDM baseband symbo stream. In sensor networks using CSMA, the acquisition preambes can require a arge overhead percentage of the overa message. Embedding synchronization information into the data information stream eiminates this channe access overhead, thus providing potentia for zero-time overhead channe access. Superposition (embedding) of the sync information, however, causes interference onto the data information, which must be removed for satisfactory BER performance. Consequenty, embedded synchronization interference canceation (EIC) is utiized, which requires accurate channe state estimation. Using coherent 4QAM- and 16QAM-OFDM moduation, channe estimation and BER performance is evauated using the COST07 mutipath fading channe mode. Less than a 1 db performance difference is found between a preambe and embedded system for short message bursts (< 1msec) and a burst data rate of greater than 1.6 Mbit/sec. The channe estimation mean square error (MSE) versus piot symbo overhead is aso determined as a function of urban and rura channe environments. 1. INTRODUCTION Orthogona frequency division mutipexing (OFDM) is being studied extensivey for various commercia standards, such as 80.11a, 80.16, asymmetric digita subscriber ine (ADSL), and wireess oca area networks (WLAN) for spectray efficient very high data rate wireess services [1]. These standards, however, are not designed to provide operationa robustness in mobie fre- Prepared through coaborative participation in the Coaborative Technoogy Aiance for Communications & Networks sponsored by the U.S. Army Research Laboratory under Cooperative Agreement DAAD The U.S. Government is authorized to reproduce and distribute reprints for Government purposes notwithstanding any copyright notation thereon. The views and concusions contained in this document are those of the authors and shoud not be interpreted as representing the officia poicies, either expressed or impied, of the Army Research Laboratory or the U.S. Government. quency hopping (F) channes used for anti-jam (AJ) operation in Army tactica communications. Wideband Army battefied communication systems wi ikey experience harsh time-varying frequency seective fading due to specuar changes in the operating environment. Current OFDM standards deveopment activities do not consider high mobiity and severe mutipath deay spreads. OFDM is, however, an attractive physica ayer waveform for ad-hoc networking due to its scaabiity and spectra efficiency, whie providing inherent robustness to mutipath fading. Most of the commercia standards are designed to operate in mutipath channes for reativey short transmission range, such that maximum expected mutipath deay spread wi be sma. In this work we adjust the OFDM waveform specifications for onger deay spread environments, with the objective of improving spectra efficiency and channe access efficiency whie under RF carrier frequency hopping. Spectra efficiency is improved by directy embedding the synchronization information into the OFDM payoad data stream. With this approach, no dedicated time sot is required for a synchronization fied. Whie providing improved bandwidth efficiency, it aso provides potentia improvement for medium access contro. For exampe, combining an embedded sync fied in a TDMA structure coud increase the number of avaiabe user sots by eiminating the dedicated synchronization fied. It aso provides improved sensor ife in a sensor network, where RF transmission time is a arge factor in sensor battery ife, and channe access efficiencies using conventiona approaches are quite ow. Eiminating the dedicated sot for acquisition/synchronization reduces transmission time and thus can ead to a dramatic improvement in battery ife for sensor nodes []. Superposition (embedding) of the sync information directy into the information symbo stream, causes interference onto the OFDM data information, which must be removed before satisfactory BER performance can be achieved. Consequenty, we devise a scheme to perform embedded interference canceation (EIC), which removes the interference caused by the embedding process onto the data information. owever, EIC requires accurate esti- 004 Genera Dynamics. A rights reserved. 1

2 Report Documentation Page Form Approved OMB No Pubic reporting burden for the coection of information is estimated to average 1 hour per response, incuding the time for reviewing instructions, searching existing data sources, gathering and maintaining the data needed, and competing and reviewing the coection of information. Send comments regarding this burden estimate or any other aspect of this coection of information, incuding suggestions for reducing this burden, to Washington eadquarters Services, Directorate for Information Operations and Reports, 115 Jefferson Davis ighway, Suite 104, Arington VA Respondents shoud be aware that notwithstanding any other provision of aw, no person sha be subject to a penaty for faiing to compy with a coection of information if it does not dispay a currenty vaid OMB contro number. 1. REPORT DATE 00 DEC 004. REPORT TYPE N/A 3. DATES COVERED - 4. TITLE AND SUBTITLE Channe Estimation Performance For Zero-Overhead Channe Access In Mobie Sensor Networks 5a. CONTRACT NUMBER 5b. GRANT NUMBER 5c. PROGRAM ELEMENT NUMBER 6. AUTOR(S) 5d. PROJECT NUMBER 5e. TASK NUMBER 5f. WORK UNIT NUMBER 7. PERFORMING ORGANIZATION NAME(S) AND ADDRESS(ES) John E. Keider, Ghassan Maaoui, Steve Gifford, and Scott Chuprun Genera Dynamics C4 Systems Scottsdae, AZ, 8557; U.S. Army Research Laboratory Adephi, MD, PERFORMING ORGANIZATION REPORT NUMBER 9. SPONSORING/MONITORING AGENCY NAME(S) AND ADDRESS(ES) 10. SPONSOR/MONITOR S ACRONYM(S) 1. DISTRIBUTION/AVAILABILITY STATEMENT Approved for pubic reease, distribution unimited 11. SPONSOR/MONITOR S REPORT NUMBER(S) 13. SUPPLEMENTARY NOTES See aso ADM001736, Proceedings for the Army Science Conference (4th) ed on 9 November - December 004 in Orando, Forida., The origina document contains coor images. 14. ABSTRACT 15. SUBJECT TERMS 16. SECURITY CLASSIFICATION OF: 17. LIMITATION OF ABSTRACT UU a. REPORT uncassified b. ABSTRACT uncassified c. TIS PAGE uncassified 18. NUMBER OF PAGES 8 19a. NAME OF RESPONSIBLE PERSON Standard Form 98 (Rev. 8-98) Prescribed by ANSI Std Z39-18

3 mation of the synchronization offsets and channe perturbations. In a frequency-fat channe, time-frequency offsets and compex scaar channe gain need to be estimated prior to EIC. In a frequency seective fading channe, time-frequency offsets and compex vector channe gain (the channe gain on a OFDM sub-carriers) must be estimated prior to EIC. In this work we assume the channe parameters change from dwe to dwe, requiring fresh channe estimation over each dwe period, whie assuming synchronization is perfect. The synchronization agorithms utiized for coarse synchronization [3] and fine synchronization of the time-frequency offsets [4][5] can be used and were found by the authors to produce itte degradation in BER performance. ence in this work, we concentrate on the effects of channe estimation error and its effect on demoduated BER performance. Piot signas are used to provide estimates of the channe state. In a frequency-seective fading channe, the number of piots spread across the frequency-domain OFDM symbo must match the minimum frequency spacing required for satisfactory interpoation of the channe at the OFDM data subcarriers subject to frequency seectivity. Utiizing a known piot spacing criteria, resuts show that sufficient BER performance can be achieved using the embedded sync with a F-OFDM system in a mutipath fading channe. We iustrate the avaiabe performance of F-OFDM systems using either 4QAM or 16QAM consteations. Continuousy superimposing (embedding) known information, aong with the unknown OFDM data information, can improve channe tracking performance in timevarying channes as shown in [6]. A per-survivor Viterbi receiver is used in [6] to provide the channe estimation and correction but is more compex than conventiona OFDM channe estimation techniques, whie the performance was found to be highy dependent on the ength and choice of PN sequence. In this work we appy embedded sync interference canceation (EIC) foowing piot symbo assisted channe estimation and correction, and use the superimposed sync information for coarse timing, frequency, and phase offset estimation. A description of the coarse synchronization agorithm can be found in [3][7]. Various piot arrangement techniques for OFDM channe estimation can be found in [8][9] where Least Square (LS) or Minimum Mean-Square (MMSE) estimation and bock- or comb-type piot arrangements are compared. In [8] comb-type piot arrangement was found to perform better, with ess compexity, than the transform domain channe estimator in [10], and in [9] was found to perform better than bock-type piot arrangement. For this work we use comb-type piot arrangement based on LS channe estimation with optima piot spacing as suggested in [11][1]. The comb-type piot arrangement Transmit Signa w/preambe K K OFDM Symbo SYNC, r(t) d(t) T p Transmit Signa w/embedded K SYNC, r(t) K OFDM Symbo(s), d(t) T s T s timing freq. est. Compex Code sampes arg X c K-1 c K-... K Correator abs ( )... * c K+1 sampes Fig. 1. Preambe- and embedded-based transmit OFDM signa structures (eft) and receiver correator. transmits known moduation on a subset of the OFDM sub-carrier frequencies in every OFDM symbo and can aso be utiized for fine timing, frequency, and phase estimation according to [4][5]. Assuming adequate synchronization is achieved, we utiize LS channe estimation using piots with spine frequency-domain interpoation [9] aong with zero-forcing equaized symbos [13]. We then compare the respective BER performance between the preambe and embedded OFDM systems when transmitted through a fading channe using the COST07 fading mode. We aso provide expressions for F rate and bandwidth efficiency between the preambe and embedded synchronization approaches. In Section, a description of the OFDM system using embedded coarse synchronization is given. In Section 3, piot symbo aided channe estimation based on combtype piot arrangement and spine interpoation is introduced. In Section 4, F rate and bandwidth efficiency capabiity are specified. In Section 5, we provide an anaysis of the BER performance of the system. In Section 6, the numerica and simuated resuts are presented, whie Section 7 concudes the paper.. OFDM TRANSMIT DATA / SYNC STRUCTURES This section presents a summary of the transmit data structures for each scheme. A more detaied description of the preambe and embedded schemes, with their performance attributes, can be found in [7]. Fig. 1 shows a bock diagram of the synchronization scheme for the preambe and embedded signa structures. The same correator can be used for both methods, and software adjustments can be made to program the correator ength according to the unique performance requirements for each respective scheme. The main parts of the synchronizer (for both methods) are a correator and a mutipier that mutipies correator outputs K 1 sampes apart. The receiver operating characteristic (ROC) performance of 1 Note: K represents the sub-correator ength, which is denoted as K p and K e for the preambe and embedded schemes, respectivey. c 1 c Genera Dynamics. A rights reserved.

4 Compex OFDM Data Compex PN SYNC Data d(t) r(t) 1 ρ X X ρ + x(t) Transmitted Data Fig.. F-OFDM embedded sync method. the correator is primariy determined by the code ength K, and the received SNR [7]. Time synchronization is achieved by comparing the correation product to a threshod proportiona to the received power. Subsequenty, once the detection threshod is crossed, the frequency offset is cacuated from the phase of the timing signa..1 Embedding Sync With OFDM Payoad Data In a preambe signaing scheme, a time sot is dedicated to the synchronization fied immediatey prior to the OFDM data symbo(s). For the embedded signaing scheme used in this work, the synchronization information is embedded across a the OFDM payoad data sampes. When adding the sync information to the OFDM payoad, the ampitude of the sync information must be sufficient such that the synchronization works we in ow signa to noise ratios. The product of the number of sync code bocks, L e, and the ength K e is set equa to the product of the number of OFDM bocks, L ofdm, and the timedomain OFDM symbo sampe ength (incuding cycic prefix), N ofdm per hop dwe period. In the frequencyhopped scenario, the product L ofdm N ofdm (or L e K e ) shoud be short to achieve high hop rates. The ampitude ratio of the sync-to-ofdm data payoad must be adjusted sufficienty high to satisfy ROC performance requirements at high hop rates. In this work, the sync information for the embedded approach can be superimposed in the frequency-domain or ineary summed to the time-domain OFDM payoad data as shown in Fig.. The OFDM data is scaed by 1 ρ before embedding the synchronization information, which is superimposed at an ampitude eve of ρ r[k], where r[k] is the origina synchronization information chip sequence. Both the sync sequence and data sequence have unity power, and when summed together, x[k] is scaed to ensure that the composite transmitted signa is aso near unity power. For a preambe scheme, during the data portion of the transmitted signa, x[k] = d[k], and during the sync portion of the signa, x[k] = r[k]. In the embedded scheme, x[k] = 1 ρd[ k] + ρ r[k]. 3. PILOT SYMBOL AIDED CANNEL ESTIMATION For this work we use comb-type piot arrangement based on LS channe estimation with optima piot spacing as suggested in [11][1]. The comb-type piot arrangement transmits known moduation on a subset of the OFDM sub-carrier frequencies in every OFDM symbo and can aso be utiized for fine timing, frequency, and phase estimation according to [4][5]. The comb-type arrangement we use is a form of the periodic arrangement described in [11]. In a F system, we assume that a sufficient hop rate can be attained such that the channe is approximatey constant over each dwe period. This impies that the symbo portion of the dwe time is ess than or equa to the channe coherence time, such that the compex vector channe gain across a OFDM sub-carriers stays fixed for each dwe period. We aso assume that the channe changes for each hop frequency, such that fresh channe estimation is required for each dwe period, and that these parameters are uncorreated from hop to hop. In the foowing, we denote a OFDM sub-carriers as X. X(i p ) is a vector of piot symbos, where i p denotes the piot sub-carrier frequency ocations, and for simpicity we denote X(i p ) as X p such that X p X. The data and nu sub-carrier symbos are denoted as X(i d ) and X(i n ), respectivey, where i d and i n denote the data and nu sub-carrier frequency ocations, respectivey. Again, for notationa simpicity, we denote X(i d ) and X(i n ) as X d and X n, where X d, X n X. Further, we denote X dp as the variabe containing both the piot and data information, such that X p, X d X dp X, where i dp contains a piot and data subcarrier ocations. In the comb-type arrangement [1], the piot frequency spacing is, and the piot time spacing is N F = 1/(f sub τ max ), (1) N T = 1/(f Dmax T s ), () respectivey, where τ max is the maximum expected deay spread and f Dmax is the maximum expected Dopper spread. In a comb-type arrangement, the transmitted piot sequence, X p, is generated by pacing N p piots uniformy across X dp as suggested in [9][11]. We define p as the compex channe gain at the piot sub-carriers. The LS estimate of the channe at the piot sub-carriers is then p Ĥ = Y p /X p, (3) where Y p is the received piot sequence through the channe at sub-carrier ocations specified by i p. We note that LS channe estimation is susceptibe to noise and ICI, but 004 Genera Dynamics. A rights reserved. 3

5 we utiize LS estimation to minimize compexity. Better channe estimation methods are avaiabe, such as MMSE [9], but are more compex. Foowing LS estimation at the piot frequencies, the channe estimates at the data sub-carriers are found using spine interpoation. The spine interpoation technique was found to provide good resuts when comparing various interpoation techniques [9]. 4. FREQUENCY OPPING RATE AND BANDWIDT EFFICIENCY We use the same frequency hopping structure as specified in [7]. Both OFDM schemes utiized 75% of N (tota number of OFDM sub-carriers) for the number of data bearing pus piot sub-carriers, represented as X dp in Section 3. The unused sub-carriers, X n, are set to nu carriers. For the preambe and embedded schemes, according to the comb-type arrangement, the piot overhead wi be determined by N T and N F. For this work N T = 1, whie the vaue of N F depends on the maximum expected deay spread, τ max, according to (1). In a preambe scheme the hop duration is [7] T h = T p + L ofdm (T s +T g ) + T sw, (4) where T p is the preambe time, T s is the OFDM symbo time, T g is the guard interva time, and T sw is the switch duration (dead time pus rise and fa times) between hop frequencies [14]. For the embedded scheme the hop duration is [7] T h = L ofdm (T s +T g ) + T sw - T g. (5) In the embedded scheme, the first guard interva is not required if we assume that the switch interva can act as a guard time between hop frequencies. In the preambe scheme, the first guard interva is required to prevent ISI between the preambe and the first OFDM symbo for that hop frequency. We define bandwidth efficiency as the transmitted bit rate divided by the RF bandwidth per hop frequency or ( N N N ) og M n p LofdmRh η = (bits/s/z), (6) B where M is the M-QAM moduation order, N is the tota number of OFDM sub-carriers, N n is the number of nu sub-carriers, N p is the number of piots, L ofdm is the number of OFDM symbos per dwe, R h = 1/T h is the hop rate in hops/s, and B = (N N n N p )f s /N is the RF bandwidth per hop, and f s is the base-band samping rate. Achievabe bandwidth efficiency, as a function of hop rate, can be determined for equivaent ROC performance by using equations (4) (6) and (13) from [7]. 5. BER ANALYSIS IN FREQUENCY SELECTIVE FADING In this section we derive an approximate anaytica expression for BER of M-QAM OFDM moduation under frequency seective fading with channe estimation error for both the preambe and embedded systems. In this work the frequency seective channe distortion is compensated with zero-forcing equaization using channe estimates as described in Section 3. Anaysis in [13] provides a cosed form approximation of the BER performance in time-varying frequency-seective fading channes. We modify the resuts in [13] for the embedded synchronization scheme. Whie our focus is to predict performance for the embedded scheme, the expressions we deveop can aso be appied to the preambe scheme by setting ρ = Signa Mode When the transmitted OFDM signa is affected by a frequency offset, the received OFDM symbo from the N-point IFFT moduator [15] is given by 1 y[ k] = [ m] X [ m] e j π k( m+ ) / N + w[ k], (7) N m X dp where X[m] and [m] are the data symbo and channe response of the m th sub-carrier, w[k] is the additive receiver noise, and X dp is as defined in Section 3. With idea timing, the output of the FFT demoduator for the th sub-carrier is 1 N 1 Z[ ] = [ m] X[ m] e N dp Z k= 0 m X = = X + I + W jπk ( m+ ) / N jπk / N + W[ ] ( 1 ρ D + ρr ) + I + W, (8) where X = 1 ρ D + ρ R, W is the frequencydomain additive noise, I and are the inter-carrier interference (ICI) and distorted channe response, respectivey, due to frequency offset and fading as found in [13]. D and R are the frequency-domain versions of variabes d and r, respectivey, as shown in Section.1. The estimated channe response can be written as ˆ = + ν, where ν denotes the channe estimation error of the th sub-carrier. Combining (8) and the estimated channe response, the zero-forcing equaized symbo is found to be ˆ ˆ Z D = = ˆ (9) ν 1 ˆ 1 ρ D ν + 1 ˆ ρ R e I + W + ˆ. 004 Genera Dynamics. A rights reserved. 4

6 Foowing embedded interference canceation (EIC) (- ρ R ) and scaing of the eftover data by 1 / 1 ρ, the estimated data symbo can be written as Dˆ ν = 1 ˆ ν D ˆ ρ R + 1 ρ I + W 1 ρ ˆ. (10) 5. Fading Channe Performance Assuming the channe response is stationary over more than one symbo, the effective SNR of the th subcarrier can be written as + ν γ =, (11) ρ 1 ν + + σ e 1 ρ where σ e is the normaized interference signa power, ( σ ) /[(1 ) I + σ W ρ σ X ], and σ I, σ W, σ X are the ICI, noise, and received signa powers, respectivey. Assuming the channe is normaized, such that E[ ] = 1, the average effective SNR is found to be [13] γ [( sin( )/ ) = π π σ e ] Ι ( a, b) + 1, (1) ea / b where Ι ( a, b) = Ei(a/b), Ei is the expectation integra, a = σ b e, and b = σ ν, representing the channe estimation error variance. Assuming the Rayeigh channe is normaized and stationary over each hop duration, the average BER of M-QAM can be approximated as P B ( M 1) 1 M og M c(1 + σ ν ) γ c(1 + σ ) γ + ( 1)(1 + σ ) ν M ν,(13) γ where γ is the average received signa-to-noise ratio and c 9.5. (1) and (13) provide an approximate anaytica soution for cacuating the BER when using either the embedded (0 < ρ < 1) or preambe (ρ = 0) schemes. 6. NUMERICAL AND SIMULATED RESULTS This section presents numerica resuts for a F- OFDM system where, N T = 1, N F = 10, L ofdm = 4 and L e = 10. We assign K p =16 and K e =18, where in (4) T p = K p T, f s = 1.333*10 6, N = 56, with N p = 0 piot subcarriers and T sw = T g = 15µs. Later we aso present the channe estimator MSE performance as a function of piot overhead and channe type. The OFDM moduator specs are the same for the preambe and embedded schemes. Fig. 3. F-OFDM 4QAM BER performance without EIC (top) and with EIC (bottom). Given these design parameters, the hop rates and bandwidth efficiencies are 1153 and 108 hops/s, and 1.59 and 1.66 bits/s/z, respectivey for the preambe and embedded schemes. The channe is assumed to be constant over each hop dwe but uncorreated between consecutive dwe intervas. Simuated BER performance resuts are presented for both F-OFDM schemes using 00k (68.8 Mbits) and 0k (6.88 Mbits) OFDM simuated data symbos in the AWGN and fading channes, respectivey. Anaytica performance estimates using (13) were found to corroborate the simuated BER performance resuts using the TU fading channe mode. 6.1 AWGN Channe EIC Performance In this section we provide simuated BER performance in AWGN with synchronization offsets. We induce a random timing offset, a frequency offset of 300 z and a carrier phase offset of 40 degrees. Fig. 3 shows the BER resuts when EIC is (bottom) and is not (top) utiized, respectivey. When EIC is utiized, the ony performance oss is due primariy to the SNR degradation from the 1 ρ data signa oss factor induced at the 004 Genera Dynamics. A rights reserved. 5

7 Fig. 4. MSE of LS channe estimator with spine interpoation as a function of ρ for γ = 1 db. transmitter, with an imperceptibe oss due to a sma residua sync estimation error. It is cear that the EIC is critica, even for the benign AWGN channe. 6. Fading Channe Estimation Error In this work, the piot spacing in time is N T = 1 and using (1), the piot spacing in frequency, assuming τ max = 10 µs, is N F 10, assuming the moduator design parameters described above. From Section 5, it is evident that the BER wi be dependent on the eve at which the embedded sync is superimposed at the transmitter. σ and σ ν wi be the primary factors affecting the embedded scheme demoduated BER. With high SNR and ow frequency offset, σ ν wi dominate because it wi detrimentay affect our abiity to perform EIC (due to noisy channe estimation). The question we answer in the foowing is whether or not ρ affects the channe estimation error as a function of γ. To address this issue, it is important to note that channe estimation quaity is determined primariy by the piot signa, which is generated in the frequency domain. In [16] optima training sequences are studied based on their MSE performance. Typicay the sub-carrier piot signa is a constant enveope (CE) signa that is orthogona to the data sub-carriers, and as ong as we maintain this property, even after embedded sync superposition, then the error variance of the channe estimate wi be unchanged when ρ > 0, and the embedded sync signa vaues across the data sub-carriers can be chosen as desired. From this we deduce the foowing constraint: Constraint 1 - The embedded sync signa, at the piot subcarrier ocations in the frequency domain, must be as- e Fig. 5. MSE of LS channe estimator as a function of γ for TU channe. signed the same vaues as the piot signa. Given that the piot moduation is constant ampitude, this ensures constant ampitude after embedded sync is superimposed onto the piot moduation. Constraint 1 simpy ensures that the estimation variance is due ony to the receiver noise, frequency-seective channe gain, and quaity of the channe estimator, and not due to additiona variance caused by a non-constant enveope signa added to the piot signa. Fig. 4 iustrates the MSE of the channe estimator, over 1000 channe reaizations at γ = 1 db, either when the embedded sync signa does or does not compy with Constraint 1 as a function of ρ. We see that, for the non- CE (Gaussian distributed) case, the MSE increases with increasing embedding factor ρ. Under Constraint 1, Fig. 5 iustrates the overa MSE as a function of SNR. Ideay, the MSE shoud decrease ineary with increasing SNR [16]. owever, in our case a MSE foor is apparent at approximatey 3*10-4. We beieve this MSE foor is due primariy to the inabiity of the spine interpoator to adequatey track the data sub-carrier frequency seectivity between the piot sub-carrier ocations, causing irreducibe estimation error at high γ. This suggests that N F < 10 may be appropriate for the TU channe. 6.3 Perfect Channe Estimation Simuated BER performance resuts are presented for both preambe (ρ = 0) and embedded (0.1 ρ 0.5) F- OFDM systems with 4QAM (see Fig. 6) and 16QAM (see Fig. 7), with perfect channe estimation. For the embedded cases of both figures, the ony apparent penaty is due to oss of signa power ( 1 ρ factor). At ρ = 0.5, a γ oss of 3 db is apparent. 004 Genera Dynamics. A rights reserved. 6

8 Fig. 6. 4QAM BER, perfect channe estimation. Fig. 8. 4QAM BER, actua channe estimation. Fig QAM BER, perfect channe estimation. Fig QAM BER, actua channe estimation. 6.4 Non-Perfect Channe Estimation Simuated BER performance resuts are presented for 4QAM (see Fig. 8) and 16QAM (see Fig. 9), using actua channe estimates. For the embedded cases, the BER penaty is due to oss of signa power ( 1 ρ factor) and due to non-perfect EIC resuting from imperfect channe estimation. Eq. (13) was found to provide a good approximation of the performance shown in Figs At ow γ, (13) is ess accurate. For perfect channe estimation, σ e dominates due to the noise variance part σ W. For actua channe estimation, σ e dominates due to the noise variance part σ W at ower to mid-range γ, but at high γ, σ ν dominates causing a BER foor, which is a direct effect of the MSE foor seen in Fig Piot Overhead and Channe Type The previous resuts show that our channe estimation method is sufficienty accurate to produce ow demoduated BER in mobie frequency-seective channes. Aso, we showed the importance of foowing Constraint 1 for minimum channe estimator MSE (ref. Fig. 4). Fig. 10 shows the estimator MSE performance versus piot symbo assisted channe estimation overhead and channe type with an embedding factor, ρ = 0.1 and a average received signa-to-noise ratio, γ = 1 db. From the resuts we see that the estimator reaches its minimum MSE for the RA and TU channes at approximatey 4% and 14%, respectivey. 004 Genera Dynamics. A rights reserved. 7

9 Fig. 10. Channe estimator MSE as a function of piot overhead and channe type for γ = 1 db snr. CONCLUSION # This paper showed that OFDM BER performance degrades ess than 1 db when using embedded synchronization (ρ = 0.1) compared to preambe synchronization (ρ = 0) in mutipath fading with embedded interference canceation. Low compexity channe estimation provides sufficienty ow mean square error to provide good demoduation performance for both 4QAM and 16QAM moduations with F-OFDM. From our resuts, hop dwe times of ess than 1 msec can be supported with this system. Aternativey, we coud consider this system appicabe to highy mobie environments where the channe coherence time is greater than 1 msec. We beieve that this system coud support higher hop rates by using more accurate channe estimation and spatia diversity. REFERENCES [1] S. Ariyavisitaku, D. Faconer, F. Adachi, and. Sari, Guest editoria wireess broadband techniques, IEEE JSAC, vo. 17, no. 10, pp , Oct., [] [3] J. Keider and S. Gifford, Synchronization for broadband OFDM mobie ad hoc networking: simuation and impementation, in Proc. of ICASSP, vo. 4, pp , 00. [4] D. Kim, S. Do,. Cho,. Choi, and K. Kim, A new joint agorithm of symbo cock adjustment for OFDM systems, IEEE Trans. on Cons. Eectron., vo. 44, no. 3, pp , Aug [5] D. Matić, N. Petrochios, A. Trindade, F. Schoute, P. Comon, and R. Prasad, OFDM synchronization based on the phase rotation of sub-carriers, in Proc. of VTC, vo., pp , May, 000. [6] P. oeher and F. Tufvesson, Channe estimation with superimposed piot sequence, in Proc. of Gobecom, pp , Dec [7] J. Keider, S. Gifford, G. Maaoui, S. Chuprun, and B. Sader, Synchronization for RF carrier frequency hopped OFDM: anaysis and simuation, in Proc. of MILCOM, Oct [8] M. sieh and C.Wei, Channe estimation for OFDM systems based on comb-type piot arrangement in frequency seective channes, IEEE Trans. Consumer Eectron., vo. 44, no. 1, pp. 17-5, Feb [9] S. Coeri, M. Ergen, A. Puri, and A. Bahai, Channe estimation based on piot arrangement in OFDM systems, IEEE Trans. Broadcasting, vo. 48, no. 3, pp , Sep. 00. [10] Y. Zhao and A. uang, A nove channe estimation method for OFDM mobie communications systems based on piot signas and transform domain processing, in Proc. of VTC, pp , May [11] M. Dong, L. Tong, and B. Sader, Optima piot pacement for channe tracking in OFDM, in Proc. of MILCOM, vo. 1, pp , Oct. 00. [1] F. Tufvesson and T. Maseng, Optimization of subchanne bandwidth for mobie OFDM systems, in D. Everitt and M. Rumsewicz, editors, Muti-access, mobiity and teetraffic advances in wireess networks, pp , Kuwer Academic Pubishers, Dordrecht, The Netherands, [13]. Cheon and D. ong, Effect of imperfect channe information in OFDM-based WLAN, IEEE Eectron. Let., vo. 38, no. 16, pp , Aug., 00. [14] D.J. Torrieri, Principes of Secure Communication Systems, nd Ed., Artech ouse, 199. [15] P.. Moose, A technique for orthogona frequency division mutipexing frequency offset correction, IEEE Trans. Commun., vo. 4, no. 10, pp , Oct., [16] T.-L. Tung and K Yao, Channe estimation and optima power aocation for a mutipe-antenna OFDM system, EURASIP JSAP, 3, pp. 1-10, 00. # The views and concusions contained in this document are those of the authors and shoud not be interpreted as representing the officia poicies, either expressed or impied, of the Army Research Laboratory or the U.S. Government. 004 Genera Dynamics. A rights reserved. 8